search

seongmin-kye / MCT

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

some issues about the result #15

Open allenhaozhu opened 2 years ago

allenhaozhu commented 2 years ago

Hi: I am trying to follow your guide in the README and train your model in miniImagenet as the following cmd: python train.py --is_train True --gpu 0 --metric instance --transductive True --flip True --drop True --n_shot 1 --n_train_class 15 Episode: 0, Train Loss: 7.484041 Episode: 100, Train Loss: 5.504799 Episode: 200, Train Loss: 5.511789

Episode: 300, Train Loss: 5.448573 Episode: 400, Train Loss: 5.385933 Episode: 500, Train Loss: 5.263885 Episode: 600, Train Loss: 5.280904 Episode: 700, Train Loss: 5.208240 Episode: 800, Train Loss: 5.384327 Episode: 900, Train Loss: 5.381695 Episode: 1000, Train Loss: 5.335180 Evaluation in Validation data Accuracy 1-shot =28.45% Evaluation in Test data

So, is there anything wrong with my condition? Could you give me some clues about that?