GitHub - Alvin0523/ros2_yolos_cpp

🚀 Overview

Implementation of YOLO-CPP by Mecatron.

⚡ YOLOs-CPP: https://github.com/Geekgineer/YOLOs-CPP
⚡ ros2_yolos_cpp: https://github.com/Geekgineer/ros2_yolos_cpp.git

ros2_yolos_cpp brings the blazing speed and unified API of YOLOs-CPP to the robot operating system. It provides composable, lifecycle-managed nodes for the entire YOLO family (v5, v8, v11, v26, etc.).

Key Features

⚡ Zero-Copy Transport: Optimized for high-throughput image pipelines using rclcpp::Subscription.
🔄 Lifecycle Management: Full support for configure, activate, deactivate, shutdown transitions.
🛠️ Composable Nodes: Run multiple models in a single container for efficient resource usage.
📦 All Tasks Supported: Detection, Segmentation, Pose, OBB, and Classification.
🏗️ Production Ready: CI/CD tested, strictly typed parameters, and standardized messages (vision_msgs).

📥 Installation

Prerequisites

Requirement	Version	Notes
C++ Compiler	C++17	GCC 9+, Clang 10+, MSVC 2019+
CMake	≥ 3.16
OpenCV (CPP)	≥ 4.5	Core, ImgProc, HighGUI
ONNX Runtime	≥ 1.16	Auto-downloaded by build script

Requirement Check

# Check GCC version
g++ --version

# Check cmake
cmake --version

# OpenCV (CPP)
pkg-config --modversion opencv4 2>/dev/null || pkg-config --modversion opencv

CUDA & cuDNN Requirement for ONNX Runtime

CUDA 12.x is the default version for ONNX Runtime GPU packages in PyPI
Other versions of CUDA please refer to ONNX_CUDA

ONNX Runtime Version	CUDA	cuDNN
1.20.x	12.x	9.x

CUDA 12.8 (WSL) & cuDNN 9.19.0

CUDA

# Clone
wget https://developer.download.nvidia.com/compute/cuda/12.8.0/local_installers/cuda_12.8.0_570.86.10_linux.run
sudo sh cuda_12.8.0_570.86.10_linux.run


# Create symlink (required for most tools to find CUDA)
sudo ln -s /usr/local/cuda-12.4 /usr/local/cuda

# Add to ~/.bashrc
echo 'export PATH=/usr/local/cuda/bin:$PATH' >> ~/.bashrc
echo 'export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
source ~/.bashrc

# Check installation
nvcc --version

cuDNN

# Install
wget https://developer.download.nvidia.com/compute/cudnn/9.19.0/local_installers/cudnn-local-repo-ubuntu2204-9.19.0_1.0-1_amd64.deb
sudo dpkg -i cudnn-local-repo-ubuntu2204-9.19.0_1.0-1_amd64.deb
sudo cp /var/cudnn-local-repo-ubuntu2204-9.19.0/cudnn-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cudnn9-cuda-12


### Installation of ONNX Runtime

```bash
# Clone
git clone https://github.com/Geekgineer/YOLOs-CPP.git
cd YOLOs-CPP

# Build (auto-downloads ONNX Runtime)
./build.sh 1.20.1 0   # CPU build 
./build.sh 1.20.1 1   # GPU build (requires CUDA - pick this)

Build the ros2_yolos_cpp package

# Create workspace
cd ros2_ws/src

# Clone package
git clone https://github.com/Geekgineer/ros2_yolos_cpp.git

# Install dependencies
cd ros2_ws
rosdep update && rosdep install --from-paths src --ignore-src -y

# Build (Release mode recommended for performance)
colcon build --packages-select ros2_yolos_cpp --cmake-args -DCMAKE_BUILD_TYPE=Release
source install/setup.bash

YOLO-CPP Package

‼️The YOLO-CPP package is used to build the onnxruntime, you may remove it when installation of the build is done (temporary)

🛠️ Usage

This package provides a launch file for each task. You must provide paths to your ONNX model and (optionally) labels file.

0. USB CAM Testing (WSL)

sudo apt install ros-humble-usb-cam

ros2 run usb_cam usb_cam_node_exe --ros-args \
  -p video_device:=/dev/video0 \
  -p image_width:=640 \
  -p image_height:=480 \
  -p framerate:=30.0 \
  -p pixel_format:=mjpeg2rgb \
  -p io_method:=mmap \
  -r __ns:=/camera

1. Object Detection

Publishes vision_msgs/Detection2DArray with bounding boxes and class IDs.

ros2 launch ros2_yolos_cpp detector.launch.py \
    model_path:=src/ros2_yolos_cpp/models/yolo11n.onnx \
    labels_path:=src/ros2_yolos_cpp/models/coco.names \
    use_gpu:=true \
    image_topic:=/camera/image_raw

2. Instance Segmentation

Publishes vision_msgs/Detection2DArray and a synchronized mask image.

ros2 launch ros2_yolos_cpp segmentor.launch.py \
    model_path:=/path/to/yolo11n-seg.onnx \
    labels_path:=/path/to/coco.names \
    image_topic:=/camera/image_raw

3. Pose Estimation

Publishes vision_msgs/Detection2DArray with keypoints.

ros2 launch ros2_yolos_cpp pose.launch.py \
    model_path:=/path/to/yolo11n-pose.onnx \
    image_topic:=/camera/image_raw

4. Oriented Bounding Boxes (OBB)

Publishes custom ros2_yolos_cpp/OBBDetection2DArray with rotated bounding boxes.

ros2 launch ros2_yolos_cpp obb.launch.py \
    model_path:=/path/to/yolo11n-obb.onnx \
    labels_path:=/path/to/dota.names \
    image_topic:=/camera/image_raw

5. Image Classification

Publishes vision_msgs/Classification.

ros2 launch ros2_yolos_cpp classifier.launch.py \
    model_path:=/path/to/yolo11n-cls.onnx \
    labels_path:=/path/to/imagenet.names \
    image_topic:=/camera/image_raw

Starting ros2_yolo_cpp

ros2_yolos_cpp is manged by life cycle
After launching the launch file:

To Activate:

# Step 1: Configure (loads model, allocates GPU memory)
ros2 lifecycle set /yolos_detector configure

# Step 2: Activate (STARTS INFERENCE → publishes detections)
ros2 lifecycle set /yolos_detector activate

To Deactivate:

# 1. Deactivate → stops inference (active → inactive)
ros2 lifecycle set /yolos_detector deactivate

# 2. Cleanup → releases GPU/memory (inactive → unconfigured)
ros2 lifecycle set /yolos_detector cleanup

# 3. Shutdown → fully terminates node (unconfigured → finalized)
ros2 lifecycle set /yolos_detector shutdown

⚙️ Configuration

Nodes can be configured via launch arguments or a YAML parameter file. See config/default_params.yaml for a template.

Parameter	Type	Default	Description
`model_path`	string	Required	Absolute path to `.onnx` model file.
`labels_path`	string	""	Path to text file with class names (one per line).
`use_gpu`	bool	`false`	Enable CUDA acceleration (requires GPU build).
`conf_threshold`	double	`0.4`	Minimum confidence score to output detection.
`nms_threshold`	double	`0.45`	IoU threshold for Non-Maximum Suppression.
`image_topic`	string	`/camera/image_raw`	Topic to subscribe for input images.
`publish_debug_image`	bool	`true`	Publish standard `sensor_msgs/Image` with visualizations.

Topics

Node Type	Subscriptions	Publications
Detector	`~/image_raw`	`~/detections` (Detection2DArray) `~/debug_image` (Image)
Segmentor	`~/image_raw`	`~/detections` (Detection2DArray) `~/masks` (Image) `~/debug_image`
Pose	`~/image_raw`	`~/detections` (Detection2DArray) `~/debug_image`
OBB	`~/image_raw`	`~/detections` (OBBDetection2DArray) `~/debug_image`
Classifier	`~/image_raw`	`~/classification` (Classification) `~/debug_image`

🐳 Docker

Run the stack without installing dependencies locally.

# Build Docker image
docker build -t ros2_yolos_cpp .

# Run with GPU support
docker run --gpus all -it --rm \
    -v /path/to/models:/models \
    ros2_yolos_cpp \
    ros2 launch ros2_yolos_cpp detector.launch.py model_path:=/models/yolov8n.onnx

Name		Name	Last commit message	Last commit date
Latest commit History 26 Commits
.github/workflows		.github/workflows
config		config
include/ros2_yolos_cpp		include/ros2_yolos_cpp
launch		launch
models		models
msg		msg
src		src
test		test
.gitignore		.gitignore
CMakeLists.txt		CMakeLists.txt
LICENSE		LICENSE
README.md		README.md
docker_build.sh		docker_build.sh
package.xml		package.xml

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

🚀 Overview

Key Features

📥 Installation

Prerequisites

Requirement Check

CUDA & cuDNN Requirement for ONNX Runtime

CUDA 12.8 (WSL) & cuDNN 9.19.0

Build the ros2_yolos_cpp package

YOLO-CPP Package

🛠️ Usage

0. USB CAM Testing (WSL)

1. Object Detection

2. Instance Segmentation

3. Pose Estimation

4. Oriented Bounding Boxes (OBB)

5. Image Classification

Starting ros2_yolo_cpp

To Activate:

To Deactivate:

⚙️ Configuration

Topics

🐳 Docker

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

🚀 Overview

Key Features

📥 Installation

Prerequisites

Requirement Check

CUDA & cuDNN Requirement for ONNX Runtime

CUDA 12.8 (WSL) & cuDNN 9.19.0

Build the ros2_yolos_cpp package

YOLO-CPP Package

🛠️ Usage

0. USB CAM Testing (WSL)

1. Object Detection

2. Instance Segmentation

3. Pose Estimation

4. Oriented Bounding Boxes (OBB)

5. Image Classification

Starting ros2_yolo_cpp

To Activate:

To Deactivate:

⚙️ Configuration

Topics

🐳 Docker

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages