First of all, thank you for your excellent work and open-source spirit. However, I encountered some issues during the data processing stage.
During the extraction of the AMASS and BEATX datasets, both of them are named HumanML3D and are placed in process_h3d_amass and process_h3d_beat, respectively. However, when computing the mean and variance (as shown in Fig. 3), only the HumanML3D extracted from BEATX is used. The computed mean and variance are then directly saved into process_h3d_amass/HumanML3D as well.
Wouldn’t this cause inconsistencies or errors? Because these two HumanML3D datasets are derived from different raw data, so their mean and variance should also differ, right?
I would really appreciate your guidance on this issue. Thank you!
First of all, thank you for your excellent work and open-source spirit. However, I encountered some issues during the data processing stage.
During the extraction of the AMASS and BEATX datasets, both of them are named HumanML3D and are placed in process_h3d_amass and process_h3d_beat, respectively. However, when computing the mean and variance (as shown in Fig. 3), only the HumanML3D extracted from BEATX is used. The computed mean and variance are then directly saved into process_h3d_amass/HumanML3D as well.
Wouldn’t this cause inconsistencies or errors? Because these two HumanML3D datasets are derived from different raw data, so their mean and variance should also differ, right?
I would really appreciate your guidance on this issue. Thank you!