search

hytseng0509 / CrossDomainFewShot

Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation (ICLR 2020 spotlight)
329 stars 62 forks source link

gnn baseline seems not working #1

Closed smallbox120 closed 4 years ago

smallbox120 commented 4 years ago

the loss seem to converge on bad state, even after the first few epochs and the validation result is equivalent to random guessing. And I waited about 40 epochs, seems to me that it won't work even if I train a few hundred more epochs. I train the model from scratch instead of pretrained parameters.

hytseng0509 commented 4 years ago

It is not tricky to train the GNN method from scratch. The code we provide here can only train the model with the pre-trained feature encoder. You can refer to the original implementation here for training the model from scratch.