BackHome19K

Dataset Description

BackHome19K is the first comprehensive seashell image dataset specifically created for ecosystem-level classification—distinguishing between shells from the Pacific and Caribbean coasts—rather than traditional taxonomic identification.

The dataset includes 19,058 high-quality images of 516 marine mollusk species, collected with the support of the School of Biology at the University of Costa Rica. It features a wide range of gastropods and bivalves, from families such as Cypraeidae, Cassidae, Ostreidae, Veneridae, and others.

This collection addresses the challenge of visually similar species across different ecosystems and supports research in geographic origin inference, morphological pattern analysis, and marine biodiversity.

Dataset Structure

The dataset follows the folder structure below:

BackHome19K/
├── train/
│   ├── Pacifico/
│   │   ├── species_1.jpg
│   │   ├── species_2.jpg
│   │   └── ...
│   └── Caribe/
│       ├── species_1.jpg
│       ├── species_2.jpg
│       └── ...
├── val/
│   ├── Pacifico/
│   │   ├── species_1.jpg
│   │   └── ...
│   └── Caribe/
│       ├── species_1.jpg
│       └── ...
└── test/
    ├── Pacifico/
    │   ├── species_1.jpg
    │   └── ...
    └── Caribe/
        ├── species_1.jpg
        └── ...

Dataset Details

• Total number of images: [19,058]
• Number of families: [84]
• Number of species: [516]
• Geographic Regions: Caribbean and Pacific from Costa Rica
• Image sources: Publicly available images from various online sources
• Image Format: JPEG/JPG/PNG
• Partitioning: The dataset is split into train, val, and test folders for supervised learning tasks.
• List of families:

Family	Family
Acmaeidae	Acteonidae
Architectonicida	Architectonicidae
Arcidae	Areneidae
Batillariidae	Buccinidae
Bullidae	Bursidae
Calliostomatidae	Calyptraeidae
Cardiidae	Carditidae
Cassidae	Cerithiidae
Chamidae	Chamidae_Pseudochama
Charoniidae	Columbellidae
Conidae	Corbulidae
Crassatellidae	Cymatiidae
Cypraeidae	Cyrenidae
Cyrenoididae	Cystiscidae
Donacidae	Dreissenidae
Ellobiidae	Fasciolariidae
Ficidae	Fissurellidae
Glycymerididae	Gryphaeidae
Harpidae	Hipponicidae
Isognomonidae	Janthinidae
Limidae	Littorinidae
Lottiidae	Lucinidae
Mactridae	Margaritidae
Melongenidae	Mitridae
Modulidae	Muricidae
Mytilidae	Nassariidae
Naticidae	Neritidae
Noetiidae	Olividae
Ostreidae	Ovulidae
Pectinidae	Personidae
Pinnidae	Pisaniidae
Planaxidae	Plicatulidae
Psammobiidae	Pteriidae
Ranellidae	Semelidae
Siphonariidae	Solecurtidae
Spondylidae	Strombidae
Tegulidae	Tellinidae
Terebridae	Tonnidae
Triviidae	Trochidae
Turbinellidae	Turbinidae
Turritellidae	Ungulinidae
Veneridae	Volutidae

Download the Dataset

You can easily download and load the dataset using the datasets library:

1. Install the Datasets library

pip install datasets

from datasets import load_dataset

dataset = load_dataset("BackHome19K", split="train")
#Replace "train" with "val" or "test" to access the other splits.

Licensing and Usage

• License: MIT License
• Usage: This dataset is intended for educational and research purposes only. Commercial use of the dataset requires prior verification of the original image sources. Respect the copyright and intellectual property rights associated with the collected images.

Acknowledgments

We thank the School of Biology at the University of Costa Rica for their guidance in species selection and validation. We also acknowledge the marine biologists who contributed to expert taxonomic verification and the collection of reference images.

We are deeply grateful to all contributors who have made marine species documentation publicly available, and to the creators of open-source tools that made this project possible.

This dataset was created to promote awareness of marine biodiversity and to support scientific research and conservation efforts. We hope it helps researchers, educators, and enthusiasts understand and protect our oceans’ ecosystems.

Citation

If you use this dataset in your research, please cite it as follows:

@dataset{BackHome19K,
  title = {BackHome19K},
  author = {FIFCO},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/FIFCO/BackHome19K}
}