laion_fmri¶
A data downloader and wrangler for the LAION-fMRI dataset. The package mirrors the bucket layout to your local disk via the official AWS CLI, keeps every accessor as a one-to-one map onto a single file in S3, and applies BIDS-entity filters so you only fetch what you need.
A typical session looks like:
from laion_fmri.config import dataset_initialize
from laion_fmri.download import download
from laion_fmri.subject import load_subject
dataset_initialize("./laion_fmri_data")
download(subject="sub-03", ses="ses-01", n_jobs=4)
sub = load_subject("sub-03")
betas = sub.get_betas(session="ses-01") # (n_trials, n_voxels), float32
The same three workflow steps – configure, inspect,
download – are also exposed as a laion-fmri shell
command, installed automatically by pip/uv:
laion-fmri config --data-dir ./laion_fmri_data
laion-fmri info
laion-fmri download --subject sub-03
Loading still happens from Python; the CLI covers configure,
inspect, and download. See Download for full
laion-fmri download semantics.
The cards below walk through each step in detail.
Pick a local data directory and persist the choice across sessions.
CC0 for the fMRI side, a short Data Use Agreement form for the stimulus images. The package handles both.
List subjects, ROIs, and bucket structure – every query reads S3 directly, no local download needed.
BIDS-entity filters, strict ses semantic with the
"averages" keyword, idempotent re-runs, and n_jobs
parallelism.
Single-trial betas, noise-ceiling maps, ROI masks, brain-space mapping, multi-subject groups, and PyTorch.
Project T1w-space values onto fsaverage surfaces or MNI volumes via the bundled FreeSurfer recon.
Five hands-on, narrated walkthroughs covering the full workflow.
Auto-generated reference for every public module, class, and function.