This code is updated - Original is NNDam and recode with paper Surveillance video synopsis framework base on tube set
You need to run Tracking first and save it in a folder. Each frame will save txt conntain Location of each obj Detection: Yolor + Deepstream (NvDCF)
frame_1.jpg
frame_1.txt
...
python extract_background.py --video_path /home1/data/congvu/deepstream-test1/video-synopsis/video_paper.mp4
python video_synopsis.py --background_path background.jpg --ROOT /home1/data/congvu/deepstream-test1/synopsis_paper --FPS 25
Attention to FPS is FPS of video input
Check in this