U-shaped transformer model in CIELAB color space. The model reconstructs the input image.
As in the
image restoration model,
only a random subset of the image patches are taught to the model, which shortens the learning time.
It prioritizes color accuracy.
