It looks like this is 3 years old and based on a really old version of PyTorch - did something happen to bump its popularity recently? Is this still SOTA? (I would assume not given how fast the field moves)
I remember seeing a video on this concept like you could tell something was an airplane shape. I wonder is there anything to gain by joining the points/adding faces on outer most dots?
Unrelated but I think really neat is monocular visual inertial odometry.