Note: Python 2.x support has officially been dropped.

Note: Unofficial Python 3.11 support is at the end of this README.

Berkeley AUTOLAB's GQCNN Package

Package Overview

The gqcnn Python package is for training and analysis of Grasp Quality Convolutional Neural Networks (GQ-CNNs). It is part of the ongoing Dexterity-Network (Dex-Net) project created and maintained by the AUTOLAB at UC Berkeley.

Installation and Usage

Please see the docs for installation and usage instructions.

Citation

If you use any part of this code in a publication, please cite the appropriate Dex-Net publication.

Updated installation of GQCNN

This repo has updated the codebase of GQCNN to newer versions and tested on Ubuntu20.04:

python 3.11

CUDA 12.2

tensorflow 2.15.0.post1

to set up the environment, follow the README in docker/updated_gpu

Download the pretrained models for Dexnet:

The link provided in the official website is obsolete, use the link below:

https://drive.google.com/file/d/1fbC0sGtVEUmAy7WPT_J-50IuIInMR9oO/view

or you can download it from huggingface using:

wget --content-disposition https://huggingface.co/WoodenHeart0214/gqcnn/resolve/main/model_zoo.zip?download=true && \
    unzip model_zoo.zip && \
    rm model_zoo.zip && \
    mv model_zoo models

Test Command

. venv/bin/activate

make gqcnn_single_object_docker

aThinkingNeal/gqcnn