ml-gsn
ml-gsn copied to clipboard
normalize_rotation
Why the code use torch.bmm or minus for Rt and Rt[:, origin_frame : origin_frame + 1]. To make the value smaller?