Scale-Balanced-Grasp

Towards Scale Balanced 6-DoF Grasp Detection in Cluttered Scenes

Haoxiang Ma, Di Huang
In CoRL'2022

Paper Video

Introduction

This repository is official PyTorch implementation for our CoRL2022 paper. The code is based on GraspNet-baseline

Environments

• Anaconda3
• Python == 3.7.9
• PyTorch == 1.6.0
• Open3D >= 0.8

Installation

Follow the installation of graspnet-baseline.

Get the code.

git clone https://github.com/mahaoxiang822/Scale-Balanced-Grasp.git
cd graspnet-baseline

Install packages via Pip.

pip install -r requirements.txt

Compile and install pointnet2 operators (code adapted from votenet).

cd pointnet2
python setup.py install

Compile and install knn operator (code adapted from pytorch_knn_cuda).

cd knn
python setup.py install

Install graspnetAPI for evaluation.

git clone https://github.com/graspnet/graspnetAPI.git
cd graspnetAPI
pip install .

Prepare Datasets

For GraspNet dataset, you can download from GraspNet

Clean scene data generation

You can generate clean data for Noisy-clean Mix by yourself.

cd dataset
sh command_generate_clean_data.sh

Tolerance Label Generation(Follow graspnet-baseline)

Tolerance labels are not included in the original dataset, and need additional generation. Make sure you have downloaded the orginal dataset from GraspNet. The generation code is in dataset/generate_tolerance_label.py. You can simply generate tolerance label by running the script: (--dataset_root and --num_workers should be specified according to your settings)

cd dataset
sh command_generate_tolerance_label.sh

Or you can download the tolerance labels from Google Drive/Baidu Pan and run:

mv tolerance.tar dataset/
cd dataset
tar -xvf tolerance.tar

Train&Test

Train

sh command_train.sh

Test

• We offer our checkpoints for inference and evaluation, you can download from Google Drive

sh command_test.sh

If you want to inference with Object Balanced Sampling, download the pretrained segmentation model and run

sh command_test_obs.sh

Evaluation

Evaluation in small-, medium- and large-scale:

python evaluate_scale.py

Top50 evaluation like Graspnet:

python evaluate.py

Citation

If any part of our paper and repository is helpful to your work, please generously cite with:

@InProceedings{Ma_2022_CoRL,
    author    = {Haoxiang, Ma and Huang, Di},
    title     = {Towards Scale Balanced 6-DoF Grasp Detection in Cluttered Scenes},
    booktitle = {Conference on Robot Learning (CoRL)},
    year      = {2022}