Effect pictures:
Test computer and System:
- Desktop setting: i9-12900KF, GPU 3090, CUDA 11.3
- System setting: Ubuntu 20.04, ROS noetic (Python 3.8)
- Test Date: 2022/12/23
- OpenPCDet Version: fork in Kin-Zhang/OpenPCDet, commit: 977cd1c
If you don't want to destroy your env/change your system version, please directly check the docker and build dockerfile. But you still have to have a computer with NVIDIA-GPU and can install cuda. Check Chinese blog Ubuntu下的NVIDIA显卡【驱动&CUDA 安装与卸载】
Dataset: Full KITTI dataset (Velodyne-64), teaser bag try onedrive link: kitti_sequence11_half.bag only 876Mb
Dependencies:
sudo apt install ros-noetic-ros-numpy ros-noetic-vision-msgs
pip install pyquaternion
And install the OpenPCDet in the env
pip install torch==1.10.0+cu113 torchvision==0.11.0+cu113 torchaudio==0.10.0 -f https://download.pytorch.org/whl/torch_stable.html
pip install spconv-cu113
sudo apt-get install python-setuptools
pip install pyquaternion install numpy==1.23 pillow==8.4
# have some potential risks
sudo chown -R $USER /usr/local/lib/python3.8/
sudo chown -R $USER /usr/local/bin/
# have some potential risks
git clone https://github.com/Kin-Zhang/OpenPCDet
cd OpenPCDet && python3 setup.py develop
Run:
mkdir -p ~/workspace/OpenPCDet_ws/src
cd ~/workspace/OpenPCDet_ws/src
git clone https://github.com/Kin-Zhang/OpenPCDet_ros.git
cd .. && catkin build
One more step check your Config file
source devel/setup.zsh
# before this step please change the model path in config file
roslaunch openpcdet 3d_object_detector.launch
Demo:
12.23.mp4
expand to see the docker usage
Dependencies, install docker and nvidia-container-toolkit, see [some issues](NVIDIA/nvidia-docker#1238)distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit
sudo systemctl restart docker
Build or pull
# choose one!
# option A: pull
docker pull zhangkin/openpcdet:ros
# option B: build by yourself
git clone https://github.com/Kin-Zhang/OpenPCDet_ros.git && cd OpenPCDet_ros
docker build -t zhangkin/openpcdet:ros .
run image to container
# if you want to display open3d, check the reference for more info
xhost +local:docker
docker run -it --net=host --gpus all -e DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix --name pcdet_ros zhangkin/openpcdet:ros /bin/zsh
# if not, simple one
docker run -it --net=host --gpus all --name pcdet_ros zhangkin/openpcdet:ros /bin/zsh
Because it need detect your setups, also you have to run the setup.py
inside the container
cd OpenPCDet && python3 setup.py develop
# After screen print: Finished processing dependencies for pcdet==0.6.0
# Test step cp model and test pcd to container:
docker cp /home/kin/workspace/OpenPCDet/tools/pv_rcnn_8369.pth pcdet_ros:/home/kin/workspace/OpenPCDet/tools/
docker cp /home/kin/workspace/OpenPCDet/tools/000002.bin pcdet_ros:/home/kin/workspace/OpenPCDet/tools
# test demo
cd tools && python3 demo.py --cfg_file cfgs/kitti_models/pv_rcnn.yaml \
--ckpt pv_rcnn_8369.pth \
--data_path 000002.bin
NOW is ROS TIME, inside container:
cd /home/kin/workspace/OpenPCDet_ws
catkin build && source devel/setup.zsh
roslaunch openpcdet 3d_object_detector.launch
-
finished
Dockerfile
-
已知的hard code 有个move_lidar_center的变量 不知道为啥之前的人设置为20,但是实际上openpcdet里无 建议TODO 看一下!Removed -
把hardcode都写到config里!
-
如果能直接标记点云可能会更好?
This repo has a Chinese blog also to read through 【点云检测】OpenPCDet 教程系列 [1] 安装 与 ROS运行
All methods and models are from: open-mmlab/OpenPCDet
Reference codes:
-
The first version of openpcdet-ros is from: Cram3r95/OpenPCDet-ROS
-
For 3d box marker drawing: Youtube AI葵 and his code
-
open3d in docker vis: open3d docs
✨✨Stargazers, positive feedback