Performance of deepcopy getAnalysisModel in analyze_faces

[rjenkin]

I'm running the reactor node on frames in a 720p video at 24fps and noticed the performance was quite slow in the analyze_faces function, specifically where it does a deepcopy of the analysis model. I removed the deep copy locally, and saw a huge performance improvement from 10 minutes down to 2 minutes for a 10 second clip at 24fps.

My question is what's the purpose of copying the analysis model and will removing this cause unwanted side-effects? In other contexts I've seen deep copying used on objects to preserve the state for that instance so was thinking it might cause issues if there's multiple faces or multiple threads running in parallel.

[rjenkin]

Actually it doesn't look like it's doing the swap

[Gourieff]

"copy.deepcopy" is needed to prevent "det_size" parameter from being attached to the analysis model, otherwise it can't be changed without reloading the model
With help of copy.deepcopy "det_size" can be changed on the fly