fateshelled / EdgeYOLO-ROS

EdgeYOLO + ROS 2 object detection package

EdgeYOLO-ROS

EdgeYOLO + ROS 2 Inference DEMO

Supported List

• ONNXRuntime C++ and Python
• TensorRT C++
• OpenVINO C++

※ TFLite inference demo code has been uploaded but not tested.

Requirements

• ROS 2 Foxy or Humble
• OpenCV
• ONNXRuntime or TensorRT or OpenVINO
• bbox_ex_msgs
• v4l2-camera (for Webcam Demo)

Build

cd ros2_ws/src
git clone https://github.com/Ar-Ray-code/bbox_ex_msgs
git clone https://github.com/fateshelled/EdgeYOLO-ROS
cd ../

# for C++
colcon build --symlink-install --packages-up-to edgeyolo_ros_cpp bbox_ex_msgs

# for python
colcon build --symlink-install --packages-up-to edgeyolo_ros_py bbox_ex_msgs

Model Download

• Model download from PINTO_model_zoo
  • use nopost model

Webcam DEMO

ONNXRuntime

# Run C++ Node
ros2 launch edgeyolo_ros_cpp edgeyolo_onnxruntime.launch.py \
      model_path:=edgeyolo_tiny_coco_416x416.onnx

# Run Python Node
ros2 launch edgeyolo_ros_py edgeyolo_onnxruntime.launch.py \
      model_path:=edgeyolo_tiny_coco_416x416.onnx

OpenVINO

# model convert
mo.py --input_model edgeyolo_tiny_coco_416x416.onnx \
      --input_shape [1,3,416,416] \
      --output_dir ./openvino_model

# Run C++ Node
ros2 launch edgeyolo_ros_cpp edgeyolo_openvino.launch.py \
      model_path:=./openvino_model/edgeyolo_tiny_coco_416x416.xml

TensorRT

# model convert
trtexec --onnx=edgeyolo_tiny_coco_416x416.onnx \
        --saveEngine=edgeyolo_tiny_coco_416x416.engine

# Run C++ Node
ros2 launch edgeyolo_ros_cpp edgeyolo_tensorrt.launch.py \
      model_path:=edgeyolo_tiny_coco_416x416.engine

Reference

• EdgeYOLO
• YOLOX-ROS
• PINTO_model_zoo