# NVlabs/MambaVision

[CVPR 2025] Official PyTorch Implementation of MambaVision: A Hybrid Mamba-Transformer Vision Backbone

Repository: https://github.com/NVlabs/MambaVision
Canonical: https://ross.abutalabs.com/products/mambavision
Homepage: https://arxiv.org/abs/2407.08083
Language: Python
License: NOASSERTION
License Family: other
Topics: deep-learning, foundation-models, image-classification, mamba, self-attention, vision-transformer, visual-recognition, hybrid-models, huggingface-transformers, transformers, instance-segmentation, object-detection, semantic-segmentation
Last push: 2026-03-11T20:34:23+00:00

## Health v2 (maintenance only)
Score: 49/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 71, release rhythm 16, longevity 58
- inputs: {"age_days": 814, "days_push": 175, "days_rel": 407, "gap_med": 115, "n_releases_24m": 2}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2224, forks 149 (observed 2026-08-28T04:06:27.825494+00:00)

## What it is
MambaVision is NVIDIA's official PyTorch implementation of a hybrid Mamba-Transformer vision backbone, published at CVPR 2025. It provides pretrained models achieving state-of-the-art accuracy-throughput tradeoffs for image classification and downstream tasks like object detection and segmentation.

## Use cases
- classify images with a high-throughput pretrained vision backbone
- extract visual features for object detection on COCO
- run semantic segmentation on ADE20K
- fine-tune a hybrid Mamba-Transformer model on my own dataset
- compare Mamba vs Vision Transformer backbones
- use a mamba-based vision model at ImageNet-21K scale

## When to choose
- you need a fast, accurate image classification backbone with strong throughput
- you want a hybrid Mamba + self-attention architecture for vision research
- you need pretrained weights for detection or segmentation downstream tasks
- you want to experiment with state-space models in computer vision

## When to avoid
- you need a permissively licensed model - it uses the NVIDIA Source Code License-NC (non-commercial)
- you need a lightweight model for CPU-only or edge deployment
- you need a general-purpose NLP or multimodal model rather than a vision backbone

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, image-processing, computer-vision
- domain: computer-vision, deep-learning, image-processing, artificial-intelligence
- platform: python, cross-platform
- tags: vision-backbone, mamba, vision-transformer, pytorch, image-classification, object-detection, semantic-segmentation, pretrained-models, nvidia, cvpr-2025, gpu

## Member repositories
- NVlabs/MambaVision (main) score 49

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:27.825494+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:44:56.925460+00:00, confidence not recorded.
  - readme: https://github.com/NVlabs/MambaVision (fetched 2026-08-28T04:06:27.825494+00:00, sha 6426e1b228ed)
  - homepage: https://arxiv.org/abs/2407.08083 (fetched 2026-08-29T10:25:42.584305+00:00, sha 88a59697cd95)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T10:25:42.593927+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T10:25:42.597445+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T10:25:42.599236+00:00, sha a1f16f915a9a)
  - registry_pypi: https://pypi.org/pypi/mambavision/json (fetched 2026-08-29T10:25:42.601500+00:00, sha bed970810160)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T10:25:42.595775+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.