SubPipe: A Submarine Pipeline Inspection Dataset for Segmentation and Visual-inertial Localization

TL;DR

This paper introduces SubPipe, an underwater dataset for visual-inertial SLAM, segmentation, and side-scan sonar object detection, recorded on an LAUV along a submarine pipeline. It provides multi-sensor data with ground-truth annotations, including RGB segmentation masks and pipeline bounding boxes, plus localization information; the dataset includes SubPipeMini as a smaller subset. The authors benchmark state-of-the-art SLAM, segmentation, and detection methods, revealing the challenges of low-texture underwater imaging and the potential of learning-based methods after domain-specific tuning. They quantify image information using delentropy and motion diversity metrics and show that underwater scenes exhibit lower information content and limited motion diversity compared to above-water datasets. The public release aims to catalyze development of robust underwater computer vision techniques.

Abstract

This paper presents SubPipe, an underwater dataset for SLAM, object detection, and image segmentation. SubPipe has been recorded using a \gls{LAUV}, operated by OceanScan MST, and carrying a sensor suite including two cameras, a side-scan sonar, and an inertial navigation system, among other sensors. The AUV has been deployed in a pipeline inspection environment with a submarine pipe partially covered by sand. The AUV's pose ground truth is estimated from the navigation sensors. The side-scan sonar and RGB images include object detection and segmentation annotations, respectively. State-of-the-art segmentation, object detection, and SLAM methods are benchmarked on SubPipe to demonstrate the dataset's challenges and opportunities for leveraging computer vision algorithms. To the authors' knowledge, this is the first annotated underwater dataset providing a real pipeline inspection scenario. The dataset and experiments are publicly available online at https://github.com/remaro-network/SubPipe-dataset

Figures (9)

• Figure 1: SubPipe has been recorded during a pipeline inspection mission with OceanScan's LAUV. The recorded data includes two monocular cameras (one monochrome camera and one RGB) with pipe segmentation annotations for the latter one; side-scan sonar images with bounding box annotations of the pipeline for object detection; temperature, altitude, and depth measurements; and the robot's pose, velocity, and acceleration.
• Figure 2: Sample images from SubPipe and other state-of-the-art underwater datasets. A sample segmentation image is also shown for those datasets that include segmentation labels.
• Figure 3: The 6 degrees of freedom of the LAUV vo:lsts_6dof.
• Figure 4: Segmentation mask generation process illustrated with two examples. Each row presents a unique example. From left to right: the original raw image, the image after histogram equalization of RGB channels, the original image after converting to grayscale and applying histogram equalization, and the final segmentation mask. The equalization process enhances contrast, aiding in the precise delineation of the pipeline, including the pipe clamp, which is annotated as part of the pipeline.
• Figure 5: Dataset metrics. The top section presents the distribution of delentropy values across various datasets, accompanied by representative images that yield the minimum and maximum delentropy. On the bottom are the motion diversity metric results. Lower delentropy and motion diversity values indicate a high degree of uniformity in image content and limited motion variety across the six degrees of freedom. This phenomenon is particularly evident in SubPipe, since pipeline inspection missions inherently limit motion and imaging variability. In contrast, datasets such as EuRoC, featuring a drone navigating freely in six degrees of freedom within an indoor environment, exhibit significantly more diversity in both imaging and motion.