lukemelas / EfficientNet-PyTorch

A PyTorch implementation of EfficientNet
Apache License 2.0
7.85k stars 1.52k forks source link

hard to converage without pretrain model on arcface #193

Open carry-xz opened 4 years ago

carry-xz commented 4 years ago

hi, thanks for your nice work When I train a face recongnition model , use EfficientNet-B0 as feature extractor, I find it's hard to convergence without pretrained model, and use pretrained model it's look like all right . I just wander why?

lukemelas commented 4 years ago

Training EfficientNets is generally very finicky, sorry to hear you're having trouble with arcface.