Skip to content

Latest commit

 

History

History
108 lines (93 loc) · 4.13 KB

README.md

File metadata and controls

108 lines (93 loc) · 4.13 KB
Raw

Robotic Continuous Grasping System by Shape Transformer-Guided Multi-Object Category-Level 6D Pose Estimation

This is the PyTorch implementation of paper Robotic Continuous Grasping System by Shape Transformer-Guided Multi-Object Category-Level 6D Pose Estimation published in IEEE Transactions on Industrial Informatics by J. Liu, W. Sun, C. Liu, X. Zhang, and Q. Fu.

intro

Grasping Demo

https://www.bilibili.com/video/BV16M4y1Q7CD or https://youtu.be/ZeGN6_DChuA

Installation

Our code has been tested with

  • Ubuntu 20.04
  • Python 3.8
  • CUDA 11.0
  • PyTorch 1.8.0

We recommend using conda to setup the environment.

If you have already installed conda, please use the following commands.

conda create -n CLGrasp python=3.8
conda activate CLGrasp
conda install ...

Build PointNet++

cd 6D-CLGrasp/pointnet2/pointnet2
python setup.py install

Build nn_distance

cd 6D-CLGrasp/lib/nn_distance
python setup.py install

Dataset

Download camera_train, camera_val, real_train, real_test, ground-truth annotations, and mesh models provided by NOCS.
Unzip and organize these files in 6D-CLGrasp/data as follows:

data
├── CAMERA
│   ├── train
│   └── val
├── Real
│   ├── train
│   └── test
├── gts
│   ├── val
│   └── real_test
└── obj_models
    ├── train
    ├── val
    ├── real_train
    └── real_test

Run python scripts to prepare the datasets.

cd 6D-CLGrasp/preprocess
python shape_data.py
python pose_data.py

Evaluation

You can download our pretrained models (camera, real) and put them in the '../train_results/CAMERA' and the '../train_results/REAL' directories, respectively. Then, you can have a quick evaluation on the CAMERA25 and REAL275 datasets using the following command. (BTW, the segmentation results '../results/maskrcnn_results' can be download from SPD)

bash eval.sh

Train

In order to train the model, remember to download the complete dataset, organize and preprocess the dataset properly at first.

# optional - train the GSENet and to get the global shapes (the pretrained global shapes can be found in '6D-CLGrasp/assets1')
python train_ae.py
python mean_shape.py

train.py is the main file for training. You can simply start training using the following command.

bash train.sh

Citation

If you find the code useful, please cite our paper.

@article{TII2023,
  author={Liu, Jian and Sun, Wei and Liu, Chongpei and Zhang, Xing and Fu, Qiang},
  journal={IEEE Transactions on Industrial Informatics},
  title={Robotic Continuous Grasping System by Shape Transformer-Guided Multi-Object Category-Level 6D Pose Estimation},
  year={2023},
  publisher={IEEE},
  doi={10.1109/TII.2023.3244348}
}

Acknowledgment

Our code is developed based on the following repositories. We thank the authors for releasing the codes.

Licence

This project is licensed under the terms of the MIT license.