Home

Awesome

VoxDet-ros

This is a ROS package for VoxDet, which simply utilizes the toolbox MMDetection of OpenMMLab. You can deploy VoxDet on real-robots using this library!

Requirements

Build your own env (Not recommended)

Use our docker (Strongly recommended, it is super easy)

Installation

  1. Clone all needed packages: this package, VoxDet libraries, Bop_toolkit, vision_msgs, and realsense-ros

    cd {ROS WORKSPACE}/src
    # this package and vision_msgs
    git clone git@github.com:Jaraxxus-Me/voxdet_ros.git
    git clone https://github.com/ros-perception/vision_msgs.git
    cd voxdet_ros
    # VoxDet library    
    git clone https://github.com/Jaraxxus-Me/VoxDet.git
    cd VoxDet
    pip install -e -v .
    # Modified BOP_tookit, use the link above
    cd bop_toolkit
    pip install -e .
    # realsense-ros
    git clone https://github.com/IntelRealSense/realsense-ros.git
    cd realsense-ros/
    git checkout `git tag | sort -V | grep -P "^2.\d+\.\d+" | tail -1`
    cd ..
    
  2. make, build, and install the ros packages

     cd ..
     catkin_make clean
     catkin_make -DCATKIN_ENABLE_TESTING=False -DCMAKE_BUILD_TYPE=Release
     catkin_make install
     catkin_make
    
  3. Install rospkg.

    pip install rospkg
    

Prepare phase1 support information

  1. You need to generate the mask and rotations for the phase 1 references the desired path look like:

    src/voxdet_ros
        VoxDet/
            data/
                1/
                    test_video (similar to the lmo and ycbv)
    	2/
    ...
    

    Note that the instance id "1", "2", can be used to switch target instance in the following demos. Then, modify the ~p1_path in the mmdetector.launch file.

  2. Use our tool to get and .npz file for each instance

    python3 src/voxdet_ros/VoxDet/tools/pre_save_p1_bop.py #change the path and object id accordingly
    

ROS Interfaces

params

topics

Usage

https://github.com/Jaraxxus-Me/voxdet_ros/assets/56875617/7a2afa18-49a3-4f1a-b21e-fc490eee9ac2

Reference

If our work inspires your research, please cite us as:

@INPROCEEDINGS{Li2023vox,       
	author={Li, Bowen and Wang, Jiashun and Hu, Yaoyu and Wang, Chen and Scherer, Sebastian},   
	booktitle={Proceedings of the Advances in Neural Information Processing Systems (NeurIPS)}, 
	title={{VoxDet: Voxel Learning for Novel Instance Detection}},
	year={2023},
	volume={},
	number={}
}