Dataset at a Glance¶

This page gives you a quick overview of everything in LAION-fMRI and helps you find the files you need.

The launch release covers 5 participants, 150 main image-viewing fMRI sessions (30 per participant), 25,052 distinct images, and 1,492 images shared by every participant. In total, 165 fMRI sessions were acquired (33 per participant); the supplemental sessions expand the shared image set to about 2,200 images and will be released later.

What’s Included¶

The dataset contains raw and preprocessed functional MRI data (fMRI Data, Preprocessing), anatomical scans and FreeSurfer reconstructions (Anatomical Data), single-trial GLMsingle beta estimates (GLMsingle Beta Estimates), the stimulus set (Stimulus Set) and derived stimulus files (Stimulus Derivatives), ROI masks (ROIs) derived from retinotopy (Retinotopy) and functional localizers (Functional Localizers), and predefined train/test splits (Train / Test Splits).

Detailed BIDS file trees, coordinate-space tables, and per-ROI summaries will be added in an upcoming documentation update.

What Files Do I Need?¶

Not everyone needs the full dataset. Start from your use case below to find the relevant files and documentation pages.

Encoding / decoding models¶

This is the most common use case (e.g. Algonauts challenge participants).

What you need	Details
Single-trial beta estimates	`derivatives/glmsingle-tedana/sub-XX/` - see GLMsingle Beta Estimates
Stimulus images & metadata	`stimuli/` - see Stimulus Set
Stimulus derivatives	Embeddings, captions, and object segmentations - see Stimulus Derivatives
Train / test splits	Predefined splits for model evaluation - see Train / Test Splits
ROI masks (optional)	`derivatives/rois/` - see ROIs

RSA / pattern similarity analyses¶

What you need	Details
Single-trial betas	`derivatives/glmsingle-tedana/sub-XX/` - see GLMsingle Beta Estimates
Stimulus metadata & categories	`stimuli/task-images_metadata.csv` - see Stimulus Set
ROI masks	`derivatives/rois/` - see ROIs

Retinotopic or localizer analyses¶

What you need	Details
Retinotopic maps	`derivatives/retinotopy/` - see Retinotopy
Functional localizer contrasts	`derivatives/localizers/` - see Functional Localizers
ROI masks	`derivatives/rois/` - see ROIs

Preprocessing from scratch¶

If you want to run your own preprocessing pipeline instead of using the provided NORDIC/tedana outputs.

What you need	Details
Raw BOLD data	`sub-XX/func/` - see fMRI Data
T1w anatomical scans	`sub-XX/anat/` - see Anatomical Data
Event timing files	`sub-XX/func/*_events.tsv` - see Experimental Design
Diffusion data (if needed)	`sub-XX/dwi/` - see Diffusion Data

Tip

For details on the preprocessing we already ran, see Preprocessing. For MRI acquisition parameters, see MRI Acquisition.

Data Formats¶

Data type	Format	Notes
MRI volumes	NIfTI (.nii.gz)	3D (anatomical) or 4D (functional/betas)
Metadata	JSON (.json)	BIDS sidecar files
Events / behavioral	TSV (.tsv)	Tab-separated, BIDS-compliant
Stimulus images	HDF5 (.h5)	Single packed file with raw JPEG bytes, row-aligned to `task-images_metadata.csv` and accessible by image name through `laion_fmri.load_stimuli()`.
Stimulus metadata	CSV	`stimuli/task-images_metadata.csv` - see Stimulus Set