that’s no wonder, it isn’t made for object detection (more to get a homography from 2 views of the same scene)