BIOSCAN-5M

Overview

As part of an ongoing worldwide effort to comprehend and monitor insect biodiversity, we present the BIOSCAN-5M Insect dataset to the machine learning community. BIOSCAN-5M is a comprehensive dataset containing multi-modal information for over 5 million insect specimens, and it significantly expands existing image-based biological datasets by including taxonomic labels, raw nucleotide barcode sequences, assigned barcode index numbers, geographical information, and specimen size.

Citation

If you make use of the BIOSCAN-5M dataset and/or its code repository, please cite the following paper:

cite as:

@inproceedings{gharaee2024bioscan5m,
    title={{BIOSCAN-5M}: A Multimodal Dataset for Insect Biodiversity},
    booktitle={Advances in Neural Information Processing Systems},
    author={Zahra Gharaee and Scott C. Lowe and ZeMing Gong and Pablo Millan Arias
        and Nicholas Pellegrino and Austin T. Wang and Joakim Bruslund Haurum
        and Iuliia Zarubiieva and Lila Kari and Dirk Steinke and Graham W. Taylor
        and Paul Fieguth and Angel X. Chang
    },
    editor={A. Globerson and L. Mackey and D. Belgrave and A. Fan and U. Paquet and J. Tomczak and C. Zhang},
    pages={36285--36313},
    publisher={Curran Associates, Inc.},
    year={2024},
    volume={37},
    url={https://proceedings.neurips.cc/paper_files/paper/2024/file/3fdbb472813041c9ecef04c20c2b1e5a-Paper-Datasets_and_Benchmarks_Track.pdf},
}

Large-Scale Foundation Model Training for Biodiversity Research

Dataset partitions

Partition	Example Splits	Description
Closed-world	train, val, test	Samples with known species names for supervised classification.
Open-world	key_unseen, val_unseen, test_unseen	Placeholder species names but known genera, enabling generalization to unseen species.
Novelty Detection	other_heldout	Unknown species and genus, suitable for open-set detection.
Pretraining	pretrain	Unlabeled data for self-/semi-supervised learning at scale.

Supported tasks

Task	Description
DNA-based Taxonomic Classification	Predict taxonomic labels from raw DNA barcode sequences.
Zero-Shot Transfer Learning	Evaluate whether unlabeled models can semantically cluster data—across modalities like image and DNA—using learned representations.
Multimodal Retrieval Learning	Retrieve matching specimens across modalities (e.g., image ↔ DNA ↔ text) via shared embeddings.

Dataset features via metadata fields

Field Group	Field(s)	Description
Image	image	RGB JPEG image of an individual insect specimen.
Indexing	processid, sampleid	Unique identifiers from BOLD and the collector.
Taxonomy	phylum, class, order, family, subfamily, genus, species	Hierarchical taxonomic classification.
Genetics	dna_bin, dna_barcode	Barcode Index Number and DNA sequence.
Geography	country, province_state, coord-lat, coord-lon	Collection location and geographic coordinates.
Specimen Size	image_measurement_value, area_fraction, scale_factor	Image-based size measures and normalization factors.
Splits & Storage	split, chunk	Data partition (e.g., train/test) and storage subdirectory.

Usage

First, download the dataset.py script to your project directory by running the following command:

wget -P /path/to/your/project_directory https://huggingface.co/datasets/bioscan-ml/BIOSCAN-5M/resolve/main/dataset.py

Once you've downloaded the script, you can use the datasets library to load the dataset. For example:

from datasets import load_dataset

ds = load_dataset("dataset.py", name="cropped_256_eval", split="validation", trust_remote_code=True)

ℹ️ Note: The CSV metadata and image package associated with the selected configuration will be automatically downloaded and extracted to ~/.cache/huggingface/datasets/downloads/extracted/ .

📑 Configurations

Each configuration loads specimen images along with associated metadata fields:

name	Available split values
cropped_256, original_256	pretrain, train, validation, test, val_unseen, test_unseen, key_unseen, other_heldout
cropped_256_pretrain, original_256_pretrain	pretrain
cropped_256_train, original_256_train	train
cropped_256_eval, original_256_eval	validation, test, val_unseen, test_unseen, key_unseen, other_heldout

ℹ️ Note: If you do not specify the split when loading the dataset, all available splits will be loaded as a dictionary.

Sample Usage

First, download the usage_demo_bioscan5m.py script to your project directory by running the following command:

wget -P /path/to/your/project_directory https://huggingface.co/datasets/bioscan-ml/BIOSCAN-5M/resolve/main/usage_demo_bioscan5m.py

This script demonstrates how to load and visualize samples from the BIOSCAN-5M dataset.

To run the script, execute:

python usage_demo_bioscan5m.py

This will display 10 dataset samples, each showing the organism image on the right, and the corresponding metadata fields on the left, including taxonomic, geographic, genetic, and size-related information.

Dataset Access

To clone this dataset repository, use the following command:

GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/bioscan-ml/BIOSCAN-5M

📦 Resources and Access

• 📄 Paper: arXiv
• 🌐 Website: BIOSCAN-5M Project Page
• 💻 GitHub: bioscan-ml/BIOSCAN-5M
• 📁 Downloads:
  • Google Drive
  • Zenodo
  • Kaggle