Confusion about the camera extrinsics about co3d HOT 2 CLOSED

Hi Rundong.
Indeed, we use Pytorch3D format. If X is a an 1x3 point in the world coordinates, then you get to camera coordinates by (X@R + T). T is added after rotation, so you can’t judge about camera location by it. The camera centre in the world coords can be computed by −T@inv(R).

We don’t guarantee that the ground plane is aligned with the coordinate system (we did not run any plane detection).
Hope it helps. Please reopen the issue otherwise.

from co3d.

Thanks for your quick reply!
However, I found −T@inv(R) has a negative third entry (i.e., z<0), indicating that the camera is below the ground plane. Am I doing something wrong or is it just a normal phenomenon?

from co3d.