Abstract

Data augmentation is crucial for image segmentation, especially in transfer learning with limited data, however it can be costly. This study examines the cost-benefit of augmentation in facade segmentation using unmanned aerial vehicles (UAV) data. We analysed how dataset size and offline augmentation impact classification quality and computation using DeepLabV3+ architecture. Expanding the dataset from 5 to 480 thousand tiles improved segmentation efficiency by an average of 3.7%. Beyond a certain point, further dataset expansion yields minimal gains, in our case, just 1%, on average, after segmentation accuracy plateaued at around 76%. These findings help avoid the computational and time costs of ineffective data augmentation.

Recommended Citation

Balak, P., Łysak, A., Choromański, K. & Luckner, M. (2025). Influence of Augmentation of UAV Collected Data on Deep Learning Based Facade Segmentation TaskIn I. Luković, S. Bjeladinović, B. Delibašić, D. Barać, N. Iivari, E. Insfran, M. Lang, H. Linger, & C. Schneider (Eds.), Empowering the Interdisciplinary Role of ISD in Addressing Contemporary Issues in Digital Transformation: How Data Science and Generative AI Contributes to ISD (ISD2025 Proceedings). Belgrade, Serbia: University of Gdańsk, Department of Business Informatics & University of Belgrade, Faculty of Organizational Sciences. ISBN: 978-83-972632-1-5. https://doi.org/10.62036/ISD.2025.64

Paper Type

Poster

DOI

10.62036/ISD.2025.64

Download

Share

COinS
 

Influence of Augmentation of UAV Collected Data on Deep Learning Based Facade Segmentation Task

Data augmentation is crucial for image segmentation, especially in transfer learning with limited data, however it can be costly. This study examines the cost-benefit of augmentation in facade segmentation using unmanned aerial vehicles (UAV) data. We analysed how dataset size and offline augmentation impact classification quality and computation using DeepLabV3+ architecture. Expanding the dataset from 5 to 480 thousand tiles improved segmentation efficiency by an average of 3.7%. Beyond a certain point, further dataset expansion yields minimal gains, in our case, just 1%, on average, after segmentation accuracy plateaued at around 76%. These findings help avoid the computational and time costs of ineffective data augmentation.