search

seongmin-kye / MCT

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

code for CIFAR_FS and FC100 #2

Closed wchixq closed 4 years ago

wchixq commented 4 years ago

Hi, many thanks for the source code. It is very helpful. Do you plan to release the settings for these smaller datasets as well? I tried to use it for CIFAR-FS but the backbone output is too small for the length-scaling module.

Thanks a lot.

seongmin-kye commented 4 years ago

Thank you for your attention! As described in paper, we use max-pooling only in even layers for CIFAR-FS and FC100. And then if you change self.fc1_pt = nn.Linear(16, 1) >> self.fc1_pt = nn.Linear(36, 1), it will be working.

I will post additional codes as soon as they are organized.

wchixq commented 4 years ago

Thanks for your prompt reply! It works well.