GTC 2020 S21852
Presenters: Thomas Muller,NVIDIA
Abstract
Neural network-based techniques have taken many fields by storm, but until recently have seen relatively little use in the field of physically-based rendering. This has begun to change. We’ll present techniques for accelerating Monte Carlo integration of light transport without introducing bias by utilizing functions learned by neural networks for variance reduction. Our techniques yield on-par or higher performance than competing machine learning-based techniques at equal sample counts and generalize beyond physically-based rendering, being applicable to other high-dimensional integration problems such as Bayesian inference and reinforcement learning.
Watch this session
Join in the conversation below.