- detect object in a video or from camera.
on my nvidia 1060 (6GB RAM ), the speed is ~20 frame per second, but gpu utilization is below 30%, I will use multiprocessing to speed up.
object label and detecion api
label data
object_detection API implemented by google
- downloadutil.py: helper function for downloading model
- fps_measure.py: utility to measure fps
- queue_seq.py: a helper class to make sure the frame after processed by multi-process is in order.
main module for detecting object
run following command
python object_detection_tf_multiprocessing.py -v /home/scott/Videos/S11E03.mp4 -p 2
- you need replace the video file with your choice
- -p determine how many image detection process you want to run. For now, each detection process will comsune 1.3 GB system RAM, ~1G Video RAM, 30% GPU(GT1060), you need decide this according to your hardware.