metadata

dataset_info:
  features:
    - name: image
      dtype: image
    - name: label
      sequence: int64
  splits:
    - name: train
      num_bytes: 51027416699.38
      num_examples: 8716
    - name: validation
      num_bytes: 14065562052.042
      num_examples: 2886
    - name: test
      num_bytes: 13973797429.51
      num_examples: 2890
  download_size: 35506863059
  dataset_size: 79066776180.93199
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
task_categories:
  - image-classification
tags:
  - marine biology
  - biodiversity
  - ecology
  - conservation
  - citizen science
  - coral reef
license: cc0-1.0
size_categories:
  - 10K<n<100K

Seatizen Atlas Image Dataset

Dataset Card

Dataset Name: Seatizen Atlas Image Dataset
Task: Multi-label image classification
Domain: Marine Biodiversity
License: cc0-1.0
Size: 14,492 annotated images

Description

The Seatizen Atlas Image Dataset is a large-scale collection of annotated underwater images designed for training and evaluating artificial intelligence models in marine biodiversity research. It is specifically tailored for multi-label image classification tasks.

This dataset was curated by selecting only classes with more than 200 annotations to facilitate the training of computer vision models. For more information about the Seatizen Atlas Dataset, please refer to the Data paper and the Zenodo repository.

Acquisition Details

Underwater data acquisition was carried out using instrumented marine platforms. These platforms can be divided into two groups: citizen platforms and scientific platforms.

Citizen platforms paddleboards, kitesurfs, snorkeling masks equipped with cameras. Images were collected by the "Seatizen" team to promote citizen science involvement in marine research.
Scientific platforms Autonomous Surface Vehicles (ASV) equipped with a camera.

Dataset Structure

The Seatizen Atlas Image Dataset is organized as follows:

Train split: 8716 images
Validation split: 2886 images
Test split: 2890 images

List of Annotated Classes

31 classes were grouped into five main categories:

Algae

Algal Assemblage
Algae Halimeda
Algae Coralline
Algae Turf

Coral

Acropora Branching
Acropora Digitate
Acropora Submassive
Acropora Tabular
Bleached Coral
Dead Coral
Living Coral
Non-acropora Millepora
Non-acropora Encrusting
Non-acropora Foliose
Non-acropora Massive
Non-acropora Coral Free
Non-acropora Submassive

Seagrass

Syringodium isoetifolium
Thalassodendron ciliatum

Habitat

Rock
Rubble
Sand

Other Organisms and Custom Classes

Atra/Leucospilota
Blurred
Fish
Homo Sapiens
Human Object
Sea Cucumber
Sea Urchin
Sponges
Useless

More detailed information about the dataset classes, including full descriptions and examples of annotated images, can be found in the supplementary material available here.

Citation

If you use the Seatizen Atlas Image Dataset in your research, please consider cite the following:

@article{contini2025seatizen,
  title={Seatizen Atlas: a collaborative dataset of underwater and aerial marine imagery},
  author={Contini, Matteo and Illien, Victor and Julien, Mohan and Ravitchandirane, Mervyn and Russias, Victor and Lazennec, Arthur and Chevrier, Thomas and Rintz, Cam Ly and Carpentier, L{\'e}anne and Gogendeau, Pierre and others},
  journal={Scientific Data},
  volume={12},
  number={1},
  pages={67},
  year={2025},
  publisher={Nature Publishing Group UK London}
}