README.md

---
base_model:
- syscv-community/sam-hq-vit-large
language:
- en
license: apache-2.0
pipeline_tag: image-segmentation
tags:
- feature-matching
- computer-vision
- co-visibility
---

# SAMatcher: Co-Visibility Modeling with Segment Anything for Robust Feature Matching

SAMatcher is a feature matching framework that formulates correspondence estimation through co-visibility modeling. Built upon the Segment Anything Model (SAM), it introduces a symmetric cross-view interaction mechanism that enables bidirectional feature exchange and cross-view semantic alignment to improve correspondence under large viewpoint, illumination, and scale variations.

- **Paper:** [SAMatcher: Co-Visibility Modeling with Segment Anything for Robust Feature Matching](https://huggingface.co/papers/2606.03406)
- **Authors:** Xu Pan, Qiyuan Ma, Jintao Zhang, He Chen, Xianwei Zheng
- **GitHub:** [https://github.com/TwSphinx54/SAMatcher](https://github.com/TwSphinx54/SAMatcher)
- **Project Page:** [https://xupan.top/Projects/samatcher](https://xupan.top/Projects/samatcher)

## Quick Start

### Installation

The project uses `uv` for dependency management. Recommended Python version is 3.12.

```bash
# Clone the repository
git clone https://github.com/TwSphinx54/SAMatcher.git
cd SAMatcher

# Install dependencies using uv
uv sync --frozen
```

### Download Weights

You can download the model weights using the Hugging Face CLI:

```bash
huggingface-cli download SSSSphinx/SAMatcher --local-dir ./weights/SAMatcher
```

Then follow the official training/evaluation scripts in the GitHub repository.

## Evaluation

To evaluate on MegaDepth, use the provided script:

```bash
bash scripts/evaluate_megadepth.sh
```

## Citation

If you find SAMatcher useful in your research, please consider citing:

```bibtex
@article{pan2024samatcher,
  title={SAMatcher: Co-Visibility Modeling with Segment Anything for Robust Feature Matching},
  author={Pan, Xu and Ma, Qiyuan and Zhang, Jintao and Chen, He and Zheng, Xianwei},
  journal={arXiv preprint arXiv:2606.03406},
  year={2024}
}
```