User-Friendly Interface for Training and Deployment of YOLO Models for Object Detection

Abstract

Introduction: The growing demand for efficient real-time object detection algorithms has led to the development of the YOLO (You Only Look Once) architecture, which has revolutionized the field of machine vision. This research introduces a user-friendly interface designed for students and researchers to easily train, validate, and deploy YOLO models.

 
Aim: The aim of this research is to develop a user-friendly software interface that enables data import, hyperparameter configuration, model training, and object detection on images, on videos, and in real-time via webcam, with a focus on educational and research applications. 

Materials and Methods: The software was developed using Python with Ultralytics YOLO, OpenCV, and PyQt5 libraries for the user interface. It consists of four modules: a training module, an image detection module, a video detection module, and a real-time detection module. Development included testing on Windows platforms with NVIDIA GPU support, using public datasets like COCO for validation. 

Results: The developed interface achieved high efficiency, with detection speeds up to 30 FPS in real-time. The modules were tested on various datasets, showing precision up to 95% on validation sets. The user interface enables intuitive navigation, and implementation on the recommended hardware configuration ensures reliable performance.

Conclusion: This research demonstrates the potential of the user-friendly interface to facilitate work with YOLO models in educational and research environments. By automating processes, the system reduces barriers for users and improves access to advanced machine vision techniques. Future work will focus on integrating newer YOLO variants and support for multiple operating systems.