Towards Robust Cross-Dataset Pothole Detection Through Multi-Dataset Pretraining
DOI:
https://doi.org/10.59395/ijadis.v7i2.1551Keywords:
Pothole Detection, Cross-dataset evaluation, Domain shift, Generalization gap, Model robustnessAbstract
Pothole detection based on deep learning has achieved high detection accuracy; however, most existing studies evaluate models using the same dataset for both training and testing, providing limited evidence of robustness under unseen data distributions. This study investigated the cross-dataset generalization capability of YOLOv8n using three publicly available pothole datasets with different visual characteristics: the Multi-Weather Pothole Dataset (MWPD), the Jaygala dataset, and the Andrew dataset. The proposed framework evaluated same-dataset and cross-dataset detection performance, quantified robustness through generalization gap analysis, examined the influence of dataset characteristics, and assessed the effectiveness of multi-dataset pretraining. Experimental results showed that the Andrew dataset achieved the highest same-dataset performance (mAP@50 = 0.816) but also exhibited the largest generalization gap (0.246), indicating limited robustness across datasets. In contrast, multi-dataset pretraining reduced the generalization gap for MWPD from 0.093 to 0.048, demonstrating improved cross-dataset robustness, although the improvement was not consistent across all datasets. These findings indicate that same-dataset accuracy alone is insufficient for evaluating model robustness and that cross-dataset evaluation provides a more realistic assessment of deployment performance in heterogeneous road environments.
Downloads
References
[1] B. P. Statistik, Statistik Transportasi Darat 2024, Jakarta, 2024. [Online]. Available: https://www.bps.go.id/id/publication/2025/12/01/ed7ff73d58fc0719ee3df145/statistik-transportasi-darat-2024.html.
[2] A. Salunke, Road Accident Detection Using AI and ML., Interantional J. Sci. Res. Eng. Manag., vol. 09, no. 05, pp. 19, 2025, doi: 10.55041/ijsrem48619. DOI: https://doi.org/10.55041/IJSREM48619
[3] A. A. Sami, S. M. N. Sakib, K. Deb, and I. H. Sarker, Improved YOLOv5-Based Real-Time Road Pavement Damage Detection in Road Infrastructure Management, Algorithms, vol. 16, no. 9, p. 452, 2023, doi: 10.3390/a16090452. DOI: https://doi.org/10.3390/a16090452
[4] K. S. Sri, M. Prasanna, K. V VENKATARAO, K. V Subrahmanyam, and P. Ramesh, Deep Learning-Based Classification of Road Conditions Using a Customized VGG-Style CNN for Pothole Detection, Int. J. Eng. Sci. Invent., vol. 14, no. 4, pp. 7380, 2025, doi: 10.35629/6734-14047380. DOI: https://doi.org/10.35629/6734-14047380
[5] K. R. Ahmed, Smart pothole detection using deep learning based on dilated convolution, Sensors, vol. 21, no. 24, 2021, doi: 10.3390/s21248406. DOI: https://doi.org/10.3390/s21248406
[6] K. W. Yuanhang Tang, Kexin Li, Research on intelligent detection of pavement damage based on CNN, Math. Model. Eng., vol. 8, no. 4, pp. 98107, 2022, doi: 10.21595/mme.2022.22918. DOI: https://doi.org/10.21595/mme.2022.22918
[7] H. Samma, S. A. Suandi, N. A. Ismail, S. Sulaiman, and L. L. Ping, Evolving Pre-Trained CNN Using Two-Layers Optimizer for Road Damage Detection from Drone Images, IEEE Access, vol. 9, pp. 158215158226, 2021, doi: 10.1109/ACCESS.2021.3131231. DOI: https://doi.org/10.1109/ACCESS.2021.3131231
[8] G. Guo and Z. Zhang, Road damage detection algorithm for improved YOLOv5, Sci. Rep., pp. 112, 2022, doi: 10.1038/s41598-022-19674-8. DOI: https://doi.org/10.1038/s41598-022-19674-8
[9] G. Yu and X. Zhou, An Improved YOLOv5 Crack Detection Method Combined with a Bottleneck Transformer, Mathematics, vol. 11, no. 10, pp. 112, 2023. DOI: https://doi.org/10.3390/math11102377
[10] J. Wang, K. Wang, and K. Li, Research on road damage recognition and classification based on improved VGG-19, Math. Model. Eng., vol. 9, no. 4, pp. 115129, 2023, doi: 10.21595/mme.2023.23455. DOI: https://doi.org/10.21595/mme.2023.23455
[11] S. Zhou et al., Enhancing autonomous pavement crack detection: Optimizing YOLOv5s algorithm with advanced deep learning techniques, Measurement, vol. 240, no. May 2024, p. 115603, 2025, doi: 10.1016/j.measurement.2024.115603. DOI: https://doi.org/10.1016/j.measurement.2024.115603
[12] J. Zeng and H. Zhong, YOLOv8 PD: an improved road damage detection algorithm based on YOLOv8n model, Sci. Rep., pp. 114, 2024, doi: 10.1038/s41598-024-62933-z. DOI: https://doi.org/10.21203/rs.3.rs-4199735/v1
[13] S. Parvin, F. Munsy, T. Rahat, A. Nahar, and K. Nur, Results in Engineering Robust multi-weather pothole detection: An enhanced YOLOv9 trained on the MWPD dataset, Results Eng., vol. 28, no. August, p. 107817, 2025, doi: 10.1016/j.rineng.2025.107817. DOI: https://doi.org/10.1016/j.rineng.2025.107817
[14] Y. Liu, S. Zhou, X. Liu, C. Hao, B. Fan, and J. Tian, Unbiased Faster R-CNN for Single-source Domain Generalized Object Detection, in 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 2883828847, doi: 10.1109/cvpr52733.2024.02724. DOI: https://doi.org/10.1109/CVPR52733.2024.02724
[15] C. Diaz-Ruiz et al., Ithaca365: Dataset and Driving Perception under Repeated and Challenging Weather Conditions, in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 2135121360, doi: 10.1109/cvpr52688.2022.02069. DOI: https://doi.org/10.1109/CVPR52688.2022.02069
[16] J. Wei, Q. W. Id, and Z. Zhao, YOLO-G: Improved YOLO for cross-domain object detection, PLoS One, vol. 19, no. 9, pp. 116, 2023, doi: 10.1371/journal.pone.0291241. DOI: https://doi.org/10.1371/journal.pone.0291241
[17] G. Li, Z. Ji, X. Qu, R. Zhou, and D. Cao, Cross-Domain Object Detection for Autonomous Driving: A Stepwise Domain Adaptative YOLO Approach, IEEE Trans. Intell. Veh., vol. 7, no. 3, pp. 603615, 2022. DOI: https://doi.org/10.1109/TIV.2022.3165353
[18] M. Jakubec, E. Lieskovska, B. Bucko, and K. Zabovska, Pothole detection in adverse weather: leveraging synthetic images and attention based object detection methods, Multimed. Tools Appl., vol. 83, no. 39, pp. 8695586982, 2024, doi: 10.1007/s11042-024-19723-6. DOI: https://doi.org/10.1007/s11042-024-19723-6
[19] S. Lamchoudi and A. Mourhir, Pothole Detection for Road Safety on Moroccan Roads, 2025 IEEE 8th Congr. Inf. Sci. Technol., pp. 336341, 2025, doi: 10.1109/cist65886.2025.11224074. DOI: https://doi.org/10.1109/CiSt65886.2025.11224074
[20] R. Buettner, S. Member, S. Bertram, and L. Fischer-brandies, The Impact of Combining Datasets on the Robustness of Deep Learning Architectures: A Cross-Dataset Analysis, IEEE Access, vol. 13, no. August, pp. 151993152009, 2025, doi: 10.1109/ACCESS.2025.3604689. DOI: https://doi.org/10.1109/ACCESS.2025.3604689
[21] A. Vijayakumar and S. Vairavasundaram, YOLO-based Object Detection Models: A Review and its Applications, Multimed. Tools Appl., vol. 83, pp. 8353583574, 2024, doi: 10.1007/s11042-024-18872-y. DOI: https://doi.org/10.1007/s11042-024-18872-y
[22] Y. Zanevych, V. Yovbak, O. Basystiuk, N. Shakhovska, S. Fedushko, and S. Argyroudis, Evaluation of Pothole Detection Performance Using Deep Learning Models Under Low-Light Conditions, Sustainability, 2024, doi: 10.3390/su162410964. DOI: https://doi.org/10.3390/su162410964
[23] M. Cantone, C. Marrocco, and A. Bria, Machine Learning in Network Intrusion Detection: A Cross-Dataset Generalization Study, IEEE Access, vol. 12, pp. 144489144508, 2024. DOI: https://doi.org/10.1109/ACCESS.2024.3472907
[24] Y. Chen et al., Achieving domain generalization for underwater object detection by domain mixup and contrastive learning, Neurocomputing, vol. 528, pp. 2034, 2021, doi: 10.1016/j.neucom.2023.01.053. DOI: https://doi.org/10.2139/ssrn.4243959
[25] S. Abdelwahed, B. Sharobim, B. Wasfey, and L. Said, Advancements in real-time road damage detection: a comprehensive survey of methodologies and datasets, J. Real-Time Image Process., vol. 22, 2025, doi: 10.1007/s11554-025-01683-1. DOI: https://doi.org/10.1007/s11554-025-01683-1
[26] Y. Zhao, Z. Zhong, N. Zhao, N. Sebe, and G. Lee, Style-Hallucinated Dual Consistency Learning: A Unified Framework for Visual Domain Generalization, Int. J. Comput. Vis., vol. 132, pp. 837853, 2022, doi: 10.1007/s11263-023-01911-w. DOI: https://doi.org/10.1007/s11263-023-01911-w
[27] Z. Demirel, S. T. Nasraldeen, . Pehlivan, S. Shoman, M. Albdairi, and A. Almusawi, Comparative Evaluation of YOLO and Gemini AI Models for Road Damage Detection and Mapping, Futur. Transp., 2025, doi: 10.3390/futuretransp5030091. DOI: https://doi.org/10.3390/futuretransp5030091
[28] O. Khare, S. Gandhi, A. Rahalkar, and S. Mane, YOLOv8-Based Visual Detection of Road Hazards: Potholes, Sewer Covers, and Manholes, in 2023 IEEE Pune Section International Conference (PuneCon), 2023, pp. 16, doi: 10.1109/punecon58714.2023.10449999. DOI: https://doi.org/10.1109/PuneCon58714.2023.10449999
[29] D. Arya et al., RDD2022: A multinational image dataset for automatic road damage detection, Geosci. Data J., vol. 11, pp. 846862, 2022, doi: 10.1002/gdj3.260. DOI: https://doi.org/10.1002/gdj3.260
[30] S. Parvin, M. Ahmed, and M. Rahman, Multi-Weather Pothole Detection Dataset (MWPD), Mendeley Data, 2024. https://data.mendeley.com/datasets/s5hx9n2jc3/2 (accessed Apr. 24, 2026).
[31] P. A. Chitale, K. Y. Kekre, H. R. Shenai, R. Karani, and J. P. Gala, Pothole Detection and Dimension Estimation System using Deep Learning (YOLO) and Image Processing, in 2020 35th International Conference on Image and Vision Computing New Zealand (IVCNZ), 2023, pp. 16, doi: 10.1109/IVCNZ51579.2020.9290547. DOI: https://doi.org/10.1109/IVCNZ51579.2020.9290547
[32] Andrew MVD, Pothole Detection Dataset, Kaggle, 2023. https://www.kaggle.com/datasets/andrewmvd/pothole-detection (accessed Apr. 28, 2026).
[33] B. Shakya, O. El-Gayar, J. Kaabi, and K. M. Ahmed, Small-Object Detection at the Edge: A Pareto-Efficient Benchmark of Lightweight YOLO Models on UAV and Overhead Datasets, IEEE Access, vol. 14, pp. 528548, 2026, doi: 10.1109/access.2025.3648968. DOI: https://doi.org/10.1109/ACCESS.2025.3648968
[34] H. M. Aljami et al., Benchmarking YOLOv8 Variants for Object Detection Efficiency on Jetson Orin NX for Edge Computing Applications, Computers, vol. 15, pp. 120, 2026. DOI: https://doi.org/10.3390/computers15020074
[35] L. Rey et al., A Performance Analysis of You Only Look Once Models for Deployment on Constrained Computational Edge Devices in Drone Applications, Electronics, vol. 14, pp. 124, 2025. DOI: https://doi.org/10.3390/electronics14030638
[36] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, You Only Look Once: Unified, Real-Time Object Detection, in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 779788, doi: 10.1109/cvpr.2016.91. DOI: https://doi.org/10.1109/CVPR.2016.91
[37] J. Terven, D.-M. Crdova-Esparza, and J.-A. Romero-Gonzlez, A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO-NAS, Mach. Learn. Knowl. Extr., vol. 5, pp. 16801716, 2023, doi: 10.3390/make5040083. DOI: https://doi.org/10.3390/make5040083
[38] A. Wang et al., YOLOv10: Real-Time End-to-End Object Detection, ArXiv, vol. abs/2405.14458, 2024, doi: 10.48550/arxiv.2405.14458. DOI: https://doi.org/10.52202/079017-3429
[39] I. Ahmed, M. Ahmad, M. Siddiqi, A. Chehri, and G. Jeon, Toward AI-Powered Edge Intelligence for Object Detection in Self-Driving Cars: Enhancing IoV Efficiency and Safety, IEEE Internet Things J., vol. 12, pp. 1699016997, 2025, doi: 10.1109/jiot.2025.3534737. DOI: https://doi.org/10.1109/JIOT.2025.3534737
[40] M. Bakirci, Enhancing vehicle detection in intelligent transportation systems via autonomous UAV platform and YOLOv8 integration, Appl. Soft Comput., vol. 164, p. 112015, 2024, doi: 10.1016/j.asoc.2024.112015. DOI: https://doi.org/10.1016/j.asoc.2024.112015
[41] M. Cao, Q.-V. Tran, N.-M. Nguyen, and K. Chang, Survey on performance of deep learning models for detecting road damages using multiple dashcam image resources, Adv. Eng. Informatics, vol. 46, p. 101182, 2020, doi: 10.1016/j.aei.2020.101182. DOI: https://doi.org/10.1016/j.aei.2020.101182
[42] T. Diwan, G. Anirudh, and J. Tembhurne, Object detection using YOLO: challenges, architectural successors, datasets and applications, Multimed. Tools Appl., vol. 82, pp. 92439275, 2022, doi: 10.1007/s11042-022-13644-y. DOI: https://doi.org/10.1007/s11042-022-13644-y
[43] M. Paiano, S. Martina, C. Giannelli, and F. Caruso, Transfer learning with generative models for object detection on limited datasets, Mach. Learn. Sci. Technol., vol. 5, 2024, doi: 10.1088/2632-2153/ad65b5. DOI: https://doi.org/10.1088/2632-2153/ad65b5
[44] X. Xu, H. Zhang, Yan, K. Liu, H. Bao, and X. Qian, TranSDet: Toward Effective Transfer Learning for Small-Object Detection, Remote. Sens., vol. 15, p. 3525, 2023, doi: 10.3390/rs15143525. DOI: https://doi.org/10.3390/rs15143525
[45] P. Goel and A. Ganatra, Unsupervised Domain Adaptation for Image Classification and Object Detection Using Guided Transfer Learning Approach and JS Divergence, Sensors (Basel)., vol. 23, 2023, doi: 10.3390/s23094436. DOI: https://doi.org/10.3390/s23094436
[46] R. Chakraborty, H. Mitra, P. Shivakumara, and U. Pal, Towards Robust Cross-Dataset Object Detection Generalization under Domain Specificity, ArXiv, vol. abs/2601.09497, 2026, doi: 10.48550/arxiv.2601.09497.
[47] Z. Xue, F. Yang, S. Rajaraman, G. Zamzmi, and S. Antani, Cross Dataset Analysis of Domain Shift in CXR Lung Region Detection, Diagnostics, vol. 13, 2023, doi: 10.3390/diagnostics13061068. DOI: https://doi.org/10.3390/diagnostics13061068
[48] Y. Gao, W. Luo, S. Zhang, N. S. Ahmad, X. Wang, and P. Goh, Benchmarking YOLOv8 to YOLOv13 for robust hand gesture recognition in humanrobot interaction, Sci. Rep., vol. 15, 2025, doi: 10.1038/s41598-025-23925-9. DOI: https://doi.org/10.1038/s41598-025-23925-9
[49] M. El-Geneedy, H. E.-D. Moustafa, H. Khater, S. Abd-Elsamee, and S. Gamel, Advanced real-time detection of acute ischemic stroke using YOLOv12, YOLOv11, and YOLO-NAS: a comparative study for multi-class classification, Sci. Rep., vol. 15, 2025, doi: 10.1038/s41598-025-18997-6. DOI: https://doi.org/10.1038/s41598-025-18997-6
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Hanifatus Sa’diyah Widihasaniputri, Oddy Virgantara Putra, Murhadi Murhadi
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
How to Cite
Share
Plum Analytics