Dataset at a Glance¶
The Active Visual Semantics (AVS) dataset combines MEG and eye-tracking data recorded while participants visually explored natural scenes, along with structural MRI scans acquired before the first MEG session. By focusing on active vision, it moves beyond the passive, fixation-enforced viewing paradigms used in most existing neuroimaging datasets.
Participants |
5 (see Participants) |
Sessions |
10 MEG + eye-tracking sessions per participant, plus one anatomical session. 10 runs in ses-01, 14 in ses-02 – ses-10; 30 trials per run, so 4,080 trials (one per scene) per participant |
Stimuli |
4,080 natural scenes from the Natural Scenes Dataset (NSD), semantically balanced (see Stimuli) |
Task |
Active viewing (4 s/scene) with a verbal scene-captioning task on 25% of trials (see Semantic Captioning Task) |
MEG |
306-channel Elekta Neuromag TRIUX, 1000 Hz (see MEG Acquisition) |
Eye tracking |
EyeLink 1000, 1000 Hz (see Eye Tracking) |
Fixation epochs |
More than 200,000 across the dataset. Shipped fixation- and saccade-locked, -0.5 to 0.8 s at 500 Hz, with per-epoch metadata |
Object labels |
Per-fixation MS-COCO / COCO-Stuff category labels, 171 categories (see Fixation Object Labeling) |
Anatomy |
Defaced individual T1 plus a ready-to-use FreeSurfer |
Release size |
22,488 files, 663.1 GiB in total – but the tree is designed for selective download (see Data Access) |
For the release layout and file formats, see Dataset Structure; for known data-quality caveats, see Known Issues; for how to get the data, see Data Access.
What Do I Need?¶
Goal |
Start here |
|---|---|
Load MEG + eye-tracking data for an analysis |
|
Understand the acquisition/preprocessing pipeline |
|
Source reconstruction / population codes |
|
Map fixations to scene objects |
|
Work out which parts of the dataset to download |
|
See example analyses (RSA, encoding, source ERFs) |
|
Full function/class reference |