This repository is the official implementation of the CGrasp model from our 3DV 2025 paper:
3D Whole-body Grasp Synthesis with Directional Controllability (https://gpaschalidis.github.io/cwgrasp/).
CGrasp is a generative model for hand grasp generation, conditioned on a given direction. The generated hand follows the specified direction.
Clone the repository using:
git clone git@github.com:gpaschalidis/CGrasp.git
cd CGraspRun the following commands:
conda create -n cgrasp python=3.9 -y
conda activate cgrasp
conda install pytorch=2.0.1 torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
pip install kaolin==0.16.0 -f https://nvidia-kaolin.s3.us-east-2.amazonaws.com/torch-2.0.1_cu117.htmlpip install -r requirements.txt- Download MANO models following the steps on the MANO repo and save them in the folder "models", following the structure below:
models
└── mano- Download the GrabNet dataset (ZIP files) from this website. After following all the instructions you should have the structure below:
GRAB
├── data
│ │
│ ├── bps.npz
│ └── obj_info.npy
│ └── sbj_info.npy
│ │
│ └── [split_name] from (test, train, val)
│ │
│ └── frame_names.npz
│ └── grabnet_[split_name].npz
│ └── data
│ └── s1
│ └── ...
│ └── s10
└── tools
│
├── object_meshes
└── subject_meshes- To test CGrasp you need the pre-trained RefineNet model from GrabNet. Download this model from the GRAB website, and move it in the folder "pretrained" as described below:
cgrasp
└── cgrasp
└── pretrained
└── refinenet.ptTo train CGrasp from scratch use the following command:
python train.py --config_path cgrasp/configs/cgrasp_cfg.yaml --data_path GRAB/data/ --rhm_path models/mano/ --save_dir saveTo try CGrasp and visualize the generated grasps together with the input grasp directions:
- First download our pre-trained model from here and place it in the folder "pretrained". The folder "pretrained" should have the following structure:
cgrasp
└── cgrasp
└── pretrained
│
├── cgrasp_cfg.yaml
├── cgrasp.pt
└── refinenet.pt- And then run the following command after picking a ".ply" file from the path "GRAB/tools/object_meshes":
python generate_grasps.py --obj_path GRAB/tools/object_meshes/banana.ply --rhm_path models/mano --grasp_type "left" --n_samples 1 --save_dir $SAVE_FOLDERYou can specify the number of the generated grasps and also the type of the grasp (left or right).
If you found this work influential or helpful for your research, please cite:
@inproceedings{paschalidis2025cwgrasp,
title = {{3D} {W}hole-Body Grasp Synthesis with Directional Controllability},
author = {Paschalidis, Georgios and Wilschut, Romana and Anti\'{c}, Dimitrije and Taheri, Omid and Tzionas, Dimitrios},
booktitle = {{International Conference on 3D Vision (3DV)}},
year = {2025}
}