Interactive Rendering of Translucent Deformable Objects

1. Interactive Rendering of Translucent Deformable Objects Tom Mertens1, Jan Kautz2, Philippe Bekaert1, Hans-Peter Seidel2, Frank Van Reeth1 1 2

2. Overview • Goal • Previous work • Translucency model • Our method • Implementation • Discussion, results and future work

3. Problem: Translucency BRDF BSSRDF

5. Previous work • Jensen et al. (SIGGRAPH ’01): • BSSRDF model • Jensen et al. (SIGGRAPH ’02): • fast, production quality rendering • Lensch et al. (PG ’02), Hao et al. (GI ’03), Carr et al. (GHW’03), Sloan et al. (SIGGRAPH’02-’03): • interactive, real-time rendering with precomputation • Our paper: • interactive rendering • varying geometry and material (no precomputation)

6. BSSRDF model • introduced by Jensen et al.(SIGGRAPH’01) • multiple scattering • materials with high albedo: marble, milk, wax, skin,… function of distance

7. BSSRDF model • introduced by Jensen et al.(SIGGRAPH’01) • multiple scattering • materials with high albedo: marble, milk, wax, skin,… function of distance

8. Integrating the BSSRDF • hierarchical approach (Jensen et al. ‘02) • decouple light and surface sampling, • decouple light sampling from geometry • 2-pass method: • irradiance sampling – integration with octree • limitation: rebuilding samples & octree • our method • integration ~ hierarchical radiosity • mesh based: beneficial for geometry updates • hierarchy = clustered triangles • form factor for BSSRDF: fast local integration

9. Our Method • boundary element method

10. Our Method • boundary element method discretized radiance discretized irradiance

11. Our Method • boundary element method discretized radiance form factor discretized irradiance

12. example

13. sample irradiance

14. pull irradiance

15. link roots

16. subdivide link

17. subdivide link again

18. gather

19. push

20. Hierarchical Evaluation • hierarchy = clustered triangles • tree hierarchy • subdivision: 4-to-1 splits • face clustering • evaluation ~ hierarchical radiosity • irradiance sampling + pull • construct link hierarchy • gather over each link • push + average at vertices • “oracle” = solid angle • interactions at different levels • speed advantage

21. Form Factor: area integral • (mid)point to triangle • semi-analytical • Taylor expansion • advantages: • fast • accurate • noiseless • indispensable for local integration • more distant: 1 sample integral over edges recursive midpoint

22. Form Factor: • point to triangle • semi-analytical • Taylor expansion • advantages: • fast • accurate • noiseless • indispensable for local integration • more distant: 1 sample

23. Form Factor: • point to triangle • semi-analytical • Taylor expansion • advantages: • fast • accurate • noiseless • indispensable for local integration • more distant: 1 sample point sampling form factor

24. stored links incremental updates promote/demote links real-time frame rate render on-the-fly instant feedback less memory overhead interactive frame rate Implementation • GPU • fresnel • tone mapping • shadow map • irradiance • point light (+ shadow) • environment map

25. Results • 5-10 fps for 10-20K tris models • dual Xeon 2.4Ghz; ATI Radeon 9700 • Demo video material change candle twist shadow leak Perlin noise deformation

30. Discussion • practical technique for interactive applications • speed advantage over previous hierarchical algorithm: • gathering in higher levels • efficient local integration • consistent hierarchy after deformation • limitation = mesh • needs hierarchy • limited by resolution • fixed topology • interactive applications often mesh-based anyway

31. Future Work • recycle radiosity techniques • adaptive meshing, high order interpolation,… • improved oracle function • varying topology • full GPU implementation • non-homogeneous media

32. Acknowledgements • Jens Vorsatz (mesh hierarchies) • P. Debevec (light probes) • funding: • European Regional Development Fund • Marie Curie doctoral fellowship