YOLOv9MIT-ROS

An MIT rewrite of YOLOv9 + ROS 2 Humble Object Detection demo

Supported List

• TensorRT C++

Requirements

• ROS 2 Humble or rolling
• OpenCV
• TensorRT
• vision_msgs
• v4l2_camera (for Webcam DEMO)

Build

cd ros2_ws/src
git clone https://github.com/fateshelled/YOLOv9MIT-ROS
cd ..

colcon build --symlink-install --packages-up-to yolov9mit_ros

Model

Download

• Releases

Convert to TensorRT Engine

# S model
wget https://github.com/fateshelled/YOLOv9MIT-ROS/releases/download/v1.0.0/v9-s.vec2box.sim.onnx
trtexec --onnx=v9-s.vec2box.sim.onnx \
        --saveEngine=v9-s.vec2box.sim.engine

# M model
wget https://github.com/fateshelled/YOLOv9MIT-ROS/releases/download/v1.0.0/v9-m.vec2box.sim.onnx
trtexec --onnx=v9-m.vec2box.sim.onnx \
        --saveEngine=v9-m.vec2box.sim.engine

# C model
wget https://github.com/fateshelled/YOLOv9MIT-ROS/releases/download/v1.0.0/v9-c.vec2box.sim.onnx
trtexec --onnx=v9-c.vec2box.sim.onnx \
        --saveEngine=v9-c.vec2box.sim.engine

Webcam Demo

export YOLOV9_MODEL_PATH="v9-c.vec2box.sim.engine"

ros2 launch yolov9mit_ros launch.xml model_path:=${YOLOV9_MODEL_PATH} video_device:=/dev/video0 imshow:=True

Topic

Subscribe

• image_raw: sensor_msgs/Image
  • Inference input image message

Publish

• yolov9mit_ros/detections: vision_msgs/msg/Detection2DArray
  • BoundingBox output
    • id (string): class name
    • bbox (vision_msgs/msg/BoundingBox2D): bounding box (center xy + width, height)
    • results (vector of vision_msgs/msg/ObjectHypothesisWithPose)
      • class_id (string): class index
      • score (double): confidence
• yolov9mit_ros/image_raw: sensor_msgs/Image
  • Image with object detection results