Closed AFangCCC closed 3 years ago
It is a commonly used trick in unsupervised ReID. Re-ranking can improve the similarity between samples, leading to a more reliable similarity. Therefore, the quality of generated pseudo-labels can be improved.
It is a commonly used trick in unsupervised ReID. Re-ranking can improve the similarity between samples, leading to a more reliable similarity. Therefore, the quality of generated pseudo-labels can be improved.
Thanks
Firstly, thanks for opening source your code! I notice that the dist matrix is processed by re_ranking() function before cluster instead of original dist matrix. But I don't understand the reason.