The execution time of get_connections increases with each next frame.

[KelLiNBY]

I used demo_video.py and I determined an increase in the processing time of the same type of videos , but only different in duration .
After that, I modified the process_frame function and added the processing time output for get_coordinate, get_connections and others in this process_frame. Similar to this:
def process_frame(cropped, heatmap_idx, model, paf_idx, output_resize_factor):
.....
tic4 = time.time()
connections = get_connections(cfg, coordinates, pafs)
print('connections : %.5f' % (time.time() - tic4))
.....
After that i run 5 min video with 29fps, the log was like this:
Processing frame: 290 (00:00:10)
processing time is 0.36798
coordinates : 0.00802
connections : 0.12001
.....
Processing frame: 3480 (00:02:00)
processing time is 36.80047
coordinates : 0.00798
connections : 32.14523
.....
Processing frame: 5833 (00:03:21)
processing time is 90.10047
coordinates : 0.0837
connections : 82.14523
.....
after that I stopped processing.
Updating python from 3.9.5 to 3.10.6 and the actual numpy and tenzorflow did not give results.

My video is almost without people and should be processed quickly.
If the script is modified and the same video starts again, then get_connections takes 80 seconds for the first frames and processing time continues to increase more than 80sec.