Curated Dataset and Inference Results for AI-Based Bridge Defect Detection Using Images and Videos

This repository contains curated dataset and inference results used in the study: "AI-powered bridge defect detection and condition assessment using drone-collected visual data." Contents:1. curated_dataset_images_annotations: Annotated images with YOLO-format labels for eight defect classes.2. curated_inference_results_images: Model predictions on test images.3. curated_dataset_bridge_videos: Raw bridge inspection videos. Defect classes:Crack, ACrack, Efflorescence, WConccor, Spalling, Wetspot, Rust, ExposedRebars. All annotations are manually created and used as ground truth for training and validation.Inference results are generated using a trained YOLO-based instance segmentation model. This dataset supports reproducibility and transparency of the proposed methodology.本仓库包含用于下述研究的精选数据集与推理结果："基于无人机采集视觉数据的AI驱动桥梁缺陷检测与状态评估"。 数据集内容如下： 1. curated_dataset_images_annotations：包含8类缺陷的带标注图像，标注采用YOLO格式； 2. curated_inference_results_images：测试图像的模型预测结果； 3. curated_dataset_bridge_videos：桥梁巡检原始视频。 缺陷类别如下：裂缝（Crack）、弧形裂缝（ACrack）、泛碱（Efflorescence）、混凝土水渍（WConccor）、剥落（Spalling）、湿斑（Wetspot）、锈蚀（Rust）、外露钢筋（ExposedRebars）。 所有标注均为人工制作，用作训练与验证的真值标签（ground truth）。推理结果由经训练的基于YOLO的实例分割模型生成。 本数据集可为所提出方法的可复现性与研究透明度提供支撑。