PY-CrackDB : a pavement crack dataset from paraguayan roads for context-aware computer vision models

Abstract

PY-CrackDB, a novel dataset of asphalt pavement images designed for developing context-aware artificial intelligence systems. The dataset contains 569 images (351 × 500 pixels), collected from national routes near Coronel Oviedo, Paraguay, and divided into 369 images with cracks and 200 without. A primary contribution of this work is its specific focus on fine fissures (< 3 mm wide), a category critical for early-stage maintenance according to Paraguayan road engineering standards. Data collection was performed under standardized conditions, and all annotations were created by civil engineering professionals and subsequently verified through a rigorous cross-review protocol to ensure accuracy. This methodological rigor resulted in a dataset that is particularly suitable for training and validating models for semantic segmentation and early defect detection, ultimately supporting the development of preventative road maintenance strategies.