search

taigw / brats17

Brain tumor segmentation for MICCAI 2017 BraTS challenge
BSD 3-Clause "New" or "Revised" License
321 stars 130 forks source link

Activation Layer Before Convolution layer in ResBlock #30

Open singh-hrituraj opened 5 years ago

singh-hrituraj commented 5 years ago

In file MSNet.py while creating (or defining) the graph for the model, I noticed that in the function that creates a residual block the activation layer is being applied before convolution. Is that a mistake or I am missing something considering we don't apply activation before Convolution?

stalhabukhari commented 5 years ago

In file MSNet.py while creating (or defining) the graph for the model, I noticed that in the function that creates a residual block the activation layer is being applied before convolution

Have been scratching my head on this for sometime now, a 'Pre-Activation ResNet' might be the intention, as explained here:

https://towardsdatascience.com/resnet-with-identity-mapping-over-1000-layers-reached-image-classification-bb50a42af03e

If otherwise, I'd like to ask the same.