I am trying to use channel-wise fully-connected layer which was introduced in paper “Context Encoders: Feature Learning by Inpainting”, however I have no idea on how to implement this in pytorch. if there someone can give me help, thanks a lot !
Pengfei_Wang (Man_813) #1
WERush (Xinge) #2
You can check the official code https://github.com/pathak22/context-encoder/blob/master/train.lua#L103. In fact, there is no channel-wise fc layer and the details of that is the conv2d+convTranspose2d.
Pengfei_Wang (Man_813) #3
thank you very much for answer me