You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I noticed that the HumanAct12 dataset used in action2motion training is in a different format compared to the original HumanAct12 from Guo's paper. The .pkl files contain additional data in rotvec format, while the original data is stored in xyz format. Could you provide some insight into how this conversion was done at your end?
Best regards,
Rory
The text was updated successfully, but these errors were encountered:
HumanAct12 is represented with SMPL angles (6D) + root location. We used this representation as is for this experiment, we didn't convert it to the HumanML3D representation.
Thank you for the prompt response. I really appreciate it. A member of the HumanAct12 team has confirmed that their original release of HumanAct12 is only released in the XYZ format. However, since HumanAct12 is a widely used dataset, it's most probable that someone else could have converted it to be represented with SMPL angles (6D) + root location.
I'm wondering if it is possible to share the source where you obtained the data or the exact conversion steps. We want to ensure that we have the same conversion so that the pre-trained weights can work as intended. I understand this will create additional work and trouble. Sorry for the inconvenience.
Thank you for the great work!
I noticed that the HumanAct12 dataset used in action2motion training is in a different format compared to the original HumanAct12 from Guo's paper. The .pkl files contain additional data in rotvec format, while the original data is stored in xyz format. Could you provide some insight into how this conversion was done at your end?
Best regards,
Rory
The text was updated successfully, but these errors were encountered: