metadata
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
- Authors: Xu Pan, Qiyuan Ma, Jintao Zhang, He Chen, Xianwei Zheng
- GitHub: https://github.com/TwSphinx54/SAMatcher
- Project Page: https://xupan.top/Projects/samatcher
Quick Start
Installation
The project uses uv for dependency management. Recommended Python version is 3.12.
# 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:
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 scripts/evaluate_megadepth.sh
Citation
If you find SAMatcher useful in your research, please consider citing:
@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}
}