| Title |
Urban road segmentation with transformers |
| Authors |
Lisauskas, Bartas ; Maskeliūnas, Rytis |
| Full Text |
|
| Is Part of |
CEUR Workshop proceedings: IVUS 2025: proceedings of the 30th international conference on information society and university studies (IVUS 2025), Kaunas, Lithuania, 15 May 2025 / edited by: I. Veitaitė, A. Lopata, T. Krilavičius, M. Woźniak.. Aachen : CEUR-WS. 2026, vol. 4213, p. 211-218.. ISSN 1613-0073 |
| Keywords [eng] |
Computer vision ; Deep learning ; Image processing ; Neural networks ; Semantic segmentation |
| Abstract [eng] |
This paper introduces a transformer-based computer vision system for segmenting different urban road scenes. Detection and understanding of objects in the environment is a critical task for autonomous vehicles or advanced self-driving robots. The approach integrates a MiT transformer-based backbone network for feature extraction with a decoder that incorporates CNN depthwise separable convolution layers, to efficiently fuse features and reduce computational cost. The system detects and separates different objects and environments into 19 semantic classes, as defined by the Cityscapes dataset. The computer vision model consists of 44.61 million parameters and reaches the mean intersection over union of the 73.95% accuracy metric with the Cityscapes dataset. The results gathered demonstrate the good ability of the model to detect different objects and environments in urban road scenes. The proposed computer vision system approach demonstrates the balance between good segmentation accuracy and efficient network structure for more reliable autonomous solutions in complex urban environments. |
| Published |
Aachen : CEUR-WS |
| Type |
Conference paper |
| Language |
English |
| Publication date |
2026 |
| CC license |
|
