We present DamCrack, a novel dataset comprising high-resolution images of a concrete dam in Spokane, Washington, USA, captured using a drone—including autonomous flights with overlapping images—and mobile devices to document complex surface deterioration. The dataset includes two damage types: cracks and spalling, with pixel-wise annotations provided in different formats. The inclusion of overlapping aerial imagery enables future photogrammetric 3D reconstruction, while the current 2D image data supports immediate computer vision tasks such as damage detection and segmentation. Some damage exhibits visual characteristics resembling alkali-silica reaction (ASR), including map-cracking patterns and surface discoloration. Captured under diverse environmental conditions—such as varying lighting, camera distances, and camera properties—the dataset specifically addresses challenging real-world scenarios where multiple damage types co-occur. DamCrack provides: (1) standardized benchmarks for 2D damage detection and segmentation algorithms, (2) high-quality imagery to support future 3D reconstruction models, and (3) annotated examples of complex, co-occurring damage patterns that contribute to advancing structural health monitoring research for critical infrastructure.
Gharehbaghi et al. (Sun,) studied this question.