MoConVQ: Unified Physics-Based Motion Control via Scalable Discrete Representations (SIGGRAPH 2024 Journal Track)
Project Page: https://moconvq.github.io/
The code provides:
- Pretrained model and code for MoConVQ representation
- Pretrained model and code for MoConGPT
- Trainging code for MoConVQ
- Trainging code for MoConGPT
Environment installation is a bit complicated, so we have prepared a script for installation, please refer to setup.cmd
Pretrained data: Download from
https://disk.pku.edu.cn/link/AAAFE3B2DDB1AC420EB5C4E0910196116F
or from OneDrive
https://1drv.ms/f/s!AsrkHbtkj4LsbqMZI08Bt9jFPJ4?e=SXkFlg
and place all file in this folder
The moconvq_base.data contains a motion encoder and a physiscs-based motion decoder.
Please refer to .\Script\track_something.py to get more information about the output of the motion encoder.
Or you can run the following command to reconstruct a kinematic motion into physics-based version
python .\Script\track_something.py base.bvh
You may use .\Script\tokenize_motion.py to convert a motion in bvh format into tokens, e.g.
python .\Script\tokenize_motion.py track.bvh -o out\tokens.txt
Run
python .\Script\tokenize_motion.py track.bvh -h
for more information
You may use .\Script\decode_token.py to decode a sequence of tokens into simulated motion, e.g.
python .\Script\decode_token.py -i 166 410 332 149 419 237 172 305 192 273 174 -o out\decode.bvh
or
python .\Script\decode_token.py -f tokens.txt -o out\decode.bvh
Run
python .\Script\decode_token.py track.bvh -h
for more information
python .\Script\unconditional_generation.py
You may use --seed argument to choose another random seed. It will generate a different motion.
python .\Script\unconditional_generation.py --seed 123
First install some additional packages:
pip install transformers sentencepiece
Then run the code:
python .\Script\text2motion_generation.py
text description can be found in the python script