Below is a list of resources related to Differentiable Rendering
Kaolin - A PyTorch 3D Deep Learning library by NVIDIA, includes several differentiable renderers and other graphics algorithms.
Tensorflow Graphics - A Tensorflow 3D Deep Learning library. Similar in scope (or perhaps ambition) to Kaolin, but isn't as far along as in development.
Redner - A differentiable monte carlo raytracer based on the Mitsuba renderer. Supports pytorch and Tensorflow. Integrates into kaolin.
DIRT - A differentiable renderer for Tensorflow that uses OpenGL's rasterizer
Neural 3D Mesh Renderer - Implements rasterization for the not-super-popular Chainer library. Has been ported to pytorch, and is included in Kaolin.
OpenDR - OpenDR is basically the original differentiable renderer. It also uses OpenGL, but doesn't interface well with any deep learning packages (afaik).
These libraries may be useful for implementing renderers or simulations
Taichi - Taichi is a library intended for implementing computer graphics applications. It is written in C++, but has python bindings, and supports automatic differentiation. Mitsuba2 is (reportedly) implemented to support automatic differentiation using Taichi.
Jax - Jax is a (seemingly) similar project to Taichi, though does not purport to be for computer graphics. They describe it as "Autograd and XLA, brought together for high-performance machine learning research."
Tim Zhang's 3D Deep Learning Paper List - Contains a list of most 3D Deep Learning papers (hasn't been updated in a while, though)