Real-Time Caustics Using Cascaded Image-Space Photon Tracing

Abstract

Caustics are formed when transparent or specular materials focus many light rays onto confined areas. Such effects are among the most challenging optical effects to render. Typical path tracing algorithms need many light ray samples to form realistic caustic highlights and/or clever strategies to sample the most relevant light paths. This paper presents an approach for rendering caustics from one specular bounce at an interactive performance by using cascaded image-space photon tracing based on both G-buffer (camera perspective) and Reflective Shadow Maps (light source perspective). A denoiser is applied to remove Monte Carlo noise before applying the caustics texture. Our results demonstrate caustics with more realistic and precise details compared to state-of-the-art Caustics Mapping, at real-time frame rates even on moderately powerful GPUs.