建筑工程施工安全隐患图像标注数据集

数据集采用三级结构：images原始图像，annotations标注文件，metadata数据集元数据与统计信息。图像数据按用途划分为训练集（train）、验证集（val）、测试集（test），图像保留原始采集尺寸，反映真实场景多样性。标注文件采用JSON格式，与图像文件同名对应。JSON结构包含以下核心字段：version（标注工具版本）、flags（标记位）、shapes（标注目标数组）、imagePath（图像相对路径）、imageData（图像Base64编码，可选）、imageHeight（图像高度）、imageWidth（图像宽度）。其中 shapes 为关键字段，每个元素描述一个隐患目标，包含 label（隐患类别）、points（二维数组坐标）、group_id（分组ID）、shape_type、flags（目标标记位）、description（隐患详细描述）。元数据目录包含 class_list.csv（类别编码表）、dataset_index.csv（样本索引表）、dataset_summary.csv（类别统计表）。This dataset adopts a three-level structure, including original images stored in the `images` directory, annotation files in the `annotations` directory, and dataset metadata and statistical information in the `metadata` directory. The image data is divided into training set (train), validation set (val) and test set (test) according to usage scenarios, and all images retain their original acquisition dimensions to reflect the diversity of real-world scenes. The annotation files are in JSON format, and each annotation file has a one-to-one correspondence with its matching image file by sharing the same name. The JSON structure contains the following core fields: `version` (annotation tool version), `flags` (general flag bits), `shapes` (array of annotation targets), `imagePath` (relative path of the image), `imageData` (Base64-encoded image data, optional), `imageHeight` (image height), and `imageWidth` (image width). Among them, `shapes` is the key field, each element of which describes a potential hazard target, and includes sub-fields such as `label` (hazard category), `points` (2D coordinate array), `group_id` (group ID), `shape_type`, `flags` (target-specific flag bits), and `description` (detailed hazard description). The metadata directory contains three files: `class_list.csv` (category encoding table), `dataset_index.csv` (sample index table), and `dataset_summary.csv` (category statistics table).

搜集汇总

背景与挑战

背景概述

该数据集是一个建筑工程施工安全隐患图像标注数据集，由上海建科工程咨询有限公司申请，属于数据加工产品。它专注于施工安全领域，通过标注图像数据来识别和分类安全隐患，适用于安全监测和风险预防应用。