def criterionHis(self, input_data, target_data, mask_src, mask_tar, index):
input_data = (self.de_norm(input_data) * 255).squeeze()
target_data = (self.de_norm(target_data) * 255).squeeze()
mask_src = mask_src.expand(1, 3, mask_src.size(2),
mask_src.size(2)).squeeze()
mask_tar = mask_tar.expand(1, 3, mask_tar.size(2),
mask_tar.size(2)).squeeze()
input_masked = input_data * mask_src
target_masked = target_data * mask_tar
input_match = histogram_matching(input_masked, target_masked, index)
input_match = self.to_var(input_match, requires_grad=False)
loss = self.criterionL1(input_masked, input_match)
return loss
def criterionHis(self, input_data, target_data, mask_src, mask_tar, index):
input_data = (self.de_norm(input_data) * 255)
target_data = (self.de_norm(target_data) * 255)
mask_src = mask_src.expand(-1, 3, mask_src.size(2), mask_src.size(3))
mask_tar = mask_tar.expand(-1, 3, mask_tar.size(2), mask_tar.size(3))
input_masked = input_data * mask_src
target_masked = target_data * mask_tar
# dstImg = (input_masked.data).cpu().clone()
# refImg = (target_masked.data).cpu().clone()
input_match = []
for (i, t, idx) in zip(input_masked, target_masked, index):
input_match.append(histogram_matching(i, t, idx))
input_match = torch.stack(input_match)
input_match = self.to_var(input_match, requires_grad=False)
loss = self.criterionL1(input_masked, input_match)
return loss