About the origin tensor in backproject function.

[AlbertHuyb]

Hello all, thanks for your nice work!

I'm reading and learning about the code implementation for feature back projection.

In line 113 of Atlas/atlas/model.py, the variable self.origin is defined as self.origin = torch.tensor([0,0,0]).view(1,3).

From my understanding, such a volume can not handle the 3D points with (x<0 || y<0 || z<0).

I wonder how does this setting deal with the 2D features which are back projected to 3D spaces with (x<0 || y<0 || z<0).

I'm really curious about it.

Thank you!