Annotation is completely occluded

nusc.render_annotation(‘52a4e38f35304e9eabc0764ea5e0b7dd’, extra_info=True)

The annotation has LIDAR points but it is not visible at all from the camera view. Is this an issue or expected behavior?

Also when I run get_sample_data with box_vis_level=BoxVisibility.ALL, it still returns the bounding box for this annotation.

Yes, we annotate it if there is one lidar or return return. Likely this is due to parallax - the lidar is mounted higher and sees “more” than the camera.
The definition of BoxVisibility.All is all(visible) and all(in_front). So all 4 corners need to be inside the image.

1 Like