Where does mean/std deviation come from? about a2j HOT 4 OPEN

@carsonswope Hi,
you can refer to this dissscusion #29 for more details of mean/std computation.
And, for the question of how the mean/std affect real time situation, honestly, I dont know, some kind of weird flex might appears with the inappropriate mean/std params, because mean/std is dataset specified, generalization ability of these parms may not be guaranteed.

from a2j.

@zhangboshen Hello,
I have the same concern about the mean/std value in the real-time situation. If I wanna use your method in a real-time application, what should I do? Can I just retrain the model and remove the mean&std during preprocessing?
Thanks!

from a2j.

@yangsen228 You are right, current mean/std parameters are sensitive to cameras. I guess the possible solution is to retrain the model for specific scenarios. And, remove the mean&std during preprocessing can make the regression problem even harder, there might be larger training errors.

from a2j.

@zhangboshen Hello,
I have the same concern about the mean/std value in the real-time situation. If I wanna use your method in a real-time application, what should I do? Can I just retrain the model and remove the mean&std during preprocessing?
Thanks!

Have you solve the problem that use this model in you own depth image or video?

from a2j.