Inquiry on the purpose on the normalization of SLAM camera 6D angular velocity

[GaigeY]

Hi @yohanshin ,

Sorry to bother you again. I'm meeting another confusing problem in SLAM result.

I noticed that you've normalized the camera 6D poses here:

cam_angvel = cam_angvel - torch.tensor([[1, 0, 0, 0, 1, 0]]).to(cam_angvel) # Normalize

This is a seldom seen treatment, as I have rarely seen anyone subtract the angle of rotation from the unit vector, and your AMASS CameraAugmentor here didn't apply this normalization strategy. May I ask if you have any basis for doing so? Or is this an empirical strategy for dealing with this?

Best wishes,

Qijun Ying

[GaigeY]

I discover that you've made same normalization to the AMASS synthetic cam_angvel here, too.