A little question about the preprocess_training_data.py.

[Learner209]

Hi, first I wanna thank you for this amazing work ! Awesome demo and well-presented paper ! During my experiments with your codebase, I have some questions about the 0_preprocess_training_data.py, in the data_npz_dir arg of main func, the default argument is ./data/processed_30fps_no_skating/, so I am guessing there is some other preprocessing undergoing (i.e. downsampling fps, remove skating frames) from the raw .npz AMASS data before entering in this main func? If that is true, are you willing to make these preprocessing steps public so we can have a more holistic understanding of this awesome project, or if I am wrong, plz point out my misunderstandings !

[brentyi]

Hi @Learner209! Our input here is preprocessed files from HuMoR, which downsamples to 30fps and removes some sequences with foot skating: https://github.com/davrempe/humor/tree/main/data

I'll update the README/script to be clearer.