lzrobots / LearningToCompare_ZSL

PyTorch code for CVPR 2018 paper: Learning to Compare: Relation Network for Few-Shot Learning (Zero-Shot Learning part)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

BatchNorm2d affect performance in testing time?

BlueAnthony opened this issue · comments

First, thank you for your awesome work!
Because I see the test_accuracy are different every execution when I repeat to execute "python miniimagenet_test_one_shot.py -w 5 -s 1", I found that the feature of the same image from feature encoder network will be different.
Should BatchNorm2d be turned off or deactivated by something like model.eval() in testing phase?
Or, I just miss something.
Thank you!

Sorry. I ask for the few-shot part, not zero-shot detection. My fault.
So, I close this issue.