hustvl/Vim

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

3913 stars · 288 forks observed · 2026-10-17

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Vision Mamba (Vim) is a PyTorch implementation of a generic vision backbone built on bidirectional Mamba state space models, published at ICML 2024. It replaces self-attention with efficient SSMs for visual representation learning, achieving better speed and memory efficiency than vision transformers like DeiT on high-resolution images.

Use cases

• train a vision backbone without self-attention
• classify images on ImageNet with Mamba blocks
• extract features from high-resolution images efficiently
• use a Mamba backbone for object detection on COCO
• semantic segmentation on ADE20k with state space models
• research alternatives to vision transformers

When to choose

• you need memory-efficient feature extraction on high-resolution images
• you are researching state space models for computer vision
• you want a DeiT alternative with lower GPU memory usage

When to avoid

• you need a battle-tested production backbone with broad ecosystem support
• you require pretrained weights for many downstream tasks out of the box
• your project depends on standard transformer tooling

Facets

library · maturity active

machine-learning deep-learning image-processing computer-vision deep-learning machine-learning python mamba state-space-models vision-backbone image-classification object-detection semantic-segmentation research gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="hustvl/Vim")