Rendervous is a project designed primarily for academic purposes. The core idea is the integration of GPU rendering capabilities including ray-tracing (Vulkan) with deep learning (Pytorch).
As a result you have a differentiable renderer that can be included in learning models and vice versa, learning models that can be included as renderer components (materials, scattering functions, parameters, etc).
- torch
- cffi
- pywin32 (Windows)
- cuda-python (if cuda device can be used)
- matplotlib (most of the offline-rendering examples)
- imgui
- glfw (for interactive examples)
|
Introducing rendervous: Creating maps in rendervous. Manipulating vectors and operations. open in colab |
|
Differentiability: Testing differentiability with simple regressions. open in colab |
|
Simple MLP: Chaining maps to build an MLP. open in colab |
|
Sensors: Examples of sensors. open in colab |
|
Transmittances: Example of transmittance computation through a volume. open in colab |
|
Reconstruction: Example of volume reconstruction from transmittance. open in colab |