vis4d.op.geometry.projection

Projection utilities.

Functions

generate_depth_map(points, intrinsics, image_hw)

Generate depth map for given pointcloud.

points_inside_image(points_coord, depths, ...)

Generate binary mask.

project_points(points, intrinsics)

Project points to pixel coordinates with given intrinsics.

unproject_points(points, depths, intrinsics)

Un-projects pixel coordinates to 3D coordinates with given intrinsics.
generate_depth_map(points, intrinsics, image_hw)[source]

Generate depth map for given pointcloud.

Parameters:

  • points (Tensor) – (N, 3) coordinates.

  • intrinsics (Tensor) – (3, 3) intrinsic camera matrices.

  • image_hw (tuple[int, int]) – (tuple[int,int]) height, width of the image

Returns:

Projected depth map of the given pointcloud.

Invalid depth has 0 values

Return type:

torch.Tensor

points_inside_image(points_coord, depths, images_hw)[source]

Generate binary mask.

Creates a mask that is true for all point coordiantes that lie inside the image,

Parameters:

  • points_coord (torch.Tensor) – 2D pixel coordinates of shape […, 2].

  • depths (torch.Tensor) – Associated depth of each 2D pixel coordinate.

  • images_hw (Tensor | tuple[int, int]) – (torch.Tensor| tuple[int, int]]) Associated tensor of image dimensions, shape […, 2] or single height, width pair.

Returns:

Binary mask of points inside an image.

Return type:

torch.Tensor

project_points(points, intrinsics)[source]

Project points to pixel coordinates with given intrinsics.

Parameters:

  • points (Tensor) – (N, 3) or (B, N, 3) 3D coordinates.

  • intrinsics (Tensor) – (3, 3) or (B, 3, 3) intrinsic camera matrices.

Returns:

(N, 2) or (B, N, 2) 2D pixel coordinates.

Return type:

torch.Tensor

Raises:

ValueError – Shape of input points is not valid for computation.

unproject_points(points, depths, intrinsics)[source]

Un-projects pixel coordinates to 3D coordinates with given intrinsics.

Parameters:

  • points (Tensor) – (N, 2) or (B, N, 2) 2D pixel coordinates.

  • depths (Tensor) – (N,) / (N, 1) or (B, N,) / (B, N, 1) depth values.

  • intrinsics (Tensor) – (3, 3) or (B, 3, 3) intrinsic camera matrices.

Returns:

(N, 3) or (B, N, 3) 3D coordinates.

Return type:

torch.Tensor

Raises:

ValueError – Shape of input points is not valid for computation.