COCO-style instance segmentation dataset for detecting **corrosion** on internal surfaces of metallic pipelines.

COCO-style instance segmentation dataset for detecting **corrosion** on internal surfaces of metallic pipelines. Images were captured in-pipe with a crawler robot (onboard lighting), then carefully annotated with polygons. One class (`corrosion`). Splits provided (train/val/test), ready for YOLOv11-seg, Detectron2, and MMDetection. License: CC BY 4.0.

面向金属管道内壁腐蚀检测的类COCO（COCO-style）实例分割数据集。该数据集的图像由搭载车载照明系统的履带式巡检机器人在管道内部拍摄采集，随后通过多边形标注完成了精细人工标注。数据集仅包含腐蚀（corrosion）一个目标类别，且已提供训练集、验证集与测试集划分方案，可直接适配YOLOv11-seg、Detectron2以及MMDetection模型。许可协议为CC BY 4.0。