NTU-RGB preprocessing steps

[dschoerk]

Hi, I'm experimenting with your code and found that data preprocessing seems to have a pretty large impact on the accuracy. For example here it seems you are filtering frames where the pose extends too much in the x-axis. Just below that, there is more filtering dependent on how many valid frames a person has.

Could you give some reasoning behind these choices and how you came up with them? Thanks in advance!