BlueRealm

BlueRealm Underwater Image Dataset

Dataset Metadata

Name: BlueRealm

Description: A comprehensive dataset designed for training and evaluating underwater image enhancement and restoration methods.

Purpose: Provide a diverse collection of underwater imagery for developing and testing algorithms related to underwater image restoration.

Dataset Composition:

Combined Dataset Size: 6,000 images (after augmentation).Original Dataset Size: 3,010 images.

2,102 underwater images from a GoPro camera.

Datasets from (Akkaynak & Treibitz, 2019) and (Li et al., 2019).

Video frames captured by a mini-ROV camera.

Image Characteristics:

Diverse underwater environments, depths, and lighting conditions.

Includes images with a color chart for accurate quality assessment.

Video frames from an ROV camera for evaluating real-time processing.

Data Augmentation: The original dataset was augmented to 6,000 images using random rotations and flips.

Reference Data: Images corrected using SeaThru method (Akkaynak & Treibitz, 2019) are provided as reference data for training and evaluation.

Data Acquisition:

GoPro camera.

Mini-ROV camera.

Publicly available datasets from previous studies.

Preprocessing:

SeaThru method applied to generate reference corrected images.

Augmentation via rotations and flips.

Availability:

The proprietary dataset (GoPro and mini ROV images) is publicly available at this link:

BlueRealm dataset

Citation:

The BlueRealm dataset has been published in the context of the following work:

C. Antoniou, S. Spanos, S. Vellas, V. Ntouskos, K. Karantzalos, (2024). StreamUR: Physics-informed Near Real-Time Underwater Image Restoration. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences.DOI: 10.5194/isprs-archives-XLVIII-3-2024-1-2024

Also cite the original datasets from (Akkaynak & Treibitz, 2019) and (Li et al., 2019), if applicable.

Contact:

Christos Antoniou - chrant.mail at gmail.com

Sotiris Spanos - spanossotiris1998 at gmail.com 

Valsamis Ntouskos - valsamis.ntouskos at unimercatorum.it; ntouskos at mail.ntua.gr

References:

Akkaynak, Derya, and Tali Treibitz. "Sea-thru: A method for removing water from underwater images." Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019.

Li, Chongyi, et al. "An underwater image enhancement benchmark dataset and beyond." IEEE transactions on image processing 29 (2019): 4376-4389.