LAION-fMRI

brain images computer vision deep learning imaging life sciences machine learning magnetic resonance imaging neuroimaging neuroscience nifti

Description

LAION-fMRI is one of the largest and most comprehensive high-resolution imaging datasets of brain responses to visually-presented photographs which we built to uncover how the human brain sees and understands the world. Five participants viewed over 25,000 unique natural images in 165 hours across 165 sessions, capturing hundreds of thousands of brain responses every second with an ultra-high-field 7-Tesla MRI scanner. This extraordinary scale means results can be replicated across individuals and generalized across a very broad range of visual experiences, overcoming critical barriers that previous datasets simply could not. The images span everything from everyday photographs (sampled from 120 million image-text pairs) to abstract shapes and visual illusions, ensuring the dataset covers the full breadth of human visual experience. Every image was measured multiple times, delivering exceptional signal quality and setting new standards for the field. Beyond functional brain scans, the dataset includes rich complementary data (eye-tracking, behavioral responses, detailed maps of each participant's visual brain architecture, and cutting-edge white matter fiber tracking), making it possibly the most deeply characterized neuroimaging resources assembled to date. LAION-fMRI provides the foundation for building next-generation AI vision models grounded in real human brain data, fundamentally bridging the gap between biological and artificial intelligence.

Update Frequency

Infrequent. Supplemental fMRI sessions and additional metadata will be released after the launch release, otherwise updates are limited to corrections.

License

CC0 1.0 for the fMRI data, derivatives, and metadata (https://creativecommons.org/publicdomain/zero/1.0/). Raw stimulus images require accepting a Data Use Agreement: https://laion-fmri.hebartlab.com/request

Documentation

https://laion-fmri.hebartlab.com

Managed By

Hebart Lab

See all datasets managed by Hebart Lab.

Contact

hebart@cbs.mpg.de

How to Cite

LAION-fMRI was accessed on DATE from https://registry.opendata.aws/laion-fmri.

Usage Examples

Tutorials

Resources on AWS

  • Description
    BIDS-organized 7T fMRI data (NIfTI) with derivatives including preprocessed data, ROIs, single-trial beta estimates from GLMsingle, retinotopy maps, functional localizers, and precision diffusion MRI tractography.
    Resource type
    S3 Bucket
    Amazon Resource Name (ARN)
    arn:aws:s3:::laion-fmri
    AWS Region
    us-west-2
    AWS CLI Access (No AWS account required)
    aws s3 ls --no-sign-request s3://laion-fmri/

Edit this dataset entry on GitHub

Tell us about your project

Home