Ar-Ray-code/YOLOX-ROS

What it solves

YOLOX-ROS bridges the gap between high-performance object detection and the ROS2 robotics ecosystem. It allows developers to integrate the YOLOX detection model into their robotics applications to identify and track objects in real-time using standard ROS2 communication protocols.

How it works

The project acts as a ROS2 node that subscribes to raw image data from a camera topic. It processes these images using the YOLOX model and publishes the detected object locations as bounding boxes. It supports multiple hardware acceleration backends, including TensorRT, OpenVINO, ONNXRuntime, and TFLite, ensuring the detection runs efficiently on various hardware platforms.

Who it’s for

This tool is designed for robotics engineers and developers working with ROS2 who need to add real-time computer vision capabilities to their robots or autonomous systems.

Highlights

• Supports multiple high-performance inference engines: TensorRT, OpenVINO, ONNXRuntime (CUDA/MIGraphX), and TFLite.
• Native ROS2 integration using standard image and bounding box message formats.
• Compatible with common detection message types like vision_msgs/Detection2DArray and bboxes_ex_msgs/BoundingBoxes.
• Ready for use in simulation environments like Unity.