Ruifeng-Zhou24/CompactCNN_for_FER

Some problems I run into while using your model

Opened this issue · 0 comments

agil27 commented
  1. There should be a BatchNorm2d after every convolution, right? or the model won't work.
  2. Why are you using VGG19 instead of your own model in the training?
  3. Why are you re-inventing all the transforms defined in torchvision? it's perplexing
  4. WIl a Adam optimizer simply work better than your complicated SGD optimizer?