Dataset Card for Phylo-Fish

ML-ready dataset containing 38 species of Teleost fishes with an average number of 200 images per species. This semi-balanced subset was sourced from a larger fish image collection from the Great Lakes Invasives Network Project (GLIN)

Dataset Details

Teleost fish images from five ichthyological research collections that participated in the Great Lakes Invasives Network Project (GLIN) (original data source site: web archive link). After obtaining the raw images from these collections (through the Fish-AIR Database), we handpicked a subset for use in this dataset, and pre-processed them by resizing and appropriately padding each image to be of a 256×256 pixel resolution. We split this subset into a training set and a validation set of ratios 80 and 20, respectively. This dataset contains 38 species of Teleost fishes with an average number of 200 images per species.

Supported Tasks and Leaderboards

[More Information Needed]

Dataset Structure

/dataset/
    train/
        <species_1>/
            <img_id 1>.JPG
            <img_id 2>.JPG
            ...
            <img_id n>.JPG
        <species_2>/
            <img_id 1>.JPG
            <img_id 2>.JPG
            ...
            <img_id n>.JPG
        ...
        <species_38>/
            <img_id 1>.JPG
            <img_id 2>.JPG
            ...
            <img_id n>.JPG
    test/
        <species_1>/
            <img_id 1>.JPG
            <img_id 2>.JPG
            ...
            <img_id n>.JPG
        <species_2>/
            <img_id 1>.JPG
            <img_id 2>.JPG
            ...
            <img_id n>.JPG
        ...
        <species_38>/
            <img_id 1>.JPG
            <img_id 2>.JPG
            ...
            <img_id n>.JPG
    fish_test.txt
    fish_train.txt
    metadata.csv

Data Instances

The train/ and test/ directories are subdivided by species (named "Genus species epithet", e.g., "Alosa chrysochloris"). Each image filename is as provided by the original source (e.g., INHS_FISH_101620.JPG).

The fish_train and fish_test text files are lists of the images used for the train and test splits, respectively; these are the filepaths from the root of the repository.

Data Fields

metadata.csv: Unified image metadata (identification, provenance, and taxonomic labels).

• filepath: Relative path to image from the root of the directory (<split>/<species>/<filename>); allows for image to be displayed in the dataset viewer alongside its associated metadata.
• filename: Image filename, as assigned by the original source data provider (ex: INHS_FISH_101620.JPG).
• scientificName: Scientific name of the fish specimen, assigned by the source (<Genus> <species epithet>).
• genus: Genus of the fish specimen, assigned by source data provider. There are 11 different genera of fish.
• family: Family of the fish specimen, assigned by source data provider. There are 8 different families of fish.
• species: Species epithet of the specimen in the image. There are 38 different species of fish.
• split: Split to which the image belongs (train or test).
• license: License under which the image is shared (CC BY-NC for all images).
• source: Provider of the image (Great Lakes Invasive Species Network).
• ownerInstitutionCode: Owner Institution Code for the specimen/image owner (GLIN is an aggregator); all entries are INHS for the Illinois Natural History Survey .
• ARKID: Unique multimedia identifier for the image as downloaded from Fish-AIR Database.
• accessURI: Fish-AIR Database access URL for original image.
• md5: The MD5 hash of the image, generated using the sum-buddy package.

Data Splits

Images are divided into a train-test split as indicated by their source folder. There is also a split column provided in metadata.csv.

There are 4,140 images for training, and 1,294 assigned to the test set.

Dataset Creation

Curation Rationale

One of the grand challenges in biology is to find features of organisms--or traits--that define groups of organisms, their genetic and developmental underpinnings, and their interactions with environmental selection pressures. Traits can be physiological, morphological, and/or behavioral (e.g., beak color, stripe pattern, and fin curvature) and are integrated products of genes and the environment. The analysis of traits is critical for predicting the effects of environmental change or genetic manipulation, and to understand the process of evolution. For example, discovering traits that are heritable across individuals, or across species on the tree of life (also referred to as the phylogeny), can identify features useful for individual recognition or species classification, respectively, and is a starting point for linking traits to underlying genetic factors. Traits with such genetic or phylogenetic signal, termed evolutionary traits, are of great interest to biologists, as the history of genetic ancestry captured by such traits can guide our understanding of how organisms vary and evolve. This understanding enables tasks such as estimating the morphological features of ancestors, how they have responded to environmental changes, or even predicting the potential future course of trait changes. However, the measurement of traits is not straightforward and often relies on subjective and labor-intensive human expertise and definitions. Hence, trait discovery has remained a highly label-scarce problem, hindering rapid scientific advancement.

Source Data

All of these images were downloaded from the Fish-AIR Database, where they were deposited by the Great Lakes Invasives Network (GLIN). All fish are museum specimen images from the Illinois Natural History Survey (INHS) Fish Collection.

Data Collection and Processing

Images were hand-selected from the initial Fish-AIR download, then cropped to 256x256 with padding the images with the ImageNet mean RGB color. Data augmentation was used when training the base VQGAN model, including random horizontal flips, spatial shifts and rotations, and brightness and contrast changes.

Who are the source data producers?

These museum specimen images and their associated taxonomic labels are from the Illinois Natural History Survey (INHS) Fish Collection.

Annotations

Species annotations are from the source data providers. The phylogeny corresponding to the dataset was obtained using the OpenTree Python package. Phylogeny processing and manipulation were done using the ETE Toolkit python package (ete3). Please see our paper for more details on this process.

Personal and Sensitive Information

None.

Considerations for Using the Data

Bias, Risks, and Limitations

Most of the specimens were collected in the Great Lakes region of the United States of America, so this is a geographically limited sample. Additionally, as museum specimens, their morphological traits are impacted by the process of preservation in alcohol and formaldehyde.

Licensing Information

The data (images and text) are licensed under Creative Commons Attribution Non-Commercial (CC BY-NC). The compilation has been marked as dedicated to the public domain by applying the CC0 Public Domain Waiver. However, images are licensed under different terms (as noted above).

Citation

Please cite both the dataset and the associated paper as noted below.

Data

@misc{<ref_code>,
  author = {Elhamod, Mohannad and Khurana, Mridul and Manogaran, Harish Babu and Uyeda, Josef C. and Balk, Meghan A. and Dahdul, Wasila and Bakiş, Yasin and Bart, Henry L. and Mabee, Paula M. and Lapp, Hilmar and Balhoff, James P. and Charpentier, Caleb and Carlyn, David and Chao, Wei-Lun and Stewart, Charles V. and Rubenstein, Daniel I. and Berger-Wolf, Tanya and Karpatne, Anuj},
  title = {Phylo-Fish},
  year = {2026},
  url = {https://huggingface.co/datasets/imageomics/phylo-fish},
  doi = {<doi once generated>},
  publisher = {Hugging Face}
}

Paper

@article{Elhamod_Discovering_Novel_Biological_2023,
    author = {Elhamod, Mohannad and Khurana, Mridul and Manogaran, Harish Babu and Uyeda, Josef C. and Balk, Meghan A. and Dahdul, Wasila and Bakiş, Yasin and Bart, Henry L. and Mabee, Paula M. and Lapp, Hilmar and Balhoff, James P. and Charpentier, Caleb and Carlyn, David and Chao, Wei-Lun and Stewart, Charles V. and Rubenstein, Daniel I. and Berger-Wolf, Tanya and Karpatne, Anuj},
    doi = {10.1145/3580305.3599808},
    journal = {Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD '23)},
    title = {{Discovering Novel Biological Traits From Images Using Phylogeny-Guided Neural Networks}},
    year = {2023}
}

Please be sure to also cite the original data sources:

Multimedia of Fish Specimen and associated metadata. Fish-AIR. Biology guided Neural Network. Tulane University Biodiversity Research Institute (https://fishair.org).

INHS collections data. 2022. http://biocoll.inhs.illinois.edu/portal/index.php.

Acknowledgements

This work was supported by the Imageomics Institute, which is funded by the US National Science Foundation's Harnessing the Data Revolution (HDR) program under Award #2118240 (Imageomics: A New Frontier of Biological Information Powered by Knowledge-Guided Machine Learning). Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Dataset Card Contact

Please open a discussion in the Community tab if you have questions regarding the dataset.