RuntimeError: Attempting to deserialize object on CUDA device 1 but torch.cuda.device_count() is 1. Please use torch.load with map_location to map your storages to an existing device. about learningtocompare_fsl

[githubwys]

i met the same problem.

you have to assign a certain gpu.

find the file miniimagenet_train_one_shot.py
from :
#feature_encoder.load_state_dict(torch.load(str("./models/miniimagenet_feature_encoder_" + str(CLASS_NUM) +"way_" + str(SAMPLE_NUM_PER_CLASS) +"shot.pkl")))
to :
feature_encoder.load_state_dict(torch.load(str("./models/miniimagenet_feature_encoder_" + str(CLASS_NUM) +"way_" + str(SAMPLE_NUM_PER_CLASS) +"shot.pkl"), map_location='cuda:0'))

also the same as the relation_network part.

hope helpful to you