Sanjay Srinivasan
Sanjay Srinivasan
I'm working on modifying the package to support depthwise conv 2d layers rn, I'll see if I can get it done soon. You can't use the standard implementation since each...
You should be able to use this code: Just add another condition branch in _apply_delete_mask: elif layer_class == 'DepthwiseConv2D:' ``` if np.all(inbound_masks): new_layer = layer else: if data_format == 'channels_first':...
The code provided is actually just for the mask propagation. So if you try to prune a depthwise conv layer, this will not work. However, if you look at this...
Checkout this PR, it has exactly what you want: https://github.com/BenWhetton/keras-surgeon/pull/27
Any updates on this? @jiaxiang-wu