[GitHub] KeyKy opened a new issue #9337: Questions about the accnn.py tools.

[GitBox <gi...@apache.org>]

KeyKy opened a new issue #9337: Questions about the accnn.py tools.
URL: https://github.com/apache/incubator-mxnet/issues/9337
 
 
   I have read the paper Efficient and Accurate Approximations of Nonlinear Convolutional Networks. However, I do not fully understand the math in accnn.py. The tool split a convolution into two convolution.
   
   ```
     def sym_handle(data, node):
       kernel = eval(node['param']['kernel'])
       pad = eval(node['param']['pad'])
       name = node['name']
   
       name1 = name + '_v'
       kernel1 = tuple((kernel[0], 1))
       pad1 = tuple((pad[0], 0))
       num_filter = W1.shape[0]
       sym1 = mx.symbol.Convolution(data=data, kernel=kernel1, pad=pad1, num_filter=num_filter, name=name1)
   
       name2 = name + '_h'
       kernel2 = tuple((1, kernel[1]))
       pad2 = tuple((0, pad[1]))
       num_filter = W2.shape[0]
       sym2 = mx.symbol.Convolution(data=sym1, kernel=kernel2, pad=pad2, num_filter=num_filter, name=name2)
       return sym2
   ```
   1. Can accnn.py apply to Convolution with stride 2 (not equal to 1)?
   

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services