yolo_ros_plugin

Just a yolo_ros_plugin

This is a YOLO ROS Plugin through usage of OpenCV, which can support sensor_msgs::Image & sensor_msgs::CompressedImage datatype. Should be sufficient for the application requirement of AIRO-Lab members. Part of the code might not be very efficient, so suggestions or even pull requests are warmly welcomed :)

Yolov4

1. Pre-requisites:

# ROS Melodic || Noetic
# OpenCV >= 4.4
# Python 3.8

### for GPU usage:
# CUDNN >= 7.0

2. Utilization:

• Define your ros message topic at here.
• Define your config, weight, and name file at here. When you clone this repo, default should have Yolov4-tiny.
• In the same launch file, define your input type (under para: input_type obviously - duh). It's your choice whether you need depth/compressed or not.
• In the same launch file, define whether or not you are using GPU. We plan to write another tutorial on setting up GPU environs -- but not today.
• Launch! You can do rqt_image_view and check topic /showoff_this_dope_image.
• If you wish to do some further shit, in the yolo.cpp, do something with the object obj_vector after detection.

Yolov8

It's the year 2023, and Jesus mother of Christ, we have come all the way through Yolov8 since the first publication of You only look once. So a python version yolo-ros-plugin with v8 is also attached in this repo.

pip install ultralytics

and run

roslaunch yolo_node yolov8.launch

We did not draw the bounding boxes for you, neither saving those information in some random object. You might have to do something by yourself based on results = self.model(self.cv_image)[0].

Maintainer

pattylo @ AIRO-LAB @ RCUAS, HKPolyU