Data Access¶

LAION-fMRI is hosted in an AWS S3 bucket sponsored by the AWS Open Data program:

The fMRI data, derivatives, dataset metadata, and stimulus-derived annotations are released openly under CC0 1.0 and can be downloaded anonymously. These files include stimulus metadata, captions, pretrained embeddings, and object segmentations.
The raw stimulus image HDF5 comes from third-party web sources and requires accepting a short Data Use Agreement.

The laion_fmri Python package handles both download paths transparently. For a quick orientation see Quickstart; this page is the full reference.

Access Requirements¶

CC0 fMRI data¶

The package shows the CC0 license once on the first download(...) call and writes a marker file so subsequent calls don’t re-prompt.

Stimulus images¶

The stimulus images require accepting a Data Use Agreement that prohibits redistribution, commercial use, and use for training general-purpose AI models. You accept it by submitting a short form (terminal or web).

The form asks for your name, institutional email, institution, optional PI/supervisor, a short research-purpose description, Terms acceptance, and Privacy-notice acknowledgement.

Download Methods¶

Python package (recommended)¶

The laion_fmri package handles both the public S3 mirror and the stimulus service. Until the PyPI package is published, install directly from GitHub:

python -m pip install "laion-fmri @ git+https://github.com/ViCCo-Group/LAION-fMRI.git@main"

Common operations:

from laion_fmri.download import download, download_stimuli

# fMRI for one subject
download(subject="sub-01")

# one session, parallel transfer
download(subject="sub-01", ses="ses-04", n_jobs=4)

# all subjects (whole-dataset mirror)
download(subject="all")

# stimuli - dataset-wide, subject-independent, requires accepting the DUA
download_stimuli()

# both at once
download(subject="sub-01", include_stimuli=True)

CLI equivalents:

mkdir -p ./laion_fmri_data
laion-fmri config --data-dir ./laion_fmri_data
laion-fmri download --subject sub-01
laion-fmri download-stimuli
laion-fmri request-access            # standalone DUA form, no download

Direct AWS CLI (public files only)¶

You can also use the AWS CLI directly. The public prefixes are read-accessible without credentials:

aws s3 sync --no-sign-request \
    s3://laion-fmri/derivatives/glmsingle-tedana/sub-01/ ./sub-01/

This skips the package’s BIDS-entity filtering and idempotency checks, but is useful if you want raw control over what’s transferred.

The raw image archive s3://laion-fmri/stimuli/task-images_stimuli.h5 is not accessible this way. Use the package or the web form for that file. The public stimulus metadata, captions, embeddings, and segmentations can be fetched anonymously.

Web form (no Python required)¶

For browser users who’d rather not install the Python package, the same DUA form is available at:

https://laion-fmri.hebartlab.com/request

The confirmation page shows the presigned download URLs directly; fetch them with curl / wget or by clicking. URLs are valid for one hour.

Data Verification¶

fMRI data: the package checks each file’s local size against the S3 size before re-fetching; download(...) is idempotent.
stimuli: download is verified against a published sha256 for both files (the manifest is served at https://laion-fmri.hebartlab.com/api/v1/manifest). On mismatch the .part file is removed and an error is raised.

Software Requirements¶

Python 3.10+
The laion_fmri package, installed from GitHub until the PyPI distribution is published: python -m pip install "laion-fmri @ git+https://github.com/ViCCo-Group/LAION-fMRI.git@main". It pulls in numpy, h5py, nibabel, pandas, awscli, and Pillow.
Raw stimulus-image downloads require the Data Use Agreement flow through download_stimuli(), laion-fmri download-stimuli, or laion-fmri request-access.

Citation¶

Until the dataset paper is available, cite the VSS 2026 conference presentation:

Zerbe, J., Roth, J., Mell, M. M., Herholz, P., Knapen, T., & Hebart, M. N. (2026). LAION-fMRI: A densely sampled 7T-fMRI dataset providing broad coverage of natural image diversity. Talk 25.11, Scene Perception Talk Session, Vision Sciences Society Annual Meeting, May 16, 2026. VSS abstract.

Also cite GLMsingle (Prince et al., 2022) if you use the provided single-trial beta estimates.

Support¶

For data access issues or questions:

Open an issue on GitHub
For stimulus image takedown, see the contact at https://laion-fmri.hebartlab.com/takedown
Check the FAQ for common questions