Learning General Descriptors for Image Matching with Regression Feedback

demo and pre-trained weight of AANet --- a dense descriptor for local feature matching.

Our work was accepted by IEEE Transactions on Circuits and Systems for Video Technology 2023, and can be accessed via manuscript.

Pre-Trained Weights

We trained our AANet with one-stage end-to-end triplet training strategy on MS-COCO, Multi-illumination and VIDIT datasets (same as LISRD) and the pre-trained weight is compressed as dna.rar

Model file

The core implementation of AANet is shown in AANet_core.py

DEMO：SIFT+AANet

We provide the demo of exporting SIFT keypoints and AANet descriptor in export_descriptor_sift.py, and it can be easily modified to other off-the-shelf detectors and matchers for evaluation.

CUDA_VISIBLE_DEVICES=0 python export_descriptor_sift.py

For more evaluation details, please refer to the HIFT and LISRD

Citation

If you are interested in this work, please cite the following work:

@ARTICLE{10102528,
  author={Rao, Yuan and Ju, Yakun and Li, Cong and Rigall, Eric and Yang, Jian and Fan, Hao and Dong, Junyu},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={Learning General Descriptors for Image Matching With Regression Feedback}, 
  year={2023},
  volume={33},
  number={11},
  pages={6693-6707},
  doi={10.1109/TCSVT.2023.3267279}}

Acknowledgments

Our work is based on LISRD and we use their code. We appreciate the previous open-source repository LISRD