Eval Specifics

[jake-austin]

Hi, I was wondering (since the eval scripts havent been released) if you can give more specifics on how you calculate Acc@5°/ 10°/ 20° for the GSO, NAVI, and ScanNet datasets. You state in the paper "we use Rodrigues’ formula to calculate relative rotation
error between the predicted/ground truth rotation matrices", and I'm wondering if "relative rotation" means just the error in the amount of predicted rotation or if this encodes error in the axis of rotation as well, or accounts for translation error

Thanks