This repository is a ROS2 package that performs object detection using YOLOv8 with ONNX.

Subscribed Topic:

• Image-stream : sensor_msgs/msg/Image
  

Published Topic:

• Image-stream with bounding box around detected objects : sensor_msgs/msg/Image
  

You can easily convert using the ultralytics python library.

pip install ultralytics

Please write down a Python file to convert a PyTorch model to ONNX format.

from ultralytics import YOLO

model_name = 'yolov8m' #@param ["yolov8n", "yolov8s", "yolov8m", "yolov8l", "yolov8x"]
input_width = 640 #@param {type:"slider", min:32, max:4096, step:32}
input_height = 480 #@param {type:"slider", min:32, max:4096, step:32}
optimize_cpu = True

model = YOLO(f"{model_name}.pt") 
model.export(format="onnx", imgsz=[input_height,input_width], optimize=optimize_cpu)

Move the ONNX file into the resource folder of the ROS2 package.

mv (yolov8m.onnx) (your_ws)/src/yolo_v8/resources

ros2 launch yolo_v8 detection.xml

• EUROC dataset experiment with Workstation CPU

  • Watch on YouTube

• Real-life situation experiment with Raspi5-robot (real-time)

  • TBD

• https://github.com/ConfusionTechnologies/ros-yolov5-node
• https://github.com/ultralytics/ultralytics

• Minje Kim ([email protected])