1. Eurographics Symposium on Rendering 2008 Yue Dong, Sylvain Lefebvre, Xin Tong, George Drettakis

2.  We introduce a new algorithm with the unique ability to restrict synthesis to a subset of the voxels, while enforcing spatial determinism ◦ Only a thick layer around the surface needs to be synthesized  Synthesize a volume from a set of pre- computed 3D candidates ◦ Carefully select in a pre-process only those candidates forming consistent triples

3.  Runs efficiently on the GPU ◦ Generates high resolution solid textures on surfaces within seconds ◦ Memory usage and synthesis time only depend on the output textured surface area ◦ Our method rapidly synthesizes new textures for the surfaces appearing when interactively breaking or cutting objects

4.  Solid textures define the texture content directly in 3D ◦ Removes the need of a planar parameterization ◦ Unique feeling that the object has been carved out of a block of matter

5.  Implicit Volume ◦ Color = f(x, y, z) ◦ Procedural texturing  Texturing and Modeling: A Procedural Approach  EBERT D., MUSGRAVE K., PEACHEY D., PERLIN K., WORLEY  Academic Press, 1994 ◦ Spectral analysis  Spectral analysis for automatic 3d texture generation  GHAZANFARPOUR D., DISCHLER J.-M.  Computers & Graphics, 1995  Generation of 3d texture using multiple 2d models analysis  GHAZANFARPOUR D., DISCHLER J.-M.  Computers & Graphics,1996  Low memory usage  Limited range of materials

6.  Explicit Volume ◦ Color = g[x, y, z] ◦ Histogram matching  Pyramid-Based texture analysis/synthesis  HEEGER D. J., BERGEN J. R.  SIGGRAPH, 1995 ◦ Stereological technique  Stereological techniques for solid textures  JAGNOW R., DORSEY J., RUSHMEIER H.  SIGGRAPH, 2004 ◦ Neighborhood matching  Texture synthesis by fixed neighborhood searching  WEI L.-Y.  PhD thesis, 2002, Stanford University  Aura 3d textures  QIN X., YANG Y.-H.  IEEE Transactions on Visualization and Computer Graphics, 2007  Solid texture synthesis from 2d exemplars  KOPF J., FU C.-W., COHEN-OR D., DEUSSEN O.,  LISCHINSKI D., WONG T.-T.  SIGGRAPH, 2007  Good Quality  Can synthesis various materials  Take long time to compute

7.  Pre-computation ◦ 3D candidates from 2D exemplars  Multi-resolution pyramid synthesis ◦ Upsample ◦ Jitter ◦ Correction

8.  Pixel : 2D / Voxel : 3D  Triple : a set of three 2D coordinates  Crossbar : a set of pixels which are crossing in three neighborhoods of size N (N = 5)

9.  We select candidate triples following two important properties ◦ A good triple must have matching colors along the crossbar of the three neighborhood  To provide color consistency ◦ A good triple must have a good coherence across all three exemplars  Which is likely to form coherent patches with other neighboring candidates

10.  A suitable candidate should be consistent across the crossbar ◦ Minimize the color difference of the crossbar ◦ Compute L 2 color difference between each pairs ◦ The sum of difference for the three pairs defines a crossbar error CB

11.  In each pixel of each exemplar ◦ Form triples using the pixel itself and two neighborhoods from the other two exemplars ◦ Select the triples producing the smallest crossbar error  To speed up the process ◦ Extract S most-similar pixel strips from each of the two exemplars, using ANN library ◦ Form S 2 triples then take 100 best triples ◦ S = 65

12.  Check whether a candidate may form coherent patches in all directions with candidates from neighboring pixels  For each coordinate within a candidate triple ◦ Verify that at least one candidate from a neighboring pixel has a continuous coordinate

13. pp+1 ExEx EyEy EzEz

14.  x C – Candidates for E x  x C k [p] – k-th candidate triple for pixel p in Ex  x C k [p] y – E y coordinate of the triple x C k [p]

15.  Iterate until having no more than 12 candidates per pixel ◦ Typically requires 2 iterations  If more candidates remain select first 12 with the smallest crossbar error  It is possible to have no candidate at all ◦ Rare in practice

16.  Candidates are not only useful for neighborhood matching, but also provide a very good initialization for the synthesis process  For each pixel ◦ One 2D neighborhood is in the plane of the exemplar ◦ Two others are orthogonal to it and intersect along a line of neighborhood size of N voxels

17.  To initialize synthesis we create such a slab using the best (first) candidate at each pixel  Using the slab directly as a 3D exemplar would be very limiting ◦ This would ignore all other candidates ◦ Uses a slab only for initialization

18.  Extended ‘Parallel Controllable Texture Synthesis’ [SIGGRAPH 2005]  Same overall structure ◦ Upsample ◦ Jitter ◦ Correction

19.  Contrary to the original scheme we perturb the result through jitter only once, after initialization ◦ If finer control is desired, jitter could be explicitly added after each upsampling step

20.  To reduce synthesis time, multi-resolution synthesis algorithms can start from an intermediate level of the image pyramid  A good initialization is key to achieve high- quality synthesis  We simply choose one of the candidate slabs and tile it in the volume ◦ Three levels above the finest (Maximum Level L – 3) ◦ Using the candidate slab from the corresponding level

21. Random InitializationSlab Initialization

22.  To explicitly introduce variations to generate variety in the result  Perturb the initial result by applying a continuous deformation, similar to a random warp

23.  J – Jittered Volume  v – Voxel coordinate  c i – Random point in space  d i – Nomalized Random direction  G = 200  A i = 0.1 ~ 0.3  σ i = 0.01 ~ 0.05

24.  It is important for A i to have larger magnitude with smaller σ i ◦ Adds stronger perturbation at small scales, while adding subtle distortions to coarser scales ◦ Small scale distortions are corrected by synthesis, introducing variety  The overall magnitude of the jitter is directly controllable by the user

25.  Each of the eight child volume cells inherits three coordinates from its parent, one for each direction

26.  Performed on all synthesized voxels simultaneously, in parallel  We compute a color by averaging the corresponding three colors from the exemplars  We visit each of its direct neighbors, and use the stored coordinate triples to gather the candidate sets

27.  P x – 3 x 2 matrix transforming a 2D offset from E x to volume space

28.  Search for the best matching candidate by distance between voxel neighborhood and 3D candidate  Distance is measured by L 2 norm on color differences ◦ Can use PCA projection to speed up the process  Replace the triple with best matching candidate ◦ triples have been pre-computed and optimized to guarantee that the color disparity between the three colors in each voxel is low  Two correction pass for every level ◦ Using sub-pass mechanism of PCTS ◦ 8 pass sub-pass

29.  We gather 12 candidates from the 33 = 27 direct neighbors, for a total of 324 candidates per voxel ◦ Too many candidates  Search for the best matching 2D candidates in each of the three directions then gather the 3D candidates only from these three best matching pixels ◦ Still a lot ◦ In practice we keep 4 2D and 12 3D candidates per exemplar pixel at coarse levels  27×4 = 108 2D candidates  3×12 = 36 3D candidates ◦ 2 2D and 4 3D candidates at the finest level

30.  Determine the entire dependency chain throughout the volume pyramid from a requested set of voxels to synthesize the smallest number of voxels ◦ Compute a synthesis mask  Mask l p ⊗Neighborhood Shape - dilation of the mask by the shape of the neighborhoods

31.  To compute a single voxel, with N = 5, 2 passes and synthesis of the 3 last levels, our scheme requires a dependency chain of 6778 voxels ◦ The size of the dependency chain grows quadratically with the number of passes

32.  Entirely in software and using the GPU to accelerate the actual synthesis  Intel Core2 6400 (2.13GHz) CPU and an NVIDIA GeForce 8800 Ultra  We sometimes add feature distance

33.  Most results in the paper are computed from a single example image repeated three times ◦ Pre-computed candidates may be shared depending on the orientation chosen for the image ◦ Typically 7 seconds for 64 2 exemplars ◦ 25 to 35 seconds for 128 2 exemplars ◦ Includes building the exemplar pyramids, computing the PCA bases and building the candidate sets ◦ 231KB memory required for a 64 2 exemplar

34.  Implemented in fragment shaders, using the OpenGL Shading Language ◦ Unfold volumes in tiled 2D textures, using three 2- channel 16 bit render targets to store the synthesized triples ◦ Pre-compute and reduce the dimensionality of all candidate 3-neighborhoods using PCA, keeping between 12 and 8 dimensions  Keep more terms at coarser levels since less variance is captured by the first dimensions ◦ Quantize the neighborhoods to 8-bits to reduce bandwidth  Stored in RGBA 8 bit textures

35.  In order to minimize memory consumption, we perform synthesis into a TileTree data structure ◦ LEFEBVRE S., DACHSBACHER C. In Proceedings of the ACM SIGGRAPH Symposium on Interactive 3D ◦ Graphics and Games (2007)

36.  When interactively cutting an object, synthesis occurs only once for the newly appearing surfaces ◦ Tile-Tree cannot be updated interactively  Store the result in a 2D texture map for display ◦ Our implementation only allows planar cuts  new surfaces are planar and are trivially parameterized onto the 2D texture synthesized when the cut occurs

39.  7.22 seconds for synthesizing the 64 3 volume from 64 2 exemplar ◦ 7 seconds for pre-computation and 220 milliseconds for synthesis ◦ Memory requirement during synthesis is 3.5MB  28.7 seconds for synthesizing the 128 3 volume from 128 2 exemplar ◦ 27 seconds for pre-computation and 1.7 seconds for synthesis ◦ ‘Solid texture synthesis from 2d exemplars’ [SIGGRAPH 2007] takes 10 to 90 minutes

41.  4.1 seconds (dragon) to 17 seconds (complex structure), excluding pre-computation  Storage of the texture data requires between 17.1MB (statue) and 54MB (complex structure) ◦ The equivalent volume resolution is 1024 3 which would require 3GB  Slower than state-of-the-art pure surface texture synthesis approaches ◦ But inherits all the properties of solid texturing

42.  On demand synthesis when cutting or breaking objects (Fig. 10) ◦ Resolution of 256 3  Initially requires 1.3MB ◦ The average time for synthesizing a 256 2 texture for a new cut is 8 ms ◦ Synthesizing a 256 2 slice of texture content requires 14.4MB  Due to the necessary padding to ensure spatial determinism

45.  We also implemented our synthesis algorithm using only standard 2D candidates ◦ Takes roughly twice the number of iterations to obtain a result of equivalent visual quality ◦ Due to the increased number of iterations, the size of the dependency chain for computing a single voxel grows from 6778 voxels with 3D candidates to 76812 voxels with 2D candidates  A factor of 11.3 in both memory usage and speed

47.  A new algorithm for solid synthesis ◦ with the unique ability to restrict synthesis to a subset of the voxels, while enforcing spatial determinism ◦ Synthesize a volume from a set of pre-computed 3D candidates ◦ GPU implementation is fast enough to provide on demand synthesis when interactively cutting or breaking objects