Ana gezinime geç Aramaya geç Ana içeriğe geç

YOLOv8x Baseline for Cross-Dataset Road-Surface Defect Detection

  • Esra Ince
  • , Muhammed Sefa Cetin
  • , Chiheb Eddine Heguehoug
  • , Muhsin Tunay Gencoglu
  • , Halit Ozen
  • Firat University

Araştırma çıktısı: Kitap/Rapor/Konferans Bildirisinde Bölüm › Konferans katkısı › Hakemli

Özet

Road-surface damage detection supports preventive maintenance planning and traffic safety, yet routine field inspection is labor-intensive and can vary with inspector judgment and operating conditions. This study establishes a reproducible baseline for road-crack and road-damage recognition using the Ultralytics YOLOv8 framework. The model is trained on three publicly available datasets while preserving the original train, validation, and test splits. Performance is evaluated with a consistent protocol on held-out validation data using AdamW optimization and moderate augmentation. Training is performed in Google Colab with mixed-precision arithmetic, and the input size is set to 640 or 768 depending on the dataset. For single-class crack detection, the baseline attains a mean Average Precision of 0.622 at an IoU threshold of 0.5 and 0.430 averaged over IoU thresholds from 0.5 to 0.95 on Dataset 01. It attains 0.688 at an IoU threshold of 0.5 and 0.464 averaged over IoU thresholds from 0.5 to 0.95 on Dataset 02. For the three-class road-damage dataset, it achieves 0.904 at an IoU threshold of 0.5 and 0.603 averaged over IoU thresholds from 0.5 to 0.95. Overall, the baseline provides strong accuracy without architectural changes and serves as a clear reference point for subsequent methodological and deployment-oriented studies.

Orijinal dilİngilizce
Ana bilgisayar yayını başlığı2026 11th International Conference on Machine Learning Technologies, ICMLT 2026
YayınlayanInstitute of Electrical and Electronics Engineers Inc.
Sayfalar453-457
Sayfa sayısı5
ISBN (Elektronik)9798331548650
DOI'lar
Yayın durumuYayınlandı - 2026
Etkinlik11th International Conference on Machine Learning Technologies, ICMLT 2026 - Berlin, Germany
Süre: 20 May 2026 → 22 May 2026

Yayın serisi

Adı2026 11th International Conference on Machine Learning Technologies, ICMLT 2026

???event.eventtypes.event.conference???

???event.eventtypes.event.conference???11th International Conference on Machine Learning Technologies, ICMLT 2026
Ülke/BölgeGermany
ŞehirBerlin
Periyot20/05/26 → 22/05/26

Bibliyografik not

Publisher Copyright:
© 2026 IEEE.

BM SKH

Bu sonuç, aşağıdaki Sürdürülebilir Kalkınma Hedefine/Hedeflerine katkıda bulunur

  1. SKH 3 - Sağlık ve Kaliteli Yaşam
    SKH 3 Sağlık ve Kaliteli Yaşam
  2. SKH 11 - Sürdürülebilir Şehirler ve Topluluklar
    SKH 11 Sürdürülebilir Şehirler ve Topluluklar

Parmak izi

YOLOv8x Baseline for Cross-Dataset Road-Surface Defect Detection' araştırma başlıklarına git. Birlikte benzersiz bir parmak izi oluştururlar.

Alıntı Yap