FIB Dataset

In this page, we describe public dataset Fish Image Bank (FIB) which we developed.

About

Fish Image Bank (FIB) Dataset is constructed by Hasegawa Laboratory. This project is supported by JST ACT-X AI powered Research Innovation / Creation entitled Development of a robust catch-recognition platform for constructing big data in the fisheries industry.


Please read the following statement carefully before use.

If you find any problems with the contents, please let us know at t-hase[at]u-fukui.ac.jp

Data description

  • FIB is a fish image dataset consisting of 405 4K images of 24 fish species.
  • The images were taken of fish distributed at markets and wholesalers in Fukui Prefecture, Japan between March 12, 2022 and September 15, 2023.
  • The backgrounds were uniformly monochromatic and illuminated to create as uniform an environment as possible.
  • Only a single fish is in an image.
  • Some unnecessary backgrounds are manually removed.
  • Multiple images of the same fish may be included in a single image.
  • Although we checked carefully, Labels are manually assigned by visual inspection of the images and may contain errors.

License

FIB Dataset by Tatsuhito Hasegawa is licensed under CC BY-SA 4.0

Citation

Please cite the following references when using the FIB dataset.

Text
Tatsuhito Hasegawa and Daichi Nakano, "Robust Fish Recognition using Foundation Models towards an Automatic Fish Resource Management", Journal of Marine Science and Engineering, vol. 12, no. 3: 488, 2024. https://doi.org/10.3390/jmse12030488

bibtex
@ARTICLE{Hasegawa2024,
title = "Robust Fish Recognition using Foundation Models towards an Automatic Fish Resource Management",
author = "Hasegawa, Tatsuhito and Nakano, Daichi",
JOURNAL = "Journal of Marine Science and Engineering",
VOLUME = 12,
YEAR = 2024,
NUMBER = 3,
ARTICLE-NUMBER = 488,
URL = "https://www.mdpi.com/2077-1312/12/3/488",
}

Sample images

fish
fish
fish
fish

Download

FIB_v1.0.zip (511MB)

Labels

Each image is labeled as follows, in COCO format.

  • Species
  • Instance mask
  • Keypoints (nine points)

Top