Visual Data from Water Pipe Inspection

This dataset contains endoscopic inspection images acquired using a soft robot designed and developed by Bendabl for pipe inspection and anomaly detection applications. The data were collected in an experimental pipe-network environment designed to simulate realistic industrial inspection scenarios. The setup included PE (polyethylene) pipes, steel pipe fittings such as connectors, 90-degree elbows, T-junctions, W-junctions, reducer couplings, and transparent acrylic pipes. Cracks were introduced only in the PE pipe sections.

Data acquisition was performed under multiple environmental and operational conditions, including horizontal, vertical, and 45-degree inclined pipe configurations, both dry and water-filled conditions, varying illumination settings, and different camera angles and inspection distances. The dataset includes s approximately 3 mm holes and radian-shaped cracks.

Images were captured using two endoscopic camera systems. Files beginning with WIN_ correspond to the endoscopic camera integrated into the soft robot and have a resolution of 1280 × 720 pixels. The remaining images were acquired using a higher-resolution endoscopic camera with a resolution of 1920 × 1080 pixels. All images are provided in .jpg format.

The dataset is organized into three standard machine-learning splits: train, val, and test. Each split contains two subfolders, crack and no-crack, corresponding to defective and non-defective pipe conditions, respectively. Additionally, the root directory contains a labels.csv file listing all image filenames and their associated binary labels, where 1 indicates the presence of a crack and 0 indicates no crack.

This dataset is intended to support research in robotic pipe inspection, computer vision, infrastructure monitoring, crack detection, and machine learning methods for industrial inspection systems.