PLIKS: A Pseudo-Linear Inverse Kinematic Solver for 3D Human Body Estimation

PLIKS: A Pseudo-Linear Inverse Kinematic Solver for 3D Human Body Estimation
Karthik Shetty, Annette Birkhold, Srikrishna Jaganathan, Norbert Strobel, Markus Kowarschik, Andreas Maier, Bernhard Egger
CVPR 2023

Installation

conda create -n pliks python=3.8
conda activate pliks

#Install PyTorch
conda install pytorch==1.9.1 torchvision==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge

#Install PyTorch3D
conda install -c fvcore -c iopath -c conda-forge fvcore iopath
conda install -c bottler nvidiacub
pip install git+https://github.com/facebookresearch/[email protected]

#Install other dependencies for visualization
pip install vtk==9.1.0 vedo=2021.0.5 opencv=3.4.2

#Install torch_geometric to run the model from the paper
pip install torch-geometric==1.7.2 torch-scatter==2.0.9 torch-sparse==0.6.12

For visualization pytorch3d==0.4.0 is required. The sparse model does not require torch_geometric.

Data

SMPL Files and Pretrained Model

• Download smpl files from the official website. Unzip and place basicModel_neutral_lbs_10_207_0_v1.0.0.pkl in model_files/.
• Download the pretrained model sparse or full.

Quick Demo

python demo.py --img_dir demo/input/* --out_dir demo/output/ --checkpoint checkpoint_sparse.pt --model MeshRegSparse

Citation

@InProceedings{Shetty_2023_CVPR,
    author    = {Shetty, Karthik and Birkhold, Annette and Jaganathan, Srikrishna and Strobel, Norbert and Kowarschik, Markus and Maier, Andreas and Egger, Bernhard},
    title     = {PLIKS: A Pseudo-Linear Inverse Kinematic Solver for 3D Human Body Estimation},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
    pages     = {574-584}
}

Acknowledgement

Code is adapted from I2L-MeshNet, SPIN, DecoMR, PARE.