Channel-wise fully connected layer missing from context encoders
libertydefense opened this issue · 2 comments
libertydefense commented
Hey,
The original paper uses a channel-wise FC layer, to connect the enconder to decoder, however in the current implementation of context_encoder.py that is completely removed, is that done on purpose?
Best
libertydefense commented
This code is full of bugs has this been tested ?
gcaria commented
Hi @libertydefense, since I'm interested in using this code, so far I've only spotted that in line 124 self.img_rows should be changed to self.img_cols.
Which other bugs have you found ?