what's the corresponding method of these functions in codes about channel-pruning HOT 3 CLOSED

I've been reading the code for two weeks now, and my understanding is that VH_decompose() performs spatial decomposition of the layers (splits them into 2 new layers with different shapes), ITQ _decompose() carries out channel decomposition @yihui-he explained a bit in issue #15. dictionary_kernel() is a function wrapper of dictionary(), which in charge of performing the lasso regression and minimizing the error reconstruction when selecting feature maps to prune.

Honestly I don't know why is there references to fully connected layers in the code, but from what I have read, fc_kernel and nonlinear_fc are not used during the execution of train.py. I have added more comment to the code in a fork I made. Feel free to check it out.

from channel-pruning.

@slothkong Thanks, you could create a pull request, so that I can merge your comments to master.

from channel-pruning.

#49

from channel-pruning.