search

lisiyao21 / Bailando

Code for CVPR 2022 paper "Bailando: 3D dance generation via Actor-Critic GPT with Choreographic Memory"
Other
382 stars 59 forks source link

Conversion to SMPL parameter (Choreographic for music in the wild, positional data --> rotational data) #56

Open h0han opened 1 month ago

h0han commented 1 month ago

Hello, thank you for your wonderful work. I'm seeking a way to take the choreography for music in the wild as a .fbx format. For this, I have tried to convert the 3D position into SMPL parameter, using the commend of python -u main_actor_critic.py --config configs/actor_critic_demo.yaml --train and rotmat=True in the config file. In this process, I have a question about the function of visulizeAndWrite() in functional.py. In the function, the dance motion data (maybe predicted 3D position sequence) is got into the get_closest_rotmat() and rotmat2aa. As I know, the SMPL model can predict its parameters using the 3D rotational data, but in the code, there is no function for converting positional data to rotational data.

So, is there a way of converting positional data to rotational data? Or do I implement the function using the IK method and so on?

If there's anything I'm misunderstanding, I would appreciate it if you could correct me.

Ancolie18 commented 1 month ago

https://github.com/wangsen1312/joints2smpl hi, this project seems can do what you want. I tried it, but meet some error. you can try it too~