Summary: 5 instances, 4 unique

Text	Count
# TODO: refactor this function	1
# TODO: remove the hack of yaw	1
# TODO: consider use torch.Tensor only	1
# TODO: ugly workaround to judge whether we are training det or seg model	2