Work

A Multimodal Optical Dataset for Underwater Image Enhancement, Detection, Segmentation, and Reconstruction

Year: 2025

Type: data-paper

Abstract: Multimodal devices utilizing optical cameras and LiDAR are crucial for precise underwater environmental perception. Enhancing and optimizing RGB images and laser point clouds through algorithms is a key focus in underwater computer vision. This research necessitates extensive underwater multivariate data for training and evaluation, along with the organization and labeling of this data. To meet these needs, we present a multimodal optical dataset for underwater image detection, segmentation, enhancement, and 3D reconstruction (MOUD). The dataset comprises over 18,000 original RGB images, 12,000 labeled images, 60 point cloud sets, and several labeled point clouds. The labeled images feature nine different target objects, including scallop, starfish, conch, holothuria, seaweed, coral, reef, abalone, and barnacle. These data were gathered from our underwater simulation scenarios. To ensure accuracy and utility, we employed a state-of-the-art image-laser underwater detector with appropriate kinematic parameters. This dataset supports training and evaluation in underwater image enhancement, detection, segmentation, and reconstruction, which are vital for precise underwater sensing. Consequently, this dataset holds significant importance for advancing underwater optical exploration technology.

Source: Scientific Data

DOI: 10.1038/s41597-025-05797-w