/marine-snow

Marine Snow Removal Benchmarking Dataset

Marine Snow Removal Benchmarking Dataset

Welcome to Marine Snow Removal Benchmarking Dataset (MSRB Dataset in short).

An example from Marine Snow Removal Benchmarking Dataset. Left: Original underwater image. Middle: Synthesized image for Task 1, small-sized marine snow artifacts. Right: Synthesized image for Task 2, various-sized marine snow artifacts.

Marine snow is one of the main degradation sources of underwater images that are caused by small particles, e.g., organic matter and sand, between the underwater scene and photosensors. We mathematically model two typical types of marine snow from the observations of real underwater images. The modeled artifacts are synthesized with underwater images to construct large-scale pairs of ground-truth and degraded images to calculate objective qualities for marine snow removal and to train a deep neural network.

References

If you use MSRB Dataset in your paper, please cite the following paper. The details for synthesizing marine snow artifacts are also described.

  1. R. Kaneko, Y. Sato, T. Ueda, H. Higashi, and Y. Tanaka, "Marine snow removal benchmarking dataset," APSIPA ASC 2023, Taipei, Taiwan, Nov. 2023, to be presented.

Dataset Descriptions

MSRB Dataset has two sub-datasets:

  • MSR Task 1: Removal of small-sized marine snow artifacts.
  • MSR Task 2: Removal of various-sized marine snow artifacts.

Each sub-dataset corresponding to a MSR Task contains 2,300 training image pairs and 400 test image pairs, all having a pixel resolution of 384 x 384. An image pair contains one original underwater image and one image containing synthesized marine snow artifacts. Each synthesized image contains 100-600 marine snow particles.

All original images are collected from flickr under a Creative Commons Attribution-NonCommercial-ShareAlike 2.0 Generic (CC BY-NC-SA 2.0) License.

MSR Task 1

In Task 1, the maximum width/height of the artifacts is restricted to 6 pixels, which correspond to roughly 1.6% of the image width/height.

MSR Task 2

The synthesized images for Task 2 have small- and large-sized marine snow artifacts. For large-sized artifacts, we set the largest width/height of marine snow to 32 pixels, which corresponds to 8.3% compared to the image width. Furthermore, the probabilities of small- and large-sized artifacts are set to 0.7 and 0.3, respectively.

Downloading MSRB Dataset

You can download MSRB Dataset from Google Drive (2.0 GB). The file is zipped. After unzipping, you can find training and test directories. In each of the directories, there are three directries MSRB1, MSRB2, and original.

The images in original are real underwater images without marine snow, i.e., ground-truth images. Those in MSRB1 and MSRB2 are images with synthesized marine snow artifacts corresponding to MSR Tasks 1 and 2, respectively. Note that the filenames of the original and synthesized images are the same for calculating objective image qualities easily.

Examples from MSRB Dataset

The images below are examples from the test data of MSRB Dataset.

Original underwater image Synthesized image for Task 1 Synthesized image for Task 2

Marine Snow Removal Benchmarking Results

The following tables are the current state-of-the-art results for marine snow removal. The average PSNRs/SSIMs are computed over the test datasets. If you would like to update the results, please let us know!

MSR Task 1

Method PSNR SSIM
Median filter (kernel size 3x3) 28.55 0.846
Median filter (kernel size 5x5) 25.98 0.711
Adaptive median filter (kernel size 3x3) 29.88 0.910
Adaptive median filter (kernel size 5x5) 28.08 0.861
U-Net 36.82 0.978
Synthesized image 32.20 0.945

MSR Task 2

Method PSNR SSIM
Median filter (kernel size 3x3) 22.81 0.770
Median filter (kernel size 5x5) 21.93 0.645
Adaptive median filter (kernel size 3x3) 23.35 0.842
Adaptive median filter (kernel size 5x5) 22.83 0.794
U-Net 30.95 0.932
Synthesized image 23.83 0.876

Restoration Results

The images below are restoration examples for both tasks.

MSR Task 1

Median filter Adaptive median filter U-Net

MSR Task 2

Median filter Adaptive median filter U-Net

Copyright

Copyright (c) 2023 Reina Kaneko, Yuya Sato, Takumi Ueda, Hiroshi Higashi, and Yuichi Tanaka.

We would like to thank all users on flickr who made original underwater images available under the CC BY-NC-SA 2.0 license. Credits of all the images in MSRB Dataset are available in this repository through img_user_id.csv. Please let us know if you have any questions and comments on this dataset.