facebookresearch/SparseConvNet

Health v2 · maintenance only

Flags: no_releases archived no_license

Adoption not part of the score

2145 stars · 334 forks observed · 2026-08-28

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

A PyTorch library for training Submanifold Sparse Convolutional Networks, providing spatially-sparse convolutions that operate efficiently on sparse input data. It supports 1-10 dimensional inputs and enables computationally efficient VGG/ResNet/DenseNet-style networks where active sites remain unchanged across layers.

Use cases

• train sparse convolutional networks on 3D point clouds
• build efficient ConvNets on sparse high-dimensional data
• process handwriting or line-like 2D data with sparse convolutions
• run semantic segmentation on LiDAR scans
• apply sparse convolutions to space-time (3+1)D data

When to choose

• your input data is spatially sparse, such as point clouds or sparse voxel grids
• you need submanifold convolutions to keep active-site counts constant across layers
• you want a PyTorch-integrated sparse convolution library

When to avoid

• you need active maintenance or new feature development, as the repository was archived in March 2025
• your data is dense and standard dense convolutions are efficient enough
• you need a library with broad community support and recent updates

Facets

library · maturity abandoned

machine-learning deep-learning image-processing deep-learning machine-learning computer-vision python sparse-convolution pytorch submanifold-convolution 3d-point-clouds convolutional-networks linux macos gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/SparseConvNet")