DefectAtlas-20: A real-world image dataset for joint material and surface-condition classification

Reliable automated surface inspection requires image datasets that reflect the conditions under which material damage occurs in practice. However, many existing defect datasets focus on individual materials, specific components, or controlled imaging environments, limiting their suitability for studying material-dependent surface defect appearance and their performance robustness under real-world variation. This article presents DefectAtlas-20, a dataset of 13,233 RGB photographs organized into 20 classes formed by combining ten commonly encountered material categories with defect and no-defect surface conditions. The materials comprise uncoated and coated metal, uncoated and coated wood, rubber, textile, cardboard, glass, cable, and hard plastic. Images were acquired using consumer mobile devices in heterogeneous indoor and outdoor field environments around Hamburg, Germany. They capture variation in illumination, viewing angle, background, object geometry, surface texture, reflectance, and usage condition. The defect classes contain naturally occurring damage, while the corresponding no-defect classes show visually comparable surfaces without visible defects. All images were collected according to a predefined protocol, manually labeled, independently reviewed, centrally quality-controlled, anonymized through metadata removal, and converted to lossless PNG format in the sRGB color space. The released data package includes 512 × 512-pixel images, class and acquisition metadata, and documentation of the collection and curation process. The dataset provides a benchmark for 20-class material-condition classification and additionally supports binary defect recognition, material classification, material-specific defect analysis, transfer learning, and domain-generalization research. Its material diversity and real-world acquisition conditions facilitate the development and evaluation of computer-vision methods for surface inspection, maintenance, refurbishment, and damage assessment.