Giters

fdbtrs / CR-FIQA

Official repository for CR-FIQA: Face Image Quality Assessment by Learning Sample Relative Classifiability

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

inference problem

carry-xz opened this issue · comments

commented

nice work！
I double that Why dropout=0.4 in training but not in inference？
QualityModel._get_model:
backbone = iresnet50(num_features=512, qs=1, use_se=False).to(f"cuda:{ctx}")

commented

It is NOT a problem. Dropout is only applied during the training mode. In the evaluation model, dropout is not applicable.