# diffren

**Repository Path**: mirrors_google-research/diffren

## Basic Information

- **Project Name**: diffren
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Apache-2.0
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2022-09-02
- **Last Updated**: 2026-03-22

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Diffren: a library for differentiable 3D rendering

*This is not an official Google product*

Diffren provides tools for 3D rendering with derivatives. Diffren may be used
to optimize 3D surfaces and camera parameters under pixel-based error terms,
especially as part of ML model training. The rendering pipeline is a
deferred-shading, rasterization-based pipeline. Any JAX function may be used
as a shader. Diffren provides a small number of sample shaders including
texture mapping, diffuse, and point-based lighting. A key feature of Diffren
is its implementation of
[rasterize-then-splat](https://openaccess.thecvf.com/content/ICCV2021/papers/Cole_Differentiable_Surface_Rendering_via_Non-Differentiable_Sampling_ICCV_2021_paper.pdf),
an approach for generating smooth derivatives at occlusion boundaries.