Awesome

TADA takes text as input and produce holistic animatable 3D avatars with high-quality geometry and texture. It enables creation of large-scale digital character assets that are ready for animation and rendering, while also being easily editable through natural language.

NEWS (2023.9.24):

• Using Omnidata normal prediction model to improve the normal&image consistency.

https://github.com/TingtingLiao/TADA/assets/45743512/f626af25-3c5c-4ab5-bbe6-7a85d95af913

https://github.com/TingtingLiao/TADA/assets/45743512/442d6617-3549-48cc-9868-e4fe0c4ba842

Install

• System requirement: Unbuntu 20.04
• Tested GPUs: RTX4090, A100, V100
• Compiler: gcc-7.5 / g++-7.5
• Python=3.9, CUDA=11.5, Pytorch=1.12.1

git clone git@github.com:TingtingLiao/TADA.git
cd TADA

conda env create --file environment.yml
conda activate tada 
pip install -r requirements.txt
 
cd smplx
python setup.py install 

# download omnidata normal and depth prediction model 
mkdir data/omnidata 
cd data/omnidata 
gdown '1Jrh-bRnJEjyMCS7f-WsaFlccfPjJPPHI&confirm=t' # omnidata_dpt_depth_v2.ckpt
gdown '1wNxVO4vVbDEMEpnAi_jwQObf2MFodcBR&confirm=t' # omnidata_dpt_normal_v2.ckpt

Data

• SMPL-X Model (Need to register, download the SMPLX_NEUTRAL_2020.npz and put it into ./data/smplx/)
• TADA Extra Data (Required) Unzip it as directory ./data
• TADA 100 Characters (Optional)
• Optional Motion Data
  • AIST, AIST++
  • TalkShow
  • MotionDiffusion

@inproceedings{aist-dance-db,
  author = {Shuhei Tsuchida and Satoru Fukayama and Masahiro Hamasaki and Masataka Goto}, 
  title = {AIST Dance Video Database: Multi-genre, Multi-dancer, and Multi-camera Database for Dance Information Processing}, 
  booktitle = {Proceedings of the 20th International Society for Music Information Retrieval Conference (ISMIR) },
  year = {2019}, 
  month = {Nov} 
}

@inproceedings{li2021learn,
  title={AI Choreographer: Music Conditioned 3D Dance Generation with AIST++}, 
  author={Ruilong Li and Shan Yang and David A. Ross and Angjoo Kanazawa},
  year={2021},
  booktitle={ICCV}
}

@inproceedings{yi2023generating,
  title={Generating Holistic 3D Human Motion from Speech},
  author={Yi, Hongwei and Liang, Hualin and Liu, Yifei and Cao, Qiong and Wen, Yandong and Bolkart, Timo and Tao, Dacheng and Black Michael J},
  booktitle={CVPR}, 
  pages={469-480},
  month={June}, 
  year={2023} 
}

@inproceedings{tevet2023human,
  title={Human Motion Diffusion Model},
  author={Guy Tevet and Sigal Raab and Brian Gordon and Yoni Shafir and Daniel Cohen-or and Amit Haim Bermano},
  booktitle={ICLR},
  year={2023},
  url={https://openreview.net/forum?id=SJ1kSyO2jwu}
}

Usage

Training

The results will be saved in $workspace. Please change it in the config/*.yaml files.

# single prompt training    
python -m apps.run --config configs/tada.yaml --text "Aladdin in Aladdin" 

# with Omnidata supervision 
python -m apps.run --config configs/tada_w_dpt.yaml --text "Aladdin in Aladdin" 

# multiple prompts training
bash scripts/run.sh data/prompt/fictional.txt 1 10 configs/tada.yaml

Animation

• Download AIST or generate motions from TalkShow and MotionDiffusion.

python -m apps.anime --subject "Abraham Lincoln" --res_dir your_result_path

Tips

• Using an appropriate learning rate for SMPL-X shape is important to learn accurate shape.
• Omnidata normal supervision can effectively enhance the overall geometry and texture consistency; however, it demands more time for optimization.

Citation

@inproceedings{liao2024tada,
  title={{TADA! Text to Animatable Digital Avatars}},
  author={Liao, Tingting and Yi, Hongwei and Xiu, Yuliang and Tang, Jiaxiang and Huang, Yangyi and Thies, Justus and Black, Michael J.},
  booktitle={International Conference on 3D Vision (3DV)},
  year={2024}
}

Related Works

• HumanNorm: multiple stage SDS loss and perceptual loss can help generate the lifelike texture.
• SemanticBoost: uses TADA's rigged avatars to demonstrate the generated motions.
• SignAvatars: uses TADA's rigged avatars to demonstrate the sign language data.
• GALA: uses TADA's avatars for asset generation.

License

This code and model are available for non-commercial scientific research purposes as defined in the LICENSE (i.e., MIT LICENSE). Note that, using TADA, you have to register SMPL-X and agree with the LICENSE of it, and it's not MIT LICENSE, you can check the LICENSE of SMPL-X from https://github.com/vchoutas/smplx/blob/main/LICENSE; Enjoy your journey of exploring more beautiful avatars in your own application.