HGC-Net: Deep anthropomorphic hand grasping in clutter

Installation

Install necessary packages:

pip install -r requirements.txt

Install pointnet++:

cd pointnet2/pointnet2/
pip install -e .

Data processing

We provide dataset for hand grasps and cluttered scenes at https://drive.google.com/file/d/1H7gvV-HS8DdbdSBww3Gva4dROQwR5WQI/view?usp=sharing Please download the data and place it in the same folder with this repo. Many thanks for @chenzyn's help in making this dataset available again!

You can run

vis_object_grasp.py

to visualize the grasp data for each object.

Collison Detection

Run

collision_detection.py

to transform grasps to the scene and filter collision grasps. It will generate a file 'scene_grasps' that contains collision free grasps for each scene. It might take several hours for collision checking. Once finished, you can run

vis_scene_grasps.py

to visualize the scene grasps.

Match Point Grasp

Run

match_point_grasp.py

to generate point labels for each point cloud (for each scene, we capture 4 images from different views and convert them to point cloud). These process can be done online but it takes time during trainning, so we do it offline. It will generate a file 'point_grasp_data' contains grasp labels.

Training

Just run

train.py & vis.py 

to train the model and visualize it.