sunset1995/DirectVoxGO

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1075 stars · 107 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

DirectVoxGO (DVGO) is a PyTorch implementation of Direct Voxel Grid Optimization for fast neural radiance field (NeRF) reconstruction, replacing NeRF's MLP with dense voxel grids for super-fast training convergence. It supports bounded, unbounded, and forward-facing scene datasets and includes a JIT-compiled CUDA extension for additional speedup.

Use cases

• reconstruct a 3D radiance field from posed images quickly
• speed up NeRF training by replacing the MLP with a voxel grid
• render fly-through videos from casually captured forward-facing scenes
• train novel view synthesis models on NeRF, NSVF, BlendedMVS, or mip-NeRF360 datasets
• evaluate novel view synthesis with SSIM and LPIPS metrics

When to choose

• you need fast NeRF-style training convergence (minutes rather than hours)
• you want a research baseline for voxel-based radiance field methods
• you have a CUDA GPU and posed multi-view images

When to avoid

• you need real-time rendering on low-end hardware without GPUs
• you want the latest radiance field methods like 3D Gaussian splatting
• you need a maintained production library - the repo is research code from CVPR 2022

Facets

library · maturity maintenance

machine-learning deep-learning graphics image-processing computer-vision graphics deep-learning machine-learning python nerf radiance-fields voxel-grid novel-view-synthesis cvpr2022 pytorch 3d-reconstruction linux gpu cuda

Member repositories

For agents

markdown · JSON · MCP: product_card(name="sunset1995/DirectVoxGO")