DamSegment
Description
DamSegment contains images acquired from a concrete dam exhibiting two distinct damage types: cracks and spalling. All images are annotated at pixel level, making the dataset suitable for structural health monitoring tasks such as damage detection, semantic segmentation, and damage classification.
Files
Steps to reproduce
Download and unzip the DamSegment dataset. The data are organized into three main tasks: Damage Classification (2,000 images: crack vs. non-crack), Damage Detection (1,500 images with YOLO and Pascal VOC bounding-box labels), and Damage Segmentation (1,500 images split into Easy/Medium/Hard, each with images, YOLO/VOC labels, and pixel-level masks for crack and spalling). Users can directly load these folders into standard deep-learning pipelines for classification, detection, and segmentation.
Institutions
- University of Kansas
Categories
Funders
- U.S. Army Engineer Research and Development CenterUnited States Army Corps of EngineersUnited States