how can we perform more data augmentation?

[sukun1045]

Hi, Sorry, I am a little bit new to this area. I am wondering how to augment the current HumanML3D data by simply concatenating the same or different motion sequences together. For example, 'jumping jack four times.' Is it possible to achieve that? And how exactly? Thank you.

[EricGuo5513]

Hi, now worries. This may bring up disconinuity at the transition between two motions. In our text2motion work, we choose to randomly cut off a few frames at the beginning/end, which cause small variation for augmentation. I guess one thing you could do is slow down/speed up the motions like the work MotionPuzzle.

[sukun1045]

Thanks for your reply. I am also wondering if I try to do some kind of augmentation by changing the fps from 20 to another number like 60, how should I do that in the data processing?