3D/2D Affine Transformation
Module: cutoop.transform.
functioncalculate_2d_projections(coordinates_3d: ndarray, intrinsics_K: ndarray)
Parameters:
- coordinates_3d – 3, N
- intrinsics_K – K matrix 3, 3 (the return value of data_types.CameraIntrinsicsBase.to_matrix() )
Returns: projected_coordinates: N, 2
functiondepth2xyz(depth_img: ndarray, intrinsics: CameraIntrinsicsBase)
Transform depth_img pixel value to 3D coordinates under cv space, using camera intrinsics.
About different camera space:
- cv space: x goes right, y goes down, and you look through the positive direction of z axis
- blender camera space: x goes right, y goes up, and you look through the negative direction of z axis
Parameters: depth_img – 2D matrix with shape (H, W)
Returns: coordinates of each pixel with shape (H, W, 3)
functionpixel2xyz(h: int, w: int, pixel: ndarray, intrinsics: CameraIntrinsicsBase)
Transform (pixel0, pixel1) to normalized 3D vector under cv space, using camera intrinsics.
Parameters:
- h – height of the actual image
- w – width of the actual image
functionsRTdestruct(rts_Nx4x4: ndarray): tuple[ndarray, ndarray, ndarray]
Destruct multiple 4x4 affine transformation into scales, rotations and translations.
Returns: scales, rotations and translations each of shape N, N, 3, 3, N, 3
functiontoAffine(scale_N: ndarray, rot_Nx3x3: ndarray, tr_Nx3: ndarray)
return affine transformation: N, 4, 4
functiontransform_coordinates_3d(coordinates: ndarray, sRT: ndarray)
Apply 3D affine transformation to pointcloud.
Parameters:
- coordinates – ndarray of shape 3, N
- sRT – ndarray of shape 4, 4
Returns: new pointcloud of shape 3, N