tfp.experimental.nn.util.make_convolution_transpose_fn_with_dilation

Like tf.nn.conv2d except applies batch of kernels to batch of x.

This version tends to be fastest on GPU. It implements the transposed convolution as a regular convolution of an image that is dilated by interleaving rows and columns of zeros equal to the number of strides.

filter_shape ...
strides ...
padding ...
rank ...
dilations ...
dtype ...
validate_args ...
name ...

convolution_transpose_fn A callable that takes an input Tensor and kernel and applies the transpose convolution operation.