magicleap/SuperPointPretrainedNetwork

Rubustness against Rotation

adamas-v opened this issue · 1 comments

seem like that the pretrained model privided is not rubust agsinst Rotation,Right?

Yes, SuperPoint+NN can't handle in-plane ratation well, maybe because there are no extreme in-plane rotation in the training examples. In superpoint paper, the author mentioned that they avoid sampling extreme in-plane rotations as they are rarely seen in HPatches.