Giters

yuchaoli / KSE

Exploiting Kernel Sparsity and Entropy for Interpretable CNN Compression

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Confusion about training and clustering.

Coolde opened this issue · comments

commented

Ingenious method and excellent experiment results, thanks for your wonderful work! I am looking forward to your reply for my two questions.

  1. The code shows that your KSE network is trained from scratch, not pruned from NORMAL model. I don't know if I understand the right.
  2. In the code, you cluster all filters which should be clustered as the same number of clusters simultaneously, rather than cluster every input channel respectively. I am not sure if they are equivalent in theory.
    Thank you in advance .