Feb 19 (Fri): "Deep Learning for Variance Reduction in Monte Carlo Rendering," Steve Bako, ECE PhD Defense
Physically based rendering is widespread due to its ability to create compelling, photorealistic images. Generating these images requires evaluating a multidimensional integral and is typically done through Monte Carlo (MC) samples of the scene's light transport. However, these samples are computationally costly and many are needed to obtain a converged result, otherwise there is objectionable noise in the final image. We propose to apply machine learning at three key locations of the rendering pipeline: pre-rendering, intra-rendering, and post-rendering to avoid the inaccurate approximations, costly initialization, and fallible heuristics of previous MC variance reduction techniques.
For pre-rendering, we introduce the first deep network for prefiltering that can accurately capture the appearance of large environments with complex geometry and materials for level of detail. Within the context of intra-rendering, we demonstrate the first offline deep network that can guide sampling for almost the entirety of rendering for an arbitrary scene. Finally, we bring deep learning to post-rendering MC denoising where our network predicts the filter kernel around each pixel, can even directly filter the image, and works on general scenes and varying levels of noise without retraining.
Steve Bako is a PhD candidate in the ECE department and is advised by Professor Pradeep Sen. His main research interests include computer graphics, computational photography, and machine learning. Specifically, he has worked on bringing deep learning to various physically-based rendering applications.
Hosted by: Pradeep Sen (Chair)
Submitted by: Steve Bako <email@example.com>