README.md

---
license: apache-2.0
tags:
- computer-vision
- feature-matching
- onnx
- efficient-loftr
pipeline_tag: image-feature-extraction
---

# EfficientLoFTR ONNX Weights

This repository contains the ONNX-optimized weights for **EfficientLoFTR**, a model used for finding matching points between pairs of images. 

By converting the original PyTorch model weights into the ONNX format, these files allow you to run fast feature-matching inference on both CPU and GPU without needing to install the heavy PyTorch framework.

## Available Files
* **`eloftr_outdoor_full.onnx`**: The standard version of the model, optimized for the best matching quality.
* **`eloftr_outdoor_opt.onnx`**: An efficiency-focused version of the model, optimized for faster inference speed.

---

## How to Use

The easiest way to load and use these files is through the **[spatialhub](https://github.com/pankajkaushik12/spatialhub)** Python library.


## Original Citation
If you use these models in academic work, please cite the original authors:
```bibtex
@inproceedings{wang2022efficientloftr,
  title={EfficientLoFTR: Semi-Dense Local Feature Matching with Sparse Transformers},
  author={Wang, Yanzhao and Geng, Yuwei and Jiang, Zheng and Zhao, Yihong and Jin, Shisheng and Lin, Siyu and Han, Feng},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2022}
}
```

**Note**: This repository provides pre-converted weights for inference purposes.