Skip to content
hepiteau.
← All projects

3D · C++ / CUDA · Python

Sparse Voxels

- present · Pre-alpha · PyPI package

A sparse voxel octree toolkit for representing and rendering 3D data. It combines a C++20/CUDA core with Python bindings for spatial queries, ray casting, and volume rendering. PyTorch integration makes density and color fields differentiable, connecting my work in 3D graphics with learning-based reconstruction.

Source code ↗Documentation ↗
A sparse voxel sphere rendered with green, cyan, and purple color gradients
A sparse voxel sphere rendered with green, cyan, and purple color gradients. Illustration from the project repository.

What it does

  • Build sparse octrees from occupied voxel coordinates and query points or rays against them.
  • Render density and color payloads on the GPU, with Python access to the C++/CUDA core.
  • Optimize density and color tensors through PyTorch autograd while keeping tree topology separate from payload data.

Project status

This is an actively developed pre-alpha toolkit. The documentation explains the supported CPU and CUDA workflows, installation requirements, and the scope of differentiability.