This repo is the official implementation for the ICCV2023 paper "COOP: Decoupling and Coupling of Whole-Body Grasping Pose Generation".

COOP is a novel framework to synthesize life-like whole-body poses that cover the widest range of human grasping capabilities.

For more details please refer to the paper COOP: Decoupling and Coupling of Whole-Body Grasping Pose Generation.

Installation

To install the dependencies please follow the next steps:

• Clone this repository:

  git clone https://github.com/zhengyanzhao1997/COOP
  cd COOP

• Install the dependencies by the following command:

  pip install -r requirements.txt
  

Getting started

BNet and HNet

• we have released the checkpoints of BNet and HNet in the folders as below.

    COOP
    ├── src
    │   ├── models
    │       ├── Bnet
    │       └── Hnet

Download SMPLX and MANO models

• Download body models following the steps on the SMPLX repo.

• Download body models following the steps on the MANO repo.

• Please put the SMPLX and MANO models in the folders as below.

    COOP
    ├── model
    │   ├── smplx
    │   └── mano

Generate whole-body grasps for test split.

cd src

Then

python infer/infer.py --rh-work-dir ../model/Hnet/ --body-work-dir../model/Bnet/  --data-file ../data --save-file [your save path]

You can also specify the test object name, subject shape and the position of the test obejct:

python ... --object apple --subject s1 --x 0.4 --z 0.4 --y 0.4

The visualize result (.html) and the generated mesh file will be saved at your save path.

Training your own COOP.

coming soon

Citation

@InProceedings{Zheng_2023_ICCV,
    author    = {Zheng, Yanzhao and Shi, Yunzhou and Cui, Yuhao and Zhao, Zhongzhou and Luo, Zhiling and Zhou, Wei},
    title     = {COOP: Decoupling and Coupling of Whole-Body Grasping Pose Generation},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023},
    pages     = {2163-2173}
}