pytorch-classification

pytorch-classification copied to clipboard

I have modified the imagenet.py by adding the lines after the "create model" codes: num_ftrs = model.fc.in_features model.fc = nn.Linear(num_ftrs,200) changing the net to train on tiny imagenet. But it has a problem: training top1 accuracy is about 79.6110 and top5 accuracy is about 91.6950; while validation accuracys are only 0.7(top1) and 1.47(top5). Do you have any ideas on this weird problem?

https://github.com/tjmoon0104/pytorch-tiny-imagenet/blob/master/val_format.py