YOLOv5 ROS

This is a ROS interface for using YOLOv5 for real time object detection on a ROS image topic. It supports inference on multiple deep learning frameworks used in the official YOLOv5 repository.

Installation

Dependencies

This package is built and tested on Ubuntu 20.04 LTS and ROS Noetic with Python 3.8.

Clone the packages to ROS workspace and install requirement for YOLOv5 submodule:

cd <ros_workspace>/src
git clone https://github.com/mats-robotics/detection_msgs.git
git clone --recurse-submodules https://github.com/mats-robotics/yolov5_ros.git 
cd yolov5_ros/src/yolov5
pip install -r requirements.txt # install the requirements for yolov5

Build the ROS package:

cd <ros_workspace>
catkin build yolov5_ros # build the ROS package

Make the Python script executable

cd <ros_workspace>/src/yolov5_ros/src
chmod +x detect.py

Basic usage

Change the parameter for input_image_topic in launch/yolov5.launch to any ROS topic with message type of sensor_msgs/Image or sensor_msgs/CompressedImage. Other parameters can be modified or used as is.

Launch the node:

roslaunch yolov5_ros yolov5.launch

Using custom weights and dataset (Working)

Put your weights into yolov5_ros/src/yolov5
Put the yaml file for your dataset classes into yolov5_ros/src/yolov5/data
Change related ROS parameters in yolov5.launch: weights, data

Known Issue

If roslaunch yolov5_ros yolov5.launch command, Then you able to following Error : AssertionError: Invalid CUDA '-device cuda' requested, use '-device cpu' or pass valid CUDA device(s) Then check your pytorch version (depend on your cuda version)

Reference

YOLOv5 official repository: https://github.com/ultralytics/yolov5
YOLOv3 ROS PyTorch: https://github.com/eriklindernoren/PyTorch-YOLOv3
Darknet ROS: https://github.com/leggedrobotics/darknet_ros