Advancing Autonomous Driving: CNNs vs. Transformers for Semantic Image Segmentation

Shrivas, Ayush Kumar (2025) Advancing Autonomous Driving: CNNs vs. Transformers for Semantic Image Segmentation. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (5MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

This study evaluates CNN-based and hybrid transformer-based models for semantic image segmentation in autonomous vehicles, which addressing the need for accurate, real-time environmental understanding. Using the Cityscapes dataset, five models—UNet, SegNet, PSPNet (CNN-based), SegFormer, and UNETR (transformer-based) were compared under a unified pipeline. Results show that the transformer-based SegFormer model achieved the highest performance (87% pixel accuracy, 75% mIoU, 86% Dice Score), excelling in complex urban scenarios, while CNNs were more efficient, they were less precise in challenging scenes. The research demonstrates the superior segmentation capabilities of transformer-based models for autonomous driving but notes their higher computational demands, highlighting the importance of further optimization and broader datasets for real-world deployment.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Abgaz, Yalemisew UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science H Social Sciences > HE Transportation and Communications > Urban Transportation
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	03 Jul 2026 11:03
Last Modified:	03 Jul 2026 11:03
URI:	https://norma.ncirl.ie/id/eprint/9463

Actions (login required)

View Item