When I use an SFU action such as “0017_WushuKicks001.fbx” as a reference action, and use “fbx_importer.py” to transform it, it can be seen that its coordinate system is different from that of the CMU action. The next time I redirected the motion using “retarget_motion.py”, the motion went horribly wrong.
CMU：
SFU：
How can I import the fbx data of SFU?
I also found some fbx datasets for CMU and he also showed the same results as SFU. .
The example come from NVIDIA-Omniverse/IsaacGymEnvs: Isaac Gym Reinforcement Learning Environments (github.com)
Hi there,
Each motion capture can use its own skeleton topology and orientations, so we’ll need to populate the retargeting configuration accordingly to each skeleton. The retarget_motion.py script provided reads configuration data from retarget_cmu_to_amp.json, which will need to be updated to include data for the SFU dataset or your fbx CMU dataset. I have attached the SFU t-pose file in SkeletonState format, which can be loaded in as part of the configuration as well. It may also be helpful to visualize the source and target skeletons to determine their orientations.
sfu_tpose.npy (2.4 KB)
