Closed prhbrt closed 7 years ago
Hi @prinsherbert, unfortunately that function is not part of matconvnet (which uses standard convolution to implement fully connected layers). It might be worth contacting the author directly to see if he can help out?
@albanie It seems that you are correct, thanks!
I am not familiar with Matconvnet, but am porting a network to Keras. The Matlab code for this project refers to creating a fully connected layer here:
https://bitbucket.org/jaderberg/eccv2014_textspotting/src/581855bba39e662d66b05392b864dfbdd15d4142/matconvnet/cudaconvnet_to_mconvnet.m?at=master&fileviewer=file-view-default#cudaconvnet_to_mconvnet.m-56
I could not, however, find anything related to
fc_layer
in either the matconvnet or the eccv2014_textspotting repository. I would like to know some of the details to see how I should interpret the weights (which dimension is encoded how and possibly other details).Where is
fc_layer
of Matconvnet defined?