A Large Scale Fish Dataset

Authors: O. Ulucan, D. Karakaya, M. Turkan
Affiliation: Department of Electrical and Electronics Engineering, Izmir University of Economics, Izmir, Turkey
Corresponding Author: M. Turkan

Paper : A Large-Scale Dataset for Fish Segmentation and Classification

General Introduction

This dataset includes nine distinct seafood varieties sourced from a supermarket in Izmir, Turkey. It was gathered through a collaborative initiative between Izmir University of Economics and industry partners. Furthermore, the dataset, along with the associated research findings, was presented at ASYU 2020.

The seafood types included are gilt-head bream, red sea bream, sea bass, red mullet, horse mackerel, black sea sprat, striped red mullet, trout, and shrimp. Therefore, this diverse collection offers valuable insights for research and industry applications.

If you use this dataset in your work, please consider to cite:

@inproceedings{ulucan2020large,
title={A Large-Scale Dataset for Fish Segmentation and Classification},
author={Ulucan, Oguzhan and Karakaya, Diclehan and Turkan, Mehmet},
booktitle={2020 Innovations in Intelligent Systems and Applications Conference (ASYU)},
pages={1–5},
year={2020},
organization={IEEE}
}

O.Ulucan, D.Karakaya, and M.Turkan.(2020) A large-scale dataset for fish segmentation and classification.
In Conf. Innovations Intell. Syst. Appli. (ASYU)

Purpose of the work

We gathered this dataset to perform tasks such as dividing images into different parts, extracting important characteristics, and categorizing them into various groups. Additionally, we used it to compare several common methods, including Semantic Segmentation, Convolutional Neural Networks, and Bag of Features. The results of all the experiments clearly demonstrate that our dataset is highly suitable for these purposes.

Data Gathering Equipment and Data Augmentation

The dataset’s images were captured using two different cameras: the Kodak Easyshare Z650 and the Samsung ST60. Consequently, the images had varying resolutions of 2832 x 2128 and 1024 x 768, respectively. To ensure uniformity and facilitate analysis, we resized all images to 590 x 445 while maintaining their original aspect ratios. Furthermore, to enhance the dataset, we augmented the labels by flipping and rotating the images.

To begin with, the images were captured using the Kodak Easyshare Z650 and the Samsung ST60, resulting in varying resolutions of 2832 x 2128 and 1024 x 768, respectively. However, to ensure uniformity and facilitate analysis, all images were resized to 590 x 445 while maintaining their original aspect ratio. Additionally, we augmented the dataset labels by flipping and rotating the images, which further enriched the dataset.

Upon completion of the augmentation process, the total number of images for each class was standardized to 2000, comprising 1000 RGB fish images and 1000 pair-wise ground truth labels.

Description of the dataset

The dataset comprises nine distinct types of seafood, each represented in the “Fish_Dataset” file along with its corresponding ground truth labels. Each class includes 1,000 augmented images, accompanied by their respective pair-wise augmented ground truths. Additionally, images for each class are sequentially arranged from “00000.png” to “01000.png.”

For instance, to access the ground truth images for shrimp, follow the path “Fish -> Shrimp -> Shrimp GT.” This organization ensures easy access to the desired images and their associated ground truth labels.