Questions regarding motion files.
Closed this issue · 2 comments
In PHC(and omnih2o), the motion file is in the format of pkl file, which may contains the whole AMASS dataset. But in this codebase, we can only use single npy motion file? If I want to do AMP for H1, I can only use yaml file right?
Not exactly.
Our data pipeline is a bit different from PHC and derived works, but it also supports pre-processing the YAML files for faster loading.
It's explained in the data section in the readme. We have a data packaging script that creates a MotionLib object and loads the YAML file. Once loaded, it pickles the MotionLib object and saves to disk. We save this as .pth.
This method allows you to quickly re-load the motions and ensures you have all the data from the YAML file (for example, textual descriptions for the motions).
I understand now. Thank you!