The semantic axes of a camera are
- x: right (pytorch3d: left)
- y: bottom (pytorch3d: top)
- z: front (pytorch3d: front)
The semantic axes of an object are
- x: left (pytorch3d: left)
- y: back (pytorch3d: top)
- z: top (pytorch3d: front)
This leads to the cam_tform4x4_obj for a camera looking straight at the front of an object of:
[ [1, 0, 0, 0],
[0, 0, -1, 0],
[0, 1, 0, +dist],
[0, 0, 0, 0],
]
This transformation can be understood as +270 or -90 degrees rotation around the x axis.
Open3d uses as default cam_tform4x4_obj:
[ [1, 0, 0, 0],
[0, -1, 0, 0],
[0, 0, -1, +dist],
[0, 0, 0, 0],
]