xiexh20 / behave-dataset

Code to access BEHAVE dataset, CVPR'22
https://virtualhumans.mpi-inf.mpg.de/behave/
Other
141 stars 6 forks source link

Question: Detailed Computation of SMPL v2v (cm) and obj. v2v (cm) #29

Closed jwings1 closed 11 months ago

jwings1 commented 11 months ago

Hello Maintainers,

I hope this message finds you well. I've been using your dataset for a while now and I've found it to be extremely helpful for my research. However, I've run into a snag with some specific aspects. Specifically, I would like to understand how SMPL v2v (cm) and obj. v2v (cm) are computed: could you elaborate on the underlying preprocessing (e.g. PA) or postprocessing (e.g. averaging) steps involved? What is the metric used?

Thank you in advance for your time and help.

Best Regards, Lorenzo

xiexh20 commented 11 months ago

Hi Lorenzo,

Are you referring to the evaluation metrics used by the BEHAVE method? i.e. the numbers reported in table 2 in our main paper.

We perform Procrustes alignment by combining both human and object meshes together. You can find a reference here. After alignment, the v2v error is simply this function

Best,