boreshkinai / smpl-ik

SMPL-IK

Learned Morphology-Aware Inverse Kinematics for AI Driven Artistic Workflows

https://arxiv.org/pdf/2208.08274.pdf

https://www.youtube.com/watch?v=FixF406owB4

Create workspace and clone this repository

mkdir workspace

cd workspace

git clone [e-mail]:boreshkinai/smpl-ik.git

Setup : Docker

docker build -f Dockerfile -t smpl-ik:$USER .

nvidia-docker run -p 18888:8888 -p 16006:6006 -v ~/workspace/smpl-ik:/workspace/smpl-ik -t -d --shm-size="1g" --name smpl-ik_$USER smpl-ik:$USER

go inside docker container

docker exec -i -t smpl-ik_$USER  /bin/bash 

Training

Train SMPL-IK model on H36M

python run.py --config=configs/experiments/smplik_h36m.yaml

Train SMPL-IK model on AMASS

python run.py --config=configs/experiments/smplik_amass.yaml

Train SMPL-SI model

python run.py --config=configs/experiments/smpl_si.yaml

Setup : Conda

conda create --name smplik python=3.8

conda activate smplik 

pip install -r requirements.txt -f https://download.pytorch.org/whl/torch_stable.html

To use notebooks (optional) :

conda install jupyter

Citation

If you use SMPL-IK in any context, please cite the following paper:

@misc{voleti2022smplik,
  doi = {10.48550/ARXIV.2208.08274},
  url = {https://arxiv.org/abs/2208.08274},
  author = {Voleti, Vikram and Oreshkin, Boris N. and Bocquelet, Florent and Harvey, Félix G. and Ménard, Louis-Simon and Pal, Christopher},
  title = {SMPL-IK: Learned Morphology-Aware Inverse Kinematics for AI Driven Artistic Workflows},
  publisher = {arXiv},
  year = {2022}
}