Image-based Multi-Damage Detection in Tunnels: a Deep Learning dataset for Structural Health Monitoring

This paper introduces a novel dataset of high-resolution panoramic images from two Italian tunnels, specifically designed for structural health monitoring (SHM). Its innovation lies in extensive annotations of key structural damages—cracks, corrosion, spalling, seepage, and damaged joints—created using the Segment Anything Model (SAM) for pixel-level segmentation and bounding box annotations, formatted in COCO-style. This comprehensive dataset supports various computer vision tasks, including classification, instance segmentation, and object detection. By enabling the benchmarking of advanced deep learning models, our work provides an essential resource for auto-mated damage detection, significantly advancing research and practical infrastructure maintenance. © 2025, European Council on Computing in Construction (EC3)