提供百度云盘地址下载

数据集格式：labelme格式(不包含mask文件，仅仅包含jpg图片和对应的json文件)

图片数量(jpg文件个数)：9398

标注数量(json文件个数)：9398

标注类别数：18

标注类别名称:["steel_corrosion","exposed_reinforcement_steel","infiltration","staining","crack","efflorescence","concrete_leaching","disaggregation_of_concrete","deep_crack","embankment_erosion","expansion_joint","rutting_pavement_failure","pothole_asphalt","faunal_biodegradation","fire_damaged","guardrail_damaged","formwork_niche","broken_drain"]

每个类别标注的框数：

steel_corrosion (钢筋锈蚀) count = 1249

exposed_reinforcement_steel (钢筋裸露) count = 3755

infiltration (渗漏) count = 1807

staining (污渍) count = 5198

crack (裂缝) count = 796

efflorescence (泛碱) count = 5466

concrete_leaching (混凝土渗析) count = 6762

disaggregation_of_concrete (混凝土离析) count = 1681

deep_crack (深裂缝) count = 183

embankment_erosion (堤岸侵蚀) count = 464

expansion_joint (伸缩缝) count = 462

rutting_pavement_failure (路面车辙破坏) count = 977

pothole_asphalt (沥青坑洞) count = 683

faunal_biodegradation (生物降解) count = 1802

fire_damaged (火灾损坏) count = 83

guardrail_damaged (护栏损坏) count = 415

formwork_niche (模板孔洞) count = 993

broken_drain (排水管损坏) count = 201

总框数：32977

使用标注工具：labelme=5.5.0

应用领域：桥梁、隧道或建筑等混凝土结构以及相关基础设施的病害与缺陷检测类别。

所在github仓库：firc-dataset

图片分辨率：640x640

标注规则：对类别进行画多边形框polygon

重要说明：可以将数据集用labelme打开编辑，json数据集需自己转成mask或者yolo格式或者coco格式作语义分割或者实例分割

特别声明：本数据集不对训练的模型或者权重文件精度作任何保证

图片预览：

标注例子：

原图（随机抽16张图展示）：

标注绘制结果：

labelme编辑图实例：