We aim to improve the performance of regressing hand keypoints and segmenting
pixel-level hand masks under new imaging conditions (e.g., outdoors) when we
only have labeled images taken under very different conditions (e.g., indoors).
In the real world, it is important that the model t