I can't seem to get the encoder decoder pair to preserve color. See example input and output below. Normalizing the input to imagenet helps very slightly, as observed #4
I would imagine that the DINOV3 embedding not to preserve colors, because they are not designed for this, but then what guidance would the decoder need to generate correct colors 9whether it is for reconstruction or generation? What am I missing?
I can't seem to get the encoder decoder pair to preserve color. See example input and output below. Normalizing the input to imagenet helps very slightly, as observed #4
I would imagine that the DINOV3 embedding not to preserve colors, because they are not designed for this, but then what guidance would the decoder need to generate correct colors 9whether it is for reconstruction or generation? What am I missing?