View-dependent Radiance Caching

Radiance caching is used to accelerate global illumination computations, exploiting the spatial coherence of indirect illumination on surfaces. We propose a new radiance caching approach capable of more correctly reconstructing inter-reflections between glossy surfaces, all while improving performance compared to previous approaches. Contrary to previous works, our view-dependent radiance caching scheme does not heavily rely fundamentally on basis-space representations such as spherical harmonics, and can directly treat outgoing radiance at surfaces instead of incoming radiance distributions. We introduce a new view-dependent record placement strategy and adapt recent Hessian-based error metrics to our view-dependent records [8]. To do so, we derive and compute more accurate derivatives of radiance at surfaces in the scene.