Real-Time, All-Frequency Shadows in Dynamic Scenes Thomas Annen * Zhao Dong * Tom Mertens † Philippe Bekaert † Hans-Peter Seidel * Jan Kautz ‡ *MPI Informatik Germany † Hasselt University tUL - IBBT, EDM, Belgium ‡ University College London UK Some materials are get from the author and this paper is presented by CG, Huang
Outline Introduction Relate work Convolution Generation of Area Lights for Environment Maps Limitations Result
Outline Introduction Relate work Convolution Generation of Area Lights for Environment Maps Limitations Result
Introduction Enable real-time, all-frequency shadows in dynamic scenes. Support area light as well as wnviroment lighting. The key contribution is renderng plausible soft shadow. Enviroment-lit scenes can be rendered.
Outline Introduction Relate work Convolution Generation of Area Lights for Environment Maps Limitations Result 1.Soft Shadows 2.Convolution 3.Precomputation and Simplification 4.Environment map sampling
Soft shadows Early work on shadow mapping extensions image-based rendering to average hard shadow.[Chen and Williams 93; Agrawala et al. 00] Classic shadow volumn method was extended to soft shadows.[Assarsson and Akenine- Moller 03]
Convolution Soler and Sillion[98] propose an image-based shadow algorithm based on convolution. Don’t support self-shadowing. Variance shadow maps[Donnelly and Lauritzen 06] Convolution shadow maps[Annen et al. 07]
Precomputation and simplification PRT [Sloan et al. 02] calculate and stroes an illumination-invariant transport solution off- line and uses it for real-time relighting. Challenging to support fully dynamic scenes with arbitrary illumination.
Environment map sampling Agarwal et al.[03] proposed an efficient point sampling strategy for environment maps. Arbree et al. Use disk-shaped light sources to approximation. This paper approximate an environment with a collection of square light sources.
Outline Introduction Relate work Convolution Generation of Area Lights for Environment Maps Limitations Result
c L Convolution shadow map x R 3 p R 2 P = T(x) Shadow function: s(x):=f(d(x),z(p)) Binary result: – 1 if d(x)<=z(p) – 0 else x p d(x)d(x) z(p)z(p)
Shadow test function: s(x) What kind of function is s(x) ? Heaviside Step Function: H(t) Shadow term for x’ c L x p d(x’)d(x’) z(p)z(p) x’x’
Approximate shadow test with Fourier series Convolution shadow map c1c1 +c c c c 16
Convolution shadow map Step function becomes sum of weighted sin() Series is separable! c1c1 +c c c c 16
Convolution Bulid on convolution-based methods. Simulate penumbrae by filtering shadows depending on the configuration of blocker, receiver, and light source.
CSM order reduction Annen et al[07] using a Fourier series to construct the f, but it’s prone to some artifacts and shadows at contact points may too bright.
Outline Introduction Relate work Convolution Generation of Area Lights for Environment Maps Limitations Result
Generation of Area Lights for Environment Maps
Outline Introduction Relate work Convolution Illumination with Soft Shadows Limitations Result 1.Ringing Suppression 2.Textured light sources
Outline Introduction Relate work Convolution Generation of Area Lights for Environment Maps Limitations Result Conclusions and Future work 1. DirectX Dual-Core AMD 2.2GHz 3. NVIDIA GeForce 8800 GTX graphics card
Result Buddha scene with 70k face MM: Mipmaps SAT: Summed area table
Result
Performance of this paper and image quality depend on: – choice of prefilter – Number of area lights – Shadow map size
Result Demonstrate the effect of the sharpening function G().
Result Shows the influence of the number of light sources used for approximating the environment map.
Outline Introduction Relate work Convolution Generation of Area Lights for Environment Maps Limitations Result Conclusions and Future work 1.Based on convolution. 2.Fast enough to render many area light sources simul- taneously. 3.Provide plausible results, even though they are not entirely physically correct. 4.At future work, intend to explore the use area lights for indirect illumination.