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.
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.