taki0112 / partial_conv-tensorflow

Simple Tensorflow implementation of "Partial Convolution based Padding" (partialconv)

PartialConv-Tensorflow

Simple Tensorflow implementation of Partial Convolution based Padding

How to use

  # typical convolution layer with zero padding
  x = conv(x, channels, kernel=3, stride=2, use_bias=True, padding='SAME', scope='conv')
  
  # partial convolution based padding
  x = partial_conv(x, channels, kernel=3, stride=2, use_bias=True, padding='SAME', scope='conv')

Results

Activation map

ImageNet Classification

Segmentation (DeepLab V3+)

Author

Junho Kim

partial_conv-tensorflow's Issues

Partial Conv based padding in Keras

Hi, thank you for the nice work. is it possible to provide keras implementation? using this function along with Lambda does not work and throwing an error "AttributeError: 'tuple' object has no attribute 'layer'"

A question about padding schemes

Hello. Thank you for your work.
My question is do you think this partial convolution idea can work with other padding schemes such as reflect?
Thank you.