Source code for【ICCV2023】Prototypical Mixing and Retrieval-based Refinement for Label Noise-resistant Image Retrieval
CUB200, CARS196, CIFAR can be downloaded from https://paperswithcode.com/. And we provide CARS98N dataset in the /CARS_98N folder.
After modify the data path in the 'run.py', use this command in the terminal to train the retrieval model: 'python run.py --train', and the checkpoint, generated hash code will be stored
After training, modidy the checkpoint file path in the 'run.py', use this command in the terminal to evaluate the trained model: 'python run.py --evaluate'
If you find our work or codebase useful in your research, please cite:
@inproceedings{yang2023prototypical,
title={Prototypical Mixing and Retrieval-based Refinement for Label Noise-resistant Image Retrieval},
author={Yang, Xinlong and Wang, Haixin and Sun, Jinan and Zhang, Shikun and Chen, Chong and Hua, Xian-Sheng and Luo, Xiao},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={11239--11249},
year={2023}
}