Home

To RI INFINITY and Beyond - NCCA People A-Z

Contents

1. Walt Disney Feature Animation surface SIG2k zerobias float opacity 1 float orientation 0 uniform color C color 1 1 1 normal Nn normalize N uniform float val float val orientation 0 1 0 if Nn I lt 0 point is front facing O val else backfacing O 1 val I QT Ci Oi The opacity field is used to handle partial transparency or to remove geometry from the shadow computation Also one could add more smarts to the shader so portions of the geometry could be selectively removed e g via texture maps or uv coordinates The orientation is used to tell the shader to focus on front facing when orientation 0 geometry or back facing when orientation 1 Shadow pass one use the zero bias shader with orientation set to 0 rendering with the zdepth dspydriver This makes everything front facing opaque and everything back facing transparent The result a zdepth map containing only the front facing geometry See Figure 3 2 top left Shadow pass two do the same thing but with orientation set to 1 rendering out to a different zdepth file This makes back facing geometry opaque and front facing geometry transparent Saving out a second version of the zdepth information containing only the back facing geometry See Figure 3 2 top right The net effect of these two passes should basically be the equivalent of using the newer PRMan min m
2. e The environment function can take the optional arguments blur width and filter which perform the same functions as for texture Environment maps typically sample the mirror direction as computed by the Shading Language built in function reflect For example normal Nf normalize faceforward N I vector R normalize reflect I N color Crefl color environment envmapname R 30 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES Note that the environment is indexed by direction only not position Thus not only is the environment map created from the point of view of a single location but all lookups are also made from that point Alternatively one can think of the environment map as being a reflection of a cube of infinite size Either way two points with identical mirror directions will look up the same direction in the environment map This is most noticeable for flat surfaces which tend to have all of their points index the same spot on the environment map This is an obvious and objectionable artifact especially for surfaces like floors whose reflections are very sensitive to position We can partially overcome this difficulty with the following strategy 1 We assume that the environment map exists on the interior of a sphere with a finite and known radius as well as a known center Example if the environment map is of a room interior we choose a sphere radius representative of the room size Ev
3. vector V normalize I V is the view vector color spec 0 Cditt 07 Jt collect Specular drittuse light float Kspec Ks vector nL varying normal nSN normalize surface normal vector S nSN T Cross product of the tangent along the hair and surface normal vector N hair T S N hair is a normal for the hair oriented away from the surface vector norm hair float 1 clamp nSN T 0 1 Dot of surface normal and T used for blending float clump darkening 1 0 float T Dot nLh 0 float I Dot e 0 float Alpha 0 114 CHAPTER 5 FUR IN STUART LITTLE float Beta 0 float Kajiya 0 float darkening 1 0 Varying color final c values from light uniform float nonspecular 0 uniform color SpecularColor 1 When the hair is exactly perpendicular to the surface use the surface normal when the hair is exactly tangent to the surface use the hair normal Otherwise blend between the two normals in a linear fashion norm hair norm hair 1 nSN 1 1 N hair normalize norm hair Make the specular only hit in certain parts of the hair v is along the length of the hair ud Kspec min smoothstep start spec start spec spec size fade v 1 smoothstep end spec end spec spec size fade v Loop over lights catch highlights as if this was a thin cylinder Specular illumination model from James T Kajiya and Timothy L Kay 19
4. Anthony A Apodaca and Larry Gritz Morgan Kaufmann 1999 ISBN 1 55860 618 1 A comprehensive up to date and advanced treatment of all things RenderMan This book covers everything from basic RIB syntax to the geometric primitives to advanced topics like shader antialiasing and volumetric effects For any RenderMan professional this is the book that will be sitting open on your desk most of the time Buy it Love it Tell your friends The RenderMan Interface Specification Version 3 2 Pixar 2000 Yes you read that right version 3 2 As I write these course notes in March it s not quite done yet But we re intending to have it ready in plenty of time for SIGGRAPH The RI Spec 3 2 is the updated bible of the RenderMan C and RIB API s and the Render Man Shading Language It reflects all of the changes and extensions made informally by PRMan and BMRT over the years put together in a single clean document 1V Lecturers Larry Gritz leads the rendering research team at Pixar and 15 chief architect of the RenderMan Interface Standard Larry has been at Pixar since 1995 developing new rendering techniques and systems and periodically serving as a technical director His film credits include Joy Story Geri s Game A Bug s Life and Toy Story 2 Larry s research interests include global illumination shading languages and systems and rendering of hideously complex scenes Prior to joining Pixar Larry authored the pop
5. Then applying the Mcp_obj2NDC transformation matrix of the stored keyframe to convert this point to the keyframe s NDC space Then the map is read using those NDC coordinates Here are some of the high points for f cameraProjection color f cameraProjection varying point P uniform matrix Mcp obj2NDC uniform float keyframes uniform float arraysize uniform float frame 72 CHAPTER 3 RENDERING ISSUES ON DINOSAUR uniform string cpTexFormat uniform color Cerror uniform float debug point Pobj transform current object P Transform from object space to the camera projection NDC space Ptmp transform Mcp obj2NDC 1i Pobj texture xcomp Ptmp ycomp Ptemp return C Things to watch out for Keep in mind that the CP is only good for the field of view of the keyframe A single camera doesn t always capture all of the visible areas of the scene So as a camera pans and trucks there will be null areas areas outside the painted texture These null areas are basically points in the scene that weren t visible from the CP keyframe So there isn t any Pref information available for them In these situations multiple keyframes multiple projections are needed Figure 3 5 Multiple projections Blue unique mappings Green overlapping mappings Red un mapped areas Figure 3 5 shows multiple projections are needed because the camera move has revealed new areas that
6. commercial software _ frame data J one frame only siggraph 2000 sony picture imageworks advanced renderman fur figure 2 The general pipeline consists of two major processes those that happen once for each character and those that happen once for each frame The former consist of gathering all of the data which controls the look and contouring of the fur and calculating as much as possible to be stored in a series of data files The second process consists of the actual creation of the fur on the character from these data files at the time of the render Most of the data for the first set of processes to design the look of the hair is managed through MEL scripts in Maya and attributes controlled through StudioPaint The instancer is then run to compile all of the appropriate data files into a instanced hair data file which specifies the location of each hair on the character This step only has to be calculated one time at the initial setup of the character or if there is a change to the fur data files It is important in the design of the system to leave as much of the large dataset creation until the latest possible point in the pipeline So the actual RiCurve primitives are not grown onto the surfaces until the time of the render when RenderMan notices the DSO references built into the rib file and calls the interpolator DSO The DSO then generates the rib data whic
7. float angle statements Ordinary illuminance loops will execute their body for every nonambient light source The named category extension to the illuminance syntax causes the statements to be executed only for a subset of light sources Light shaders can specify the categories to which they belong by declaring a string parameter named category this name has two underscores whose value is a comma separated list of categories into which the light shader falls When the illuminance statement contains a string parameter category the loop will only consider lights for which the category is among those listed in 40 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES its comma separated category list If the illuminance category begins with a character then only lights not containing that category will be considered For example float uvcontrib 0 illuminance uvlight P Nf PI 2 float uv 0 lightsource uv uv uvcontrib uv Ci uvcontrib uvglowcolor will look specifically for lights containing the string uvlight in their category list and will execute those lights summing their uv output parameter An example light shader that computes ultraviolet intensity might be light uvpointlight float intensity 1 uvintensity 0 5 color lr ghtceolor I point from point shader 0 0 0 output varying float uv 0 string Category uylrght illuminate from Gl dntensity 11ightecol
8. float debug 0 float printMatrix 0 point p uniform string cp CmapFormatStr concat cp mapBaseName d cp mapSuffix color C vector V normalize I normal Nf normalize faceforward normalize N I if printMatrix 1 dump out the object gt NDC matrix for this camera t printMatrix 05 else must already have the matrix information so do projections This file must define global ARRAYSIZE and arrays keyframes and Mobject2NDC include matrixdata h Get original P before displacement if it exists if usePref 1 p Pref else if displacement Porig p 1 f cameraProjection p Mcp obj2NDC keyframes ARRAYSIZE frame cp CmapFormatStr Cerror debug Oi Os Ci O1 C Ka ambient Kd diffuse Nf specularcolor Ks specular Nf V roughness clamp Ci color 0 color 1 H II 70 CHAPTER 3 RENDERING ISSUES ON DINOSAUR else not debug For the camera projections to work the camera transformations for each of the keyframes need to be known This is by done setting the shader s parameter printMatrix to true and then rendering the keyframe using the shader The matrix information will be dumped out to stdout via a printf statement As the printf statement will be the same for the entire render there is no need to render the entire frame Just kill the render after a few of the lines start printing
9. like a 2 dimensional ribbon always oriented towards the camera In comparison to long tubes these primitives render extremely efficiently However we still need to solve the problem of how to properly shade these primitives as if they were thin hairs 5 5 SHADING 101 shading fur other diffuse hair shading models more mathmatically correct did not look good when fur was oriented towards the light without a global illumination solution of some sort didn t look very natural not intuitive to light by a TD who is used to lighting surfaces siggraph 2000 sony picture imageworks advanced renderman fur figure 17 When using a more conventional lighting model for the fur essentially a Lambert shading model applied to a very small cylinder we came across a few complications First a diffuse model which integrates the contribution of the light as if the normal pointed in all directions around the cylinder leaves fur dark when oriented towards the light This may be accurate but does not produce images that are very appealing In our testing we felt that the reason that this model didn t work as well was because a lot of the light that fur receives is a results of the bounce of the light off other fur in the character which could be modeled with some sort of global illumination solution but would be very expensive The final and for us the most significant reason that we chose to develop a new diffuse lightin
10. ness 0 5 bottom left Anisotropic reflection with xroughness 0 5 yroughness 0 15 bot tom right In all cases xair normalize dPdu is the angle between the surface normal and the direction of the light source e 0 is the angle between the surface normal and the vector in the direction the light is reflected 1 e toward the viewer e 7 and 1 are the two perpendicular tangent directions on the surface e a and a the standard deviations of the slope in the and 7 directions respectively We will call these x roughness and yroughness e fiis the unit surface normal normalize N e histhe half angle between the incident and reflection rays 1 H normalize normalize normalize L Listing 2 12 lists the function LocI11umWardAnisotropic which implements the anisotropic specular component of Larson s model This function can be used instead of an ordinary specu lar call It differs from specular in that it takes two roughness values one for the direction aligned with surface tangent xdir and the other for the perpendicular direction Figure 2 8 shows this model applied to a teapot When comparing the original equation to our SL implementation you may wonder where the factor of 1 7 went and why there appears to be an extra factor of L N This is not an error Greg Ward Larson s paper describes the BRDF which is only part of the kernel of the light integral whereas shaders describe t
11. since those prim itives are probably going to be seen through the transparent foreground it is often quite surprising to see the renderer slow down as much as it does when the usually extremely efficient occlusion culling is essentially disabled by transparent foreground layers 1 4 4 Displacement Bounds Displacement shaders can move grid vertices and there is no built in constraint on the distance that they can be moved However recall that shading happens halfway through the rendering pipeline with bounding splitting and dicing happening prior to the evaluation of those displacements In fact the renderer relies heavily on its ability to accurately yet tightly bound primitives so that they can be placed into the correct bucket If a displacement pushes a grid vertex outside of its original bounding box it will likely mean that the grid is also in the wrong bucket Typically this results in a large hole in the object corresponding to the bucket where the grid should have been considered but wasn t This is avoided by supplying the renderer a bound on the size of the displacement generated by the shader From the shader writer s point of view this number represents the worst case displace ment magnitude the largest distance than any vertex might travel given the calculations inherent in the displacement shader itself From the renderer s point of view this number represents the padding that must be given to every boundin
12. 0 1 ard color MaterialThinPlastic normal Nf vector V color basecolor float Ka Kd Kt Ks roughness return basecolor Ka ambient Kd diffuse Nf Kt diffuse Nf Ks specular Nf V roughness 2 2 Reflections We have covered materials that are linear combinations of Lambertian diffuse and specular com ponents However many surfaces are polished to a sufficient degree that one can see coherent reflections of the surrounding environment This section will discuss two ways of simulating this phenomena and show several applications People often assume that mirror like reflections require ray tracing But not all renderers support ray tracing and in fact those renderers are typically much faster than ray tracers In addition there are situations where even ray tracing does not help For example if you are compositing a CG object into a live action shot you may want the object to reflect its environment This is not possible even with ray tracing because the environment does not exist in the CG world Of course one could laboriously model all the objects in the live action scene but this seems like too much work for a few reflections Luckily RenderMan Shading Language provides support for faking these effects with texture maps even for renderers that do not support any ray tracing In this case we can take a multipass approach first rendering the scene from the points of view of the reflectors then using these first
13. 2 13 LocI1llumGlossy function a nonphysical replacement for specular that makes a uniformly bright specular highlight LocIllumGlossy a possible replacement for specular having more uniformly bright core and a sharper falloff It s a nice specular function to use for something made of glass or liquid Inputs roughness related to the size of the highlight larger is bigger sharpness 1 is infinitely sharp 0 is very dull color LocIllumGlossy normal N vector V float roughness sharpness color C 0 float w 18 1 sharpness extern point P illuminance P N PI 2 Must declare extern L amp Cl because we re in a function extern vector L extern color Cl float nonspec 0 lightsource __nonspecular nonspec if nonspec lt 1 vector normalize normalize 1 1 C 1 nonspec smoothstep 72 w 72 w 0 1 roughness return C Listing 2 14 Ambient light source shader light ambientlight float intensity 1 color IsgHbtoolor dy Cl intensity lightoolor doesn t depend on position L 0 no light direction Listing 2 15 distantlight a light shader for infinitely distant light sources with parallel rays light distantlight float intensity 1 color lightcolor 1 point from point shader 0 0 0 point to point shader 0 0 1 solar to from 0 CL e Intensity Lligh
14. All of the image artifacts that computer graphics researchers had had to live with up to this point were unacceptable if the images were going to fool the critical eyes of the movie going masses In particular Reyes was designed to overcome the following problems with existing rendering algorithms e Vast visual complexity Photographs of the real world contain millions of objects and every object has minute details that make it look real CG images must contain the same complexity if they are to blend seamlessly with live camera work e Motion blur Photographs of moving objects naturally exhibit a blur due to the camera shutter being open for a period of time while the object moves through the field of view Decades of stop motion special effects failed to take this into account and the audience noticed e Speed and memory limitations Motion pictures contain over 100 000 frames and are filmed in a matter of days or weeks Fast computers are expensive and there is never enough mem ory particularly in retrospect Implementability on special purpose hardware was a clear necessity The resulting design brought together existing work on curved surface primitives and scanline algorithms with revolutionary new work in flexible shading and stochastic antialiasing to create a renderer that could produce images that truly looked like photographs It was first used in a film titled Young Sherlock Holmes in 1985 drew rave reviews for its use in The Ab
15. DINOSAUR fama LS ee Figure 3 13 Basic Aladar model The lines are trying to give a sense of the model s patch bound aries C Disney Enterprises Inc One of these reasons was so changes to one layer would not impact on another Generally there were at least two displacement layers on a character One layer created larger features like folds of skin Another displacement layer created the pebbling and scales on the dinosaur s skin Another reason was so each layer could be controlled individually That way we could play with the balance between different types of displacements by number tweaking in the shader Doing such tweaking in the shader is much faster than trying to repaint as the painter has to try to estimate relative grey levels So that one displacement map could both displace in and out middle grey 0 5 was picked as the center of the map Values less then 0 5 would displace inwards While values greater than 0 5 in the maps would displace the surface outwards A big benefit is the displacement bound is half in size What this means is that you get the same amount of detail for half the displacement bound By having the displacement bound smaller the rendering times should be smaller and there should be less opportunity for displacement cracks Also this way of painting is conceptually easier for the painters as they could either cut in or out rather then just always expanding the model outwards The downs
16. E 2E Igt 2E I t yt E 2 14 120 r 0 I Dt 2 E Dt E E r 0 for which the value s of be solved using the quadratic equation This solution is performed by the raysphere function which in turn is used by our enhanced Environment routine see Listing 2 5 Note that Environment also returns an alpha value from the environment map allowing us to composite multiple environment maps together 2 2 REFLECTIONS 3 Listing 2 5 Environment function replaces a simple environment call giving more accurate reflections by tracing against an environment sphere of finite radius raysphere calculate the intersection of ray E I with a sphere centered at the origin and with radius r We return the number of intersections found 0 1 or 2 and place the distances to the intersections in t0 tl always with t0 lt t1 Ignore any hits closer than eps float raysphere point E vector I Origin and unit direction of the ray float r radius of sphere float eps epsilon ignore closer hits output float tO t1 distances to intersection Set up a quadratic equation note that a 1 if I is normalized float b 2 vector E I float c vector E vector r r float discrim b b 4 c float solutions if discrim gt 0 Two solutions discrim sqrt discrim tO discrim b 2 if tO gt eps tl discrim b 2 solutio
17. Lo JRenderimne7Arnlacts 4 ROS Se 16 2 The Secret Life of Lights and Surfaces 23 2 1 Built in local illumination models 23 Zo IRCUCCHOUS 4 5 4 41S e Ee ee hee 28 2 3 Illuminance Loops How diffuse and specular Work 37 2 4 Identifying Lights with Special Properties 38 2 5 Custom Material Descriptions 40 2 0 HNES OUNCES Gre as e Ee eO oa e CEU Eu eub ose uk die UR DT D 44 3 Rendering Issues on Dinosaur 39 OL ACPO SHA OW DIAS OR ey Sei a Co SE Lodi aep grt d 59 2 2 Camera EIO CCUONS doge oes e Ra X ule EM EUR ORC d 67 3 3 Using RenderMan 76 4 The Artistry of Shading Human Skin 83 Ad ASO PTOACH e EE Ue US Gre a Gee ee Batt andit ok ee 83 25 Conceptual Structure 4 sue da ve Sedi we ke BS Oe RUP eee d GSR A 84 2 5 Vii plementavoOn a gs al dese m dot db ete do be IR poe d des 84 AES I D C ot che Sou ce ff ai at elias Se a Soa aie Be ete 85 5 Fur in Stuart Little 89 NEC OU ns ee ee iene ELECTI 89 23 2 Overview e 4 4 oe S Rocha EVE ERE RES ER E ede Ge 90 5 3 FurInstantiation the interpolator 9 JA CUMPS ues 3 3 POE RUE SOC hws EA
18. R float Kr blur DECLARE ENVPARAMS color 0 float alpha if envname if envspace NDC C ReflMap envname P blur alpha else C Environment envname envspace en vrad P R blur alpha return Kr C with a prepared flat reflection map you need only pass NDC as the envspace parameter and the SampleEnvironment function will correctly access your reflection map 2 2 5 Fresnel and Shiny Plastic All materials are more reflective at glancing angles than face on in fact materials approach 100 reflectivity at grazing angles For polished metals the face on reflectivity is so high that this differ ence in reflectivity with angle is not very noticeable You can convince yourself of this by examining an ordinary mirror it appears nearly equally shiny when you view your reflection head on versus when you look at the mirror at an angle So we tend to ignore the angular effect on reflectivity assuming for simplicity that polished metals are equally reflective in all directions as we have in the previous section Dielectrics such as plastic ceramics and glass are much less reflective than metals overall and especially so when viewed face on Therefore their higher reflectivity at grazing angles is a much more significant visual detail see Figure 2 5 The formulas that relate angle to reflectivity of such materials are known as the Fresnel equations A function that calculates the Fresnel terms is
19. Rsp envrad 1 0e 4 t0 t1 gt 0 Rsp vector Psp t0 Rsp alpha float environment envname 3 Rsp blur blur fill 1 return color environment envname Rsp blur blur 32 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES 2 2 2 Creating Cube Face Environment Maps The creation of cube face environment maps is straightforward First six reflection images are created by rendering the scene from the point of view of the reflective object in each of six orthogonal directions in world space given in Table 2 1 Table 2 1 The six view directions in an environment map Axis toward top Axis toward right px positive x nx negative x Dy ny pz nz Next these six views which should be rendered using a square 90 field of view to completely cover all directions are combined into a single environment map This can be done from the RIB file MakeCubeFaceEnvironment px nx py ny pz nz envfile Here px nx and so on are the names of the files containing the individual face images and envfile is the name of the file where you would like the final environment map to be placed Alternatively most renderers will have a separate program that will assemble an environment map from the six individual views In the case of PRMan this can be done with the txmake program txmake envcube px nx py ny pz nz envfile 2 2 3 Flat Surface Reflection Maps For the special case of flat objects such as floors or flat mi
20. a lot of memory In the past when memory was at a premium it was often extremely important to optimize the bucket size and maximum grid size to limit the potentially large visible point and micropolygon list memory consumption On modern computers it is rare that these data structures are sufficiently large to concern us and large limits are perfectly acceptable The default values for bucket size 16 x 16 pixel buckets and maximum grid size 256 micropolygons per grid work well except under the most extreme situations 1 4 3 Transparency Partially transparent objects cause no difficulty to the algorithm generally however they can have two effects on the efficiency of the implementation First transparent objects clearly affect the mem ory consumption of the visible point lists Due to the mathematical constraints of the compositing algebra used by PRMan it is not possible to composite together the various partially transparent layers that are held in the visible point list of a sample until the sample is entirely complete No tice that an opaque layer can immediately truncate a list but in the presence of large amounts of 1 4 RENDERING ATTRIBUTES AND OPTIONS 15 transparency many potentially visible layers must be kept around Second and more importantly transparent layers do not contribute to the occlusion culling of future primitives which means that more primitives are diced and shaded than usual Although this should be obvious
21. are determined to be in shadow if they are farther away than the value in the shadow map at the appropriate pixel In order to antialias this depth comparison given that averaging depths is a nonsensical operation because it implies that there is geometry in some halfway place where it didn t actually exist several depths from the shadow map in neighboring pixels are stochastically sampled and the shadowing result is the percentage of the tests that succeeded 1 2 4 Hiding After shading the shaded grid is sent to the hidden surface evaluation routine First the grid is busted into individual micropolygons Each micropolygon then goes through a miniature version of the main primitive loop It is bounded checked for being on screen and back faced culled if appropriate Next the bound determines in which pixels this micropolygon might appear In each such pixel a stochastic sampling algorithm tests the micropolygon to see if it covers any of the several predetermined point sample locations of that pixel For any samples that are covered the color and opacity of the micropolygon as well as its depth are recorded as a visible point Depending on the shading interpolation method chosen for that primitive the visible point color may be a Gouraud interpolation of the four micropolygon corner colors or it may simply be a copy of one of the corners Each sample location keeps a list of visible points sorted by depth Of course keeping more than
22. aren t really using true area lights We propose that we could achieve this look by thresholding the specular highlight In other words we modify a Blinn Phong specular term c f Shader Listing 2 10 from Cl pow max 0 Nn H 1 roughness to the thresholded version Cl smoothstep eO el pow max 0 Nn H 1 roughness With an appropriately chosen eO and 1 the specular highlight will appear to be smaller have a sharp transition and be fully bright inside the transition region Shader Listing 2 13 is a full implementation of this proposal with some magic constants chosen empirically by the authors Finally Figure 2 9 compares this glossy specular highlight to a standard plastic like specular function In that example we used roughness 0 1 sharpness 0 5 2 6 Light Sources Previous sections discuss how surface shaders respond to the light energy that arrives at the surface reflecting in different ways to give the appearance of different types of materials Now it is time to move to light shaders which allow the shader author similarly detailed control over the operation of the light sources themselves 2 6 1 Non Geometric Light Shaders Light source shaders are syntactically similar to surface shaders The primary difference is that the shader type is called Light rather than surface and that a somewhat different set of built in variables is available The variables available inside light shaders are listed in Table 2
23. been diced and shaded but are not relevant to the current bucket The visible point lists have been reduced to only those that are part of the current bucket a small fraction of the lists required for an entire image Thus we have traded visible point list memory for geo 12 CHAPTER 1 HOW PHOTOREALISTIC RENDERMAN WORKS metric database memory and in all but the most pathological cases this trade off wins by orders of magnitude 1 3 2 Occlusion Culling As described so far the Reyes algorithm processes primitives in arbitrary order within a bucket In the preceeding discussion we mentioned that this might put a primitive through the dicing shading hiding pipeline that will eventually turn out to be obscured by a later primitive that is in front of it If the dicing and shading of these objects takes a lot of computation time which it generally does in a photorealistic rendering with visually complex shaders this time is wasted As stated this problem is not unique to Reyes it happens to nearly every Z buffer algorithm but it is still annoying The enhanced Reyes algorithm significantly reduces this inefficiency by a process known as occlusion culling The primitive bound and sort routine is changed to also sort each bucket s primitives by depth This way objects close to the camera are taken from the sorted list and processed first while farther objects are processed later Simultaneously the hider keeps track of a simple hierarchi
24. comments on the source code in Listing 2 7 To allow easy specification of the many environment related parameters we define macros ENVPARAMS DECLARE_ENVPARAMS and DE CLARE DEFAULTED ENVPARAMS which are macros containing the parameter names declara tions and declarations with default values respectively These macros allow us to succinctly include them in any shader as we have done in shader shinymetal Listing 2 8 With this routine in our toolbox we can make a straightforward shader that can be used for shiny metals like mirrors chrome or copper see Listing 2 8 If you are using the shader for a flat object Listing 2 6 Re 1Map function performs simple reflection mapping color ReflMap string reflname point P float blur output float alpha Transform to the space of the environment map point Pndc transform NDC P float x xcomp Pndc y ycomp Pndc alpha float texture reflname 3 x y blur blur fill 1 return color texture reflname x y blur blur 34 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES Listing 2 7 SampleEnvironment function makes calls to either or both of Environment or ReflMap as needed define ENVPARAMS envname envspace envrad define DECLARE ENVPARAMS X string envname envspace uniform float envrad define DECLARE DEFAULTED ENVPARAMS string envname envspace world y uniform float envrad 100 color SampleEnvironment point P vector
25. frame 10 of the sequence would consist of a weighted average of the last 6 frames of the contact pass This enabled a single footprint to leave an impression that would dissapear quickly but not instantaneously in the final deformations The amount of blur applied to the images can also be animated over time generating footprints that soften as they decompress This new sequence is generated as a series of deformation textures The final step is to use those deformation textures as a displacement map to offset the Pref surface along the normal or another arbitrary vector to create the P surface Naturally you can do other interesting things with this displacement map such as add noise or other features to create interesting wrinkles to make the P surface appear more natural But once the two surfaces are created the shader described above will calculate the difference between the two surfaces and should yield fairly natural looking results when used in combination with careful lighting 5 8 OTHER TRICKS TEXTUREINFO 111 5 8 Other Tricks textureinfo This is simply a handy RenderMan technique which has been available since the addition of the textureinfo command in version 3 8 of PRMan It is mentioned here because it has applica tion in the procedural deformation system when applying the deformation textures back onto the Pref geometry if you use this system you can simply project them through the depth file and they magic
26. grids to have different numbers of micropolygons along their common edge see Figure 1 4 and this difference in micropolygon size across an edge is the source of some shading artifacts that are described in Section 1 5 4 In most other rendering algorithms a shading calculation occurs at every hidden surface sample so raising the antialiasing rate increases the number of shading samples as well Because the vast majority of calculations in a modern renderer are in the shading and hidden surface calculations increasing PixelSamples therefore has a direct linear effect on rendering time In Reyes these two calculations are decoupled because shading rate affects only micropolygon dicing not hidden surface evaluation Antialiasing can be increased without spending any additional time shading so raising the number of pixel samples in a Reyes image will make a much smaller impact on rendering time than in other algorithms often in the range of percentage points instead of factors Conversely adjusting the shading rate will have a large impact on rendering time in images where shading dominates the calculation 1 4 2 Bucket Size and Maximum Grid Size The bucket size option obviously controls the number of pixels that make up a bucket and inversely controls the number of buckets that make up an image The most obvious effect of this control is 14 CHAPTER 1 HOW PHOTOREALISTIC RENDERMAN WORKS Figure 1 4 Adaptive parametric subdivision leads t
27. if needed illuminance N1 radians illum width Cil Kd fne mydirruse Cl L NI illuminance Porig N2 radians illum width C12 Kd fne mydiffuse Cl L N2 Difference in lighting acts as brightness control p GCLO cp CC Bens Go ey Oi 1 0 if debug 1 output the texture no lighting CL else if debug 2 output texture with P s lighting Ci Cil else if debug 3 output texture with Pref s lighting Ci Ci2 5 11 SHADERS 117 5 11 3 Contact Shadow Shader Render contact shadow based on depth data derived from a light placed onto the surface which catches the contact shadow by Rob Engle and Jim Berney surface SLE Contact string shadowname the name of the texture file float samples 10 samples to take per Z lookup float influence 1 0 world space distance in which effect is visible float gamma 0 5 controls ramp on of effect over distance float maxdist 10000 how far is considered infinity get a matrix which transforms from current space to the camera space used when rendering the shadow map uniform matrix matNl textureinfo shadowname viewingmatrix matNl get a matrix which transforms from current space to the Screen space 1 1 used when rendering the shadow map uniform matrix matNP textureinfo
28. into two containers each with fewer members In either case the idea is to create subprimitives that are simpler and smaller on screen and more likely to be small enough when they are examined This technique is often called divide and conquer The resulting subprimitives are then dropped independently into the top of the loop in no par ticular order to be themselves bound cull tested and size tested Eventually the progeny of the original primitive will pass the size test and can move on to the next phase called dicing Dicing converts the small primitive into a common data format called a grid A grid is a tessellation of the primitive into a rectangular array of quadrilateral facets known as micropolygons see Figure 1 2 Because of the geometry of the grid each facet is actually a tiny bilinear patch but we call it a micropolygon nonetheless The vertices of these facets are the points that will be shaded later so the facets themselves must be very small in order for the renderer to create the highly detailed visually complex shading that we have come to expect Generally the facets will be on the order of one pixel in area primitives regardless of original type are converted into grids that look essentially the same to the remainder of the rendering pipeline At this stage each grid retains all of the primitive s attributes and any primitive vertex variables that were attached to the primitive have been correctly in
29. just the frontmost element of the list is only necessary if there is transparency involved Once all the primitives that cover a pixel have been processed the visible point lists for each sample can be composited together and the resulting final sample colors and opacities blended to gether using the reconstruction filter to generate final pixel colors Because good reconstruction kernels span multiple pixels the final color of each pixel depends on the samples not merely in that pixel but in neighboring pixels as well The pixels are sent to the display system to be put into a file or onto a framebuffer 1 2 BASIC GEOMETRIC PIPELINE 9 1 2 5 Motion Blur and Depth of Field Interestingly very few changes need to be made to the basic Reyes rendering pipeline to support several of the most interesting and unique features of PRMan One of the most often used advanced features is motion blur Any primitive may be motion blurred either by a moving transformation or by a moving deformation or both In the former case the primitive is defined as a single set of control points with multiple transformation matrices in the later case the primitive actually contains multiple sets of control points In either case the moving primitive when diced becomes a moving grid with positional data for the beginning and ending of the motion path and eventually a set of moving micropolygons The only significant change to the main rendering pipeline necessary t
30. models sets and cameras The purpose is for staging how the character relate to each other and where everything will be in frame 76 CHAPTER 3 RENDERING ISSUES ON DINOSAUR 3 3 Using RenderMan as a modeler How would you build a detailed model of a 20 foot Dinosaur The model will need to have different kinds of scales for skin pads on the bottom of its feet worn toenails Oh did I mention that it would need to be able to hold up to being scrutinized at less then a foot away and at resolutions up to 2K In other words this model has to be hyper real Once you are done with that model there are another 41 more to go On Dinosaur we had 42 different characters totaling 18 different species The need for this super high detail was that there was a strong desire to have character interac tions If you have a three foot high lemur walking or sitting on a sixty foot high dinosaur and the camera is fairly close up on the smaller character the detail on the larger character really needs to hold up Exactly how high a detail are we talking about Check out Figures 3 9 3 12 Figure 3 9 Basic model of Carnotaur c Disney Enterprises Inc Figure 3 9 shows the basic model of the carnotaur This is what the model looked like coming out from the modeling department Now check out Figure 3 10 This figure has the displacement maps and displacement shader applied to the model Do you notice any differences Check out all those horns The
31. of Shading Human Skin Mitch Prater Pixar Animation Studios mitch pixar com 4 1 Approach 4 1 1 Artistic v s Scientific Past attempts at simulating human skin were often based on direct simulation of the physical prop erties of skin This approach begins with study and data collection of the illumination properties of skin and ends with models describing the interaction of light with and within the various physical components that make up skin and how all those combine to yield the final result Depending on the level of accuracy of the data collection the models derived from the data and the simulation environment this can result in a simulation of skin that has some very realistic looking qualities This is all very good solid fundamental science And it helps us a great deal to understand the nature of skin But to an artist this approach seems a bit like examining a person s DNA in order to figure out how many freckles are on his nose Except for certain schools of modern art to an artist it s all about the end result and the ability to control that result rather than the process used to arrive there even if the end result is something that s supposed to look as realistic as possible If a painter is very skilled he can create a representation of skin that is extremely realistic even though paint on a canvas has vitually nothing to do with the physical processes of light interacting with skin The reason this is possible is a
32. percieved blotchiness needed to be decreased dramatically Shininess and many illumination characteristics were similarly manipulated to maintain a consistent and perceptually orthogonal effect relative to other features All of the skin qualities were set to default values according to the race age and sex of skin that was to be created These default values were all locally modifiable through the application of paint to the individual human model In this way local variation such as the application of makeup wrinkles oiliness whiskers pore characteristics sweat and other skin features were added to each individual The quality and detail of these paintings coupled with the level of control afforded by the skin model were a major factor in the realism of the skin 4 3 3 Composition All the layers were calculated and illuminated individually Once the final result for each character istic layer was determined the layers were composited together using standard compositing algebra to arrive at the final skin result Generally the opacity used was derived directly from the paint pass used to control a particular layer In the case of layers under the skin this opacity value was further modulated by a function that described the variation in the density of the skin 4 4 Results The following images show the results of this work implemented on human characters in the film Toy Story 2 The first image is of Al McWhiggin sans his distin
33. shaders that help us do the job efficiently One such syntactic construct is the solar statement solar vector axis float spreadangle statements The effect of the solar statement is to send light to every Ps from the same direction given by axis The result is that rays from such a light are parallel as if the source was infinitely far away An example of such a source would be the sun 46 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES Listing 2 15 is an example of the solar statement The solar statement implicitly sets the L variable to its first argument there is no need to set L yourself Furthermore the solar state ment will compare this direction to the light gathering cone of the corresponding illuminance statement in the surface shader If the light s L direction is outside the range of angles that the 11 luminance statement is gathering the block of statements within the solar construct will not be executed The spreadangle parameter is usually set to zero indicating that the source subtends an infinites imal angle and that the rays are truly parallel Values for spreadangle greater than zero indicate that a plethora of light rays arrive at each Ps from a range of directions instead of a single ray from a particular direction Such lights are known as broad solar lights and are analogous to very distant but very large area lights for example the sun actually subtends a 1 2 degree angle when seen from Earth The
34. than 400 patches Also some of the maps for these layers can be quite large Fortunately disk space 15 cheap Another price with using such severe displacements is that rendering times will start to pick up when the camera starts getting real close to these displacements So it is a good idea to try to 82 CHAPTER 3 RENDERING ISSUES ON DINOSAUR Figure 3 15 Plio with wounded Bruton c Disney Enterprises Inc maximize the efficiency of PRMan by getting the displacement bounds as tight as possible It has been my observation that it is common enough to find displacement maps that don t go all the way to pure black or pure white So frequently it is worth going through the maps to find out what their maximum ranges really are Then use this in factoring the displacement bounds for each of the patches Wrap up We did what we did because of the hyper real detail requirements that were needed for this movie Take the lemurs for instance they had pads on their hands and feet These pads had to hold up to being viewed at 2k Plus they interacted with characters twenty times their size The detail on these much larger characters had to hold up when these smaller creatures touched sat and walked on top of these larger ones All of these characters abused PRMan displacements like they have never been abused before Remember the carnotaur So there you have it a small taste of the rendering related issues on Dinosaur Chapter 4 The Artistry
35. the radial bspline texture filter This newer filter does do a pretty good job of filtering the textures for displacements It is a good general solution with very little user intervention So if you don t have anything else to use then you really should be using this texture filter when reading texture maps for displacements Instead of displacement mapping there is always bump mapping But there may still have been some of the buzzing due to the same problem Also we would have lost all of our nice profile detail The bump mapping would have made the profiles smooth and details would shrink and grow depending on the orientation of the camera Don t forget those horns on the carnotaur With bump mapping we would have lost them too No for us displacements were the way to go So it made sense to add the rest of the detail like the Dinosaur s scales and folds of skin with PRMan displacements By going with the displacements we were able to tweak the height of the scales or even particular scales on a shot by shot basis Sometimes the directors just wanted the detail lessened or accented in certain shots Such changes were accomplished very easily through the use of displacement by either changing the displacement scalar value for the shader or using a mask texture layer to alter the amount of displacements for that area On Dinosaur all of the texturing was done in layers There were several reasons for this 80 CHAPTER 3 RENDERING ISSUES ON
36. the shader to evaluate the true displacement magnitude for each grid on a grid by grid basis and then store those values with the grid as the exact displacement bound The disadvantage of this technique is that it requires the primitive to be shaded twice once solely to determine the displacement magnitude and then again later to generate the color when the grid is processed normally in its new bucket Thus it is a simple space time trade off 16 CHAPTER 1 HOW PHOTOREALISTIC RENDERMAN WORKS This technique is enabled by the ext remedisplacement attribute which specifies a thresh old raster distance If the projected raster size of the displacement bound for a primitive exceeds the extreme displacement limit for that primitive the extra shading calculations are done to ensure economy of memory If it does not then the extra time is not spent under the assumption that for such a small distance the memory usage is transient enough to be inconsequential 1 4 6 Motion Factor When objects move quickly across the screen they become blurry Such objects are indistinct both because their features are spread out over a large region and because their speed makes it difficult for our eyes to track them As a result it is not necessary to shade them with particularly high fidelity as the detail will just be lost in the motion blur Moreover every micropolygon of the primitive will have a very large bounding box corresponding to the length of the str
37. to perform weighted average of the maps so one map could have emphasis over the other This was needed on certain geometry when large blurs were being applied during shadow lookup The net effect of the pure fifty fifty average was the shadow map was getting pushed back too far from the front causing hot spots By controlling the importance between the two shadow the problem was alleviated Figure 3 2 bottom right shows the color pass using the shadow map from Figure 3 2 bottom left Here is the listing for a C program SIG_zmean c Its purpose is to apply a straight average between two shadow maps creating a third shadow map SIG2k zmean c For SIGGRAPH 2000 RenderMan course Example program that reads in two PRman zdepth files and creates a new zdepth file that is the average of the two This is used to create a shadow map that takes advantage of Laurent Charbonnel s zero shadow bias technique Author Tal Lancaster Date 03 04 2000 include lt stdio h gt include stdlib h include sys stat h include fcntl h typedef struct zfileHeader t int magic type of file short width width of map short height height of map char reserved 128 2 transformation matrices zfileHeader t main int argc char argv int fin mapl fin map2 input maps int fout map output map zfileHeader t zheadl zhead2 map headers fl
38. will be handled in a future bucket Primitive C is in the current bucket because its bounding box touches it as shown but once split you can see that both child primitives will fall into future buckets Primitive D will be diced and shaded in the current bucket but some of the micropolygons generated will be held for sampling until the next bucket is processed Eventually there are no more primitives in the current bucket s list because they all have either been sampled or transferred to future buckets At that point all of the visible point lists in that bucket can be resolved and the pixels for that bucket displayed This is why PhotoRealistic RenderMan creates output pixels in little blocks rather than in scanlines like many algorithms Each block is a bucket The algorithm does not require that the buckets be processed in a particular order but in practice the implementation still uses a scanline style order processing buckets one horizontal row at a time left to right across the row and rows from top to bottom down the image The major effect of this pipeline change is the utilization of memory The entire database is now read into memory and sorted into buckets before any significant amount of rendering is done The vast majority of the geometric database is stored in the relatively compact form of per bucket lists full of high level geometric primitives Some memory is also used for per bucket lists of microp olygons that have already
39. 1 11 12 siggraph 2000 sony picture imageworks advanced renderman fur figure 27 The second thing that needs to be taken into account to make the deformation of the surface appear realistic is the lighting The approach we take is to measure the lighting intensity on the P geometry measure the lighting on the Pref geometry and use the equation illustrated in the previous figure to calculate the brightness change to the plate This will correctly approximate what would have happened on the set had the same deformation taken place This approach generally worked fairly well as long as it wasn t taken to the extreme The res olution of the input texture to the system was only the resolution of the film scans of the negative and there is some amount of softening that takes place while projecting and rendering the surface in RenderMan Furthermore if the stretching is very significant issues such as the film grain of the plate and other details cannot be ignored We also found it useful to add a set of color correction parameters that could control how the brightness changes to the plate were specifically applied so that the actual geometry deformation may not need to be re animated to make a particular part of the deformation appear a little lighter or darker 110 CHAPTER 5 FUR IN STUART LITTLE other tricks deformation system procedural deformations create a depth map of the character s feet from below
40. 2 The goal of a light shader is primarily to determine the radiant energy C1 and light direction L of light impinging on Ps from this source In addition the light may compute additional quantities and store 2 6 LIGHT SOURCES 45 Figure 2 9 Comparing the glossy versus plastic specular illumination models Table 2 2 Global variables available inside light shaders Variables are read only except where noted point Ps Position of the point on the surface that requested data from the light shader The vector giving the direction of outgoing light from the source to the point being shaded Ps This variable can be set explic itly by the shader but is generally set implicitly by the illu minate or solar statements color Cl The light color of the energy being emitted by the source Set ting this variable is the primary purpose of a light shader them in its output variables which can be read and acted upon by illuminance loops using the Lightsource statement The most basic type of light is an ambient source which responds in a way that is independent of position Such a shader is shown in Shader Listing 2 14 The ambientlight shader simply sets C1 to a constant value determined by its parameters It also explicitly sets L to zero indicating to the renderer that there is no directionality to the light For directional light although we could explicitly set L there are some syntactic structures for emitting light in light
41. 89 Rendering Fur with Three Dimensional Textures Computer Graphics 23 3 271 280 illuminance P norm hair radians illum width nL normalize L T Dot n e Tbe T DOG e cuv Alpha acos T Dot nL Beta acos T Dot Kajiya T Dot nL T Dot e sin Alpha sin Beta calculate diffuse component if clump dark strength gt 0 0 clump darkening 1 Coolumpodark strength abs clamp nL normalize 1 clump vect 1 0 else clump darkening 1 0 get light source parameters if lightsource nonspecular nonspecular 0 nonspecular 0 M ligHtsource SpecularColor SpecularColor 0 Specularcolor s color 1 Cspec 1 nonspecular SpecularColor clump darkening SPEC1 Cl pow Kajiya 1 roughness1 SPEC2 Cl pow Kajiya 1 roughness2 5 11 SHADERS 115 Cdiff clump darkening fnc diffuselgt Cl L norm hair darkening clamp harr col var 0 1 darkening 1 smoothstep var fade end var fade start abs luminance Kd Cdiff darkening final c mix rootcolof trxpeolor v darkening Ci Ka ambient Kd Cdiff static ambient final c v Kspec Cspec specularcolor Cr Clamp Ci color 0 color Ll Oi Os Ci Oi Ci 5 11 2 Deformation Shader deformation surface shader projects a texture through the camera onto the Pref object and deforms it to the P position addit
42. Advanced RenderMan 2 To RI INFINITY and Beyond SIGGRAPH 2000 Course 40 course organizer Larry Gritz Pixar Animation Studios Lecturers Tony Apodaca Pixar Animation Studios Larry Gritz Pixar Animation Studios Tal Lancaster Walt Disney Feature Animation Mitch Prater Pixar Animation Studios Rob Bredow Sony Pictures Imageworks July 2000 ii Desired Background This course is for graphics programmers and technical directors Thorough knowledge of 3D image synthesis computer graphics illumination models and previous experience with the RenderMan Shading Language is a must Students must be facile in C The course is not for those with weak stomachs for examining code Suggested Reading Material The RenderMan Companion A Programmer s Guide to Realistic Computer Graphics Steve Upstill Addison Wesley 1990 ISBN 0 201 50868 0 This is the basic textbook for RenderMan and should have been read and understood by anyone attending this course Answers to all of the typical and many of the extremely advanced questions about RenderMan are found within its pages Its failings are that it does not cover RIB only the C interface and also that it only covers the 10 year old RenderMan Interface 3 1 with no mention of all the changes to the standard and to the products over the past several years Nonetheless it still contains a wealth of information not found anyplace else Advanced RenderMan Creating CGI for Motion Pictures
43. ENVPARAMS OT s Ci ESI 2 3 ILLUMINANCE LOOPS OR HOW DIFFUSE AND SPECULAR WORK 37 Figure 2 6 MaterialShinyMetal left and MaterialShinyPlastic right 2 3 illuminance Loops or How diffuse and specu lar Work We have seen classes of materials that can be described by various relative weightings and uses of the built in diffuse and specular functions In point of fact these are just examples of possible illumination models and it is quite useful to write alternative ones In order to do that however the shader needs access to the lights how many are there where are they and how bright are they The key to such functionality is a special syntactic structure in Shading Language called an illuminance loop illuminance point position statements illuminance point position vector axis float angle statements The illuminance statement loops over all light sources visible from a particular position In the first form all lights are considered and in the second form only those lights whose directions are within angle of axis typically angle n 2 and axis N which indicates that all light sources in the visible hemisphere from P should be considered For each light source the statements are executed during which two additional variables are defined L is the vector that points to the light source and C1 is the color representing the incoming energy from that light source Perhaps the m
44. GHTS AND SURFACES be found in Blinn 1982 Treatments of iridescence can be found in Smits and Meyer 1989 and Gondek et al 1994 An excellent overall discussion of surface physics including refraction and the Fresnel equations derived in all their gory detail can be found in Hall 1989 This book contains equations and pseudocode for many of the more popular local illumination models Unfortunately it has come to our attention that this book is now out of print Glassner s book also is an excellent reference on BRDFs Additional papers discussing local illumination models include Blinn and Newell 1976 Blinn 1977 Cook and Torrance 1981 Whitted and Cook 1985 Whitted and Cook 1988 Hall 1986 Nakamae et al 1990 He et al 1991 Westin et al 1992 Schlick 1993 Hanrahan and Krueger 1993 Lafortune et al 1997 Goldman 1997 Shadow maps are discussed in Williams 1978 and Reeves et al 1987 2 6 LIGHT SOURCES 53 Listing 2 11 LocIllumOrenNayar implements a BRDF for diffuse but rough surfaces col Loc lar N c Oren and Nayar s generalization of Lambert s reflection model The roughness parameter gives the standard deviation of angle orientations of the presumed surface grooves When roughness 0 the model is identical to Lambertian reflection Or IllumOrenNayar normal N vector V float roughness Surface roughness coefficients for Oren N
45. HOTOREALISTIC RENDERMAN WORKS The biggest disadvantage of shading before hiding 15 that objects are shaded before it is known whether they will eventually be hidden from view If the scene has a large depth complexity large amounts of geometry might be shaded and then subsequently covered over by objects closer to the camera That would be a large waste of compute time In fact it is very common for this to occur in Z buffer renderings of complicated scenes This disadvantage is addressed in the enhanced algorithm described in Section 1 3 2 1 2 7 Memory Considerations In this pipeline each stage of processing converts a primitive into a finer and more detailed version Its representation in memory gets larger as it is split diced busted and sampled However notice also that every primitive is processed independently and has no interaction with other primitives in the system Even sibling subprimitives are handled completely independently For this reason the geometric database can be streamed through the pipeline just as a geometric database is streamed through typical Z buffer hardware There is no long term storage or buffering of a global database except for the queue of split primitives waiting to be bounded which is rarely large and therefore there is almost no memory used by the algorithm With a single exception the visible point lists As stated earlier no visible point list can be processed until it is known that all of the primi
46. S Next the bounding box is checked to see if the primitive is actually on screen The camera description in the graphics state gives us the viewing volume a 3D volume of space that contains everything that the camera can see In the case of perspective projections this is a rectangular pyramid bounded on the sides by the perspective projection of the screen window sometimes called the screen space viewport in graphics texts and truncated at front and back by the near and far clipping planes for an orthographic projection this is a simple rectangular box It is important to note at this point that Reyes does not do any global illumination or global intervisibility calculations of any kind For this reason any primitive that is not at least partially within the viewing volume cannot contribute to the image in any way and therefore is immediately culled trivially rejected Also any one sided primitives that are determined to be entirely back facing can be culled at this stage because they also cannot contribute to the image If the primitive is at least partially on screen its size is tested If it is deemed too large on screen according to a metric described later it is split into smaller primitives For most parametric primitives this means cutting the primitive in two or possibly four along the central parametric line s For primitives that are containers of simpler primitives such as polyhedra splitting may mean roughly dividing
47. a ee ee ee ns 02 ded obe NAR ee fe Bay eae deu wk i VOR TU 100 IE RELENE MP 103 vii viii CONTENTS 5 7 Other Tricks Deformation System 107 IS Other Ticks fexturciniO nasa c ope hex 111 s SII TREES 112 S T0 One ls OD qoum n E b IIR De eer AER REESE RAE rr RE a 112 JEE SMde Se we os Ae hth ae ie ae ase Sele ea oy ee 112 Bibliography 119 Preface Welcome to Advanced RenderMan 2 To RI INFINITY and Beyond How s that for a goofy title This course covers the theory and practice of using RenderMan to do the highest quality computer graphics animation production This is the sixth SIGGRAPH course that we have taught on the use of RenderMan and it is quite advanced Students are expected to understand all of the basics of using RenderMan We assume that you have all read the venerable RenderMan Companion and certainly hope that you have attented our previous SIGGRAPH courses and read the book Advanced RenderMan Creating CGI for Motion Pictures Morgan Kaufmann 1999 We ve taught several SIGGRAPH courses before Though this year s course resembles the 1998 and 1999 courses Advanced RenderMan Beyond the Companion in detail and advanced level we are delighted to be teaching a course that has all new topics and even some new
48. ad grid artifacts in second derivative calculations PRMan version 3 9 has a better differential estimator that makes many of these artifacts disappear but it is still possible to confuse it at inflection points 1 2 3 Texturing The number and size of texture maps that are typically used in photorealistic scenes are so large that it is impractical to keep more than a small portion of them in memory at one time For example a typical frame in a full screen CGI animation might access texture maps approaching 10 Gb in total size Fortunately because this data is not all needed at the highest possible resolution in the same frame mip maps can be used to limit this to 200 Mb of texture data actually read However even this is more memory than can or needs to be dedicated to such transient data PRMan has a very sophisticated texture caching system that cycles texture data in as necessary and keeps the total in core memory devoted to texture to under 10 Mb in all but extreme cases The proprietary texture file format is organized into 2D tiles of texture data that are strategically stored for fast access by the texture cache which optimizes both cache hit rates and disk I O performance Shadows are implemented using shadow maps that are sampled with percentage closer filtering Reeves et al 1987 In this scheme grid vertices are projected into the view of the shadow casting light source using shadow camera viewing information stored in the map They
49. ally always end up in the right position even with animated cameras It also has other convenient uses including projecting multiple textures onto a single object for recreating surface detail from multiple still photographs one last trick textureinfo project through a tx file you want to project a texture through an arbitrary camera but don t want to deal with doing the projection yourself and passing all those camera parameters to renderman render a tiny tx file from the camera of interest with no geometry turned on or a sequence of images for an animated camera siggraph 2000 sony picture imageworks advanced renderman fur figure 29 one last trick textureinfo project through a tx file use this snippet to project the texture texname through the projection matrix used to render pr j tx texturerirnto pro tx projectionmatrix cur2tex texP transform cur2tex Pref xcomp texP 1 0 2 0 ycomp texP 1 0 2 0 texture texname x y siggraph 2000 sony picture imageworks advanced renderman fur figure 3 0 112 CHAPTER 5 FUR IN STUART LITTLE 5 9 Credits The majority of the development of the fur technology for Stuart Little was performed by Clint Hanson and Armin Bruderlin Armin also contributed in a large way to these course notes through his paper A Method to Generate Wet and Broken up Animal Fur which was used nearly verbatim to describe the cl
50. and hills This is modeled mathematically as a collection of microfacets 2 5 CUSTOM MATERIAL DESCRIPTIONS 4 having a statistical distribution of relative directions Their results indicated that many real world rough materials like clay could be more accurately modeled using the following equation L2 05 0 5 r PE cos 0 A B max 0 cos sina tan where 2 10 0 5 0 0522 0233 2 0 45 c 0 09 a max 0 0 min 0 and the terms mean e pis the reflectivity of the surface Kd Cs e Fo 15 the energy arriving at the surface from the light C1 e 0 is the angle between the surface normal and the direction of the light source e 0 is the angle between the surface normal and the vector in the direction the light is reflected 1 toward the viewer e 9 is the angle about the normal between the incoming and reflected light directions e o is the standard deviation of the angle distribution of the microfacets in radians Larger values represent more rough surfaces smaller values represent smoother surfaces If 0 the surface is perfectly smooth and this function reduces to a simple Lambertian reflectance model We ll call this parameter roughness These equations are easily translated into an illuminance loop as shown in Listing 2 11 Figure 2 7 shows a teapot with the Oren Nayar reflectance model The left teapot
51. ant amounts of work occur on other buckets The most important difference in the pipeline is in the bounding step which now also sorts the primitives based on which buckets they affect that is which buckets the bounding box overlaps If a primitive is not visible in the current bucket of interest it is put onto a list for the first bucket where it will matter and is thereby held in its most compact form until truly needed 1 3 ENHANCED GEOMETRIC PIPELINE 11 completed buckets current bucket future buckets Figure 1 3 When the renderer processes the primitives that are on the list for the current bucket their size and positions determine their fates After this the algorithm proceeds in the obvious way Buckets are processed one at a time Objects are removed from the list for the current bucket and either split or diced Split primitives might be added back to the list or might be added to the lists of future buckets depending on their bounding boxes Diced primitives go through the normal shading pipeline and are busted During busting the micropolygons are bound and similarly bucket sorted Micropolygons that are not in the current bucket of interest are not sampled until the appropriate bucket is being processed Figure 1 3 shows four primitives whose disposition is different Primitive A will be diced shaded and sampled in the current bucket Primitive B needs to be split and half will return to the current bucket while half
52. are in need of mapping The red dots represent areas that still haven t been mapped yet As long as the camera never sees those areas there is no need for them to be painted The blue areas represent areas that are unique to each map and so present no worries when using multiple keyframe 3 2 CAMERA PROJECTIONS 73 projections However the green areas represent areas that are shared between the maps It is these areas that care is needed to blend or impose a precedence in the keyframes between the maps to provide a smooth transition Figure 3 6 Top left Frame 1 of crescent moon dune CG set Top right Frame 1 of crescent moon dune Middle left Frame 81 of crescent moon dune CG set Middle right Frame 81 without zdepth map option Bottom Corrected CP using depth maps All images Disney Enterprises Inc A similar problem occurs when the scene s geometry hides other portions of the ground plane See Figure 3 6 Say the top left image was used as a keyframe The points hidden by the front dune and the dune itself share the same NDC projections So if the camera moved to something like in the middle left image a camera projection will start to produce a duplicate of the front dune 74 CHAPTER 3 RENDERING ISSUES ON DINOSAUR as seen in the middle right image A solution for this case is to render depth maps when making the keyframes Then use the depth maps to aid in determining how to project the maps later when rendering with the pain
53. ariability was grossly controlled with settings to adjust the overall characteristics as well as with paint to provide detailed local control The features themselves were chosen to be easily identifiable both physically and aesthetically and to be mutually orthogonal Any interaction of the controls as is often the case with a prurely physically based implementation would make the controls completely impractical These were probably the most crucial aspects of the skin that all characteristics were defined to have a direct physical and aesthetic meaning that all characteristics were visually orthog onal to each other and that each was specified using a normalized range that remained consistent in its effect throughout its range and the ranges of other features 4 2 2 Layering Each feature applies to a specific layer of the skin whether within the skin itself or on its surface The layering reflects the actual physical layering of the components themselves Since all the char acteristics were carefully constructed using a normalized range and to be orthogonal to each other there was no difficulty in specifying and controlling each characteristic individually even though in reality they may reside in the same physical layer Other features would occlude features in lower layers depending on their physical properties of opacity and the degree to which they were applied This layering of features applied to the illumination of each feature as we
54. as is more of an art than science There are many factors to con sider Large scale objects typically will need larger shadow bias values than smaller objects Also the shape of the object and its orientation to the light will effect the shadow bias settings needed to get the shadow to look better If the object or light change their orientations too much relative to each other the bias too will need to be changed accordingly aka the shadow bias may need ani mating Typically the actual shadow bias number is determined by a trial and error method Since the object s orientation to the light could effect the shadow bias settings each shadow light in the scene may need to have its shadow bias values changed differently On Dinosaur we were locked off on PRMan 3 7 We never had access to any of the newer shadow generation options like midpoint which minimizes the need for shadow bias tweaking So shadow bias tweaking was a fact of life on our production Laurent Charbonnel a Look Development TD on Dinosaur came up with a simple but ingenious method for allowing the lighters to ignore the shadow bias much of the time This technique was called Zero Shadow Bias as much of the time one could leave the shadow bias setting at 0 Zero Shadow Bias is a good example of adding more intelligence to the shadow generation process See PRMan 3 8 release notes http www pixar com products rendermandocs toolkit Toolkit rnotes 3 8 html shadowmap for
55. ass the light shader can determine if a particular surface point is in shadow by comparing its distance to the light against that stored in the shadow map If it matches the depth in the shadow map it is the closest surface to the light in that direction so the object receives light If the point in question 1s farther than indicated by the shadow map it indicates that some other object was closer to the light when the shadow map was created In such a case the point in question is known to be in shadow Figure 2 10 shows a simple scene with and without shadows as well as the depth map that was used to produce the shadows Figure 2 10 Shadow depth maps A simple scene with and without shadows left The shadow map is just a depth image rendered from the point of view of the light source right To visualize the map we assign white to near depths black to far depths Shading Language gives us a handy built in function to access shadow maps 2 6 LIGHT SOURCES 49 float shadow string shadowmapname point Ptest The shadow function tests the point Ptest in current space against the shadow map file specified by shadowmapname The return value is 0 0 if Ptest is unoccluded and 1 0 if Ptest is occluded in shadow according to the map The return value may also be between 0 and 1 indicating that the point is in partial shadow this is very handy for soft shadows Like texture and environment the shadow call has sev
56. assign color and opacity to each grid vertex is of very little consequence to the Reyes pipeline itself but of course is of central importance to the users of PRMan PRMan s shading system is an interpreter for the RenderMan Shading Language which reads the byte codes in the s10 file previously created by the Shading Language compiler Hanrahan and Lawson 1990 Notice that the data structure that the interpreter operates on a grid is a bunch of large rectangular arrays of floating point numbers For this reason it may not be surprising that the interpreter actually executes as a virtual SIMD Single Instruction Multiple Data vector math unit Such vector pipelines were typical in 1980 s vintage supercomputers The run time interpreter reads shader instructions one at a time and executes the operator on all grid vertices It is important to contrast this scheme with the alternative which would run the entire shader on the first grid vertex then run it on the second grid vertex and so on The advantages of the breadth first solution is efficiency We are able to make near optimal use of the pipelined floating point units that appear in modern processors and we have excellent cache performance due to strong locality of reference In addition for uniform variables the grid has exactly one data value not nvertices values As a result the interpreter is able to compute operations on uniform data exactly once on each grid saving significantly
57. ax depth points in the shadows Shadow pass is getting the min depth values and shadow pass 2 is getting the max depth values So instead of using the surface shader one could accomplish pass one by setting the hider option Hider hidden jitter 0 depthfilter min Then for pass2 set it to Hider hidden jitter 0 depthfilter max Example surface shader illustrating Laurent Charbonnel s zero bias shadow 3 1 ZERO SHADOW BIAS 63 If this is all that is required then this is all that would be needed and there would be no need for the surface shader However if there 1s a desire that partial geometry be removed for the shadow calculations then a surface shader containing the necessary code for this object removal would still be needed in the shadow pass a n Figure 3 2 Top left shadow pass 1 frontfacing top right shadow pass 2 backfacing bottom left averaged result of shadow pass and 2 bottom right Image rendered using zero shadow bias Step3 Before the color pass process the two zdepth files to create a third zdepth file See Figure 3 2 bottom left for an average of the shadow maps from Figures 3 2 top left and right The exact form of processing could be whatever On Dinosaur it was typically a fifty fifty average between the two maps which should produce something very similar to the newer PRMan mid 64 CHAPTER 3 RENDERING ISSUES ON DINOSAUR point depth shadow file But there were controls
58. ayar s formula float sigma2 roughness roughness float 1 0 5 sigma2 sigma2 0 33 float B 0 45 gigma2 sigma2 0 09 Useful precomputed quantities float theta acos V N Angle between V and N vector V perp normalize V N V N Part of V perpendicu to N Accumulate incoming radiance from lights in color C 0 extern point P illuminance P N PI 2 Must declare extern L amp Cl because we re in a function extern vector 1 extern color Cl float nondiff 0 lig tso rce nondifftuse nondiff if e vector LN normalize 1 float cos theta_i LN N float cos_phi_diff V perp N normalize LN os theta 1 float theta 1 acos cos theta 1 float alpha max theta i theta r float beta min theta 1 theta C lenondi tt Cl ces theta B max 0 cos phi diff sin alpha tan beta return C 54 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES Listing 2 12 LocI1llumWardAnisotropic Greg Ward Larson s anisotropic specular illumina tion model oo ob ok ob ob OR FF FH HF HF HF F d Greg Ward Larson s anisotropic specular local illumination model The derivation and formulae can be found in Ward Gregory J Measuring and Modeling Anisotropic Reflection ACM Computer Graphics 26 2 Proceedings of Siggraph 92 pp 265 272 July 1992 Notice that compared t
59. cal data structure that describes how much of the bucket has been covered by opaque objects and at what depths Once the bucket is completely covered by opaque objects any primitive which is entirely behind that covering is occluded Because it cannot be visible it can be culled before the expensive dicing and shading occurs in the case of procedural primitives before they are even loaded into the database By processing primitives in front to back order we maximize the probability that at least some objects will be occluded and culled This optimization provides a two to ten times speedup in the rendering times of typical high resolution film frames 1 3 3 Network Rendering In the enhanced Reyes algorithm most of the computation dicing shading hiding and filtering takes place once the primitives have been sorted into buckets Moreover except for a few details discussed later those bucket calculations are generally independent of each other For this reason buckets can often be processed independently and this implies that there is an opportunity to exploit parallelism PRMan does this by implementing a large grain multiprocessor parallelism scheme known as NetRenderMan With NetRenderMan a parallelism control client program dispatches work in the form of bucket requests to multiple independent rendering server processes Server processes handle all of the calculation necessary to create the pixels for the requested bucket then
60. clumping of the final fur animated area clumping Solution create additional clumping areas on neighboring patches if u v radii of current clumping area exceed the boundaries of the patch siggraph 2000 sony picture imageworks advanced renderman fur figure 15 The solution presented here utilizes our surface seam and topology generator module whenever 100 CHAPTER 5 FUR IN STUART LITTLE a new particle hits a surface the clump method checks at each frame whether the u v radii exceed the boundaries of that surface if so it calculates the u v center and new u v radii of this clumping area with respect to the affected neighboring patches These secondary clumping areas on affected neighboring patches are then included in the calculations above to determine which clumps fall into clumping areas animated area clumping siggraph 2000 sony picture imageworks advanced renderman fur figure 16 Image 02000 by Columbia Pictures Industries Inc All Rights Reserved 5 5 Shading In order to render large amounts of hair quickly and efficiently the geometric model of each hair is kept simple As explained above a hair is represented by a parametric curve with few control vertices the default is four We use the RenderMan RiCurves primitive to render the final hairs This primitive consists of many small micro polygons falling along a specified hair curve with their normals pointing directly towards the camera
61. ctive facial hair The second image is of the toy cleaner otherwise known as Geri from Geri s Game 86 CHAPTER 4 THE ARTISTRY OF SHADING HUMAN SKIN Figure 4 1 Al Image 2000 Disney Pixar All Rights Reserved 4 4 RESULTS 87 Figure 4 2 The Cleaner Image 2000 Disney Pixar All Rights Reserved 88 CHAPTER 4 THE ARTISTRY OF SHADING HUMAN SKIN Chapter 5 Fur in Stuart Little Rob Bredow Sony Pictures Imageworks rob_bredow yahoo com Abstract Realistic fur instantiation and shading is a challenge that was undertaken during recent work on the film Stuart Little This talk will detail some of the techniques used to create Stuart Little s furry coat and how RenderMan was used to light and shade the mouse s hair 5 1 Goals e Must be cute and realistic e Retain detailed looking fur regardless of size on screen e Realistic lighting model needs flexible controls for cases where realism is not desired handle multiple lighting environments day night sourcey diffuse wet shiny dirty e Fur interaction with environment wind other objects self clothing e Multiple furred characters on screen at one time e Efficient enough for nearly 500 shots 89 90 CHAPTER 5 FUR IN STUART LITTLE 5 2 Design Overview design overview Y RenderMan DSO i ter final interpolated hairs hair shader Legend optional data file i
62. cur zval cur zval map2 go through maps looking for the max and min values for j 0 j zheadl height j for i 0 i lt zheadl width i if cur_zval_mapl lt min map1 min cur zval if cur_zval_mapl gt max map1 max cur_zval_mapl if cur zval map2 min map2 min map2 cur zval map2 if cur_zval_mapl gt max map2 max map2 cur zval map2 cur zval 1 cur zval_map2 reset the position so can go through again cur zval zvals cur zval map2 zvals map2 for j 0 j zheadl height j for i 0 i lt zheadl width i zval mapl cur zval mapl zval map2 cur zval map2 zval map3 zval_mapl zval map2 2 0f 66 CHAPTER 3 RENDERING ISSUES ON DINOSAUR if zval map3 gt 1e6 amp amp zval map3 gt 0 5 max 1 Just a check as the second map could produce a very large depth value throwing off the mean so want to clamp it zval map3 max mapl Reusing the data to hold the new map cur zval mapl zval cur zval cur zval 2 write fout map zvals mapl mapsize else fprintf stderr Not zfliles or zfriles not same size n close fin_mapl1 close fin_map2 close fout map The following structure from this program may need some explaining typedef struct zfileHeader t int magic type of fi
63. der won t know about any of the changes Now is probably a good time to note that this function will only work with versions of PRMan 3 8 and later due to the changes in matrix casts If you wanted to have this work under PRMan 3 7 something like the following would be needed uniform matrix Mobject2NDC Mobj 1 MNDC Here is an example of what a matrixdata h file would look like Define the number of keyframes to use Unless f _cameraProjections has been properly modified to handle multiple keyframes this number should always be one Tu 3 2 CAMERA PROJECTIONS 71 Figure 3 4 Slideprojector NDC projection Showing a point P in the scene being transformed to NDC space Pndc The transformation matrix that does this is noted in Mobject2NDC define ARRAYSIZE 1 store the keyframes that were used In this case frame one was the keyframe uniform float keyframes ARRAYSIZE 1 The matrix array of the various keyframes In this example the matrix was for frame 1 E uniform matrix Mcp obj2NDC ARRAYSIZE frame 1 Mcp obj2NDC 0 1 1798 0 108326 0 545303 0 545303 0 15267 2 26076 0 294529 0 294529 0 239958 0 136075 0 784792 0 784791 0 54674 0 62762 0 987516 0 988516 The main workhorse for the camera projection shader is cameraProjection Its goal is to perform the projection and then return a color It does this by transforming the point to object space
64. dered approximations To help alleviate this problem PRMan has shadow bias controls When reading the shadow maps these controls help to create a fudge factor that pushes the depth values back away from the front of the surface It is important to find the right fudge factor to apply Because a shadow bias that is too small will cause the surface to look dirty due to inaccurate self shadowing see Figure 3 1 top This is due to round off error A surface in the shadow map alternates believing that it is in and out of shadow because it is approximately represented in the map instead of exactly One also has to take care that the shadow bias isn t too large or the shadow will get pushed back so much that it detaches from the object entirely see Figure 3 1 bottom right Figure 3 1 shows the results with various shadow bias settings In In the top two images the bias is too small and so the images appear dirty In the bottom right image the bias is so large that the shadow has detached from the cylinder entirely In the bottom left image the bias is about right Frequently the ideal shadow should read about midpoint of the object from the view of the light casting the shadow This is what the goal should be most of the time to get one s shadows to read correctly The PRMan midpoint method does precisely this which is why one very rarely needs to worry about setting the shadow bias when using this method Finding just the right shadow bi
65. djacent grid vertices are moved in different directions wildly or subtly the area of the micropolygon connecting them will change Generally this change is so small that the micropolygon is still safely in the range expected of shading rate However if the displacement function has strong high frequencies adjacent grid 20 CHAPTER I HOW PHOTOREALISTIC RENDERMAN WORKS vertices might move quite differently For example an embossing shader might leave some vertices alone while moving vertices inside the embossed figure quite a distance The micropolygons whose corners move very differently will change size radically and sometimes will be badly bent or twisted see Figure 1 7 Twisted micropolygons have unusual normal vectors and this alone may be enough to cause shading artifacts For example highly specular surfaces are very sensitive to normal vector orienta tion so a micropolygon that is twisted in an unusual direction may catch an unexpected highlight More common however is that the stretching of the micropolygons will itself be visible in the final image An individual flat shaded micropolygon creates a constant colored region in the image With a standard shading rate of around a pixel every pixel gets a different micropolygon and the flat shading is not visible But large stretched or long twisted micropolygons will cover many pixels and the constant colored region will be evident Sometimes this takes the form of alternating dark a
66. e 66 pages 108 114 Greene 1986b Environment mapping and other applications of world projections IEEE Computer Graphics and Applications 6 11 Hall R 1986 A characterization of illumination models and shading techniques The Visual Computer 2 5 268 77 Hall 1989 Illumination and Color in Computer Generated Imagery Springer Verlag New York includes C code for radiosity algorithms Hanrahan P and Krueger W 1993 Reflection from layered surfaces due to subsurface scattering In Kajiya J T editor Computer Graphics SIGGRAPH 93 Proceedings volume 27 pages 165 174 119 120 BIBLIOGRAPHY Hanrahan P and Lawson J 1990 A language for shading and lighting calculations In Baskett F editor Computer Graphics SIGGRAPH 90 Proceedings volume 24 pages 289 298 He X D Torrance K E Sillion F X and Greenberg D P 1991 A comprehensive physical model for light reflection In Sederberg T W editor Computer Graphics SIGGRAPH 91 Proceedings volume 25 pages 175 186 Kajiya J T 1985 Anisotropic reflection models In Barsky B A editor Computer Graphics SIGGRAPH 65 Proceedings volume 19 pages 15 21 Lafortune E P F Foo S C Torrance E and Greenberg D 1997 Non linear approximation of reflectance functions In Whitted T editor SIGGRAPH 97 Conference Proceedings Annual Conference Series pages 117 126 ACM SIGGRAPH Addison We
67. e the lights Beware e When rendering the shadow map only include objects that will actually cast shadows on themselves or other objects Objects that only receive but do not cast shadows such as walls or floors can be eliminated from the shadow map pass entirely This saves rendering time when creating the shadow map and also eliminates the possibility that poorly chosen bias will cause these objects to incorrectly self shadow since they aren t in the maps anyway e Some renderers may create shadow map files directly Others may create only depth maps or z files that require an additional step to transform them into full fledged shadow maps much as an extra step is often required to turn ordinary image files into texture maps For example when using PRMan z files must be converted into shadow maps as follows txmake shadow shadow z shadow sm This command invokes the txmake program PRMan s texture conversion utility to read the raw depth map file shadow 2 and write the shadowmap file shadow sm It is also possible that some renderers including BMRT but not PRMan support automatic ray cast shadows which do not require shadow maps at all In the case of BMRT the following RIB attribute causes subsequently declared LightSource and AreaLightSource lights to automatically be shadowed Attribute light shadows on There are also controls that let you specify which geometric objects cast shadows consult the BMRT Us
68. e implementations of diffuse and specular re spond to _nondiffuse and _nonspecular in this manner although as we explained earlier the implementation of specular is not quite what is in Listing 2 10 2 4 IDENTIFYING LIGHTS WITH SPECIAL PROPERTIES 39 Listing 2 10 The implementation of the built in diffuse and specular functions includ ing controls to ignore nondiffuse and nonspecular lights color diffuse normal Nn extern point P color 0 illuminance Nn PI 2 float nondiff 0 lightsource _ mnondsrtrusge nondiff Cl 1 nondiff Nn normalize L return C color specular normal Nn vector V float roughness extern point P color 0 illuminance Nn PI 2 float nonspec 0 lightsource _nonspecular nonspec vector H normalize normalize L V C Cl 1 nonspec pow max 0 Nn H 1 roughness return C This message passing mechanism may be used more generally to pass all sorts of extra in formation from the lights to the surfaces For example a light may include an output parameter giving its ultraviolet illumination and special surface shaders may respond to this parameter by exhibiting fluorescence There is an additional means of controlling illuminance loops with an optional light cate gory specifier illuminance string category point position statements illuminance string category point position vector axis
69. e that is either too small or too large e width isa float that multiplies the estimates of the rate of change of Ptest used for antialiasing the shadow map lookup This parameter functions analogously to the width parameter to texture or environment Its use is largely obsolete and we recom mend using blur to make soft shadow edges rather than width Figure 2 11 Adding blur to shadow map lookups can give a penumbra effect 50 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES Figure 2 12 Selecting shadow bias Too small a bias value will result in incorrect self shadowing left Notice the darker dirtier look compared to Figures 2 11 or 2 10 Too much bias can also introduce artifacts such as the appearance of floating objects or the detached shadow at the bottom of the cylinder right The Ptest parameter determines the point at which to determine how much light is shadowed but how does the renderer know the point of origin of the light When the renderer creates a shadow map it also stores in the shadow file the origin of the camera at the time that the shadow map was made in other words the emitting point The shadow function knows to look for this information in the shadow map file Notice that since the shadow origin comes from the shadow map file rather than the light shader it s permissible and often useful for the shadows to be cast from an entirely different position than the point from whic
70. e use of zdepth maps for looking for object occulsion Some parameters that are used frame the current frame being rendered cp mapBaseName The prefix of the camera projection maps Utimately the shader expects that the name of the maps follow a form like testcp 53 tx For this example the base name is testcp cp mapSuffix The tail of the the maps name In the example given for cp mapBaseName the suffix would be tx usePref This value should be true if Pref is provided for the geometry Wo N CAMERA PROJECTIONS in arrays in the matrixdata h Tal Lancaster 02 19 00 talrmr pacbell net This tells the shader to ignore P and use the alternate points instead The Pref points themselves Cerror This is the color to return if the points being rendered are outside any of the painted maps debug Used to invoke the debugging options in the f cameraProjection function printMatrix Set this to true to dump the camera matrix of the current frame to stdout This should be done for each camera projection keyframe Then have these matrix values stored include f SIG2k cameraProjection h surface SIG2k cameraProjection float Kd e 0 5 Ka s 0 05 KS 0 rouguness s 0 15 color specularcolor 1 float frame 0 string cp mapBaseName string cp _ mapSuffix float usePref 0 varying point Pref 0 color Cerror color 1 0 0
71. eak which means that fine tessellations will lead to large numbers of micropolygons that linger a long time in memory as they are sampled by the many buckets along their path The solution to this problem is to enlarge the shading rate of primitives if they move rapidly It is possible for the modeler to do this of course but it is often easier for the renderer to determine the speed of the model and then scale the shading rates of each primitive consistently The attribute that controls this calculation is a GeometricApproximation flag known as mot ionfactor For obscure reasons mot ionfactor gives a magnification factor on shading rate per every 16 pixels of blurring Experience has shown that a motion factor of 1 0 is appropriate for a large range of images The same argument applies equally to depth of field blur and in the current implementation motionfactor despite its name also operates on primitives with large depth of field blurs as well 1 5 Rendering Artifacts Just as an in depth understanding of the Reyes pipeline helps you understand the reason for and utility of various rendering options and attributes it also helps one understand the causes and so lutions for various types of geometric rendering artifacts that can occur while using PhotoRealistic RenderMan 1 5 1 Eye Splits Sometimes PhotoRealistic RenderMan will print the error message Cannot split primitive at eye plane usually after appearing to stall for qui
72. ealism and Physics in Computer Graphics Upstill S 1990 The RenderMan Companion A Programmer s Guide to Realistic Computer Graphics Addison Wesley Ward G J 1992 Measuring and modeling anisotropic reflection In Catmull E E editor Computer Graphics SIGGRAPH 92 Proceedings volume 26 pages 265 272 Westin S H Arvo J R and Torrance K E 1992 Predicting reflectance functions from complex surfaces In Catmull E E editor Computer Graphics SIGGRAPH 92 Proceedings volume 26 pages 255 264 Whitted T and Cook R L 1985 A comprehensive shading model In SIGGRAPH 85 Image Rendering Tricks seminar notes Whitted T and Cook R L 19883 A comprehensive shading model In Joy K I Grant C W Max N L and Hatfield L editors Tutorial Computer Graphics Image Synthesis pages 232 243 Computer Society Press Williams L 1978 Casting curved shadows on curved surfaces In Computer Graphics SIGGRAPH 78 Proceedings volume 12 pages 270 274
73. ect naturally the backside of the object will be made dark because of the shadow call This makes it impossible to wrap light onto the backside of the object 106 CHAPTER 5 FUR IN STUART LITTLE rim lighting shadow solution wrap 120 we solve by slightly reducing the size of the geometry during the shadow map calculations siggraph 2000 sony picture imageworks advanced renderman fur figure 23 Fortunately RenderMan affords a few nice abilities to get around this problem One way to solve this problem is by reducing the size of your geometry for the shadow map calculations This way the object will continue to cast a shadow albeit a slightly smaller shadow and the areas that need to catch the wrapped lighting are not occluded by the shadow map As a practical matter shrinking an arbitrary object is not always an easy task It can be done by displacing the object inwards during the shadow map calculations but this can be very expensive In the case of our hair we wrote opacity controls into the hair shader that were used to drop the opacity of the hair to 0 0 a certain percentage of the way down their length It is important to note that the surface shader for an object is respected during shadow map calculation and if you set the opacity of a shading point to 0 0 it will not register in the shadow map The value which is considered transparent can be set with the following rib command Option limits zthresho
74. ell M E 1976 Texture and reflection in computer generated images Communications of the ACM 19 542 546 Cook R L Carpenter L and Catmull E 1987 The Reyes image rendering architecture In Stone M C editor Computer Graphics SIGGRAPH 57 Proceedings pages 95 102 Cook R L Porter T and Carpenter L 1984 Distributed ray tracing In Computer Graphics SIGGRAPH 64 Proceedings volume 18 pages 137 45 Monte Carlo distribution of rays to get gloss translucency penumbras depth of field motion blur Cook R L and Torrance K E 1981 A reflectance model for computer graphics volume 15 pages 307 316 Cook R L and Torrance K E 1982 A reflectance model for computer graphics ACM Transactions on Graphics 1 1 7 24 Glassner A e 1989 An Introduction to Ray Tracing Academic Press Goldman D B 1997 Fake fur rendering In Whitted T editor SIGGRAPH 97 Conference Proceedings Annual Conference Series pages 127 134 ACM SIGGRAPH Addison Wesley ISBN 0 89791 896 7 Gondek J S Meyer G W and Newman J G 1994 Wavelength dependent reflectance functions In Glassner A editor Proceedings of SIGGRAPH 94 Orlando Florida July 24 29 1994 Computer Graphics Proceedings Annual Conference Series pages 213 220 ACM SIGGRAPH ACM Press ISBN 0 89791 667 0 Greene 1986a Applications of world projections In Green M editor Proceedings of Graphics Interfac
75. en if we have assembled an envi ronment map from six rectangular faces because it is indexed by direction only for simplicity we can just as easily think of it as a spherical map 2 Instead of indexing the environment by direction only we define a ray using the position and mirror direction of the point then calculate the intersection of this ray with our aforemen tioned environment sphere 3 The intersection with the environment sphere is then used as the environment lookup direction Thus if a simple environment lookup is like ray tracing against a sphere of infinite radius then the scheme above is simply ray tracing against a sphere of a radius appropriate to the actual scene in the environment map As a subproblem we must be able to intersect an environment sphere with a ray A general treatment of ray object intersections can be found in Glassner 1989 but the ray sphere case is particularly simple If a ray is described by end point E and unit direction vector expressed in the coordinate system of a sphere centered at its local origin and with radius r then any point along the ray can be described as It for free parameter t This ray intersects the sphere anyplace that E It Because the length of a vector is the square root of its dot product with itself then r Expanding the x y z components for the dot product calculation yields E I t Ey E Lt r 0
76. en the time comes to dice them their own individual size on screen determines the tessellation rate As aresult it is often the case that the adjacent grids resulting from adjacent subprimitives will project to different sizes on the screen and as a result will tessellate at different rates during dicing Tessellated micropolygons are rectangular in the parametric space of the original primitive and all of the micropolygons in a single grid will be the same size in that parametric space but due to to the differing tessellation rates the micropolygons that make up the adjacent grids will have different sizes in parametric space One of the artifacts that results from this difference is that filtering calculations based on para metric size will change discontinuously across a grid boundary In Chapter 11 of Advanced Ren derMan various filtering techniques have been discussed that use parametric size as part of the calculation of filter width This type of size discontinuity will result in filtering discontinuities which can be visible in the final image For example in simple texturing the texture filter size defaults to the micropolygon size The resulting texture can have visible changes in sharpness over the surface of an otherwise smooth primitive If the result of a texture call is used as a displacement magnitude a displacement crack can result Section 1 5 2 Recent versions of PRMan have significantly reduced problems such as these by the
77. er is divided by the shading rate to determine the number of micropolygons that must make up the grid However the dicing tessellation is always done in such a manner as to create within a single grid micropolygons that are of identically sized rectangles in the parametric space of the primitive For this reason it is not possible for the resulting micropolygons in a grid to all be exactly the same size in raster space and therefore they will only approximate the shading rate requested of the object Some will be slightly larger others slightly smaller than desired Notice too that any adjacent sibling primitive will be independently estimated and therefore the number of micropolygons that are required for it may easily be different even if the sibling primitive is the same size in parametric space In fact this is by design as the Reyes algorithm fundamentally takes advantage of adaptive subdivision to create micropolygons that are approximately equal in size in raster space independent of their size in parametric space That way objects farther away from the camera will create a smaller number of equally sized micropolygons instead of creating a sea of inefficient nanopolygons Conversely objects very close to the camera will create a large number of micropolygons in order to cover the screen with sufficient shading samples to capture the visual detail that is required of the close up view For this reason it is very common for two adjacent
78. er s Manual for details Chapter 17 of Advanced RenderMan Creating CGI for Motion Pictures Ray Tracing in PRMan also discusses extensions to Shading Language that allow for ray cast shadow checks in light shaders Further Reading Early simple local illumination models for computer graphics used Lambertian reflectance Bui Tuong Phong Phong 1975 proposed using a specular illumination model L R and also noted that the appearance of faceted objects could be improved by interpolating the vertex normals Phong s reflection model is still commonly used in simple renderers particularly those implemented in hard ware Blinn reformulated this model as H with defined as the angle halfway between L This gives superior results but for some reason few renderers or graphics boards bother to use this improved version The fundamentals of environment mapping can be found in Greene 19862 and Greene 1986b The anisotropic specular illumination model that we use came from Ward 1992 The reader is directed to that work for more information on the derivation details and use of Greg Ward Larson s model Additional anisotropic local illumination models can be found in Kajiya 1985 and Poulin and Fournier 1990 Oren and Nayar s generalization of Lambert s law can be found in Oren and Nayar 1994 A similar reflection model for simulation of clouds dusty and rough surfaces can 32 CHAPTER 2 THE SECRET LIFE OF LI
79. eral optional arguments that can be specified as token value pairs e blur takes a float and controls the amount of blurring at the shadow edges as if to simulate the penumbra resulting from an area light source see Figure 2 11 A value of blur 0 makes perfectly sharp shadows larger values blur the edges It is strongly advised to add some blur as perfectly sharp shadows look unnatural and can also reveal the limited resolution of the shadow map e samples is a float specifying the number of samples used to test the shadow map Shadow maps are antialiased by supersampling so although having larger numbers of samples is more expensive they can reduce the graininess in the blurry regions We recommend a minimum of 16 samples and for blurry shadows it may be quite reasonable to use 64 samples or more e bias isa float that shifts the apparent depth of the objects from the light The shadow map is just an approximation and often not a very good one Because of numerical impre cisions in the rendering process and the limited resolution of the shadow map it is possible for the shadow map lookups to incorrectly indicate that a surface is in partial shadow even if the object is indeed the closest to the light The solution we use is to add a fudge factor to the lookup to make sure that objects are pushed out of their own shadows Selecting an appropriate bias value can be tricky Figure 2 12 shows what can go wrong if you select a valu
80. ere noted point Ps Position of the point on the surface that requested data from the light shader The surface normal of the light source geometry at P float u The 2D parametric coordinates of P on the light source geometry vector The partial derivatives 1 tangents of the light source dPdu vec geometry at P tor dPdv vector L The vector giving the direction of outgoing light from the source to the point being shaded Ps This variable can be set explic itly by the shader but is generally set implicitly by the illu minate or solar statements color The light color of the energy being emitted by the source Set ting this variable is the primary purpose of a light shader Positions normals and parameters on the light only make sense if the light is an area light source defined by a geometric primitive If the light is an ordinary point source rather than an area light these variables are undefined Note that PRMan does not support area lights Therefore in that renderer the light geometry variables are undefined and should not be trusted to contain any meaningful data An example light shader that could be used for a simple area light source is given in Listing 2 18 Note the similarity to the pointlight shader the main difference is that rather than using a from parameter as the light position we illuminate from the position P that the renderer chose for us by sampling the area light
81. erring until the next section metals that are polished to the point that they have visible reflections of surrounding objects we will concentrate for now on roughened metallic surfaces without coherent reflections Cook and Torrance Cook and Torrance 1981 Cook and Torrance 1982 realized that an im portant difference between plastics and metals is the effect of the base color of the material on the specular component Many materials including paint and colored plastic are composed of a transparent substrate with embedded pigment particles see Figure 2 1 The outer clear surface boundary both reflects light specularly without affecting its color and transmits light into the me dia that is permeated by pigment deposits Some of the transmitted light 1s scattered back diffusely after being filtered by the pigment color This white highlight contributes greatly to the perception of the material as plastic Figure 2 1 Cross section of a plastic like surface Homogeneous materials including metals lack a transparent outer layer that would specularly reflect light without attenuating its color Therefore in these materials all reflected light including specular is scaled by the material s base color This largely contributes to the metallic appearance We implement this look in MaterialRoughMetal Listing 2 3 2 1 3 Backlighting So far the materials we have simulated have all been assumed to be of substantial thickness In other w
82. es These particles originate from one or more emitters whose attributes determine for instance the rate and spread of the particles Once a particle hits a surface patch a circular clumping area is created on the patch at that u v location with clump percent clump rate and radius determined by a creation expression Similar to static clump size this radius is converted into a corresponding u radius and v radius Runtime expressions then define clump percent and rate to determine how quickly and how much the fur gets wet The above image shows a snapshot of animated clumping areas where each area is visualized as a cone the most recent cone is highlighted In this case one particle emitter is used and has been animated to follow the curve above the head 98 CHAPTER 5 FUR IN STUART LITTLE animated area clumping each animated area is made up of one or more actual clumps clumping is first done globally and then restricted at each frame to clumps which fall within the animated clumping area siggraph 2000 sony picture imageworks advanced renderman fur figure 13 Once the animated clumping areas are defined for each frame of an animation hairs which fall within these clumping areas are clumped Clumping areas are usually much bigger than a clump thus containing several individual clumps In our implementation we first apply static area clumping to the whole model as described above and then restrict the clumping
83. esults when fur is turned on It also accounts for shading differences between individual hairs and along the length of each hair For the wet fur look we change two aspects of our regular hair shading process First we increase the amount of specular on the fur Second we account for clumping in the shading model Geometrically as explained earlier we model fur in clumps to simulate what actually happens when fur gets wet In the shading model for each hair we calculate for each light what side of the clump the hair is on with respect to the light s position and then either darken or brighten the hair based on this Thus hairs on the side of a clump facing the light are brighter then hairs on a clump farther from the light 5 6 RIM LIGHTING 103 shading fur optimization pixel samples 7 7 sync filter high detail opr eee gt ae shading rate 7 low detail siggraph 2000 sony picture imageworks advanced renderman fur figure 19 We learned a couple of things through experimentation about optimizing our fur renders The level of detail requirements along the length of the fur from one end to another were very low Our fur did not consist of many curves or sharp highlights so we could use a very high shading rate and simply allow RenderMan to interpolate the shaded values between the samples This substantially reduced the number of times our shader code
84. etermine which blurry micropolygons they see 1 2 6 Shading Before Hiding Notice that this geometric pipeline has a feature that few other renderers share The shading calcula tions are done before the hidden surface algorithm is run In normal scanline renderers polygons are depth sorted the visible polygons are identified and those polygons are clipped to create spans that cover portions of a scanline The end points of those spans are shaded and then painted into pixels In ray tracing renderers pixel sample positions are turned into rays and the objects that are hit by and therefore visible from these rays are the only things that are shaded Radiosity renderers often resolve colors independently of a particular viewpoint but nonetheless compute object inter visibility as a prerequisite to energy transfer Hardware Z buffer algorithms do usually shade before hiding as Reyes does however they generally only compute true shading at polygon vertices not at the interiors of polygons One of the significant advantages of shading before hiding is that displacement shading is possi ble This is because the final locations of the vertices are not needed by the hider until after shading has completed and therefore the shader is free to move the points around without the hider ever knowing In other algorithms if the shader moved the vertices after the hider had resolved surfaces it would invalidate the hider s results 10 CHAPTER 1 HOW P
85. exact mechanism for handling these cases may be implementation dependent differing from renderer to renderer For lights that have a definite finitely close position there is another construct to use illuminate point from statements This form of the illuminate statement indicates that light is emitted from position from and is radiated in all directions As before illuminate implicitly setsL Ps from Listing 2 16 shows an example of a simple light shader that radiates light in all directions from a point from which defaults to the origin of light shader space Dividing the intensity by L L which is the square of the length of L results in what is known as 1 r falloff In other words the energy of light impinging on a surface falls off with the square of the distance between the surface and the light source A second form of illuminate also specifies a particular cone of light emission given by an axis and angle illuminate point from vector axis float angle statements An example use of this construct can be found in the standard spot light shader shown in List ing 2 17 The illuminate construct will prevent light from being emitted in directions outside the cone In addition the shader computes 1 r distance falloff applies a cosine falloff to directions away from the central axis and smoothly fades the light out at the edges of the cone Both forms of illuminate will check the corresponding illuminance s
86. face shader with clumping and specular model by Clint Hanson and Armin Bruderlin color tne diffuselgt color Cin Light Colour point Lin Light Position point Nin Surface Normal color Cout Cin vector LN NN float Atten normalize the stuff LN normalize vector Lin NN normalize vector Nin 5 11 SHADERS diffuse calculation Atten max 0 0 LN NN Cout Atten return Cout 113 define luminance c comp c 0 0 299 comp c 1 0 587 comp c 2 0 114 surface arr fur Hair Shading float Ka 0 0287 float Kd 0 77 float Ks 1 285 float roughnessl 0 008 float SPEC1 0 01 float roughness2 0 016 float SPEC2 0 003 float start spec 0 3 float end spec 0 95 float spec size fade 0 1 float illum width 180 Float var_fade_start 0 005 float var fade end 0 001 float clump dark strength 0 0 Hair Color color rootcolor color 9714 9714 9714 color tipeolor color 2 52195 2325 125 3 color specularcolor color l tipcelor 7 2 Color Static ambient color 04057 0 057 045057 3 Variables Passed from the rib uniform float hair col var 0 0 uniform float hair _ length 0 0 uniform normal surface_normal normal 1 varying vector clump vect vector 0 uniform float hair id 0 0 Watch Out Across Patches vector T normalize dPdv tangent along length of hair
87. g model for our fur was that it was not intuitive for a TD to light Most lighting TD s have become very good at lighting surfaces and the rules by which those surfaces behave So our lighting model was designed to be lit in a similar way that you would light a surface while retaining variation over the length of the fur which is essential to get a high quality look 102 CHAPTER 5 FUR IN STUART LITTLE shading fur obtaining a shading normal S o 2 on when the hair is exactly perpendicular to the surface use surface normal when the hair 1s exactly tangent to the surface use N hair otherwise blend between the two normals in a linear fashion siggraph 2000 sony picture imageworks advanced renderman fur figure 18 In order to obtain a shading normal at the current point on the hair we mix the surface normal vector at the base of the hair with the normal vector at the current point on the hair The amount with which each of these vectors contributes to the mix is based on the angle between the tangent vector at the current point on the hair and the surface normal vector at the base of the hair The smaller this angle the more the surface normal contributes to the shading normal We then use a Lambertian model to calculate the intensity of the hair at that point using this shading normal This has the benefit of allowing the user to light the underlying skin surface and then get very predictable r
88. g box calculation prior to shading to protect against vertices leaving their boxes The renderer grows the primitive bounding box by this value which means that the primitive is diced and shaded in a bucket earlier than it would normally be processed This often leads to micropolygons that are created long before their buckets need them which then hang around in bucket micropolygon lists wasting memory or primitives that are shaded before it is discovered that they are off screen Because of these computational and memory inefficiencies of the expanded bounds it is important that the displacement bounds be as tight as possible to limit the damage 1 4 5 Extreme Displacement Sometimes the renderer is stuck with large displacement bounds either because the object really does displace a large distance or because the camera is looking extremely closely at the object and the displacements appear very large on screen In extreme cases the renderer can lose huge amounts of memory to micropolygon lists that contain most of the geometric database In cases such as these a better option is available Notice that the problem with the displacement bound is that it is a worst case estimate over the primitive as a whole whereas the small portion of the primitive represented by a single small grid usually does not contain the worst case displacement and actually could get away with a much smaller tighter bound The solution to this dilemma is to actually run
89. geometry This shader illuminates only points on the outside of the light source geometry It would also be fine to simply use pointlight if you wanted the area light geometry to illuminate in all directions 48 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES 2 6 3 Shadows Notice that light shaders are strictly do it yourself projects If you want color you have to specify it If you want falloff you need to code it in the case of distantlight we have no distance based falloff for spot Light and pointlight we used 1 r falloff Similarly if you want the lights to be shadowed that also needs to be in the shader A number of shadowing algorithms have been developed over the years and their relative merits depend greatly on the overall rendering architectures So as not to make undue demands on the renderer the RenderMan standard provides for the lowest common denominator shadow maps Shadow maps are simple relatively cheap very flexible and can work with just about any rendering architecture The shadow map algorithm works in the following manner Before rendering the main image we will render separate images from the vantage points of the lights Rather than render RGB color images these light source views will record depth only hence the name depth map An example depth map can be seen in Figure 2 10 Once these depth maps have been created we can render the main image from the point of view of the camera In this p
90. h contains an RiCurve for each hair contoured to the surface appropriately and taking into account the character s animation The hair is then shaded with a standard RenderMan shader and the output image is generated 5 3 FUR INSTANTIATION THE INTERPOLATOR 9 5 3 Fur Instantiation the interpolator hair root on surface siggraph 2000 sony picture imageworks advanced renderman fur figure 3 The interpolator performs the function of growing the RiCurve primitives from all of the input data e instanced hair data The instancer specifies the position of each of the individual hairs for each patch in terms of the u v coordinate of the surface e animated nurbs patches These patches form the skin of the character and are used to position the hair in the correct 3d space and orientation according to the surface normal and the surface derivative information at the position of interest item hair data from Maya and optionally animated area clumping information This data is used to generate the 4 point curve taking into account the combing information and the other fur attributes specified in the data files 92 CHAPTER 5 FUR IN STUART LITTLE 5 4 Clumping clumping siggraph 2000 sony picture imageworks advanced renderman fur figure 4 Clumping of hairs can occur when the fur gets wet due to the surface tension of water The overall effect is that sets of neighboring hairs tend to join into a single poin
91. h the light shader illuminates Listing 2 19 shows a modification of the spotlight shader that uses a shadow map This light shader is still pretty simple but the entirety of Chapter 14 of Advanced RenderMan Creating CGI for Motion Pictures discusses more exotic features in light shaders Here are some tips to keep in mind when rendering shadow maps e Select an appropriate shadow map resolution It s not uncommon to use 2k x 2k or even higher resolution shadow maps for film work e View the scene through the shadow camera before making the map Make sure that the field of view is as small as possible so as to maximize the effective resolution of the objects in the shadow map Try to avoid your objects being small in the shadow map frame surrounded by lots of empty unused pixels e Remember that depth maps must be one unjittered depth sample per pixel In other words the RIB file for the shadow map rendering ought to contain the following options PixelSamples 1 1 PixelFilter box 1 1 Hider hidden jitter 0 Display shadow z zfile z ShadingRate 4 e In shadow maps only depth is needed not color To save time rendering shadow maps remove all Surface calls and increase the number given for ShadingRate for example 2 6 LIGHT SOURCES 51 as above If you have surface shaders that displace significantly and those bumps need to self shadow you may be forced to run the surface shaders anyway though you can still remov
92. he data from 78 CHAPTER 3 RENDERING ISSUES ON DINOSAUR Figure 3 11 Carnotaur with displacements and color C Disney Enterprises Inc some of the displacement maps They ended up so many polygons that the modeling program and machine which had 2G of RAM choked up Even at this point the model was far from being fully processed Let s say you did manage to get all of this super high detail into the model and you manage to get it through your pipeline and can actually render it What do you do when the model starts to alias This is geometric aliasing that we are talking about Under PRMan you have no recourse but to drop the shading rate and bump up the pixel samples Surely bumping up some already very high rendering times Whereas leaving this detail to displacement one can apply anti aliasing shading tricks to remove the required detail Using displacement maps can be thought of as a form of lazy TD s level of detail The multi res maps of PRMan pick just the right size detailed map for you automatically and quickly So all of the higher frequencies in the maps will be removed first As long as the model is going to be at least a couple of hundred pixels in size displacement mapping can provide a very convenient level of detail automatically Now is a good time to bring up an important point Everybody knows that PRMan s texture look up function is great at anti aliasing Whether the map is being used for color bump displacement e
93. he result of that integral This is something that must be kept in mind when coding traditional BRDFs in illuminance loops 44 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES 2 5 3 Glossy Specular Highlights The previous subsections listed Shading Language implementations of two local illumination mod els that are physically based simulations of the way light reflects off certain material types Many times however we want an effect that achieves a particular look without resorting to simulation The look itself may or may not match a real world material but the computations are completely ad hoc This subsection presents a local illumination model that achieves a useful look but that is decidedly not based on the actual behavior of light It works well for glossy materials such as finished ceramics glass or wet materials We note that specular highlights are a lot like mirror reflections of the light source The obser vation that they are big fuzzy circles on rough objects and small sharp circles on smooth objects is a combination of the blurry reflection model and the fact that the real light sources are not infinites imal points as we often use in compute graphics but extended area sources such as light bulbs the sun and so on For polished glossy surfaces we would like to get a clear distinct reflection of those bright area sources even if there isn t a noticeable mirror reflection of the rest of the environment and even if we
94. her more sophisticated local illumination func tions When we start writing our own local illumination functions we will use the convention of naming them with the prefix LocIllum and will typically name them after their inventors e g LocIllumCookTorrance But first let us see what effects we can get with just the built in specular anddiffuse functions 2 1 1 Matte Surfaces The typical combination of the built in diffuse and specular functions formalized in MaterialPlastic is great at making materials that look like manufactured plastic such as toys But many objects that you model will not be made from materials that feature a prominent specular highlight One could of course simply call MaterialPlastic passing Ks 0 However that seems wasteful to call the potentially expensive specular function only to mul tiply it by zero Our solution is to create a separate simpler MaterialMatte that only calculates the ambient and Lambertian via dif fuse contributions without a specular highlight as shown in Listing 2 2 Listing 2 2 MaterialMatte computes the color of the surface using a simple Lambertian BRDF color MaterialMatte normal Nf color basecolor float Ka Kd return basecolor Ka ambient Kd diffuse Nf 26 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES 2 1 2 Rough metallic surfaces Another class of surfaces not well modeled by the MaterialPlastic function is that of metals Def
95. house software Nor does it require a full blown 3D paint program A regular 2D program like Photoshop will work just fine Now the painter paints on top of the rendered camera projection keyframe images One must be careful so that the major features in the painting match up with the major features on the geometry This is why painting should happen on top of the rendered frame as it contains the 2D projected information of the scene If this isn t done then the paint won t project to the geometry correctly which will be very apparent once the camera and perspective start to change Figure 3 3 right is an example of a painted camera projection keyframe image 5 Rendering CG set 2nd pass Now with the keyframes painted the shot can be rendered using these maps with the camera projections applied to the CG set SIG2k_cameraProjection sl is a simple but working camera projection shader Actually the brains of this shader is found in the function f cameraProjection listed later Here are some of the high points of a camera projection shader Example surface shader illustrating camera projections technique for SIGGRAPH 2000 RenderMan course Simplistic version Inspired from shaders by Robert Beech and Mark Hammel The function that this version calls f cameraProjection only will work with 1 camera projection keyframe It lacks the sophistication of blending between the camera projection keyframe maps Nor does it mak
96. hts would be undesirable Therefore we would like to construct our diffuse function so that it can ignore light sources that have been tagged as being nondiffuse that is the source itself should only contribute specular highlights We can do this by exploiting the message passing mechanism of Shading Language wherein the surface shader may peek at the parameters of the light shader float lightsource string paramname output type result The lightsource function which may only be called from within an illuminance loop searches for a parameter of the light source named paramname If such a parameter is found and if its type matches that of the variable result then its value will be stored in the result and 1ightsource function will return 1 0 If no such parameter is found the variable result will be unchanged and the return value of Light source will be zero We can use this mechanism as follows Let us assume rather arbitrarily that any light that we wish to contribute only to specular highlights and that therefore should be ignored by the diffuse function will contain in its parameter list an output float parameter named _nondiffuse Similarly we can use an output float parameter named _nonspecular to indicate that a particular light should not contribute to specular highlights Then implementations of diffuse and specular as shown in Listing 2 10 would respond properly to these controls In both PRMan and BMRT th
97. ide is that now the painter only has 128 levels to paint in a given direction rather than the usual 256 There are still 256 levels from bottom to top of course but only 128 to cut in 128 to cut out There were lots of attribute maps to control various effects like dust wetness calluses scratches and wounds These characters were suppose to look like they lived in their environment So they had dirt in between their scales They were covered with dust They had worn toenails and calluses on their knees and elbows and other worn areas They had scratches and scars from old injuries 3 3 USING RENDERMAN AS A MODELER 81 For immediate injuries that happened during the film there was the wound layer The wounds were another displacement layer The purpose of this layer was to give the effect of torn and bitten flesh It could remove features from the other displacement maps Figure 3 14 Wounded Bruton after Carnotaur attack Disney Enterprises Inc Downside However there is a price to be paid for abusing displacements like we did on Dinosaur One penalty was with disk space The textures consumed Gigabytes of disk space The paint program that we used created a texture map for every patch As we painted using lots of different layers the number of texture maps was tremendous Take our example of a 400 patch model x 5 layers which brings us to 2000 maps for just one model Some models had up to 11 layers and some models had a lot more
98. image numbered two is the area seen by the second keyframe The white area numbered one represents the area seen by the first keyframe Disney Enterprises Inc As mentioned earlier it doesn t matter what is painted as long as the CG set has the key features 3 2 CAMERA PROJECTIONS 75 that represent what is to be painted There are other uses for this technique than just painting full digital sets Another way that this technique was used on Dinosaur was to help clean up some of the live action plates Frequently these plates had to be cleaned to remove various unwanted objects You know to remove things that didn t exist 60 million years ago like tracking balls and roads Traditionally these frames are cleaned up by hand one frame at a time Under the right conditions camera projections can make this task much easier Say for example that the camera move is just a slow truck out Create an exact model set from the shot footage if you don t already have one Use the plates from the shot footage to paint keyframe images Then the only plates that need to be cleaned are the ones used to paint those keyframe images Figure 3 8 left shows the original plate around frame 154 of a 280 frame shot Figure 3 8 right is has been cleaned by rendering the CG set using frame 280 as the CP keyframe obviouslly Figure 3 8 right hasn t been comped with the final background but you get the idea In other words only frame 280 was cleaned u
99. included in Shading Language void fresnel vector I normal N float eta output float Kr Kt output vector R T According to Snell s law and the Fresnel equations fresnel computes the reflection and transmission direction vectors R and T respectively as well as the scaling factors for reflected and transmitted light Kr and Kt The I parameter is the normalized incident ray N is the 2 2 REFLECTIONS 35 Listing 2 8 MaterialShinyMetal shinymetal shader Compute the color of the surface using a simple metal like BRDF To give a metallic appearance both diffuse and specular components are scaled by the color of the metal It is recommended that Kd 0 1 gt 0 5 and roughness gt 0 15 to give a believable metal lic appearance color MaterialShinyMetal normal Nf color basecolor float Ka Kd Ks roughness Kr blur DECLARE ENVPARAMS extern point P extern vector I vector IN normalize I V IN vector R reflect IN Nf return basecolor Ka ambient Kd diffuse Nf Ks specular Nf V roughness SampleEnvironment P R Kr blur ENVPARAMS surface shinymetal float Ka 1 Kd 0 1 Ks 1 roughness 0 2 float Kr 0 8 blur 0 DECLARE DEFAULTED ENVPARAMS normal Nf faceforward normalize N I Ci MaterialShinyMetal Nf Cs Ka Kd Ks roughness Kr blur ENVPARAMS Or Cr Oig normalized surface normal and eta 15 the ratio of
100. introduction of smooth derivatives The derivatives and the parametric size values that are available to shaders now describe a smoothly varying idealized parametric size for the micropolygon That is the values do not exactly match the true size of the micropolygon in parametric space but instead track closely the desired parametric size given the shading rate requested in some sense compensating for the compromises that needed to be made to accommodate binary dicing or other tessellation constraints at the time of dicing These smoothly varying parametric size estimates ameliorate texture filter size mismatches both in built in shading functions and in antialiased procedural textures Generally the fact that micropolygons are not exactly the size that they advertise is a small issue and where it is an issue it can be compensated for by minor modifications to the shader 1 5 5 Conclusion The Reyes rendering architecture is so general and flexible that new primitives new graphics algo rithms and new effects features have been added modularly to the existing structure almost conti nously for over 15 years The system has evolved from an experimental testbed with a seemingly unattainable dream of handling tens of thousands of primitives into a robust production system that regularly handles images a hundred times more complex than that The speed and memory enhance ments that have been added may appear to have been short term requirements as Mo
101. ionally calculates the changes to shading on the surface measured by the change in diffuse lighting from the Pref to P additional contrast and color controls are left as an exersize for the user by Rob Bredow and Scott Stokdyk color nno myditrruse color Cl vector L normal N normal Nn vector Ln Nn normalize Ln normalize L return Cl max Ln Nn 0 void tno projyectCurrentCamera point lt P output float X Ys point Pnde transform NDC X Y xcomp Pndc ycomp Pndc 116 CHAPTER 5 FUR IN STUART LITTLE surface srf deformation string texname Texture to project float debug 0 0 deformed lit image 1 texture deformed with no lighting 2 output lighting of the P object 3 output lighting of the Pref object float Kd 1 Surface Kd for lighting calculations varying point Pref point shader 0 0 0 float x color CiO Cil Ci2 normal N1 N2 float illum width 180 point Porig Pref ENG pre ectCurrentCamera Pref x if texname 10 texture texname x y else CiO 0 5 Calculate shading difference between P and Porig normalize calculatenormal P N1 faceforward normalize N normalize calculatenormal Porig N2 faceforward normalize N Cri 0003519 20 OF These lighting loops can be enhanced to calculate specular or reflection maps
102. ld 0 5 0 5 0 5 Lastly and most simply you can blur the shadow map when making the shadow call If your requirements are for soft shadows anyway this is probably the best solution of all It allows for the falloff of the light to reach around the object and if there is any occlusion caused by the shadow it will be occluded softly which will be less objectionable For Stuart Little we used a combination of shortening the hair for the shadow map calculations and blurring our shadows to be able to read the appropriate amount of wrap 5 7 OTHER TRICKS DEFORMATION SYSTEM 107 rim lighting shadow solution wrap 120 we solve by slightly reducing the size of the geometry during the shadow map calculations siggraph 2000 sony picture imageworks advanced renderman fur figure 2 3 Image 02000 by Columbia Pictures Industries Inc All Rights Reserved 5 7 Other Tricks Deformation System other tricks deformation system deformation system was designed to handle Stuart s interaction with soft surfaces siggraph 2000 sony picture imageworks advanced renderman fur figure 2 5 Image 2000 by Columbia Pictures Industries Inc All Rights Reserved Stuart Little called for a number of shots where Stuart needed to interact with a soft environment For these shots we developed a system in RenderMan to create realistic animation of the surfaces 108 CHAPTER 5 FUR IN STUART LITTLE which were generall
103. le short width width of map short height height of map char reserved 128 2 transformation matrices zfileHeader t This structure reads the header of a PRman zdepth file The first field magic says what kind of file itis A zdepth file will be Ox2f0867ab The next two fields width and height give the dimensions of the depth file The last field reserved gives the two 4x4 transformation matrices world to light screen space and world to light camera space 3 1 1 Pros and cons with this technique Cons Using this technique requires two shadow passes for each light that it is to be used on It assumes closed surface order That is front facing geometry is the closest min depth to the shadow light view and back facing geometry is the farthest max depth Newer versions of PRMan shadow depth options like midpoint help to alleviate the need for shadow bias tweaking Pros This technique may still be useful in situations where the regular built in shadow depth options don t give the desired results Like in cases where a midpoint isn t what is wanted but where a weighted average is desired An example of this maybe if the shadow maps have very large blurs applied to them Midpoint will not hold up so well but zero shadow bias will by varying the weight of the shadow maps 3 2 CAMERA PROJECTIONS 67 3 2 Camera Projections Camera Projection is basically a slide projector where the projector is the came
104. light is typically approximated by a low level constant non directional light contribution set by the user in a rather ad hoc manner When renderers try to accurately calculate this interreflected light in a principled manner it is known as global illumination or depending on the exact method used as radiosity path tracing Monte Carlo integration and others 2 1 BUILT IN LOCAL ILLUMINATION MODELS 25 Listing 2 1 MaterialPlastic computes a local illumination model approximating the appear ance of ordinary plastic Compute the color of the surface using a simple plastic like BRDF Typical values are Ka 1 Kd 0 8 Ks 0 5 roughness 0 1 color MaterialPlastic normal Nf color basecolor float Ka Kd Ks roughness extern vector I return basecolor Ka ambient Kd diffuse NE Ks specular Nf normalize I roughness For the remainder of this chapter functions that compute completed material appearances will be named with the prefix Material followed by a description of the material family Argu ments passed will generally include a base color surface normal viewing direction and a variety of weights and other knobs that select individual appearances from the family of materials The implementation of the Material functions will typically be to compute a weighted sum of several primitive local illumination functions Before long it will be necessary to move beyond am bient diffuse and specular to ot
105. ll since different features often had very different responses to illumination 4 3 Implementation 4 3 1 Data Collection One of the keys to realistic skin is the accurate reproduction of the natural detailed variation found in human skin This is particularly true when the skin is to be viewed from very close range as was needed in Toy Story 2 To collect a sample of this variation the facial skin of a suitable subject as placed on a scanner and a high res scan taken Various types of skin were scanned in order to acquire samples of pores and whiskers in addition to clear skin The scans were manipulated 4 4 RESULTS 85 in a paint application to create tiling textures covering a suitably large area so as to preclude any noticable repetition These texture maps formed the foundation for all the basic variation found within the skin itself 4 3 2 Manipulation While in effect all controls had to be orthogonal to achieve this result the internal functionality did not necessarily operate orthogonally In fact most of the controls for the various features were very carefully coupled together in such a way as to maintain a normalized range of effect One example of this was blotchiness In order to normalize the effect of this control and thereby decouple it from the skin color its effect on the color variation was modulated by the overall color since as the color became darker the range of color variation to achieve the same degree of
106. ly uses a slightly different proprietary formula for specular BMRT also uses a slightly nonstandard formulation of spec ular in order to more closely match PRMan So beware though the spec dictates Blinn Phong individual implementations can and do substitute other reflection models for specular color ambient Returns the contribution of so called ambient light which comes from no specific location but rather represents the low level of scattered light in a scene after bouncing from object to object Therefore those three lines we had at the end of our shaders calculate a weighted sum of ambient diffuse and specular lighting components Typically the diffuse and ambient light is filtered by the base color of the object but the specular contribution 15 not or is filtered by a separate specular color We usually assign Oi the opacity of the surface to simply be the default surface opacity Os Finally we scale the output color by the output opacity because RenderMan requires shaders to compute premultiplied opacity values When specularcolor is 1 i e white these calculations yield an appearance closely resem bling plastic Let us then formalize it with the Shading Language function MaterialPlastic in Listing 2 1 With this function in our library we could replace the usual ending lines of our shader with Ci MaterialPlastic Nf V Cs Ka Kd Ks roughness Oi Os Ci Oi In most renderers ambient
107. make themselves available for additional buckets Serial sections of the code particularly in database sorting redundant work due to primitives that overlap multiple buckets and network latency cut the overall multiprocessor efficiency to approximately 70 8096 on typical frames but nevertheless the algorithm often shows linear speedup through 8 to 10 processors Because these processes run independently of each other with no shared data structures they can run on multiple machines on the network and in fact on multiple processor architectures in a heterogeneous network with no additional loss of efficiency 1 4 RENDERING ATTRIBUTES AND OPTIONS 13 1 4 Rendering Attributes and Options With this background it is easy to understand certain previously obscure rendering attributes and options and why they affect memory and or rendering time and also why certain types of geometric models render faster or slower than others 1 4 4 Shading Rate In the RenderMan Interface the ShadingRate of an object refers to the frequency with which the primitive must be shaded actually measured by sample area in pixels in order to adequately capture its color variations For example a typical ShadingRate of 1 0 specifies one shading sample per pixel or roughly Phong shading style In the Reyes algorithm this constraint translates into micropolygon size During the dicing phase an estimate of the raster space size of the primitive is made and this numb
108. more information 2See http www pixar com products rendermandocs toolkit Toolkit AppNotes appnote 15 html for more information 3 1 ZERO SHADOW BIAS 61 Figure 3 1 Shadow bias 0 0 top left 0 05 top right 0 25 just about right bottom left 1 0 too much bottom right The zero shadow bias process produces results that are very similar to the newer PRMan mid point method However it is much more expensive then the midpoint method So generally one should consider using the built in midpoint over the Zero Shadow Bias technique But there are some situations for which this technique should be considered which will be touched upon later Here are the steps involved for the Zero Shadow Bias technique Render two zdepth passes 2 Create a third shadow map from a weighed average of the two maps 3 Render the color pass using this new processed shadow map The benefit of this method is that generally one doesn t have to worry about re rendering due to shadow bias issues One common way of attempting to decrease the numerical inaccuracies of shadow maps is to increase the map size So another benefit of the zero shadow bias method is that the shadow maps can be smaller Here is an example surface shader to be used on the shadow pass that illustrates the technique 62 CHAPTER 3 RENDERING ISSUES ON DINOSAUR technique for SIGGRAPH 2000 RenderMan course Tal Lancaster 02 05 00 talrmrepacbell net
109. mpHairCV i siggraph 2000 sony picture imageworks advanced renderman fur figure 10 The equations for calculating the clumped vertices are illustrated above The vertices of each clump hair except the root vertex are re oriented towards their corresponding clump center hair vertices from their dry combed position where the default value for numberOfCVs is 3 4 minus the root vertex and the index for the current control vertex 1 ranges from 1 3 It is noted that both rate and percent operations change the lengths of clumps hairs We have not however observed any visual artifacts even as these values are animated over the course of an animation static area clumping siggraph 2000 sony picture imageworks advanced renderman fur figure 11 Image 02000 by Columbia Pictures Industries Inc All Rights Reserved 5 4 CLUMPING 97 5 4 2 Animated Area Clumping animated area clumping animated clumping areas cones are defined through particles hitting surface patches each area has its own radius creation expression clump percent and clump rate runtime expressions siggraph 2000 sony picture imageworks advanced renderman fur figure 12 Animated area clumping is desirable if we want to simulate spouts of water or raindrops hitting the fur and making it increasingly wet To simulate this animated clumping areas are defined in an animation system through particles hitting surface patch
110. n area light The controls built into our lights had the ability to specify the end wrapping point in terms of degrees where a wrap of 90 degrees corresponded to the standard Lambertian shading model and higher values indicated more wrap This control was natural for the lighting TD s to work with and gave predictable results 5 6 RIM LIGHTING 105 For wrapped lights to calculate correctly the thrid argument to the illuminance statement must be set to at least the same degree as the wrap value for the highest light Otherwise lights will be culled out from the lighting calculations on the back of the surface and the light will not correctly wrap In our implemention the wrap parameter was set in each light and then passed into the shader using message passing where we used the customized diffuse calculation to take into account the light wrapping As an aside wrapping the light more actually makes the light contribute more energy to the scene For illustration purposes the value of the light intensity was halved to generate the previous figure rim lighting shadow problem wrap 120 problem is that everything behind the 90 mark falls into shadow thus eliminating our beautiful wrap siggraph 2000 sony picture imageworks advanced renderman fur figure 2 2 The first problem we encountered when putting these wrapped lights to practical use was with shadow maps When you have shadow maps for an obj
111. nd light triangles along the face of a displacement cliff Corners of micropolygons that are shaded for the top of the plateau hang down while corners of micropolygons that are shaded for the valley poke up interleaved like teeth of a gear Figure 1 7 Displacement stretching leads to micropolygons that are bent twisted or significantly larger than anticipated The visual artifacts of these problems can be somewhat ameliorated by using smooth shading interpolation which will blur the shading discontinuities caused by the varying normals However the geometric problems remain The primary solution is to lower the frequency content of the displacement shader so that adjacent micropolygons cannot have such wildly varying motion See Chapter 11 of Advanced RenderMan Creating CGI for Motion Pictures Shader Antialiasing for hints If this cannot be done the brute force approach is to reduce the shading rate to values such as 0 25 pixels or smaller so that even stretched micropolygons stay under 1 pixel in size However this will have significant performance impact because shading time is inversely proportional to shading 1 5 RENDERING ARTIFACTS 21 rate Fortunately shading rate is an attribute of individual primitives so the extra expense can be limited to the part of the model that requires it 1 5 4 Texture Filter Mismatches As primitives are split their subprimitives proceed through the rendering pipeline independently and wh
112. needed to be executed and sped up our renders However the value for the pixel samples needed to be quite high We had lots of very thin overlapping curve primitives and without adequate pixel sampling they would alias badly For most every fur render our pixel samples needed to be 7 in both directions We tried different filters as well and found visually that the sinc filter kept more sharpness and did not introduce any objectionable ringing 5 6 Rim Lighting To give Stuart his characteristic hero look it was necessary to have the capability of giving him a nice strong rim light that would wrap in a soft and natural way around the character 104 CHAPTER 5 FUR IN STUART LITTLE rim lighting standard diffuse lighting gt gt N L 0 0 0 5 N3 L 1 0 siggraph 2000 sony picture imageworks advanced renderman fur figure 2 0 The previous slide illustrates a standard Lambertian diffuse shading model rim lighting wrapped diffuse lighting wrap 120 gt asin N L wrap 0 0 asin N L wrap asin N L wrap 0 5 asin N4 L wrap 1 0 siggraph 2000 sony picture imageworks advanced renderman fur figure 2 1 The first step in generating nicely wrapped lighting is to give the light the ability to reach beyond the 90 degree point on the objects This has the effect of softening the effect of the light on the surface simulating a
113. ngle radians 30 float conedeltaangle radians 5 float beamdistribution 2 string shadowname float samples 16 float blur 0 01 float bias 0 01 uniform vector A normalize to from uniform float cosoutside cos coneangle uniform float cosinside cos coneangle conedeltaanale illuminate from A coneangle float cosangle L A length L float atten pow cosangle beamdistribution L L atten smoothstep cosoutside cosinside cosangle if shadowname atten 1 shadow shadowname Ps samples samples blur blur bias bias Cl atten intensity lightcolor 58 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES Chapter 3 Rendering Related Issues on the Production of Disney s Dinosaur Tal Lancaster Walt Disney Feature Animation tlanefa disney com Abstract This paper is broken into three clean sections 1 Zero shadow bias 2 Camera Projections 3 Modeling with RenderMan There were plenty of other topics to choose from Like encapsulating a particle and blobby system inside a surface shader creating a 50 foot high concussion wave on an already turbulent ocean using just a flat plane for geometry ripping flesh off of characters and having them heal up with the aid of displacement shaders cycling texture maps for herd characters CG grass CG fur and tons of other stuff But as I only have one hour to present rather than the eight tha
114. nnot be classified Reyes gives up and throws the primitive away and prints the Cannot split message If that primitive was supposed to be visible the lost section will leave a hole in the image Primitives that have large displacement bounds or are moving rapidly toward the camera will exacerbate the eye splitting problem because the parametric splitting process will do very little to reduce the bounding box Indeed for primitives for which the camera is inside the displacement 18 CHAPTER 1 HOW PHOTOREALISTIC RENDERMAN WORKS bound of part of the surface or primitives whose motion path actually goes through the camera splitting can never succeed In order to reduce the artifacts due to eye split culling the key is to give the renderer the largest possible safety zone Place the near clipping plane as far forward as is possible without otherwise affecting the image The near clipping plane can be placed surprisingly far forward for most shots made with cameras that have reasonable fields of view If you don t normally set the clipping plane set it to some small but reasonable value immediately the default value of 1e 10 is just about as bad as you could get The number of splitting iterations that will be permitted can be controlled with a rendering option and if you permit a few more you can sometimes help handling of large but otherwise simple primitives Beware however the displacement and motion cases because upping the limit
115. ns 2 else tO discrim b 2 solutions t0 gt eps 1 0 else if discrim 0 One solution on the edge tO b 2 solutions t0 gt eps 1 0 else Imaginary solution gt no intersection solutions 0 return solutions Environment A replacement for ordinary environment lookups this function ray traces against an environment sphere of known finite radius Inputs are envname filename of environment map envspace name of space environment map was made in envrad approximate supposed radius of environment sphere P R position and direction of traced ray blur amount of additional blur to add to environment map Outputs return value the color of incoming environment light alpha opacity of environment map lookup in the direction R Warning the environment call itself takes derivatives causing p PNE if called inside a loop or varying conditional Be cautious color Environment string envname envspace uniform float envrad point P vector R float blur output float alpha Transform to the space of the environment map point Psp transform envspace P vector Rsp normalize vtransform envspace R uniform float r2 envrad envrad Clamp the position to be inside the environment sphere if vector Psp vector Psp gt r2 Psp point envrad normalize vector float t0 t1 if raysphere Psp
116. o adjacent grids that are different sizes paramet rically and micropolygons that approximate the desired shading rate to regulate the amount of memory devoted to visible point lists Smaller buckets are more memory efficient because less memory is devoted to visible point lists Less obviously smaller buckets partition the geometric database into larger numbers of shorter sorted primitive lists with some consequential decrease in sorting time However this small effect is usually offset by the increase in certain per bucket overhead The maximum grid size option controls dicing by imposing an upper limit on the number of micropolygons that may occur in a single grid Larger grids are more efficient to shade because they maximize vector pipelining However larger grids also increase the amount of memory that can be devoted to shader global and local variable registers which are allocated in rectangular arrays the size of a grid More interestingly however the maximum grid size creates a loose upper bound on the pixel area that a grid may cover on screen a grid is unlikely to be much larger than the product of the maximum grid size and the shading rate of the grid This is important in relation to the bucket size because grids that are larger than a bucket will tend to create large numbers of micropolygons that fall outside of the current bucket and that must be stored in lists for future buckets Micropolygons that linger in such lists can use
117. o support this type of motion is that bounding box computations must include the entire motion path of the object The hidden surface algorithm modifications necessary to handle motion blur are implemented using the stochastic sampling algorithm first described by Cook et al in 1984 The hidden surface algorithm s point sample locations are each augmented with a unique sample time As each micropolygon 15 sampled it is translated along its motion path to the position required for each sample s time PRMan only shades moving primitives at the start of their motion and only supports linear mo tion of primitives between their start and stop positions This means that shaded micropolygons do not change color over time and they leave constant colored streaks across the image This is incorrect particularly with respect to lighting as micropolygons will drag shadows or specular highlights around with them In practice this artifact is rarely noticed due to the fact that such objects are so blurry anyway Depth of field is handled in a very similar way The specified lens parameters and the known focusing equations make it easy to determine how large the circle of confusion is for each primitive in the scene based on its depth That value increases the bounding box for the primitive and for its micropolygons Stochastically chosen lens positions are determined for each point sample and the samples are appropriately jittered on the lens in order to d
118. o the paper the implementation below appears to be missing a factor of 1 pi and to have an extra L N term This is not an error It is because the paper s formula is for the BRDF which is only part of the kernel of the light integral whereas shaders must compute the result of the integral Inputs N unit surface normal V unit viewing direction from P toward the camera xdir a unit tangent of the surface defining the reference direction for the anisotropy xroughness the apparent roughness of the surface in xdir yroughness the roughness for the direction of the surface tangent perpendicular to xdir color LocIllumWardAnisotropic normal N vector V vector xdir float xroughness yroughness float sqr float x return x x float cos theta_r clamp N V 0 0001 1 vector X xdir xroughness vector Y N xdir yroughness color C 0 extern point P illuminance P N PI 2 Must declare extern L amp Cl because we re in a function extern vector 1 extern color Cl float nonspec 0 lightsource __nonspecular nonspec if nonspec lt 1 vector LN normalize L float cos theta_i LN N if eos theta 3 930 4 vector H normalize V LN float rho exp 2 sqr X H sqr Y H 1 H N sqrt cos theta_i cos theta_r 1 1 cos theta 1 rho return 4 xroughness yroughness 2 6 LIGHT SOURCES 55 Listing
119. o to the extra effort of computing the wavelength and angle dependency of metals Important note In MaterialShinyPlastic we set fkt 1 fkr overriding the value returned by fresnel The reasons for this are extremely subtle and well beyond the scope of this book Suffice it to say that the value that fresnel is supposed to return for Kt assumes a more rigorous simulation of light propagation than either PRMan or BMRT provides In light of these inaccuracies the intuitive notion that Kt and Kr ought to sum to 1 is about as good an approximation as one might hope for We will therefore continue to use this oversimplification Listing 2 9 shinyplastic shader is for highly polished plastic and uses fresnel so that reflections are stronger at grazing angles color MaterialShinyPlastic normal Nf color basecolor float Ka Kd Ks roughness Kr blur ior DECLARE ENVPARAMS extern vector I extern point P vector IN normalize I V IN float fkr fkt vector R T fresnel IN Nf 1 ior fkr fkt R T fkt 1 fkr return fkt basecolor Ka ambient Kd diffuse NE Ks fkr specular Nf V roughness SampleEnvironment P R fkr Kr blur ENVPARAMS surface shinyplastic float Ka 1 Kd 0 5 Ks 5 roughness 0 1 float Kt 1 blur 0 ior 1 5 DECLARE DEFAULTED ENVPARAMS normal Nf faceforward normalize N I Ci MaterialShinyPlastic Nf Cs Ka Kd Ks rough ness Kr blur ior
120. oat zval mapl zval map2 depth val from maps float zval map3 depth val for new map float zvals mapl zvals map2 depth vals from maps float cur zval mapl cur zval map2 ptrs to current map vals float min mapi min map2 min depth val for maps float max mapl max map2 max depth val for maps if arge e 4 1 fprintf stderr Usage s zdepthMapl zdepthMap2 new zdepthMapWn argv 0 exit 1 fin mapi open argv 1 RDONLY 3 1 ZERO SHADOW BIAS 65 open argv 2 O RDONLY fin map2 open argv 3 O WRONLY O CREAT O TRUNC 0666 fout map get depth map header read fin mapi amp zheadl sizeof zfileHeader t read fin map2 amp zhead2 sizeof zfileHeader t if zheadl magic 0x2f0867ab amp amp zhead2 magic Ox2f0867ab amp amp zheadl width zhead2 width amp amp zheadl height zhead2 height both maps are zdepth maps store size of maps int mapsize sizeof float zheadl width zheadl height zn du dump out header information to new map write fout map amp zheadl sizeof zfileHeader t allocate space for the two input maps zvals mapl float malloc mapsize zvals map2 float malloc mapsize read in rest of map data read fin mapl zvals mapl mapsize read fin map2 zvals map2 mapsize leo 1e6 min _mapl max mapi min map2 max map2 zvals mapl zvals map2
121. ojection keyframes If the camera animation changes so that the camera at the picked frames is no longer in the same position as when the associated keyframes were painted then this work will be lost So it is a good idea to not start working with camera projections until the camera animation has been finalized 3 Renderer Ok you need to use a renderer that is capable of dealing with the camera projection transfor mations RenderMan will do Once the camera projection keyframe is picked this key or keys 15 rendered out creating keyframe images It really isn t important what surface shaders are used Just something that will give a sense of the contours in the scene So even a plastic or matte surface shader could be used These rendered keyframe images are what will be painted on Figure 3 3 left is such an example It 1s the CG set rendered using a matte surface shader to be painted on The rendering of the keyframes 15 the first pass Later a second rendering pass of all of the scene s frames will be rendered using the painted texture maps with applied camera projec tions 4 Paint program and maybe a painter Based on a WDFA internal presentation by Mark Hammel 68 CHAPTER 3 RENDERING ISSUES ON DINOSAUR Figure 3 3 Left Firey shore CG set Used for complete digital background Right Final CG background painted on groundplane model C Disney Enterprises Inc Camera projections don t require any special in
122. or 7 AL s cus uv uvintensity L s L Don t worry too much that you haven t seen light shaders yet Before the end of this chapter this example will be crystal clear 2 5 Custom Material Descriptions We have seen that the implementation of diffuse is that of a Lambertian shading model that approximates a rough surface that scatters light equally in all directions Similarly specular implements a Blinn Phong scattering function according to the RI spec As we ve seen different weighted combinations of these two functions can yield materials that look like a variety of plastics and metals Now that we understand how they operate we may use the illuminance construct ourselves to create custom primitive local illumination models Past ACM SIGGRAPH proceedings are a treasure trove of ideas for more complex and realistic local illumination models In this section we will examine three local illumination models two physically based and one ad hoc and construct shader functions that implement them 2 5 1 Rough Surfaces Lambert s law models a perfectly smooth surface that reflects light equally in all directions This is very much an oversimplification of the behavior of materials In the proceedings of SIGGRAPH 94 Michael Oren and Shree K Nayar described a surface scattering model for rough surfaces Their model and others including Beckman Blinn and Cook Torrance considers rough surfaces to have microscopic grooves
123. ords lights on the same side of the surface as the viewer reflect off the surface and are seen by the camera but lights behind the surface from the point of view of the camera do not scatter light around the corner so that it contributes to the camera s view It is not by accident that such backlighting is excluded from contributing Note that the dif fuse function takes the surface normal as an argument The details of its working will be revealed in Section 2 3 let us simply note here that this normal parameter is used to exclude the contribution of light sources that do not lie on the same side of the surface as the viewer But thinner materials paper lampshades blades of grass thin sheets of plastic do have ap pearances that are affected by lights behind the object These objects are translucent so lights lights 2 1 BUILT IN LOCAL ILLUMINATION MODELS 27 Listing 2 3 MaterialRoughMetal calculates the color of the surface using a simple metal like BRDF Compute the color of the surface using a simple metal like BRDF To give a metallic appearance both diffuse and specular components are scaled by the color of the metal It is recommended that Kd lt 0 1 Ks gt 0 5 and roughness gt 0 15 to give a believable metal lic appearance color MaterialRoughMetal normal Nf color basecolor float Ka Kd Ks roughness extern vector I return basecolor Ka ambient Kd diffuse Nf Ks spec
124. ore s law allows us to run the program on computers that are faster and have more memory without any additional programming However this is shortsighted for our appetite for complexity has scaled too as fast or faster than Moore s Law allows Undoubtedly in 15 more years when computers with a terabyte of main memory are common and optical processors chew up 1 billion primitives without flinching we will still be using the Reyes architecture to compute our holofilms 22 CHAPTER 1 HOW PHOTOREALISTIC RENDERMAN WORKS Chapter 2 The Secret Life of Lights and Surfaces Larry Gritz Pixar Animation Studios lg pixar com Abstract In this section we will explore how surfaces respond to light as well as how light sources themselves operate which are critical aspects of the overall appearance of materials In doing so we will build up a variety of library routines that implement different material appearances and light properties 2 1 Built in local illumination models You ve probably seen countless shaders that ended with the following lines of code or something remarkably close Ci Ct Ka ambient Kd diffuse Nf specularcolor Ks specular Nf normalize I roughness Oi Os Ci Oi Three functions are used here that have not been previously described color diffuse vector N Calculates light widely and uniformly scattered as it bounces from a light source off of the surface Diffuse reflectivity is generall
125. ost straightforward example of the use of illuminance loops is the implemen tation of the dif fuse function 38 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES color diffuse normal Nn extern point P color 0 illuminance Nn PI 2 C Cl Nn normalize L return C Briefly this function is looping over all light sources Each light s contribution is computed using a Lambertian shading model that is the light reflected diffusely is the light arriving from the source multiplied by the dot product of the normal with the direction of the light The contributions of all lights are summed and that sum is returned as the result of diffuse 2 4 Identifying Lights with Special Properties In reality surfaces simply scatter light The scattering function or BRDF which stands for bidi rectional reflection distribution function may be quite complex Dividing the scattering function into diffuse versus specular or considering the BRDF to be a weighted sum of the two is simply a convenient oversimplification Many other simplifications and abstractions are possible In the physical world light scattering is a property of the surface material not a property of the light itself or of its source Nonetheless in computer graphics it is often convenient and desirable to place light sources whose purpose is solely to provide a highlight or alternatively to provide a soft fill light where specular highlig
126. out Here is the code for the f printMatrix function void f prazntMatrax t current object transformation matrix transform object P same as transform Mobj P ay uniform matrix Mobj matrix object 1 current gt NDC transformation matrix transform NDC P same as transform MNDC P Pr uniform matrix MNDC matrix NDC 1 object gt NDC transformation matrix Pobj transform object P Pndc transform object NDC Pobj same as Pndc transform Mobject2NDC Pobj ia uniform matrix Mobject2NDC 1 Mobj MNDC printf Camera matrix m n Mobject2NDO The goal of this f printMatrix is to dump out the object to NDC space matrix for a given keyframe The reason for this is that a camera projection is simply an NDC transformation Fig ure 3 4 shows a Point P being transformed into NDC space Pndc The variable Mobject2NDC in the f printMatrix contains this transformation This is what is dumped out by the function and is used later to perform the camera projection Note the grey area in Figure 3 4 is what this keyframe maps to If the camera moves in any way such that it sees outside this view then this new view will contain unmapped areas In such a situation multiple keyframes will be needed The matrix that is dumped out is then stored in the matrixdata h include file Note that the shader will need to be recompiled whenever this data file changes Otherwise the sha
127. over the redundant calculations that would be necessary in a depth first implementation However life is never all gravy When a conditional or loop instruction is reached the results may require that not all points on the grid enter the protected block Because of this the SIMD controller has run flags which identify which grid vertices are active and which are inactive for the current instruction For any instruction where the run flag vector has at least one bit on the instruction is executed but only on those grid vertices that require it Other operators such as else break and continue manipulate the run flags to ensure that the SIMD execution accurately simulates the depth first execution Another advantage of the SIMD execution model is that neighborhood information is available for most grid vertices excepting those on the grid edges which means that differentials can be computed These differentials the difference between values at adjacent grid vertices substitute for derivatives in all of the Shading Language operators that require derivative information Those operators known generically as area operators include Du calculatenormal texture and their related functions Notice that grid vertices on edges actually on two edges not all four have no neighbors so their differential information is estimated PRMan version 3 8 and lower estimated 8 CHAPTER 1 HOW PHOTOREALISTIC RENDERMAN WORKS this poorly which led to b
128. p by hand Camera projection took care of the rest Figure 3 8 Left Original plate around middle of camera move Right Cleaned up frame via camera projection Disney Enterprises Inc Camera projections don t have to be used exclusively for painting color maps This process could also be used to create maps for displacement shaders Also transparency could be used to help muddle up the geometry too The pluses of using camera projections CP is less work then 3D paint because only one or small number paint job is needed Also it will match up exactly to Layout because it is using the actual camera from the scene It doesn t have the same hang ups as 3D painting such as perspective distortions of geometry With camera projection the camera view is perspective while painting under 3D 15 usually orthographic In orthographic it can be hard to determine how features should distort in different views This is especially true when the sets are very large and the perspective distortion is huge Don t forget camera projections usually require less work than traditionally cleaning up live action plates In conclusion CP is a very simple technique that makes certain tasks much easier But it is very limited in its application It doesn t do well when the perspective shift is extreme There are just times where painting the scene with 3D painting will be needed 4Layout is the department in the production pipeline that puts together the
129. passes as special texture maps when rendering the final view from the main camera 2 2 1 Environment Maps Environment maps take images of six axis aligned directions from a particular point like the six faces of a cube and allow you to look up texture on those maps indexed by a direction vector thus simulating reflection An example of an unwrapped environment map is shown in Figure 2 3 Accessing an environment map from inside your shader is straightforward with the built in en vironment function type environment string filename vector The environment function is quite analogous to the cexture call in several ways 2 2 REFLECTIONS 29 CUN he e F 4U F T am gt gt AIR i TAL P tn Y y ae 5 r AA E A Figure 2 3 An example environment map e The return type can be explicitly cast to either float or color If you do not explicitly cast the results the compiler will try to infer the return type which could lead to ambiguous situations e A float in brackets immediately following the filename indicates a starting channel default 15 to start with channel 0 e For environment maps the texture coordinates consist of a direction vector As with tex ture derivatives of this vector will be used for automatic filtering of the environment map lookup Optionally four vectors may be given to bound the angle range and in that case no derivatives will be taken
130. procedure to hairs within the animated clumping areas at each frame i e to clumps which fall within any clumping area so that final hairs of clumps outside clumping areas are normally produced as dry hairs To determine whether a clump falls within a clumping area the clump method checks at each frame whether the u v distance between the clump center hair of the clump and the center of the clumping area is within the u v radii of the clumping area For clumps in overlapping clumping areas the values for clump percent and rate are added to generate wetter fur The above image shows the resulting clumped hairs in the areas defined in the previous image 5 4 CLUMPING 99 animated area clumping Problem straddling clumping areas across patches siggraph 2000 sony picture imageworks advanced renderman fur figure 1 4 One problem with animated clumping areas has not been addressed yet they can straddle surface patch boundaries as shown in this illustration which shows a top view of a clumping area grey colored cone the center of this clumping area black dot is on patch P1 but the area straddles onto patch P2 and P3 across a T junction seams are drawn as bold lines Since our clumping areas are associated with the surface which contains the center of the clumping area 1 e the position where the particle hit portions of a clumping area straddling neighboring patches are ignored This would lead to discontinuities in
131. ra One could just project images on a wall plane But that is pretty flat and boring Things get a little more interesting when you apply more complex geometry The essence of camera projection is first projecting 3D geometry points to 2D texture points Then apply a texture to the 3D scene independent of the camera motion So camera projections provide a quick way to apply a texture map to a 3D scene They can be thought of as a form of point referencing Pref These projections are also a way to get changes in perspective in 2D paintings Various studios have used camera projections for some time now On Dinosaur camera pro jections were used in about 60 shots The following is a simplified description of how they were used What is needed for camera projection 1 Geometry A 3D set is needed There needs to be something for the textures to project onto The 3D set needs to contain the major features that are going to be painted So if there is going to be a hill in the foreground painted there had better be a model of the hill in the 3D set 2 Camera A frame or frames needs to be picked which best represents a particular scene Frequently this frame is one that has as much of the scene as visible as possible Whatever frame s is picked the camera transformations are noted These frames are known as the camera projec tion keyframes This technique is based entirely on mapping textures to a particular camera view the camera pr
132. rack Unfortunately only careful coding of displacement shaders can eliminate this type of cracking Notice that patch cracks cannot happen on the interiors of grids Because grid micropolygons explicitly share vertices it is not possible for such neighbor micropolygons to have cracks between them For this reason patch cracks will only occur along boundaries of grids and therefore along parametric edges In PRMan versions prior to 3 9 patch cracks could occur on any grid edge including those that resulted from splitting a patch into subpatches Later versions of PRMan have a crack avoidance algorithm that glues all such subpatches together Therefore modern versions of PRMan will only exhibit patch cracks along boundaries of original primitives not along arbitrary grid boundary 1 5 3 Displacement Stretching Another problem that displacement shaders might create is stretching of micropolygons This is caused when displacement shaders move the vertices of a micropolygon apart so that it no longer obeys the constraint that it is approximately the size specified by the shading rate In the process of dicing the renderer estimates the size of the primitive on screen and makes a grid that has micropolygons that approximately match the shading rate Shading then occurs on the vertices of the grid the corners of the micropolygons Displacement shaders are permitted to move grid vertices but there is no constraint on where they are moved If two a
133. rcials and short films Prior to joining Pixar he taught computer graphics and worked in graphics and thermal simulation Rob Bredow served as senior technical director at Sony Pictures Imageworks where he just com pleted work on Stuart Little Prior to his stint at Sony Imageworks Rob was director of R amp D at VisionArt Design and Animation Inc where he worked on such projects as Star Trek DS9 In dependence Day and Godzilla Currently he is working as Visual Effects Supervisor on the film Megiddo due to be released early 2001 Schedule Welcome and Introduction Larry Gritz How PhotoRealistic RenderMan Works Tony Apodaca Break The Secret Life of Lights and Surfaces Larry Gritz Lunch Rendering Related Issues on the Production of Disney s Dinosaur Tal Lancaster The Artistry of Shading Human Skin Mitch Prater Break The Artistry of Shading Human Skin continued Fur in Stuart Little Rob Bredow Roundtable Discussion Q amp A All panelists Finished go to a party 8 30 AM 8 45 AM 10 00 AM 10 15 AM 12 00 PM 1 30 PM 2 30 PM 3 00 PM 3 15 PM 3 45 PM 4 45 PM 5 00 PM vi Contents Preface 1 1 How PhotoRealistic RenderMan Works 3 D ub oo Ga ea Bow AROMAS Be Ree Re Ge 3 2 Basic Geometrc Pipeline odo RE bea 4 1 3 Enhanced Geometric 1 10 1 4 Rendering Attributes and Options 13
134. re are three extra rows of horns that are not in the model and don t forget about all those scales Figure 3 11 is with the color layer added Ok now check out Figure 3 12 This is how close we had to get and they still hold up Look at the detail on each of those scales Attempting to reach this level of detail with traditional modeling techniques would have been very daunting Our models got heavy very quickly Four hundred NURBS patches on a model was not uncommon Some of them even had more than double that amount Even so the models still were far from the required detail Figure 3 13 shows the basic Aladar model The lines on the model attempt to give a sense of the 3 3 USING RENDERMAN AS A MODELER 77 Figure 3 10 Greyscale Carnotaur with displacements Disney Enterprises Inc patch boundaries to give a feel of the size and number of patches on the model Even with those 380 patches the look is still lacking considering the camera may and does get to within inches from his face For us the solution was to have our models contain the grosser features and then use PRMan displacement shaders to add the finer features that would be harder to do in modeling It made sense to do it this way We had to show several versions of these finer grain details for each model Through the use of displacement maps these finer grain details can be changed logarithmically faster than they could be done in a modeling program A texture painter jus
135. refractive index of the medium containing I to that on the opposite side of the surface We can use fresnel to attenuate the relative contributions of environmental and diffuse re flectances This is demonstrated in the MaterialShinyPlastic and shinyplastic both shown in Listing 2 9 The index of refraction of air is very close to 1 0 water is 1 33 ordinary window glass is approx imately 1 5 and most plastics are close to this value In computer graphics we typically assume that 1 5 is a reasonable refractive index for a very wide range of plastics and glasses Therefore Figure 2 5 The ratio of reflected to transmitted light at a material boundary changes with the angle of the incident light ray 36 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES a reasonable value for eta parameter passed to fresnel is 1 1 5 Nearly any optics text book will list the indices of refraction for a variety of real materials should you wish to be more physically accurate Of course for real materials the refractive index is wavelength dependent We tend to ignore this effect since we have such an ad hoc approach to spectral sampling and color spaces anyway Although metals also have wavelength dependent indices of refraction fresnel effects and angle dependent spectral reflectivities they are much less visually noticeable in most metals so we usually just pretend that shiny metals reflect uniformly You are of course perfectly free to g
136. result of something every artist knows fundamentally that we will see what they want us to see as long as they are able to capture the fundamental essence of how we percieve something and recreate it So my approach was to study skin in order to develop a sense of what makes skin be percieved as skin and to reproduce those qualities in a way that gave us control over them as well as incorpo rating physical characteristics when doing so would enhance the perception 83 84 CHAPTER 4 THE ARTISTRY OF SHADING HUMAN SKIN 4 1 2 Reference Material Unlike a scientific approach where I would have read lots of research papers disected skin to un derstand how it s put together and taken lots of reflectivity measurements of its various constituents I took the artistic approach and just looked at lots of pictures that conveyed various aspects of skin or of skin as a whole very well From this I determined what qualities we wanted and how they would all fit together into a single conceptual structure 4 2 Conceptual Structure 4 2 1 Controllable Features The skin had to be applicable to the full range of race sex and age The properties of the skin itself had to be manipulatable to accomodate this variety as well as allow the addition of optional features such as makeup whiskers and sweat On a practical level the skin had to also change between different types of skin allowing a smooth transition into the mouth and ears All of this v
137. rrors there is an even easier and more efficient method for producing reflections which also solves the problem of environment maps being inaccurate for flat objects For the example of a flat mirror we can observe that the image in the reflection would be identical to the image that you would get if you put another copy of the room on the other side of the mirror reflected about the plane of the mirror This geometric principle is illustrated in Figure 2 4 Once we create this reflection map we can turn it into a texture and index it from our shader Because the pixels in the reflection map correspond exactly to the reflected image in the same pixels of the main image we access the texture map by the texture s pixel coordinates not the s coordi nates of the mirror We can do this by projecting P into NDC space This is done in the Re 1Map function in Listing 2 6 2 2 4 General Reflections and Shiny Surfaces We would prefer to write our shaders so that we may use either reflection or environment maps Therefore we can combine both into a single routine SampleEnvironment given in List 2 2 REFLECTIONS 33 Figure 2 4 Creating a mirrored scene for generating a reflection map On the top left a camera views a scene that includes a mirror Below is the image it produces On the top right the camera instead views the mirror scene Notice that the image it produces below contains the required reflection ing 2 7 A few
138. s chapter explains how PhotoRealistic RenderMan works on the inside and how it sometimes fails to work as well 1 1 History Pixar s PhotoRealistic RenderMan renderer is an implementation of a scanline rendering algorithm known as the Reyes architecture Reyes was developed in the mid 1980s by the Computer Graphics Research Group at Lucasfilm now Pixar with the specific goal of creating a rendering algorithm that would be applicable to creating special effects for motion pictures The algorithm was first described by Cook et al in their 1987 Siggraph paper The Reyes Image Rendering Architecture This section is reprinted from Advanced RenderMan Creating CGI for Motion Pictures by Anthony A Apodaca and Larry Gritz ISBN 1 55860 618 1 published by and copyright 2000 by Morgan Kaufmann Publishers all rights reserved and is used with permission Hey Why don t you go get the whole book instead of reading a couple out of context excerpts 4 CHAPTER 1 HOW PHOTOREALISTIC RENDERMAN WORKS They developed this novel rendering algorithm because they felt that the other algorithms generally in use at the time polygon Z buffer algorithms polygon scanline algorithms and ray tracers had various flaws and constraints that really limited their use in this venue Images used in motion picture special effects need to be photorealistic that is appear of such high quality that the audience would believe they were filmed with a real camera
139. s which create a power of two number of micropolygons along each edge Although these micropolygons are smaller than are required by shading rate alone binary dicing ensures that adjacent grids have tessellations that are powers of two 1 5 RENDERING ARTIFACTS 19 Pasted vertices Cracks lt i Figure 1 6 Tessellation differences result in patch cracks multiples of each other generally a single factor of two Thus alternating vertices will coincide and the extra vertices on one side are easily found and pasted to the other surface Another way that patch cracks can happen is when the displacement shader that operates on grid vertices gets different results on one grid from another If a common vertex displaces differently on two grids the displacement literally rips the surface apart leaving a crack between One common reason for such differing results is displacement mapping using texture filter sizes that are mis matched see Section 1 5 4 The texture call then returns a slightly different value on the two grids and one grid displaces to a different height than the other Another common reason is displacement occurring along slightly different vectors For example the results of calculatenormal are almost guaranteed to be different on the left edge of one grid and on the right edge of the adjacent grid If displacement occurs along these differing vectors the vertices will obviously go to different places opening a c
140. shadowname projectionmatrix matNP transform the ground plane point into texture coordinates 9 needed to look up the point in the shadow map point screenP transform matNP P float ss xcomp screenP 1 0 5 float tt 1 ycomp screenP 0 5 Lf ese 0 LI oe sq ee S a point being shaded is outside the region of the depth map Ci color 0 else get the distance from the shadow camera to the closest object as recorded in the shadow map float mapdist texture shadowname ss tt samples samples transform the point on the ground plane into the shadow camera space in order to get the distance from the shadow camera to the ground plane point cameraP transform matNl P the difference between the two distances is used to calculate the contact shadow effect float distance mapdist zcomp cameraP distance smoothstep 0 1 distance influence distance pow distance gamma convert into a color white shadow Qn e Cl 0 drstance 118 CHAPTER 5 FUR IN STUART LITTLE Bibliography Apodaca A A and Gritz L 1999 Advanced RenderMan Creating CGI for Motion Pictures Morgan Kaufmann Blinn J F 1977 Models of light reflection for computer synthesized pictures volume 11 pages 192 198 Blinn J F 1982 Light reflection functions for simulation of clouds and dusty surfaces volume 16 pages 21 29 Blinn J F and New
141. sley ISBN 0 89791 896 7 Nakamae E Kaneda K Okamoto T and Nishita T 1990 A lighting model aiming at drive simulators In Baskett F editor Computer Graphics SIGGRAPH 90 Proceedings volume 24 pages 395 404 Oren M and Nayar S K 1994 Generalization of lambert s reflectance model In Glassner A editor Pro ceedings of SIGGRAPH 94 Orlando Florida July 24 29 1994 Computer Graphics Proceedings Annual Conference Series pages 239 246 ACM SIGGRAPH ACM Press ISBN 0 89791 667 0 Phong B T 1975 Illumination for computer generated pictures Communications of the ACM 18 6 311 317 Pixar 1989 The RenderMan Interface Version 3 1 Poulin P and Fournier A 1990 A model for anisotropic reflection In Baskett F editor Computer Graphics SIGGRAPH 90 Proceedings volume 24 pages 273 282 Reeves W T Salesin D H and Cook R L 1987 Rendering antialiased shadows with depth maps In Stone M C editor Computer Graphics SIGGRAPH 87 Proceedings volume 21 pages 283 291 Schlick C 1993 A customizable reflectance model for everyday rendering In Cohen M F Puech C and Sillion F editors Fourth Eurographics Workshop on Rendering pages 73 84 Eurographics held in Paris France 14 16 June 1993 Smits B E and Meyer G M 1989 Newton colors Simulating interference phenomena in realistic image synthesis In Eurographics Workshop on Photosimulation R
142. speakers Topics this year include the inner workings of PRMan the operation of lights and surfaces and how you can use that knowledge to create custom lights and materials advanced aspects of fur and hair rendering skin and artistic effects and absurd displacement shading Not a single topic is repeated from previous SIGGRAPH course A couple of speakers are veterans but three are first time SIGGRAPH RenderMan course speakers We always strive to make a course that we would have been happy to have attended ourselves and so far this year is definitely shaping up to meet or exceed our expectations Thanks for coming We hope you will enjoy the show CONTENTS Chapter 1 How PhotoRealistic RenderMan Works and what you can do about it Tony Apodaca Pixar Animation Studios aaa pixar com Abstract Beginning and maybe even intermediate RenderMan users can use the abstract model of a renderer which is provided by the RenderMan Interface and described by the RenderMan Companion and numerous previous Siggraph courses to build their models and write their shaders The whole point of the RenderMan Religion is that this should possible And the vast majority of the time this will lead to success However those who are doing the most subtle manipulations of the renderer making the kind of images that the students in this course would be proud of need to have a deeper understanding of what is going on inside their chosen renderer Thi
143. specifically does not depend on the orientation of the surface as it spins around the normal vector To help visualize the situation consider the following diagram Imagine rotating the material around the normal vector If the reflectivity in a particular direction is independent of the surface orientation then the material is said to be isotropic On the other hand if the material reflects preferentially depending on surface orientation then it is anisotropic Anisotropic materials are not uncommon Various manufacturing processes can produce ma terials with microscopic grooves that are all aligned to a particular direction picture the surface being covered with tiny half cylinders oriented in parallel or otherwise coherently This gives rise to anisotropic BRDFs A number of papers have been written about anisotropic reflection models including Kajiya 1985 Poulin and Fournier 1990 Greg Ward Larson described an anisotropic reflection model in his SIGGRAPH 92 paper Mea suring and Modeling Anisotropic Reflection Ward 1992 In this paper anisotropic specular reflection was given as 2 x D h g Dod UR 2 h Og ex v cos cos 6 ATAz ay d 1 where 2 5 CUSTOM MATERIAL DESCRIPTIONS 43 Figure 2 8 Examples of the Ward anisotropic reflection model Isotropic reflection with xrough ness yroughness 0 3 top Anisotropic reflection with xroughness 0 15 yrough
144. t creating a kind of cone shaped super hair or circular clump We have implemented two kinds of clumping techniques static area clumping and animated area clumping The former generates hair clumps in fixed pre defined areas on the model whereas the latter allows for the areas that are clumped to change over time In both cases we provide parameters which can be animated to achieve various degrees of dry to wet fur looks which can also be animated over time 5 4 CLUMPING 93 5 4 1 Static Area Clumping static area clumping Four required clumping parameters clump density clump size clump percent clump rate siggraph 2000 sony picture imageworks advanced renderman fur figure 5 There are four required clumping input parameters clump density clump size clump percent and clump rate Clump density specifies how many clumps should be generated per square area Our clump method translates this density into an actual number of clumps depending on the size of each surface patch We call the center hair of each clump the clump center hair and all the other member hairs of that clump which are attracted to this clump center hair clump hairs Clump size defines the area of a clump in world space To determine clump membership of each final hair our clump method first converts the specified clump size into a u radius and v radius component in parametric surface space at each clump center hair location It
145. t doesn t have the same issues that a modeler has The texture painter just paints the pattern that they want So any change in the features that is wanted happens as quickly as the painter can paint it well ok paint and then render Of course the painter does not have as much control as the modeler which is why we still used the modeling process to achieve a good approximation of the final character If one attempts to put all of the details in the model directly other considerations need to be given when the model starts to bloat These are mainly connected to the increased I O in dealing with the bloated model A very heavy model will slow down interactively for the animators increase processing time for muscle and skinning increase load on the paint program 1 may only be able to load in less and less of the model at a time longer render times etc At one point we weren t so sure about using the displacements so there were some tests done to put more detail in the model One of them involved adding detail to one of the model s head In the end the head had 10X the number patches and it still didn t match up to the detail that we were getting with displacements Even if the time involved in building the model didn t matter and ignoring the other issues associated with bloated models it still may have been impossible to have built such a hyper real detail model at all For fun somebody tried to convert a model to polygons based on t
146. t would be needed I have picked these three to give a taste of what we did Also I hope that these are generally applicable and will inspire you to new directions So without further ado 3 1 Zero shadow bias It is common practice to post process the color images that one gets out of RenderMan People come up all kinds of reasons to manipulate their images However you never hear of people mucking around with their shadow maps It seems as though most people blissfully leave their shadow maps alone but are eager to find ways to process their color pass 59 60 CHAPTER 3 RENDERING ISSUES ON DINOSAUR Now consider the possibilities of processing shadow maps This section illustrates a technique we used on Dinosaur that entailed processing our shadow maps before using them in the color pass render The result was that our lighters rarely needed to be concerned with setting the shadow bias for any of their lights Beginning with PhotoRealistic RenderMan PRMan 3 8 there are some choices on what data the shadow maps should contain The current choices for pixel depth values are minimum maxi mum average and midpoint Before all these choices pre PRMan 3 8 days shadow maps contained only minimum depth values One of the biggest problems with the minimum depth maps was with self shadowing For numerous reasons numerical inaccuracies being the main culprit incorrect shadows would get produced So minimum depth maps should be consi
147. tatement from the surface shader which specified a cone axis and angle from which to gather light If the from position is outside this angle range the body of statements inside illuminate construct will be skipped thus saving computation for lights whose results would be ignored In a light shader Shader space is the coordinate system that was in effect at the point that the LightSource statement appeared in the RIB file 2 6 LIGHT SOURCES 47 The lights presented here are very simple examples of light shaders only meant to illustrate the solar and illuminate constructs For high quality image generation many more controls would be necessary including more flexible controls over the light s cross sectional shape direc tional and distance falloff and so on Such controls are discussed in detail in Chapter 14 of Advanced RenderMan Creating CGI for Motion Pictures Lighting Controls for Computer Cinematography 2 6 2 Area Light Sources Area light sources are those that are associated with geometry Table 2 3 lists the variables available inside area light sources Many of the variables available to the area light shader describe a position on the light that has been selected by the renderer as the point at which the light shader is sampled You need not worry about how this position is selected the renderer will do it for you Table 2 3 Global variables available inside area light shaders Variables are read only except wh
148. tc the texture function does indeed anti alias the values it reads in very well There is a problem though when these texture values are used for displacements Something more is needed when dealing with high frequency texture maps to displace geometry Why you may ask You just said that the texture function anti aliases my texture just fine The crux of the problem is that small changes in displacement maps can effect the surface s reaction to light non linearly The regular filter from the texture call albeit it a nicely anti aliased one doesn t address this issue For this the reason one is apt to see buzzing when high frequency maps are used for displacement Laurent Charbonnel a Lookdev TD on Dinosaur came up with a great function that we used 3 3 USING RENDERMAN AS A MODELER 79 Figure 3 12 Carnotaur eyeshot in rain C Disney Enterprises Inc in looking up our texture maps in PRMan In the way that it filters this function tries to lessen the impact that minute displacement shifts will have on the surface s reaction to light With Laurent s function we can control the amount of buzzing that is acceptable on a per patch basis and a per shot basis as lighting changes affect displacement aliasing The result is this function removed A LOT of buzzing in our textures Thus making displacements viable for our production Keep in mind that on Dinosaur we used PRMan 3 7 So we didn t have access to newer features in PRMan like
149. tcolor 56 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES Listing 2 16 pointlight radiates light in all directions from a particular point light pointlight float intensity 1 color lvghtcolor s 3 point from point shader 0 0 0 7 illuminate from GI intensity 7 b gt s Listing 2 17 spot light radiates a cone of light in a particular direction light spotlight float intensity 1 color JIxghtoolor is point from point shader 0 0 0 point to point shader 0 0 1 float coneangle radians 30 float conedeltaangle radians 5 float beamdistribution 2 uniform vector A normalize to from uniform float cosoutside cos coneangle uniform float cosinside cos coneangle conedeltaanale illuminate from A coneangle float cosangle L A length L float atten pow cosangle beamdistribution L L atten smoothstep cosoutside cosinside cosangle Cl atten intensity lightcolor Listing 2 18 arealight is a simple area light shader Lrzght arealight float intensity 1 color lrghtcolor 17 illuminate P N PI 2 CL e intensity 7 CG laghteoler 2 6 LIGHT SOURCES 57 Listing 2 19 shadowspot is just like spotlight but casts shadows using a shadow depth map light shadowspot float intensity 1 Color lightcolor 1 point from point shader 0 0 0 point to point shader 0 0 1 float conea
150. te a while This error message is a result of perhaps the worst single algorithmic limitation of the modified Reyes algorithm In order to correctly estimate the shading rate required for a primitive the primitive must first be projected into raster space in order to evaluate its size Additionally recall that the first step in the geometric pipeline is to bound the primitive and sort it into buckets based on its position in raster space which requires the same projection The problem is that the mathematics of perspective projection only works for positions that are in front of the camera It is not possible to project points that are behind the camera For this reason the renderer must divide the primitive into areas that are in front of and areas that are behind the camera 1 5 RENDERING ARTIFACTS 17 Most rendering algorithms if they require this projection at all would clip the primitive against the near clipping plane hence the name and throw away the bad regions However the entire Reyes geometric and shading pipelines require subprimitives that are rectangular in parametric space which can be split and diced cleanly Clipping does not create such primitives and cannot be used Instead Reyes simply splits the primitive hoping that portions of the smaller subprimitives will be easier to classify and resolve Figure 1 5 shows the situation Notice that primitives which lie entirely forward of the eye plane are projectable so can be accep
151. te machine and geometric primitives whose attributes are defined by the then current version of the graphics state machine The hierarchical graphics state 1 2 BASIC GEOMETRIC PIPELINE 5 shaders textures application ss Parser HI calls shaded grids RI asw bust amp API bound gprims micropolygons multiple sample we N gprims in Y visible points too NY composite split amp filter N 1 pixels unshaded grids dice image Figure 1 1 The Reyes rendering pipeline machine is kept in a stack based data structure within the RI layer Whenever a geometric primitive arrives the current top of the stack set of attributes is attached to the primitive before it proceeds into the main Reyes geometric processing engine 1 2 1 Splitting Loop The first thing that Reyes does to arriving primitives is bound them The renderer computes a camera space axis aligned bounding box that is guaranteed to contain the entire primitive Ren derMan does not contain any unbounded primitives such as infinite planes so this is generally straightforward Most RenderMan primitives have the convex hull property which means that the primitive is entirely contained within the volume outlined by the vertices themselves Primitives that don t have this property such as Catmull Rom patches are converted to equivalent primitives that do such as Bezier patches 6 CHAPTER 1 HOW PHOTOREALISTIC RENDERMAN WORK
152. ted Primitives which lie entirely behind the near clipping plane can be trivially culled It is only primitives which span both planes that cannot be classified and are split The region between the planes can be called the safety zone If a split line lies entirely within this zone in 3D of course both children of the bad primitive are classifiable which is the situation we are hoping for If the split line straddles a plane at least we will shave some of the bad primitive away and the remaining job we hope is slightly easier Reyes splits as smartly as it can attempting to classify subprimitives Eye Plane Near Clipping Plane L Ld Projected Spans Planes s Requires Split 7 L I L f L 1 I I 1 L L I L I I I I 1 L L 1 L L 1 RES I I I Fite ci oo 6626 01 ve a 4 Near Clipable 2 Projectable Safety Zone Figure 1 5 The geometric relationship of the near and eye planes gives rise to three categories of primitives the cullable the projectable and the spanners which require splitting Unhappily there are geometric situations where the splitting simply doesn t work in a reasonable number of steps Reasonable is defined as a small integer because each split doubles the number of subprimitives so even 10 attempts creates 210 1024 primitives If after splitting the maximum permitted number of times the primitive still ca
153. ted keyframe maps This is what was done for the bottom image An example of another situation to watch out for say the scene has a very long truck out and the last frame of the scene is used as the CP keyframe The very early frames in this situation will have a very low res quality This is because the texture map will be using a very small sub portion of the CP keyframe texture map A solution for this situation is to have a second CP keyframe near the beginning and then blend between these two maps or impose a map precedence Figure 3 7 shows an example of a severe pullback The top left image a shows the first frame of the camera animation The top right image b shows the final frame of the animation Figure 3 7 a represents just a few percent of b So if just one keyframe was picked b 19 the earlier frames in the animation will be very low quality because they represent such a small area of the image A solution is shown in the bottom image where a frame earlier in the animation is picked as another keyframe the white area numbered 1 This way the earlier frames can use a map with more detail for their level Then the later frames can make use of the second keyframe Care will be needed when transitioning between the maps This may include blending along the borders of the maps so they will merge seamlessly Figure 3 7 Top left Beginning frame of camera pullback Top right Final frame 367 of camera pull back Bottom This
154. terpolated onto every vertex in the grid but the vertices have not yet been shaded dice bicubic subpatch micropolygon grid Figure 1 2 A primitive that is small enough will be diced into a grid of tiny bilinear facets known as micropolygons 1 2 BASIC GEOMETRIC PIPELINE 7 1 2 2 Shading The grid is then passed into the shading system to be shaded PRMan first evaluates the displacement shader which may move grid vertices and or recompute shading normals Then the surface shader is evaluated In a typical surface shader there are calls to diffuse or illuminance somewhere in the code These routines require the evaluation of all of the light shaders attached to the object The light shaders run as coroutines of the surface shader the first time that they are needed but their results are cached and reused if they are accessed again for example in a subsequent specular call When the surface shader finishes it has computed the color and opacity of every grid vertex Finally the atmosphere shader is evaluated making adjustments to the vertex color and opacity to simulate fog or other volumetric effects The end result of all of these evaluations is that the grid now has final color and opacity assigned to each vertex and the vertices themselves might be in different places than when originally diced The other attributes of the original primitive are now mostly superfluous The details of the method that the shading system uses to
155. then checks for each hair whether it falls within the radii of a clump center hair if so that clump center hair s index is stored with the hair if not the hair is not part of any clump i e is not a clump hair Our clump method also has an optional clump size noise parameter to produce random variations in the size of the clumps Furthermore texture maps for all four parameters can be created and specified by the user one per surface patch to provide local control for instance if clump size maps are specified the global clump size input parameter is multiplied for a particular clump clump center hair at u v on a surface patch with the corresponding normalized s t value in the clump size feature map for that surface patch Also a clump area feature map can be provided to limit clumping to specified areas on surface patches rather than the whole model 94 CHAPTER 5 FUR IN STUART LITTLE static area clumping unaffected hairs clump center hair clump boundary siggraph 2000 sony picture imageworks advanced renderman fur figure 6 static area clumping clump percent KAA 0 0 0 3 0 6 siggraph 2000 sony picture imageworks advanced renderman fur figure 7 Clump percent controls the amount the hair is drawn to the clump center hair This is done in a linear fashion where the endpoint of the hair is effected 100 and the base of the hair where it is connected to the patch is never moved Higher clump percent
156. tives that cover its pixel have in fact been processed Because the streaming version of Reyes cannot know that any given pixel is done until the last primitive is rendered it must store all the visible point lists for the entire image until the very end The visible point lists therefore contain a point sampled representation of the entire geometric database and consequently are quite large Strike that They are absolutely huge many gigabytes for a typical high resolution film frame Mon strously humongous As a result the algorithm simply would not be usable if implemented in this way Memory sensitive enhancements are required to make the algorithm practical 1 3 Enhanced Geometric Pipeline The original Reyes paper recognized that the memory issue was a problem even more so in 1985 than it is now So it provided a mechanism for limiting memory use and other mechanisms have been added since which together make the algorithm much leaner than most other algorithms 1 3 4 Bucketing In order to alleviate the visible point memory problem a modified Reyes algorithm recognizes that the key to limiting the overall size of the visible point memory is to know that certain pixels are done before having to process the entire database Those pixels can then be finished and freed early This is accomplished by dividing the image into small rectangular pixel regions known as buckets which will be processed one by one to completion before signific
157. ular Blue Moon Rendering Tools a shareware RenderMan compliant renderer that supports ray tracing and radiosity Larry has a BS from Cornell University and MS and PhD degrees from the George Washington University Larry is co author of Advanced RenderMan Creating CGI for Motion Pictures Morgan Kaufmann 1999 Tony Apodaca is a senior technical director at Pixar Animation Studios In the 20th century he was director of Graphics R amp D at Pixar was co creator of the RenderMan Interface Specification and lead the development of PhotoRealistic RenderMan Tony has been at Pixar since 1986 where his film credits include work from Red s Dream through A Bug s Life He received a Scientific and Technical Academy Award from the Academy of Motion Picture Arts and Sciences for work on PhotoRealistic RenderMan and with Larry was co author of Advanced RenderMan Creating CGI for Motion Pictures He holds an MEng degree in computer and systems engineering from Rensselaer Polytechnic Institute Tal Lancaster has been a technical director for Walt Disney Feature Animation since 1995 concen trating on RenderMan shader development His film credits include Fantasia 2000 and Dinosaurs Tal also maintains the RenderMan Repository www renderman org Mitch Prater is a shading TD at Pixar and has been writing shaders for production and product uses for over 10 years Mitch s film credits include Toy Story A Bug s Life Toy Story 2 and numerous comme
158. ular Nf normalize I roughness Figure 2 2 The finishes left to right MaterialPlastic MaterialRoughMetal and Ma terialMatte applied to a vase shine through the object albeit usually at a lower intensity than the reflections of the lights in front of the object Note the difference between translucency the diffuse transmission of very scattered light through a thickness of material and transparency which means one can see a coherent image through the object Ordinary paper is translucent whereas glass is transparent Therefore since diffuse Nf sums the Lambertian scattering of lights on the viewer s side of the surface then the lights from the back side should be described by diffuse Nf In fact this works exactly as we might hope Thus making a material translucent is as easy as adding an additional contribution of dif fuse oriented in the backwards direction and presumably with a different and smaller weight denoted by Kt This is exemplified by the MaterialThinPlas tic function of Listing 2 4 28 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES Listing 2 4 MaterialThinPlastic implements a simple thin plastic like BRDF Compute the color of the surface using a simple thin plastic like BRDF We call it thin because at includes a transmission compo nent to allow light from the back_ of the surface to affect the appear ance Typical values are Ka 1 Kd 0 8 Kt 0 2 Ks 0 5 roughness
159. umping techniques presented herein Additionally others who have contributed to the Sony Imageworks hair fur pipeline are Alan Davidson Hiro Miyoshi Bruce Navsky Rob En gle Jay Redd Amit Agrawal as well as the Imageworks software digital character and production groups The author also wishes to acknowledge the support of Sony Imageworks for the publication of these notes It should also be noted that some of the fur methods discussed in this course are techniques for which Sony Imageworks is currently applying for patents All figures and illustrations not indicated otherwise are Copyright 2000 by Sony Pictures Image works all rights reserved 5 10 Conclusion There were several key concepts that made the fur pipeline for Stuart Little a workable solution Delaying the creation of the largest portion of the dataset until render time was an efficient decision made early in the pipeline design that proved itself in terms of storage and processing time For the clumping process the idea of using both a realistic modeling technique combined with shading techniques yielded a strong wet fur look And finally the decision to use a lighting model which was based more on a TD s familiarity with lighting surfaces than what would be the most mathmat ically accurate seemed to give more palatable results in less time which improved the look of the film 5 11 Shaders 5 11 1 Complete Fur Shader with Clumping and Specular Model fur sur
160. use srf contact shader to render the soft surface white where the character is in close proximity to the surface use a compositing system to time average several contact images together into a displacement map use the displacement map to create the Pref surface for deformation shading siggraph 2000 sony picture imageworks advanced renderman fur figure 2 8 Creating the position of the deformed or the P object through entirely procedural methods is an interesting task which was developed in RenderMan The first step is to create the depth map of the character s feet or the part of the body that needs to do the interaction with the surface from the bottom Next use the depth map as a data file to drive a shader named srf contact sl which is included as an appendix to these notes srf_contact is applied to the surface to be deformed and will turn to a 1 0 value when the character is within a specified distance from the surface and will gradually fade off as the character gets further away from the surface along the direction vector of the camera used to render the depth file In our case this process created a set of images with little white feet that would fade on and off as Stuart s feet got closer to and further from the surface This element is named the contact pass The next step in the process is to sample together several of these images into a new sequence We used some time filtering techniques by which
161. uses a rough ness coefficient of 0 5 while the right uses a roughness coefficient of 0 which has reflectance iden tical to Lambertian shading Notice that as roughness increases a strong back scattering effect 1s present The object begins to act as a retro reflector so that light that comes from the same direction as the viewer bounces back at a rate nearly independent of surface orientation It is this type of backscattering that accounts for the appearance of the full moon Everyone has noticed that a full moon when the viewer and sun are nearly at the same angle to the moon looks like a flat disk rather than like a ball The moon is not Lambertian it is more closely modeled as a rough surface and this reflection model is a good approximation to the behavior of lunar dust 42 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES e Figure 2 7 When the light source is directly behind the viewer the Oren Nayar model left acts much more as a retroreflector compared to the Lambertian model right 2 5 2 Anisotropic Metal The MaterialRoughMetal function described earlier does an adequate job of simulating the appearance of a metal object that is rough enough that coherent reflections from ray tracing or environment maps are not necessary However it does make the simplifying assumption that the metal scatters light only according to the the angular relationship between the surface normal the eye and the light source It
162. values yield more wet looking fur 5 4 CLUMPING 95 static area clumping clump rate AA 0 0 0 3 0 6 clump percent 1 0 siggraph 2000 sony picture imageworks advanced renderman fur figure 8 Clump rate is the control used to dial in how much the fur is curved near the base to match the clump center hair This is also interpolated over the length of the fur so the base of the fur is never effected by the clumping procedure Dialing this parameter up yields dramatic increases in the wetness look of the character as the individual hairs are drawn more strongly to the clump centers static area clumping once only clumping calculations turn some hairs within static clumping areas into clump center hairs clump density calculate clump u v radii for each clump center hair clump size for all other hairs check whether they lie within clump radii of any clump if so make these hairs clump hairs and assign clump center hair clump percent and clump rate to this hair siggraph 2000 sony picture imageworks advanced renderman fur figure 9 96 CHAPTER 5 FUR IN STUART LITTLE static area clumping for each frame clumping calculations generate all clump hairs default for numberOfCVs is 3 and i ranges from 1 3 fract 2 1 numberOfCVs delta clumpPercent fract clumpRate 1 fract clumpHairCV i 2 clumpHairCV i delta clumpCenterHairC V 1 clu
163. will just let the renderer waste exponentially more time before it gives up Also make sure that displacement bounds for primitives near the camera for example the ground plane are as tight as possible and that the shader is coded so that the displacement itself is as small as possible If you place the camera so close to some primitive that the displacement bound is a significant portion of the image size there will be no end of trouble with both eye splits and displacement stretching discussed later In fly overs of the Grand Canyon the Canyon should be modeled not implemented as a displacement map of a flat plane It will also help to keep the camera as high off the ground as possible without ruining the com position of the shot Reyes simply has lots of trouble with worm s eye views And make sure that no object is flying through the camera you wouldn t do that in live action photography would you Generally if you pretend that the CG camera has a physical lens that keeps objects in the scene at least a certain distance away and respect that border you will have fewer problems with eye splits 1 5 2 Patch Cracks Patch cracks are tiny holes in the surface of objects that are caused by various errors in the approxi mation of primitives by their tessellations we ll use the term loosely to include tessellation created cracks on primitives other than patches Patch cracks usually appear as scattered pinholes in the surface although the
164. y already in the plate photography These surfaces needed to be manipulated to appear to be deforming under Stuart s feet and body other tricks deformation shader basic concept shade the point at both the P and the Pref surface and calculate the difference in Pref surface lighting between them P surface J siggraph 2000 sony picture imageworks advanced renderman fur figure 2 6 The basic concept behind the system is to create two geometries The first which represents the surface as it is represented in the plate If the shot is moving or the surface was animating in the sequence that surface may be matchmoved frame by frame For purposes of the discussion of the shader we will refer to this as the Pref surface The second geometry represents the position of the surface when in contact with Stuart We generate this surface in a variety of ways one of which will be discussed later but the basic idea 15 that the P surface is the deformed surface To warp the plate from the Pref to the P position is simply a matter of projecting the plate through the scene s camera onto the Pref geometry and then applying that texture to the P shading point see the shader in the appendix for implemenation details 5 7 OTHER TRICKS DEFORMATION SYSTEM 109 other tricks deformation shader lighting calculation texture 0 5 Cil lighting at point P 12 lighting at point Pref texture
165. y approximated by Lambert s law nlights XO Cl max 0 N Li oc l This section is substantially reprinted from Advanced RenderMan Creating CGI for Motion Pictures by Anthony A Apodaca and Larry Gritz ISBN 1 55860 618 1 published by and copyright 2000 by Morgan Kaufmann Publishers all rights reserved and is used with permission Hey Why don t you go get the whole book instead of reading a couple out of context excerpts 23 24 CHAPTER 2 THE SECRET LIFE OF LIGHTS AND SURFACES where for each of the i light sources L is the unit vector pointing toward the light C l is the light color and is the unit normal of the surface The max function ensures that lights with N L lt 0 those behind the surface do not contribute to the calculation color specular vector N V float roughness Computes so called specular lighting which refers to the way that glossier surfaces have no ticeable bright spots or highlights resulting from the narrower in angle scattering of light off the surface A typical formula for such scattering might be the Blinn Phong model nlights Cl max 0 N H toughness i 1 where H is the vector halfway between the viewing direction and the direction of the light source 1 normalize normalize I normalize L The equation above is for the Blinn Phong reflection model which is what is dictated by the RenderMan Interface Specification PRMan actual
166. y can sometimes appear as lines of pinholes or as small slits Importantly they always appear along parametric lines of primitives They are recognizably different from other holes created by clipping culling or bucketing errors which are usually larger often triangular and occur in view dependent places like on silhouettes or aligned with bucket boundaries Patch cracks occur because when objects are defined by sets of individual primitives the con nectedness of those primitives is only implied by the fact that they abut that they have edges which have vertices in common There is no way in the RenderMan Interface to explicitly state that sepa rate primitives have edges that should be glued together Therefore as the primitives go through the geometric pipeline independently there is a chance that the mathematical operations that occur on one version of the edge will deviate from those on the other version of the edge and the vertices will diverge If they do a crack occurs between them The deviations can happen in several ways One is the tessellation of the edge by adjacent grids being done with micropolygons of different sizes Figure 1 6 shows that such tessellations naturally create intermediate grid vertices that do not match and there are tiny holes between the grids For many years PRMan has had a switch that eliminated most occurrences of this type of crack Known as binary dicing it requires that every grid have tessellation
167. yss in 1989 and in 1993 was given an Academy Award for contributions to the film industry 1 2 Basic Geometric Pipeline The Reyes algorithm is a geometric pipeline not entirely unlike those found in modern day hardware graphics engines What sets it apart is the specific types of geometric operations that occur in the pipeline and the way that as the data streams through the system it gains and retains enough geometric and appearance fidelity that the final result will have very high image quality Figure 1 1 shows the basic block diagram of the architecture The first step of course is loading the scene description from the modeler Typically the scene description is in a RIB file loaded from disk In that case the RIB file is read by a RIB parser which calls the appropriate RI routine for each line of the RIB file Notice that since the RIB file is a simple metafile of the RI API it is extremely easy to parse The only minor complexity arises from the handling of parameterlist data which is dependent on the parameter type declarations that appear earlier in the RIB file Alternatively a program that is linked to the renderer can call the RI API directly in which case the parser is simply bypassed The second step is the processing of the RenderMan Interface calls themselves This stage of the pipeline maintains the hierarchical graphics state machine RI calls fall into two classes attributes or options that manipulate the graphics sta

Download Pdf Manuals

Related Search

Related Contents

TCH 250-150 Manual  Téléchargez votre magazine  TP-8800 - 白山工業株式会社  LG Electronics 560T Flat Panel Television User Manual    Side-Power MAXI 31/32/33/34/40/43/44  INTERFELL 100 P (ES/FR/PT) rev.01  MATÉRIA TÉCNICA APTTA BRASIL SENSORES MAGNETO  Gigabyte MIB T5140    

Copyright © All rights reserved. Failed to retrieve file