OmniControl: Control Any Joint at Any Time for Human Motion Generation
Yiming Xie, Varun Jampani, Lei Zhong, Deqing Sun, Huaizu Jiang

@article{xie2023omnicontrol,
  title={Omnicontrol: Control any joint at any time for human motion generation},
  author={Xie, Yiming and Jampani, Varun and Zhong, Lei and Sun, Deqing and Jiang, Huaizu},
  journal={arXiv preprint arXiv:2310.08580},
  year={2023}
}

📢 10/Dec/23 - First release

• Code for training and inference.
• Pretrained model on HumanML3D.
• Evaluation code and metrics.
• Pretrained model with different training strategies.
• Evaluation for cross combination of joints.
• Pretrained model on KIT-ML.

This code requires:

• Python 3.7
• conda3 or miniconda3
• CUDA capable GPU (one is enough)

Install ffmpeg (if not already installed):

sudo apt update
sudo apt install ffmpeg

For windows use this instead.

Setup conda env:

conda env create -f environment.yml
conda activate omnicontrol
python -m spacy download en_core_web_sm
pip install git https://github.com/openai/CLIP.git

Download dependencies:

bash prepare/download_smpl_files.sh
bash prepare/download_glove.sh
bash prepare/download_t2m_evaluators.sh

HumanML3D - Follow the instructions in HumanML3D, then copy the result dataset to our repository:

cp -r ../HumanML3D/HumanML3D ./dataset/HumanML3D

KIT - Download from HumanML3D (no processing needed this time) and the place result in ./dataset/KIT-ML

Download the model(s) you wish to use, then unzip and place them in ./save/.

HumanML3D

model_humanml

cd save
gdown --id 1oTkBtArc3xjqkYD6Id7LksrTOn3e1Zud
unzip omnicontrol_ckpt.zip -d .
cd ..

Check the manually defined spatial control signals in text_control_example. You can define your own inputs following this file.

python -m sample.generate --model_path ./save/omnicontrol_ckpt/model_humanml3d.pt --num_repetitions 1

We randomly sample spatial control signals from the ground-truth motions of HumanML3D dataset.

python -m sample.generate --model_path ./save/omnicontrol_ckpt/model_humanml3d.pt --num_repetitions 1 --text_prompt ''

You may also define:

• --device id.
• --seed to sample different prompts.
• --motion_length (text-to-motion only) in seconds (maximum is 9.8[sec]).

Running those will get you:

• results.npy file with text prompts and xyz positions of the generated animation
• sample##_rep##.mp4 - a stick figure animation for each generated motion.

It will look something like this:

You can stop here, or render the SMPL mesh using the following script.

This part is directly borrowed from MDM.
To create SMPL mesh per frame run:

python -m visualize.render_mesh --input_path /path/to/mp4/stick/figure/file

This script outputs:

• sample##_rep##_smpl_params.npy - SMPL parameters (thetas, root translations, vertices and faces)
• sample##_rep##_obj - Mesh per frame in .obj format.

Notes:

• The .obj can be integrated into Blender/Maya/3DS-MAX and rendered using them.
• This script is running SMPLify and needs GPU as well (can be specified with the --device flag).
• Important - Do not change the original .mp4 path before running the script.

Notes for 3d makers:

• You have two ways to animate the sequence:
  1. Use the SMPL add-on and the theta parameters saved to sample##_rep##_smpl_params.npy (we always use beta=0 and the gender-neutral model).
  2. A more straightforward way is using the mesh data itself. All meshes have the same topology (SMPL), so you just need to keyframe vertex locations. Since the OBJs are not preserving vertices order, we also save this data to the sample##_rep##_smpl_params.npy file for your convenience.

HumanML3D
Download the pretrained MDM model. The model is from MDM. Then place it in ./save/. Or you can download the pretrained model via

cd save
gdown --id 1XS_kp1JszAxgZBq9SL8Y5JscVVqJ2c7H
cd ..

You can train your own model via

python -m train.train_mdm --save_dir save/my_omnicontrol --dataset humanml --num_steps 400000 --batch_size 64 --resume_checkpoint ./save/model000475000.pt --lr 1e-5

HumanML3D

• Takes about 45 hours (on a single GPU). You can take multi-GPUs to evaluate each setting in parallel to accelerate this process.
• The output of this script for the pre-trained models is provided in the checkpoints file.

./eval_omnicontrol_all.sh ./save/omnicontrol_ckpt/model_humanml3d.pt 

Or you can evaluate each setting separately, e.g., root joint (0) with dense spatial control signal (100).
It takes about 1.5 hours.

./eval_omnicontrol.sh ./save/omnicontrol_ckpt/model_humanml3d.pt 0 100

Spatial Guidance. (./diffusion/gaussian_diffusion.py#L450)
Realism Guidance. (./model/cmdm.py#L158)

Our code is based on MDM.
The motion visualization is based on MLD and TMOS. We also thank the following works: guided-diffusion, MotionCLIP, text-to-motion, actor, joints2smpl, MoDi.

This code is distributed under an MIT LICENSE.
Note that our code depends on other libraries, including CLIP, SMPL, SMPL-X, PyTorch3D, and uses datasets that each have their own respective licenses that must also be followed.