Comment by bravura

If this is supposed to be used for deep-learning, shouldn't all the transforms be GPU-accelerated torch functions?