How far out are we from Object Detection that can identify at different scales? - i.e. angled, facing towards, rotated, skewed, etc...

We seem to have good OD that can create horizontal bounding boxes, but these bounding boxes seem to be generic estimates.

Even rectangle detection with these models can't identify the angle or skew of a rectangle in a frame (we get the same generic bounding box)

OpenCV seemed to have models awhile ago that could do this just fine.