Open afroowoo opened 4 years ago
Hi, when reading a KITTI label, it is converted from camera to lidar coordinates with:
https://github.com/poodarchu/Det3D/blob/9fca2d94e4a138b6a2cd5d3fe4f4de38eb760306/det3d/core/bbox/box_np_ops.py#L925-L930
In this code, shouldn't the rotation also be converted from camera to lidar coordinates, for example r_lidar = r_camera @ r_rect @ velo2cam?
r_lidar = r_camera @ r_rect @ velo2cam
Skipping this transform causes a small rotational misalignment of the ground truth. Is this intended?
Hi, when reading a KITTI label, it is converted from camera to lidar coordinates with:
https://github.com/poodarchu/Det3D/blob/9fca2d94e4a138b6a2cd5d3fe4f4de38eb760306/det3d/core/bbox/box_np_ops.py#L925-L930
In this code, shouldn't the rotation also be converted from camera to lidar coordinates, for example
r_lidar = r_camera @ r_rect @ velo2cam
?Skipping this transform causes a small rotational misalignment of the ground truth. Is this intended?