Camera viepoint

[udrs]

Camera viepoint

[xheon]

The sample has ID 596b8941-3eb6-4f49-a7b1-8881118c07f2/0014, i.e refers to scene 596b8941-3eb6-4f49-a7b1-8881118c07f2
You can find all the camera poses in the provided "front3d-camposes.zip".

Since the reconstruction happens in camera-space the view point does not change. In order to move the reconstruction from grid space to camera space you can use:

# define camera2frustum
camera2grid = torch.tensor([[33.3333, 0.0000, 0.0000, 114.7484],
                                [0.0000, 33.3333, 0.0000, 85.8808],
                                [0.0000, 0.0000, 33.3333, -13.3333],
                                [0.0000, 0.0000, 0.0000, 1.0000]])
 # get inverse of camera2frustum
 grid2camera = torch.inverse(camera2grid)

(obtained from lib/structures/frustum.py - compute_camera2frustum_transform() )

Hope that helps.

[udrs]

Thank you for your great project and reply!