Home

Redner is a differentiable physically-based Monte Carlo renderer. It takes a 3D scene, including geometry, materials, camera, light sources, represented by PyTorch tensors, and outputs an image, also represented as a PyTorch tensor. It provides necessary machinery for propagating the gradients of the output image to the scene parameters. For the theory behind redner, please consult our paper "Differentiable Monte Carlo Ray Tracing through Edge Sampling". This page is a tutorial for using redner. It is still work in progress. Please let us know what to improve through email ([email protected]) or Github issues.

Installation

Installation Troubleshooting

Tutorial 1: optimizing a single triangle

Tutorial 2: pose estimation

Tutorial 3: materials and textures (WIP)

Tutorial 4: fast deferred rendering

Tutorial 5: coarse-to-fine estimation

Tutorial 6: adversarial examples (WIP)