We found another deviation to the original implementation. The tanh non-linearity is only applied to the mask output, rather than to the generated image, i.e. tanh(m(x)) rather than tanh(m(x)+x). The mistake happens on this line: https://github.com/orobix/Visual-Feature-Attribution-Using-Wasserstein-GANs-Pytorch/blob/master/src/train.py#L63
I know you don't have the capacities to fix this at the moment, and I also don't know if I will have the time. So I am mainly posting this issue to document the problem.
Best wishes,
Christian
We found another deviation to the original implementation. The
tanhnon-linearity is only applied to the mask output, rather than to the generated image, i.e.tanh(m(x))rather thantanh(m(x)+x). The mistake happens on this line: https://github.com/orobix/Visual-Feature-Attribution-Using-Wasserstein-GANs-Pytorch/blob/master/src/train.py#L63I know you don't have the capacities to fix this at the moment, and I also don't know if I will have the time. So I am mainly posting this issue to document the problem.
Best wishes,
Christian