YOLO TensorRT ROS2 Package

This ROS2 package provides an integration of YOLONAS with TensorRT for high-performance object detection. The package allows flexibility in specifying input topics and model names through launch parameters.

Demonstration - Object Detection with Custom Model

Check out the following gifs demonstrating the YOLONAS TensorRT ROS2 package in action:

YOLO NAS TRT in Real-Time:

1. Flower Detection in Lab

2. Flower Detection in the Field

Quick Start

To quickly get started with the YOLO TensorRT ROS2 package, follow these steps:

1. Install prerequisites:

   Before using this package, ensure you have the following dependencies installed:

• ROS2 Humble
• TensorRT tensorrt==8.6.1.post1
• Pycuda pycuda==2024.1
• OpenCV
• PIL (Python Imaging Library)
• vision_msgs
• cv_bridge

pip install tensorrt==8.6.1.post1 --extra-index-url https://pypi.nvidia.com
pip install pycuda==2024.1
pip install opencv-python
pip install pillow
sudo apt-get install ros-humble-vision-msgs
sudo apt-get install ros-humble-cv-bridge

2. Clone the repository and install dependencies:

   cd ~/ros2_ws/src
   git clone https://github.com/djetshu/yolonas_trt_ros2.git
   cd ~/ros2_ws
   rosdep install --from-paths src --ignore-src -r -y

3. Build the package:

   colcon build --packages-select yolo_py

4. Source the workspace:

   source ~/ros2_ws/install/setup.bash

5. Download the TensorRT model and ROSBAG files from this link.

6. Move the YOLO NAS TRT model to the following directory:

   mv /path/to/downloaded/yolo_nas_m_flower.trt ~/ros2_ws/src/yolonas_trt_ros2/models/

7. Modify the class names to your custom classes' names in ros2_ws/src/yolonas_trt_ros2/yolo_py/dataset.py.

8. Launch the YOLO node with the custom model:

   ros2 launch yolo_py yolo_trt.launch.py input_topic:=/oak/rgb/image_raw model_name:=yolo_nas_m_flower.trt

Usage

To use this package, you need to launch the yolo_trt node. The package includes a launch file that allows you to specify the input topic and the model name.

Launch Parameters

• rviz_config: Path to the RViz config file. Default is set to the RViz config provided in the package.
• rviz_en: Enable or disable RViz. Default is true.
• input_topic: The input topic for YOLO. Default is /camera/image.
• model_name: The name of the YOLO model to use. Default is yolo_nas_m.trt.

Nodes

YOLO Node

This node subscribes to an image topic, processes the image with YOLO using TensorRT, and publishes the detection results.

• Subscribed Topics:

  • /image (sensor_msgs/Image): Input image topic.

• Published Topics:

  • /detection (vision_msgs/Detection2DArray): Detection results.
  • /detection_image (sensor_msgs/Image): Image with detection results.

Example Launch Commands

To launch the YOLO node:

ros2 launch yolo_py yolo_trt.launch.py input_topic:=/your_custom_topic model_name:=your_custom_model.trt

Contact Information

For inquiries, collaboration opportunities, or questions feel free to contact:

• Email: [email protected]

Previous Step: Training YoloNAS and Conversion to TensorRT

Consult this previous repository to train YoloNAS and convert to TensorRT model.