comprehensible data preparation scripts - Githubissues

mateuszwyszynski / PoseNDF

Implementation of Pose-NDF: Modeling Human Pose Manifolds with Neural Distance Fields

Other

0 stars 0 forks source link

comprehensible data preparation scripts #2

Open mateuszwyszynski opened 7 months ago

mateuszwyszynski commented 7 months ago

The current way the data scripts (like data/prepare_data.py) work is very messy. Among many problems:

there are hard-coded paths from the original paper author's machine
the data files which are nested more than one level are not automatically read by the data/prepare_data.py script. One has to manually move these files. This should be automated

Overall the current data preparation process is extremely unclear. There are unanswered questions like:

[ ] which datasets from the AMASS should be downloaded
[ ] how to train a model using a smaller subset of datasets
[ ] after downloading datasets, how exactly the data should be organised for the scripts to work