Trying same video 3 times, but results video not same

First, Thank you for Git repo! It is very usefull

but i have some problem

I tried same video(mp4) default option, but results(save_out) was everytime different

For example, my first trying model didn't detect chair, only detect person

but second same video, model detect chair which didn't detect before and person results get poor(sit-down pose, but output standing)