ShikunLi / Sel-CL

CVPR 2022: Selective-Supervised Contrastive Learning with Noisy Labels
89 stars 17 forks source link

Question about knn evaluation on representations #1

Closed Hzzone closed 2 years ago

Hzzone commented 2 years ago

Thanks for your work! I wonder whether the clean labels has involved the knn evaluation for the results of Tables 2 and 4. Looking forward to your reply!

ShikunLi commented 2 years ago

Yes, we use clean data to evaluate representations in order to better display how algorithm is degraded when there are noisy labels.