I’m curious, when the base model was pretrained with RGB data, what was the final loss value at the end of the training? thanks.