The pytorch implementation of UGNCL: Uncertainty-Guided Noisy Correspondence Learning for Effective Cross-Modal Matching
We referred to the implementations of NCR and DECL to build up our codebase.
For the expanded version of UGCNL to journal, we will open TRAINING CODE after it's accepted.
If UGNCL is useful for your research, please cite the following paper:
@inproceedings{UGNCL,
author = {Zha, Quanxing and Liu, Xin and Cheung, Yiu-ming and Xu, Xing and Wang, Nannan and Cao, Jianjia},
title = {UGNCL: Uncertainty-Guided Noisy Correspondence Learning for Efficient Cross-Modal Matching},
year = {2024},
publisher = {Association for Computing Machinery},
booktitle = {Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval},
pages = {852–861},
numpages = {10},
doi = {10.1145/3626772.3657806}
}