Closed lxxcontinue closed 1 year ago
In order to construct unpaired data, for unpaired training on the ISTD train dataset, the shadow and shadow free images were randomly sampled? But such random sampling may yield paired data. ISTD have different shadow images for the same shadow-free image. Was I wrong about the way to sample the unpaired data
We followed the data loader of MaskShadowGAN if self.unaligned: item_B = self.transform(Image.open(self.files_B[random.randint(0, len(self.files_B) - 1)]))
This random sampling method may sample ground truth, so it may not be true unpaired data. Maybe this situation can be ignored?
ground truth
May have a chance to sample the ground truth for training, but there are 1330 training images for AISTD/ISTD, I think the chance is small.
If you want totally unpaired, you can try the USR dataset proposed by MaskShadowGAN.
OK ! Thank you for your response : )
In order to construct unpaired data, for unpaired training on the ISTD train dataset, the shadow and shadow free images were randomly sampled? But such random sampling may yield paired data. ISTD have different shadow images for the same shadow-free image. Was I wrong about the way to sample the unpaired data