tornadomeet / ResNet

Reproduce ResNet-v2(Identity Mappings in Deep Residual Networks) with MXNet
Apache License 2.0
556 stars 199 forks source link

Why does the ResNext use original Residual Block instead of pre-activation residual block? #53

Open Itsanewday opened 2 years ago

Itsanewday commented 2 years ago

In the Identity Mapping, the author claimed that pre-activation is better than post-activation, but i find that in the ResNext , the residual blocks are the original post activation residual blocks, like conv->bn->relu->conv->bn->sum->relu instead of bn->relu->conv->bn->relu->conv->sum. Can you help we? Thanks a lot