Perspektive Camera does not extract focal_lenght and principal_point from calibration matrix

[WeberSamuel]

🐛 Bugs / Unexpected behaviors

opencv_from_cameras_projection does not return correct calibration matrix if PerspektiveCamera was configured using a calibration matrix.

Instructions To Reproduce the Issue:

from pytorch3d.utils.camera_conversions import opencv_from_cameras_projection
from pytorch3d.renderer import PerspectiveCameras

calibration_matrix = torch.tensor(
    [
        [
            [150.0, 0.0, 300.0, 0],
            [0.0, 150.0, 300.0, 0],
            [0.0, 0.0, 0.0, 1],
            [0.0, 0.0, 1.0, 0],
        ],
        [
            [51515.0, 51.0, 12.0, 0],
            [486.0, 1.0, 545.0, 0],
            [896.0, 41.0, 1.0, 1],
            [13.0, 61.0, 1.0, 14],
        ],
    ]
)
cams = PerspectiveCameras(K=calibration_matrix)
camera_matrix = opencv_from_cameras_projection(cams, image_size=torch.tensor([[600, 600]]))[2]
assert torch.any(camera_matrix[0] != camera_matrix[1]) # fails as both are equal

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[bottler]

PerspectiveCameras with explicit K is not supported in opencv_from_cameras_projection, which is trying to use the unused principal_point and focal_length members.

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions]

This issue was closed because it has been stalled for 5 days with no activity.