Capture The Flag - Developing a Pipeline for Binary Flag Detection

Malinovski, Stefan (2024) Capture The Flag - Developing a Pipeline for Binary Flag Detection. Bachelor's Thesis, Computing Science.

Preview	Text bCS2024StefanMalinovski.pdf Download (36MB) | Preview
	Text toestemming_ Stefan Malinovski _ degree programme_ Computing Science.pdf Restricted to Registered users only Download (150kB)

Abstract

Flag detection systems can have many useful applications in the world of safety and security. Recent advances in machine learning enable the development of these systems using deep learning approaches. However, the effectiveness of such methods heavily depends on the availability of large amounts of data. This bachelor’s thesis presents the development of a flag detection system using a YOLOv9 object detection model. Annotated flag images from the Open Images V7 dataset serve as the initial training data. To further address the lack of large-scale annotated datasets, various data enhancement methods are investigated with the goal of improving the model’s performance and robustness. Synthetic data is produced by a Stable Diffusion XL image generation model, as well as by programmatically altering stock photos of flags and pasting them onto natural image backgrounds. Augmented data is created by taking samples from the base dataset and applying transformations to them to create new images. The results shown in this paper confirm that these methods are viable answers to the lack of data and can be used to train a fast and accurate flag detection model.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Azzopardi, G.
Degree programme:	Computing Science
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	08 Jul 2024 12:05
Last Modified:	08 Jul 2024 12:05
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/33088

Actions (login required)

View Item