Closing the Domain Gap: Can Pseudo-Labels from Synthetic UAV Data Enable Real-World Flood Segmentation?

We present a novel methodology for generating and filtering synthetic Unmanned Aerial Vehicle (UAV) flood imagery to enhance the generalization capabilities of segmentation models. Our framework combines text-to-image synthesis and image inpainting, using curated prompts and real-world segmentation...

Full description

Saved in:
Bibliographic Details
Main Authors: Georgios Simantiris, Konstantinos Bacharidis, Costas Panagiotakis
Format: Article
Language:English
Published: MDPI AG 2025-06-01
Series:Sensors
Subjects:
Online Access:https://www.mdpi.com/1424-8220/25/12/3586
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849425989567774720
author Georgios Simantiris
Konstantinos Bacharidis
Costas Panagiotakis
author_facet Georgios Simantiris
Konstantinos Bacharidis
Costas Panagiotakis
author_sort Georgios Simantiris
collection DOAJ
description We present a novel methodology for generating and filtering synthetic Unmanned Aerial Vehicle (UAV) flood imagery to enhance the generalization capabilities of segmentation models. Our framework combines text-to-image synthesis and image inpainting, using curated prompts and real-world segmentation masks to produce diverse and realistic flood scenes. To overcome the lack of human annotations, we employ an unsupervised pseudo-labeling method that generates segmentation masks based on floodwater appearance characteristics. We further introduce a filtering stage based on outlier detection in feature space to improve the realism of the synthetic dataset. Experimental results on five state-of-the-art flood segmentation models show that synthetic data can closely match real data in training performance, and combining both sources improves model robustness by 1–7%. Finally, we investigate the impact of prompt design on the visual fidelity of generated images and provide qualitative and quantitative evidence of distributional similarity between real and synthetic data.
format Article
id doaj-art-642c22d62a6847b98a26e8a6a1ef6c06
institution Kabale University
issn 1424-8220
language English
publishDate 2025-06-01
publisher MDPI AG
record_format Article
series Sensors
spelling doaj-art-642c22d62a6847b98a26e8a6a1ef6c062025-08-20T03:29:35ZengMDPI AGSensors1424-82202025-06-012512358610.3390/s25123586Closing the Domain Gap: Can Pseudo-Labels from Synthetic UAV Data Enable Real-World Flood Segmentation?Georgios Simantiris0Konstantinos Bacharidis1Costas Panagiotakis2Department of Management Science and Technology, Hellenic Mediterranean University, 72100 Agios Nikolaos, GreeceDepartment of Management Science and Technology, Hellenic Mediterranean University, 72100 Agios Nikolaos, GreeceDepartment of Management Science and Technology, Hellenic Mediterranean University, 72100 Agios Nikolaos, GreeceWe present a novel methodology for generating and filtering synthetic Unmanned Aerial Vehicle (UAV) flood imagery to enhance the generalization capabilities of segmentation models. Our framework combines text-to-image synthesis and image inpainting, using curated prompts and real-world segmentation masks to produce diverse and realistic flood scenes. To overcome the lack of human annotations, we employ an unsupervised pseudo-labeling method that generates segmentation masks based on floodwater appearance characteristics. We further introduce a filtering stage based on outlier detection in feature space to improve the realism of the synthetic dataset. Experimental results on five state-of-the-art flood segmentation models show that synthetic data can closely match real data in training performance, and combining both sources improves model robustness by 1–7%. Finally, we investigate the impact of prompt design on the visual fidelity of generated images and provide qualitative and quantitative evidence of distributional similarity between real and synthetic data.https://www.mdpi.com/1424-8220/25/12/3586unsupervised image segmentationdeep learningpseudo-labelsflood segmentationunmanned aerial vehiclesimage inpainting
spellingShingle Georgios Simantiris
Konstantinos Bacharidis
Costas Panagiotakis
Closing the Domain Gap: Can Pseudo-Labels from Synthetic UAV Data Enable Real-World Flood Segmentation?
Sensors
unsupervised image segmentation
deep learning
pseudo-labels
flood segmentation
unmanned aerial vehicles
image inpainting
title Closing the Domain Gap: Can Pseudo-Labels from Synthetic UAV Data Enable Real-World Flood Segmentation?
title_full Closing the Domain Gap: Can Pseudo-Labels from Synthetic UAV Data Enable Real-World Flood Segmentation?
title_fullStr Closing the Domain Gap: Can Pseudo-Labels from Synthetic UAV Data Enable Real-World Flood Segmentation?
title_full_unstemmed Closing the Domain Gap: Can Pseudo-Labels from Synthetic UAV Data Enable Real-World Flood Segmentation?
title_short Closing the Domain Gap: Can Pseudo-Labels from Synthetic UAV Data Enable Real-World Flood Segmentation?
title_sort closing the domain gap can pseudo labels from synthetic uav data enable real world flood segmentation
topic unsupervised image segmentation
deep learning
pseudo-labels
flood segmentation
unmanned aerial vehicles
image inpainting
url https://www.mdpi.com/1424-8220/25/12/3586
work_keys_str_mv AT georgiossimantiris closingthedomaingapcanpseudolabelsfromsyntheticuavdataenablerealworldfloodsegmentation
AT konstantinosbacharidis closingthedomaingapcanpseudolabelsfromsyntheticuavdataenablerealworldfloodsegmentation
AT costaspanagiotakis closingthedomaingapcanpseudolabelsfromsyntheticuavdataenablerealworldfloodsegmentation