Pytorch implementation of ConvMatch for AAAI'23 paper "ConvMatch: Rethinking Network Design for Two-View Correspondence Learning", by Shihua Zhang and Jiayi Ma.

This paper focuses on establishing correspondences between two images. We design a correspondence learning network called ConvMatch that for the first time can leverage convolutional neural network (CNN) as the backbone to capture better context, thus avoiding the complex design of extra blocks. Specifically, with the observation that sparse motion vectors and dense motion field can be converted into each other with interpolating and sampling, we regularize the putative motion vectors by estimating dense motion field implicitly, then rectify the errors caused by outliers in local areas with CNN, and finally obtain correct motion vectors from the rectified motion field.

This repo contains the code and data for essential matrix estimation described in our AAAI paper.

If you find this project useful, please cite:

@inproceedings{zhang2023convmatch,
  title={ConvMatch: Rethinking Network Design for Two-View Correspondence Learning},
  author={Zhang, Shihua and Ma, Jiayi},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2023}
}

Please use Python 3.6, opencv-contrib-python (3.4.0.12) and Pytorch (>= 1.7.0). Other dependencies should be easily installed through pip or conda.

For a quick start, clone the repo and download the pretrained model.

git clone https://github.com/SuhZhang/ConvMatch 
cd ConvMatch 

Then download the pretrained models from here.

Then run the feature matching with demo ConvMatch.

cd ./demo && python demo.py

First download YFCC100M dataset.

bash download_data.sh raw_data raw_data_yfcc.tar.gz 0 8
tar -xvf raw_data_yfcc.tar.gz

Download SUN3D testing (1.1G) and training (31G) dataset if you need.

bash download_data.sh raw_sun3d_test raw_sun3d_test.tar.gz 0 2
tar -xvf raw_sun3d_test.tar.gz
bash download_data.sh raw_sun3d_train raw_sun3d_train.tar.gz 0 63
tar -xvf raw_sun3d_train.tar.gz

Then generate matches for YFCC100M and SUN3D (only testing) with SIFT.

cd ../dump_match
python extract_feature.py
python yfcc.py
python extract_feature.py --input_path=../raw_data/sun3d_test
python sun3d.py

Generate SUN3D training data if you need by following the same procedure and uncommenting corresponding lines in sun3d.py.

We provide the model trained on YFCC100M and SUN3D described in our AAAI paper. Run the test script to get similar results in our paper (the generated putative matches are different once regenerating the data).

cd ./test 
python test.py

You can change the default settings for test in ./test/config.py.

After generating dataset for YFCC100M, run the tranining script.

cd ./core 
python main.py

You can change the default settings for network structure and training process in ./core/config.py.

The provided models are trained using SIFT. You had better retrain the model if you want to use ConvMatch with your own local feature, such as RootSIFT, SuperPoint and etc.

You can follow the provided example scirpts in ./dump_match to generate dataset for your own local feature or data.

This code is borrowed from OANet and CLNet. If using the part of code related to data generation, testing and evaluation, please cite these papers.

@inproceedings{zhang2019learning,
  title={Learning two-view correspondences and geometry using order-aware network},
  author={Zhang, Jiahui and Sun, Dawei and Luo, Zixin and Yao, Anbang and Zhou, Lei and Shen, Tianwei and Chen, Yurong and Quan, Long and Liao, Hongen},
  booktitle={Proceedings of the IEEE/CVF international conference on computer vision},
  pages={5845--5854},
  year={2019}
}
@inproceedings{zhao2021progressive,
  title={Progressive correspondence pruning by consensus learning},
  author={Zhao, Chen and Ge, Yixiao and Zhu, Feng and Zhao, Rui and Li, Hongsheng and Salzmann, Mathieu},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={6464--6473},
  year={2021}
}