gstreamer_live_overlay_object_detector.py Bounding boxes in wrong place and slower than I expected
When running on my desktop ubuntu 16.04 LTS:
python3 gstreamer_live_overlay_object_detector.py --video_file vtest.avi
The bounding boxes are in the wrong position and it is not realtime (my own mobilenetSSD V2 model running on ncnn runs in realtime)
You can get the vtest.avi from opencv https://github.com/opencv/opencv/tree/master/samples/data
I think the problem is that the python code is not scaling the bounding boxes to the input image ratio. The static image bounding box drawer doesn't seem to exhibit this problem.