Crack Roboflow DTI UNB

The dataset is designed for instance segmentation tasks related to detecting cracks in concrete structure inspection. It is a valuable resource for training CNNs to identify and segment cracks in various surfaces accurately. It was assembled from three different sources: 1. The platform Roboflow-Universe developed by Dwyer et al (2022): We used the Roboflow-Universe-Crack (RUC) dataset (Nadar, 2022) with a total of 1551 samples. 2. The archives of the Department of Transportation and Infrastructure (DTI) from New Brunswick: We picked 540 original crack images, each of which underwent annotation using the CVAT annotation tool (Sekachev et al., 2020), ensuring precision and uniformity of labelling. 3. The Crack500 dataset (Yang et al., 2020; Zhang et al., 2016) We re-annotated and used 56 samples.

本数据集由三个不同来源汇集而成，专为混凝土结构巡检场景下的裂缝检测实例分割任务设计，是训练卷积神经网络（Convolutional Neural Network, CNN）以精准识别并分割各类表面裂缝的宝贵研究资源。其数据来自以下三个渠道：1. 由Dwyer等人（2022）开发的Roboflow-Universe平台：我们采用了该平台中的Roboflow-Universe-Crack（RUC）数据集（Nadar, 2022），共计1551个样本；2. 新不伦瑞克省交通与基础设施部（Department of Transportation and Infrastructure, DTI）档案库：我们选取了540张原始裂缝图像，所有图像均通过CVAT标注工具（Sekachev等，2020）完成标注，保障了标注的精准性与统一性；3. Crack500数据集（Yang等，2020；Zhang等，2016）：我们对其中56个样本进行了重新标注并投入使用。

搜集汇总

背景与挑战

背景概述

Crack Roboflow DTI UNB是一个用于混凝土结构裂缝检测的实例分割数据集，包含来自Roboflow-Universe-Crack、DTI档案和Crack500数据集的合并数据，总计2147个样本。数据集以YOLO格式提供，大小为296.5 MB，适用于训练卷积神经网络（CNN）进行裂缝检测和分割任务。