To reproduce the ProtoNet results for (n_way = 5, n_shot=5):
python train.py --exp_id protonet.exp_im1 --model protonet --gpu 0
The result is 65.13% (valid), 66.52% (test)
To reproduce the ProtoNet results for (n_way = 5, n_shot=1):
python train.py --exp_id protonet.exp_im2 --model protonet --gpu 0
The result is 48.12% (valid), 48.91% (test)
mini-imagenet dataset pretrain
python train.py --pretype train --model dve --exp_id dve.im --gpu 1 --pretrain_dir saved_models/dve
train the model
python train.py --type train --model dve --exp_id dve.im --gpu 1 --pretrain_dir saved_models/dve