Cityscape-Adverse: Benchmarking Robustness of Semantic Segmentation With Realistic Scene Modifications via Diffusion-Based Image Editing

Recent advancements in generative AI, particularly diffusion-based image editing, have enabled the transformation of images into highly realistic scenes using only text instructions. This technology offers significant potential for generating diverse synthetic datasets to evaluate model robustness i...

Full description

Saved in:
Bibliographic Details
Main Authors: Naufal Suryanto, Andro Aprila Adiputra, Ahmada Yusril Kadiptya, Thi-Thu-Huong Le, Derry Pratama, Yongsu Kim, Howon Kim
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10870179/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850177752198545408
author Naufal Suryanto
Andro Aprila Adiputra
Ahmada Yusril Kadiptya
Thi-Thu-Huong Le
Derry Pratama
Yongsu Kim
Howon Kim
author_facet Naufal Suryanto
Andro Aprila Adiputra
Ahmada Yusril Kadiptya
Thi-Thu-Huong Le
Derry Pratama
Yongsu Kim
Howon Kim
author_sort Naufal Suryanto
collection DOAJ
description Recent advancements in generative AI, particularly diffusion-based image editing, have enabled the transformation of images into highly realistic scenes using only text instructions. This technology offers significant potential for generating diverse synthetic datasets to evaluate model robustness in real-world applications, especially in the era of autonomous driving. In this paper, we introduce Cityscape-Adverse, a benchmark that employs diffusion-based image editing to simulate eight adverse conditions, including variations in weather, lighting, and seasons, while preserving the original semantic labels. We evaluate the reliability of diffusion-based models in generating realistic scene modifications and assess the performance of state-of-the-art CNN and Transformer-based semantic segmentation models under these challenging conditions. Additionally, we analyze which modifications have the greatest impact on model performance and explore how training on synthetic datasets can improve robustness in real-world adverse scenarios. Our results demonstrate that all tested models, particularly CNN-based architectures, experienced significant performance degradation under extreme conditions, while Transformer-based models exhibited greater resilience. We verify that models trained on Cityscape-Adverse show significantly enhanced resilience when applied to unseen domains. Code and datasets will be released at <uri>https://github.com/naufalso/cityscape-adverse</uri>.
format Article
id doaj-art-43af9a6e7d1e47dcb3cd9e0a1b9a3d93
institution OA Journals
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-43af9a6e7d1e47dcb3cd9e0a1b9a3d932025-08-20T02:18:55ZengIEEEIEEE Access2169-35362025-01-0113699216994010.1109/ACCESS.2025.353798110870179Cityscape-Adverse: Benchmarking Robustness of Semantic Segmentation With Realistic Scene Modifications via Diffusion-Based Image EditingNaufal Suryanto0https://orcid.org/0000-0002-0396-5938Andro Aprila Adiputra1Ahmada Yusril Kadiptya2https://orcid.org/0009-0008-8612-9812Thi-Thu-Huong Le3https://orcid.org/0000-0002-8366-9396Derry Pratama4https://orcid.org/0000-0003-2246-3017Yongsu Kim5Howon Kim6https://orcid.org/0000-0001-8475-7294IoT Research Center, Pusan National University, Busan, Republic of KoreaSchool of Computer Science and Engineering, Pusan National University, Busan, Republic of KoreaSchool of Computer Science and Engineering, Pusan National University, Busan, Republic of KoreaBlockchain Platform Research Center, Pusan National University, Busan, Republic of KoreaSchool of Computer Science and Engineering, Pusan National University, Busan, Republic of KoreaSmartM2M, Busan, Republic of KoreaSchool of Computer Science and Engineering, Pusan National University, Busan, Republic of KoreaRecent advancements in generative AI, particularly diffusion-based image editing, have enabled the transformation of images into highly realistic scenes using only text instructions. This technology offers significant potential for generating diverse synthetic datasets to evaluate model robustness in real-world applications, especially in the era of autonomous driving. In this paper, we introduce Cityscape-Adverse, a benchmark that employs diffusion-based image editing to simulate eight adverse conditions, including variations in weather, lighting, and seasons, while preserving the original semantic labels. We evaluate the reliability of diffusion-based models in generating realistic scene modifications and assess the performance of state-of-the-art CNN and Transformer-based semantic segmentation models under these challenging conditions. Additionally, we analyze which modifications have the greatest impact on model performance and explore how training on synthetic datasets can improve robustness in real-world adverse scenarios. Our results demonstrate that all tested models, particularly CNN-based architectures, experienced significant performance degradation under extreme conditions, while Transformer-based models exhibited greater resilience. We verify that models trained on Cityscape-Adverse show significantly enhanced resilience when applied to unseen domains. Code and datasets will be released at <uri>https://github.com/naufalso/cityscape-adverse</uri>.https://ieeexplore.ieee.org/document/10870179/Adverse conditionsbenchmarkdatasetdiffusion-based image editinggenerative AImodel robustness
spellingShingle Naufal Suryanto
Andro Aprila Adiputra
Ahmada Yusril Kadiptya
Thi-Thu-Huong Le
Derry Pratama
Yongsu Kim
Howon Kim
Cityscape-Adverse: Benchmarking Robustness of Semantic Segmentation With Realistic Scene Modifications via Diffusion-Based Image Editing
IEEE Access
Adverse conditions
benchmark
dataset
diffusion-based image editing
generative AI
model robustness
title Cityscape-Adverse: Benchmarking Robustness of Semantic Segmentation With Realistic Scene Modifications via Diffusion-Based Image Editing
title_full Cityscape-Adverse: Benchmarking Robustness of Semantic Segmentation With Realistic Scene Modifications via Diffusion-Based Image Editing
title_fullStr Cityscape-Adverse: Benchmarking Robustness of Semantic Segmentation With Realistic Scene Modifications via Diffusion-Based Image Editing
title_full_unstemmed Cityscape-Adverse: Benchmarking Robustness of Semantic Segmentation With Realistic Scene Modifications via Diffusion-Based Image Editing
title_short Cityscape-Adverse: Benchmarking Robustness of Semantic Segmentation With Realistic Scene Modifications via Diffusion-Based Image Editing
title_sort cityscape adverse benchmarking robustness of semantic segmentation with realistic scene modifications via diffusion based image editing
topic Adverse conditions
benchmark
dataset
diffusion-based image editing
generative AI
model robustness
url https://ieeexplore.ieee.org/document/10870179/
work_keys_str_mv AT naufalsuryanto cityscapeadversebenchmarkingrobustnessofsemanticsegmentationwithrealisticscenemodificationsviadiffusionbasedimageediting
AT androaprilaadiputra cityscapeadversebenchmarkingrobustnessofsemanticsegmentationwithrealisticscenemodificationsviadiffusionbasedimageediting
AT ahmadayusrilkadiptya cityscapeadversebenchmarkingrobustnessofsemanticsegmentationwithrealisticscenemodificationsviadiffusionbasedimageediting
AT thithuhuongle cityscapeadversebenchmarkingrobustnessofsemanticsegmentationwithrealisticscenemodificationsviadiffusionbasedimageediting
AT derrypratama cityscapeadversebenchmarkingrobustnessofsemanticsegmentationwithrealisticscenemodificationsviadiffusionbasedimageediting
AT yongsukim cityscapeadversebenchmarkingrobustnessofsemanticsegmentationwithrealisticscenemodificationsviadiffusionbasedimageediting
AT howonkim cityscapeadversebenchmarkingrobustnessofsemanticsegmentationwithrealisticscenemodificationsviadiffusionbasedimageediting