The TF documentation has an output_shape parameter in tf.conv2d_transpose. Why is this needed? Don't the strides, filter size and padding parameters of the layer decide the output shape of that layer, similar to how it is decided during convolution?
This question was already asked on TF github and received an answer:
output_shape is needed because the shape of the output can't necessarily be computed from the shape of the input, specifically if the output is smaller than the filter and we're using VALID padding so the input is an empty image. However, this degenerate case is unimportant most of the time, so it'd be reasonable to make the Python wrapper compute output_shape automatically if it isn't set.
It makes sense to read the whole thread.
If you assume the following notation, output = o, input = i, kernel = k, stride = s, padding = p, the shape of the output will be:
来源：https://stackoverflow.com/questions/43624625/why-do-we-have-to-specify-output-shape-during-deconvolution-in-tensorflow