search

zyh-uaiaaaa / Erasing-Attention-Consistency

Official implementation of the ECCV2022 paper: Learn From All: Erasing Attention Consistency for Noisy Label Facial Expression Recognition
78 stars 15 forks source link

Help reproduce rafdb #20

Closed chenhao-user closed 1 year ago

chenhao-user commented 1 year ago

thanks for your great work! I can reproduce the accuracy of the rafdb dataset when there is label noise. But I can't reproduce the SOTA results of rafdb（89.99）, can you provide some details of the experiment, such as hyperparameters and random number seed settings?

zyh-uaiaaaa commented 1 year ago

I have re-run my code without making any changes. Using the pre-trained ResNet-50 model, it easily achieved an accuracy of 90.35% and the best accuracy recorded was 90.51%, without any label noise. Could you please either re-run my code or provide me with your log file to assist me in identifying the reason behind this?

chenhao-user commented 1 year ago

It's my mistake. I previously modified the batch size, which resulted in the previous replication of ResNet-18 yielding only 83% accuracy. Now, I have successfully replicated the accuracy of both networks. Thanks!