search

nkolot / ProHMR

Repository for "Probabilistic Modeling for Human Mesh Recovery"
Other
258 stars 23 forks source link

is there a way to convert the predicted 3d points to bvh? #14

Open visonpon opened 2 years ago

visonpon commented 2 years ago

Hi, @nkolot ,I come to notice that there exist work that convert 3d keypoints to bvh(here), but it only contain the human3.6 and cmu data structure, I have tried to modify theopenpose_skeleton.py follow theh36m_skeleton.py andcmu_skeleton.py, but didn't work yet. have you also tried to get bvh from the predicted 3d points? hope you can give some advices, thanks~

nkolot commented 2 years ago

Can you elaborate more on what is the issue you are facing? I haven't tried to work with bvh, but it looks that you can use the openpose_skeleton.py if you get the first 25 keypoints that we provide as output. These by definition correspond to the OpenPose keypoints.

visonpon commented 2 years ago

Hi @nkolot , since in 3d animation, the usual data format is bvh ,so it's comman to convert 3d keypoints to euler and save it as bvh. I have done the openpose_skeleton.py successfully just now and convert the 3d points to final bvh, althouth the smpl fitting results look great, but the actual 3d results have some serious problems like root drifting.

akk-123 commented 2 years ago

hi @visonpon can you share how to convert 3d keypoints to bvh?