search

xiexh20 / behave-dataset

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

Mismatch between number of frames for object_fit_all.pkl and smpl_fit_all.pkl #30

Closed connorzl closed 9 months ago

connorzl commented 11 months ago

Hi, I noticed that for some sequences, the number of frames in object_fit_all.pkl and smpl_fit_all.pkl is not the same.

I downloaded the SMPL and object parameters here: https://datasets.d2.mpi-inf.mpg.de/cvpr22behave/behave-30fps-params-v1.tar

Is this intended? The number of object frames is always less than the number of human frames, so should I only be processing up to the number of object frames?

xiexh20 commented 11 months ago

Hi,

Thank you for your interests. Can you point out which sequence(s) have this problem? I will dig it a bit deepeer and figure out the problem. Thanks!

connorzl commented 10 months ago

Hi,

Thanks for the response! The two sequences that have this problem (number of object frames != number of human frames) are Date02_Sub02_backpack_twohand and Date06_Sub07_backpack_back

I found this by comparing the number of frames contained in "smpl_fit_all.npz" and "object_fit_all.npz" for each scene.

xiexh20 commented 10 months ago

Thank you for pointing them out. Indeed these two sequences have mismatch between SMPL and object frames, this is because some frames were removed due to poor object registrations. But the SMPL parameters were not updated to also reflect the removal.

Yes, you can just process up to the number of object frames.

I have update these mismatched sequences, everything should be fine now.

Best, Xianghui