#6464th of 70 in Open models
mambavision
Hybrid Mamba-Transformer vision backbone with pretrained ImageNet checkpoints
- Stars
- 2.3k
- License
- custom license
- Last commit
- Mar 2026
- Last release
- Jul 2025
- Language
- Python
2,108 on 21 Apr 20262,256 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Official PyTorch implementation of MambaVision, a hierarchical vision backbone that mixes Mamba-style mixer blocks with self-attention. It ships ImageNet-1K and 21K classification checkpoints from 31.8M to 739.6M parameters, loadable via the `mambavision` pip package or Hugging Face with `trust_remote_code`. Code and models for COCO detection (Cascade Mask R-CNN) and ADE20K segmentation (UPerNet) are included.
Who it is for: Computer vision researchers and engineers needing a pretrained backbone
Strengths
- Checkpoints from 31.8M to 739.6M params; L3-512-21K reports 88.1% ImageNet top-1
- Accepts any input resolution; returns 4-stage multi-scale features for downstream tasks
- Loads via pip package or Hugging Face; Colab notebook provided
- Includes detection and segmentation configs, logs and checkpoints
Weaknesses
- Vision backbone only; no chat, serving API or UI
- License is not a standard SPDX identifier; business use goes through NVIDIA licensing
- Examples assume CUDA; CPU or Apple Silicon use is not documented
- Detection and segmentation live in separate subfolders with their own setup
What it needs
- GPU required
- Docker
- Needs PyTorch, timm, transformers
- Models: PyTorch .pth.tar checkpoints, Hugging Face transformers
Quick answers
Is mambavision open source?
Its code is public on GitHub under its own license terms; read them before you use it at work.
Can you run mambavision with Docker?
Yes. Its repository has a Docker setup: a Dockerfile, a Compose file or a published image.
Does mambavision need a GPU?
Yes, it needs a GPU.
Also in Open models
See all 70| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 94 out of 100 |
| 2 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 88 out of 100 |
| 3 | rf-detrReal-time transformer models for object detection, segmentation and keypoints | 79 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 78 out of 100 |
| #5 | yolov5PyTorch YOLOv5 object detection, segmentation and classification models with training code | 76 out of 100 |