1 backproject_depth_with_p3d

rendering.unproject.backproject_depth_with_p3d(
    depth,
    cameras,
    valid_mask,
    *,
    stride=1,
    max_points=None,
)

Back-project a single depth map using PyTorch3D cameras and a validity mask.

1.1 Parameters

Name	Type	Description	Default
depth	torch.Tensor	Tensor["H", "W"] metric depth (metres) in the camera frame.	required
cameras	PerspectiveCameras	Matching pytorch3d.renderer.PerspectiveCameras.	required
valid_mask	torch.Tensor	Tensor["H", "W"] boolean mask for usable pixels (e.g. zclose/zfar filtering from CandidateDepths).	required
stride	int	Optional subsampling stride in pixel space.	1
max_points	int | None	Optional cap on returned points (random subset when exceeded).	None

1.2 Returns

Name	Type	Description
	torch.Tensor	Tensor["N", 3"] of world-frame points for valid pixels only.