should have different BatchNorm config for training and testing?

vinthony / ghost-free-shadow-removal

[AAAI 2020] Towards Ghost-free Shadow Removal via Dual Hierarchical Aggregation Network and Shadow Matting GAN

https://arxiv.org/abs/1911.08718

297 stars 59 forks source link

should have different BatchNorm config for training and testing? #30

Open centosrhel opened 3 years ago

centosrhel commented 3 years ago

Hi, thanks for your good job, but I have some confusion about net config. It seems that you had set the same BN layer between training and testing mode. Shouldn't we use training=True or training=False for BN layer, during training and testing phase respectively?