Set a video frame's side data to detection bounding boxes for use with the drawbox filter

[z-khan]

Scenario

Displaying object detection bounding boxes on a video frame using the drawbox filter.
By default, this filter draws one box with the specified x,y,w,h parameters. For example, we can add a drawbox filter to a graph in PyAV as:

drawbox_filter = graph.add( 'drawbox', f"x={int(box[0])}:y={int(box[1])}:w={int(box[4])}:h={int(box[3])}" )

However, when there are multiple boxes e.g. in case of object detection output, the filter allows the boxes to be sourced from the side data of the frame using the optional box_source parameter:

drawbox_filter = graph.add( 'drawbox', "box_source=side_data_detection_bboxes")

Will it be possible to set the side data of a video frame with an array of bounding boxes to make use of the filter?

References

drawbox filter documentation
drawbox filter source
AV_FRAME_DATA_DETECTION_BBOXES side data