Redner is a differentiable renderer. It takes a 3D scene, including geometry, materials, camera, light sources, represented by PyTorch/TensorFlow tensors, and outputs an image, also represented as a PyTorch/TensorFlow tensor. It provides necessary machinery for correctly 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 (tzumao@mit.edu) or Github issues.

How to load an object and render it in redner.
PyTorch TensorFlow

How to optimize the pose of an object using redner.
PyTorch TensorFlow

How to do local lighting using deferred rendering in redner.
PyTorch TensorFlow

How to blend the rendering output with a background image, and why you should do it in the linear color space.
PyTorch TensorFlow

How to do physically-based rendering using path tracing in redner.
PyTorch TensorFlow

Redner's material and light source models.
PyTorch TensorFlow

Redner's camera models.
PyTorch TensorFlow

Batch rendering in redner.
PyTorch TensorFlow

How to fit a face to a target image using a PCA-based 3D morphable model.
This tutorial doesn't run on Colab since you'll need to agree with license terms of the Basel face model.
PyTorch TensorFlow

https://redner.readthedocs.io/en/latest/

See the sidebar.