1. 1/45 A Fast Rendering Method for Clouds Illuminated by Lightning Taking into Account Multiple Scattering Yoshinori Dobashi (Hokkaido University) Yoshihiro Enjyo (Hokkaido University) Tsuyoshi Yamamoto (Hokkaido University) Tomoyuki Nishita (The University of Tokyo)

2. 2/45 Overview Introduction Previous Work Basic Idea of Our Method Precomputation of Basis Intensities Real-time Rendering Process Examples Conclusions & Future Work

3. 3/45 Introduction (1/3) Realistic Image Synthesis  Visual Assessment, flight simulators, etc…  Assuming under fine weather conditions Simulation under bad weather conditions  Rainfalls, windstorm, lightning, clouds, etc…  Flight simulation/computer games  Real-time rendering  Reality

4. 4/45 Introduction (2/3) Example image clouds illuminated by lightning

5. 5/45 Introduction (3/3) Multiple scattering  Enhancing reality  High computation cost Difficult to achieve real-time rendering clouds viewpoint lightning

6. 6/45 Features of Proposed Method Real-time rendering of clouds illuminated by lightning  Arbitrary shapes of lightning  Anisotropic multiple scattering  Consisting of preprocess and real-time process  Use of wavelet transform for drastically accelerating intensity calculation Assumption: Static clouds

7. 7/45 Overview Introduction Previous Work Basic Idea of Our Method Precomputation of Basis Intensities Real-time Rendering Process Examples Conclusions & Future Work

8. 8/45 Previous Work on Lightning Visual simulation of lightning [ Reed94]  Use of random numbers to create shape of lightning strokes Probabilistic modeling of lightning [Kruszewski99]  Allowing user to control shapes Physically-based animation of lightning [Kim04]  Calculating propagation of lightning via physical simulation Focusing on modeling shape of lightning Illumination on clouds are not considered.

9. 9/45 Previous Work on Fast Rendering of Clouds/Smoke Rendering of clouds using GPU [Dobashi00] Clouds/atmosphere illuminated by lightning  Use of GPU for acceleration  Single scattering only Precomputed radiance transfer [Sloan02]  Real-time rendering of clouds as well as glossy objects  Assuming light sources far from clouds  Lihgtning is not considered. [Dobashi01]

10. 10/45 Overview Introduction Previous Work Basic Idea of Our Method Precomputation of Basis Intensities Real-time Rendering Process Examples Conclusions & Future Work

11. 11/45 Basic Idea of Our Method (1/7) Shape of lightning  Polylines generated by Reed’s method [Reed94] Density distribution of clouds  Represented by 3D grid  Use of Dobashi’s method [Dobashi00] lightning clouds

12. 12/45 Basic Idea of Our Method (2/7) Intensity calculation of clouds  Point sources on strokes  Illuminations due to each point source  Multiple scattering  Accumulation of illuminations due to all point sources  Preprocess and real-time process point source illuminate lightning clouds

13. 13/45 Basic Idea of Our Method (3/7) simulation space grid for clouds grid for simulation space Two separate grids clouds

14. 14/45 Basic Idea of Our Method (4/7) Virtual point sources at each grid point clouds simulation space grid for clouds grid for simulation space virtual point source

15. 15/45 Basic Idea of Our Method (5/7) Preprocess ： computation of basis intensities virtual point source (intensity = 1.0) intensity calculation taking into account multiple scattering save

16. 16/45 Basic Idea of Our Method (6/7) Preprocess ： computation of basis intensities

17. 17/45 Basic Idea of Our Method (6/7) Preprocess ： computation of basis intensities

18. 18/45 basis intensities Basic Idea of Our Method (6/7) Preprocess ： computation of basis intensities Repeating for all grid points

19. 19/45 Basic Idea of Our Method (7/7) Real-time process ： rendering of clouds basis intensities intensity of virtual point sources multiplication accumulation  generating lightning Fast but cost proportional to number of grid points Further acceleration by using wavelets × × × ×

20. 20/45 Overview Introduction Previous Work Basic Idea of Our Method Precomputation of Basis Intensities Real-time Rendering Process Examples Conclusions & Future Work

21. 21/45 Subdivision of Simulation Space Contribution of virtual point source clouds r intensity due to point source 1.0 r2r2 virtual point source simulation space

22. 22/45 Subdivision of Simulation Space Adaptive subdivision clouds rlrl r l+1 simulation space

23. 23/45 Precompuation of Basis Intensities Intensity calculation due to virtual point source virtual source l grid point i : intensity at grid point i due to virtual source l b il n cld : number of grid points for clouds b il calculation of multiple scattering (b 1l, b 2l, …, b n cld, l )

24. 24/45 Precompuation of Basis Intensities Intensity calculation due to virtual point source virtual source l grid point i basis intensities due to virtual light l b il calculation of multiple scattering n cld : number of grid points for clouds (b 1l, b 2l, …, b n cld, l )

25. 25/45 Precompuation of Basis Intensities Intensity calculation due to virtual point source virtual source l grid point i n cld : number of grid points for clouds b il calculation of multiple scattering wavelet transform (b 1l, b 2l, …, b n cld, l ) (B 1l, B 2l, …, B n cld, l ) basis intensities due to virtual light l

26. 26/45 Precompuation of Basis Intensities basis intensity at grid point i due to virtual source l number of basis functions (= number of grid points for clouds) coefficient after wavelet transform basis function (Haar basis) Basis intensities due to virtual source l

27. 27/45 Basis intensities due to virtual source l Many of B ij are nearly zero. Discarding if |B il | <  (  : threshold) Precompuation of Basis Intensities

28. 28/45 Basis intensities due to virtual source l Many of B ij are nearly zero. Discarding if |B il | <  (  : threshold) Precompuation of Basis Intensities n l << n cld Reducing memory requirement Accelerating intensity calculation

29. 29/45 Overview Introduction Previous Work Basic Idea of Our Method Precomputation of Basis Intensities Real-time Rendering Process Examples Conclusions & Future Work

30. 30/45 Real-time Rendering Process (1/7) Calculation of intensities of virtual source generating lightningplacing point sources on strokes distributing to neighboring grid points

31. 31/45 Real-time Rendering Process (2/7) Calculation of intensity of virtual source virtual source l intensity e l

32. 32/45 Real-time Rendering Process (3/7) Weighted sum of basis intensities basis intensities  e1e1 e2e2 e3e3 e4e4 final image

33. 33/45 Real-time Rendering Process (4/7) Final intensity I i at grid point i for clouds number of virtual sources with non-zero intensities intensity I i grid point i virtual source l intensity e l basis intensity

34. 34/45 Real-time Rendering Process (5/7) Intensity I i at grid point i for clouds basis intensity coefficient after wavelet transform number of virtual sources with non-zero intensities basis intensity

35. 35/45 Real-time Rendering Process (6/7) Intensity I i at grid point i for clouds 1. Calculation of H k 2. Inverse wavelet transform wavelet transform of intensity distribution of clouds

36. 36/45 with wavelet Real-time Rendering Process (7/7) Calculation of Intensities at all grid points for clouds without wavelet (k = 1, 2,…, n l ) (i = 1, 2,…, n cld ) n l << n cld (1/10) + inverse wavelet transform O(n cld )

37. 37/45 Overview Introduction Previous Work Basic Idea of Our Method Precomputation of Basis Intensities Real-time Rendering Process Examples Conclusions & Future Work

38. 38/45 Examples Simulation condition simulation space 20x20x7 Memory requirement for basis intensities  without wavelet compression ： 1137.5 MB  with wavelet compression ： 86 MB compression to 1/13 intensity calculation ： 32x32x4 cloud volume: 128x128x16

39. 39/45 Previous Method vs. Proposed Method same visual quality previous method[Max94] proposed method

40. 40/45 Previous Method vs. Proposed Method 0 10% previous proposed difference image (100 % = 255 difference in intensity)

41. 41/45 Previous Method vs. Proposed Method Computation time  Precomputation: 60 min (proposed method only)  Real-time process: computer: Pentium Ⅳ 3.5GHz, GeForece 7800 GTX 35 times faster 提案手法 0.2 秒 Proposed method 0.06 sec. Previous method 2.12 sec.

42. 42/45 Various examples

43. 43/45 Demo

44. 44/45 Animation

45. 45/45 Conclusions Real-time rendering of clouds illuminated by lightning  Taking into account multiple scattering  Precomputation of basis intensities  Acceleration by using Haar wavelet transform  16 frames per second on CPU

46. 46/45 Future Work Further acceleration using GPU Real-time rendering of realistic rain Dynamic clouds

47. 47/45 Basic Idea of Our Method Process flow intensity calculation due to virtual point source wavelet transform basis intensities generating lightning intensity of virtual point source weighted sum of basis intensities inverse wavelet transform display (volume rendering) preprocess real-time process

48. 48/45 Real-time Rendering Process Intensity I i at grid point i for clouds Haar basis coefficient after WT n l << n cld Reducing memory requrement Accelerating intensity calculation 1. Calculation of H k 2. Inverse WT

49. 49/45 Single Scattering vs. Multiple Scattering single scattering onlywith multiple scattering

50. 50/45 Previous Method vs. Proposed Method difference image previous proposed

51. 51/45 Precompuation of Basis Intensities Intensity calculation due to virtual point source virtual source l grid point i : intensity at grid point i due to virtual source l b il (b 1l, b 2l, …, b n cld, l ) n cld wavelet transform (B 1l, B 2l, …, B n cld, l ) : number of grid points for clouds b il calculation of multiple scattering

52. 52/45 Introduction (3/3) Clouds illuminated by lightning  Single scattering only  Not real-time Multiple scattering  Enhancing reality  High computation cost Difficult to achieve real-time rendering clouds viewpoint lightning

53. 53/45 Introduction (2/3) Example image glow around strokes shape of lightning clouds illuminated by lightning

54. 54/45 Single Scattering vs. Multiple Scattering Clouds become brighter single scattering only with multiple scattering

55. 55/45 Previous Work on Fast Rendering of Clouds/Smoke Use of GPU for fast computation of multiple scattering [Daubert03] [Szimary05]  Precomputing attenuation of light between all pairs of points  Sun light only (computational cost is proportional to number of light sources. )  Lightning is not considered.

56. 56/45 Subdivision of Simulation Space Contribution of virtual point source clouds simulation space contribution ： large virtual point source contribution ： small higher resolution lower resolution