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
Hello, it's so excited to see your work,especially the demo using htc controllers and headset motions input, which is very amazing for the VR game developers!
And when I try to use the VR headset and controllers' motion inputs to try to generate the avatar poser with the pretrained model provided, I met a problem of alignment issue, finding that the three input motions from quest 2 might having different coordinates system, I'm not very sure with this tho, for both rotations and positions compared to AvatarPoser's one. Since it seems that avatar poser for synthetic data AMASS get their coordinates by first registering on the smpl body first (forward kinematic I guess), and for only the three sparse inputs, seems the coordinates registration is no longer applicable, as the local rotations is not continuous for the required body joints by the smpl body.
So I wonder,if using the VR's motions, how should we prepare this input data to the model?
The text was updated successfully, but these errors were encountered:
Hello, it's so excited to see your work,especially the demo using htc controllers and headset motions input, which is very amazing for the VR game developers!
And when I try to use the VR headset and controllers' motion inputs to try to generate the avatar poser with the pretrained model provided, I met a problem of alignment issue, finding that the three input motions from quest 2 might having different coordinates system, I'm not very sure with this tho, for both rotations and positions compared to AvatarPoser's one. Since it seems that avatar poser for synthetic data AMASS get their coordinates by first registering on the smpl body first (forward kinematic I guess), and for only the three sparse inputs, seems the coordinates registration is no longer applicable, as the local rotations is not continuous for the required body joints by the smpl body.
So I wonder,if using the VR's motions, how should we prepare this input data to the model?
The text was updated successfully, but these errors were encountered: