I am deepening my knowledge and practice of Kdenlive - happy to have discovered its motion tracking capabilites.
Now I know the “motion tracking” effect is available, and how to use it, I wanted to know if it is possible (perhaps the question has asked before, but my search didn’t bring up any result) to make the “Analyse” process (“to see real effect”) faster by making use of multi-core/multi-threads processors - or is it already doing this?
In other words: is the setting under “Kdenlive Settings” → “Environment” → “Tasks and threads” the one to raise to positively influence the speed of the “Analyse” process ?
Or does it have no influence at all on the “Analyse” process of the “motion tracking” effect ?
If not, is a setting of that kind (multi-threaded “Analyse” ) foreseen?
You can’t just Throw Threads at any problem, having additional cores available to you can only help if the problem itself is both inherently parallelisable and its implementation solves it that way. And for some problems there will be a tradeoff where a parallel attempt at solution may not solve it “as well” as a single threaded solution which has complete knowledge of all the work done to date.
So ultimately this is going to depend on the algorithm you use for motion tracking and on the implementation in use here. I can empirically say that when melt is processing CSRT analysis it’s putting 2-and-a-bit cores to good use for me, but that might still only be one for the actual analysis. There’s a lot of numbers to crunch for this, there’s probably some speed to be gained in further optimisation of that code, but it’s always going to be “slow” compared to most other operations, that’s just the nature of the beast.