search

seongmin-kye / MCT

Pytorch implementation of Meta-Learned Confidence for Few-shot Learning
71 stars 12 forks source link

model could not convergent #4

Closed jjn037 closed 4 years ago

jjn037 commented 4 years ago

I set input image size to 224, and self.fc1_pt = nn.Linear(144, 1), but the model could not convergent。Could you give me some advice about the problem

seongmin-kye commented 4 years ago

Hmm,,,, I didn't use 224x224 image. But, I think that problem may occurs in dense classification. So, if you use more layers with max-pooling or 2 stride Conv to make compact informative feature map, dense classification could extract image inherent information.

jjn037 commented 4 years ago

thx, may be I make some mistakes ,trying to make it