Thanks for your meaningful work.I wonder if we can directly use the images in the test set for training, e.g. domain adaptation, pseudo-labeling, self-supervised, etc.