Forums - How to apply padding for dilated convolution?

1 post / 0 new
How to apply padding for dilated convolution?
jerry.wj_chen
Join Date: 5 Mar 18
Posts: 7
Posted: Wed, 2018-04-11 02:31

Hi all,

We need to deploy a network for our product on Qualcomm DSP.

I use tensorflow 1.3.0 to train the network.

The network's performance is highly depends on dilated convolution for larger receptive field.

I tried tf.nn.dilation2d, tf.nn.atrous_conv2d, tf.layers.conv2d with dilation_rate 2 and padding='SAME'

The frozen graph of the training result can be converted to DLC format successfully but the output tensor has wrong shape (no padding applied)

I read the document. It seems that SNPE doesn't support padding feature for dilated convolution, neither support pure padding operation.

Is there any alternative solutions can be applied to achieve the padding operation?

Thanks

  • Up0
  • Down0

Opinions expressed in the content posted here are the personal opinions of the original authors, and do not necessarily reflect those of Qualcomm Incorporated or its subsidiaries (“Qualcomm”). The content is provided for informational purposes only and is not meant to be an endorsement or representation by Qualcomm or any other party. This site may also provide links or references to non-Qualcomm sites and resources. Qualcomm makes no representations, warranties, or other commitments whatsoever about any non-Qualcomm sites or third-party resources that may be referenced, accessible from, or linked to this site.