BIOSCAN-5M: Multimodal Insect Images and DNA Barcodes

Provider
BIOSCAN project (Gharaee, Lowe, Gong et al.; University of Guelph, Vector Institute and partners)
License
CC-BY-3.0 (source)
Size
13 files
Stored at
Hugging Face (https://huggingface.co/datasets/bioscan-ml/BIOSCAN-5M)
Access
Free and public. No account with the data provider is needed.
Category
Life Sciences
Source
https://arxiv.org/abs/2406.12723

About BIOSCAN-5M: Multimodal Insect Images and DNA Barcodes

Multimodal data for over 5 million insect specimens from the BIOSCAN biodiversity monitoring effort, presented at NeurIPS 2024 (Datasets and Benchmarks). Each specimen has a microscope image, taxonomic labels, its raw DNA barcode nucleotide sequence, an assigned Barcode Index Number (BIN), geographic information and specimen size. Partitions support closed-world species classification (train, val, test), open-world generalization to unseen species of known genera, novelty detection (other_heldout) and a large unlabelled pretrain split for self-supervised learning. Tasks include DNA-based taxonomic classification, zero-shot transfer and cross-modal retrieval between images, DNA and text. Listing the repository (measured 2026-11-13) shows zip archives of images at 256-pixel resolution in two versions, original_256 (37.8 GB) and cropped_256 (39.1 GB), each also split into pretrain, train and eval archives, plus BIOSCAN_5M_Insect_Dataset_metadata_MultiTypes.zip (2.07 GB), a dataset.py loader and a usage demo; about 160 GB in all. Load with Hugging Face datasets using dataset.py (configurations such as cropped_256_eval), or unzip and read metadata with pandas and images with PIL. Released under CC-BY-3.0; cite Gharaee et al. 2024.

How to load BIOSCAN-5M: Multimodal Insect Images and DNA Barcodes in Python

from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="bioscan-ml/BIOSCAN-5M", filename="BIOSCAN_5M_Insect_Dataset_metadata_MultiTypes.zip", repo_type="dataset")
print(path)

Data structure

Files

Example file: BIOSCAN_5M_Insect_Dataset_metadata_MultiTypes.zip

Analyze this dataset without downloading it

On Scigantic, BIOSCAN-5M: Multimodal Insect Images and DNA Barcodes opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Life Sciences datasets