Seathru Underwater Imaging Dataset

Description:

The Seathru Underwater Imaging Dataset comprises five distinct datasets (D1-D5), each designed to address different aspects of underwater imaging. Each dataset contains:

A set of N linear images captured without any color correction.

N depth maps corresponding to each linear image.

The exact number of images (N) varies between datasets. For a detailed breakdown of each dataset’s size and specifications, please refer to the accompanying research paper.

Download Dataset

Key Features:

The five different scenes are selected based on unique underwater imaging features. These include:

Imaging distance: Variation in the distance between the camera and the scene.

Camera orientation: Whether the camera is positioned to look forward or downward.

Optical conditions: Different water clarity levels and lighting conditions to simulate real-world underwater environments.

Each dataset offers multiple overlapping images of the scenes, enabling the construction of 3D models using Structure-from-Motion (SFM) techniques. In our work, we employed Agisoft Metashape Pro to generate depth maps corresponding to the images.

Data Format:

Due to storage constraints, the original RAW files are not included in this release. Instead, we provide linear images in PNG format, scaled down to 30% of their original resolution. If you require access to the full-resolution RAW files, please reach out to us directly.

Depth Map Details:

The depth maps represent distances in meters. Important considerations for depth map interpretation include:

Zero values: These should be interpreted as NaNs (Not a Number), representing areas where depth information could not be calculated.

Depth scaling: All depth maps have been scaled in meters, allowing for easy interpretation and integration with other 3D modeling data.

Applications and Use Cases:

The Seathru dataset is particularly useful for a wide range of underwater research and machine learning applications, including:

3D Scene Reconstruction: Using SFM, researchers can build accurate 3D models of underwater environments, which are essential for robotic navigation and marine research.

Color Correction and Image Restoration: The uncorrected images can serve as a baseline for developing algorithms that perform color correction and restoration in challenging underwater conditions.

Depth Estimation Algorithms: The dataset provides a valuable resource for training machine learning models to improve depth estimation and object detection in underwater environments.

This dataset is sourced from Kaggle.