350 likes | 499 Views
Microsoft Research Asia. Precomputed Shadow Fields for Dynamic Scenes. Kun Zhou Yaohua Hu Stephen Lin Baining Guo Heung-Yeung Shum. 1. Outline. • Introduction • Method Overview • Detail • Results. • Introduction • Method Overview • Detail • Results. Introduction. Soft Shadow:
E N D
Microsoft Research Asia Precomputed Shadow Fields for Dynamic Scenes Kun Zhou Yaohua Hu Stephen Lin Baining Guo Heung-Yeung Shum 1
Outline • Introduction • Method Overview •Detail • Results
• Introduction • Method Overview • Detail • Results
Introduction Soft Shadow: multiple shadow maps -[Heckbert and Herf] 1997 Exntnd shadow volume -[Assarsson and Akenine-Moller] 2003 Fake approach -[Chan and Durand] 2003
Introduction Why this is not good? For a dynamic scene: -scene complexity -recomputed for each frame For illuminants: -point-like or small light source 5
Introduction PRT (Precomputed Radiance Transfer): based on two main assumptions: 1.all objects in the scene are non-emitters 2.the incoming light direction is independent with the position of a point x 6
Introduction PRT (Precomputed Radiance Transfer): These spherical functions can be compressed by using some basis functions: Spherical harmonics Wavelet 7
• Introduction • Method Overview • Detail • Results
Method Overview Compute Incoming radiance of each point 11
Method Overview Compute Occlusion Field for each Object 12
Method Overview Compute Occlusion Field for each Object 13
Method Overview Compute Occlusion Field for each Object 14
Method Overview Compute Occlusion Field for each Object 15
Method Overview Compute radiance field for each Light Source 16
Method Overview Compute Radiance Field for each Light Source 17
Method Overview multiply these and self-visibility at the point together => Incident radiance distribution 18
Method Overview Combine with the BRDF to get the final radiance at the point 19
Method Overview Combine with the BRDF to get the final radiance at the point 20
• Introduction • Method Overview •Detail • Results
Detail SRF and OOF Precomputation 22
Detail Data Sampling For low frequency shadow: 16 concentric spheres For all frequency shadow: 32 concentric spheres Intermediate points: trilinear interpolation with 8 nearest samples 23
Detail Data Compression How? approximate the function using basis functions 24
Detail Data Compression For low frequency shadow: Spherical harmonics 25
Detail Data Compression Spherical harmonics reconstruction 26
Detail Data Compression Spherical harmonics has two properties: 1.Rotationaly invariant 2.Integrating the product of two SH functions = a dot product of their coefficients. 27
Detail Data Compression For all frequency shadow: Wavelet Basis 28
Detail Data Compression Spherical harmonics VS Wavelet 29
Detail Incident Radiance Computation 30
Detail Soft Shadow Rendering Acceleration -Culling -Lazy Occlusion Updating occluders for a specific light source remain unchanged, the cached visibility can be reused 31
• Introduction • Method Overview • Detail • Results