This project is the open-source implementation of the paper "CDE-Learning: Camera Deviation Elimination Learning for Unsupervised Person Re-identification". Unsupervised person re-identification aims to identify the same person from non-overlapping camera views without annotated data. However, camera attribute differences (such as contrast, saturation, and resolution) often lead to deviations in target features, affecting recognition accuracy. The CDE-Learning method proposed in this project is dedicated to solving this problem through innovative module designs, effectively improving the performance of unsupervised person re-identification.
email for checkpoints " 875508913@qq.com"