Experimental Design

The LAION-fMRI dataset uses a dense-sampling design: a small number of participants are scanned across many sessions, each seeing on the order of 6,000 unique natural images. The general logic follows dense sampling designs (few subjects, many sessions, controlled image repetitions for noise ceiling estimation), but with a broader image set based on LAION-natural (derived from LAION-2B, Roth & Hebart, 2025) that goes beyond standard scene or object datasets. Each of the 5 participants completed 30 main sessions of image viewing, one eyetracking session (ses-31), and 3 supplemental fMRI sessions after a time gap of about half a year. The launch release focuses on the 30 main image-viewing sessions per participant; the supplemental sessions and their additional shared images will be released later. Counting all acquired fMRI sessions gives 165 sessions (33 per participant), plus 5 eyetracking sessions. In total, the recording took place over the course of about 1.5 years (Nov 2024 to Mar 2026) at the Max Planck Institute for Human Cognitive and Brain Sciences in Leipzig, Germany.

Experiments

Experiment	Design type	Purpose
Main image viewing (task-images, ses-01 to ses-30)	Event-related, single trials	Single-trial responses to ~6,100 natural images per subject
Precision mapping (six sessions)	Sweeping bars and rotating wedges; block-design localizers	Population receptive-field estimates across the central and peripheral visual field, plus category- and motion-selective maps
Eyetracking session (ses-31)	Mixed (calibration + image viewing + localizer + deepmreye)	Independent test set with concurrent eyetracking that includes task-deepmreye (2 runs), task-images (2 runs), task-oloc (4 runs)

Main Experiment

In the main experiment (ses-01 to ses-30), participants viewed natural images while performing a continuous recognition task (“Have you seen this image before?”). On each trial participants pressed a button to indicate whether the image was new (first presentation) or old (already seen in a previous trial or session) with their right index finger. The task keeps participants attentive during long sessions and yields behavioural data on memory performance; the primary purpose of the experiment is to obtain reliable single-trial BOLD responses to each image.

Each subject saw around 6,204 unique images across roughly 31,856 image presentations, plus around 2,583 blank trials. The image set is split into shared and subject-unique images. The 1,492 shared images are common to all subjects: 881 of them are shown 12 times (the 12-repeat set, used for noise ceiling estimation) and 611 are shown 4 times (the 4-repeat set). The remaining 4,712 unique images per subject are shown 4 times each and are different across subjects, which widens the total stimulus set seen across all participants without reducing per-subject repetition counts.

Images were distributed across the 30 sessions following two scheduling rules. First, every image was shown at least twice within the same session at some point during the experiment, so that GLMsingle has within-session repeats to work with for HRF fitting and ridge regression tuning. No image was shown more than 3 times in the same session. Second, one repeat of the same image was always kept within ±7 sessions of its twice-per-session occurrence, so that the inter-repetition gap stays manageable for the memory task.

For details on the image content and where the images come from, see Stimulus Set. For information on which images are reserved for testing, see Train / Test Splits.

Note on incomplete repetitions

Most images were shown for their full number of repetitions but 8 images per participant ended up with one repetition fewer than planned. In most cases this affects subject-unique images that were only shown 3 times and more rarely, 12-repeat images that were only shown 11 times. The exact counts per subject are:

Subject	unique (4 → 3 reps)	shared_4rep (4 → 3 reps)	shared_12rep (12 → 11 reps)
sub-01	5	0	3
sub-03	4	1	3
sub-05	5	0	3
sub-06	4	0	4
sub-07	6	0	2

For analyses that require an exact repetition count (across-session Noiseceiling4rep and Noiseceiling12rep maps), these images were excluded. For analyses that average over all available repeats (NoiseceilingAllrep), they remained. For cross-subject analyses, these images should be excluded.

Session Structure

A session always consists of 12 functional runs, plus 1 structural and 4 fieldmap acquisitions. The standard layout used for all participants from ses-08 onwards was:

[S F RRRR F RRRR F RRRR F]

where S = structural (MEGRE), F = fieldmap, R = experimental run. The structural is acquired at the start of each session, followed by a fieldmap and then groups of 4 runs separated by additional fieldmaps. Fieldmaps are repeated multiple times throughout the session to track B0 field drift over time.

Earlier sessions (ses-01 to ses-04) used a different layout in which the structural was placed in the middle of the session ([F RRRR F RR S RR F RRRR F]). From ses-05 onward, the structural was moved back to the start of the session.

Session structure overview.

Run Parameters

Duration per run	~5 min (304 s for sessions 1-5, 309 s for sessions 6-34)
Trials per run	~94-97 (87-89 image trials + 7-8 blank trials)
Runs per session	12
Total runs (per subject)	360 (12 runs × 30 main sessions)

The acquisition TR is 1.9 s, but during preprocessing the data is temporally upsampled to a TR of 1.0 s for GLMsingle. So depending on whether you look at the raw or preprocessed data, the number of timepoints per run will differ (160 or 163 raw volumes vs 304 or 310 timepoints after upsampling).

Note

For MRI acquisition parameters (TR, voxel size, etc.), see MRI Acquisition.

Trial Structure

Each trial lasts 3 seconds. On image trials, the stimulus is shown for 2.5 s followed by a 0.5 s inter-stimulus interval. Blank trials have the same duration but no image is shown. Blank trials are distributed pseudo-randomly throughout each run and serve as a baseline condition. They also help the BOLD signal return closer to baseline between image trials which improves single-trial response estimation. A fixation cross in form of a small red dot was continuously shown also during ISI and blank trials. Participants’ responses during ISI were still counted as belonging to the previous trial image. Top button pressing indicated a new image and second button just below indicated an old image.

Example trial sequence from the main experiment, showing image trials (2.5 s stimulus + 0.5 s ISI) interleaved with blank trials. The red fixation dot remains on screen throughout, including during the ISI and blank trials.

Stimulus Presentation

Stimuli were presented through a PROpixx MRI/MEG DLP LED projector (VPixx Technologies), mirroring a BenQ display at 1920 × 1080 px and 60 Hz refresh rate onto a rear projection screen inside the scanner room. Each stimulus image was rendered at 1000 × 1000 pixels, which corresponds to about 26.5 × 26.5 cm on the projection screen, or roughly 9.2 × 9.2 degrees of visual angle. The total viewing distance was 164.8-165.7 cm (3.9-4.6 cm from mirror to eye, plus 161 cm from mirror to screen).

The PsychoPy window used the default color=(0, 0, 0) in PsychoPy’s rgb colour space. This is middle grey (0.5 per channel in normalized RGB), not black. It filled the display outside the stimulus as well as the inter-stimulus and blank-trial background. For the 131 RGBA PNGs in the OOD set, transparent and partially transparent pixels were alpha-composited over this same background during presentation.

Software	PsychoPy
Display	PROpixx projector + BenQ mirrored display
Resolution	1920 × 1080 (projector); 1000 × 1000 (stimulus)
Refresh rate	60 Hz
Background	PsychoPy rgb=(0, 0, 0) (middle grey)
Viewing distance	164.8-165.7 cm
Visual angle (stimulus)	~9.2 × 9.2 degrees
Scanner synchronization	To be documented in an upcoming update
Dummy scans	0

Behavioral Data

Behavioral data from the main experiment consists of the button responses given during the recognition task, including the response (old or new) and the reaction time per trial.

Behavioral file names and column definitions will be added in an upcoming documentation update.

Precision Mapping Experiment

Retinotopic mapping and the category and motion localizers were acquired in a separate study, before the main image-viewing experiment began, and are referred to here as the precision mapping experiment. It comprises six sessions at TR 1.5 s: an initial session, ses-4BarScreenfLoc, which combined two retinotopy pilot runs with the eight fLoc runs, followed by five mapping sessions holding the remaining retinotopy runs, as well as motion localizer, resting-state, and DeepMREye acquisitions.

Session	Run sequence
ses-4BarScreenfLoc	4bar ×2, fLoc ×8
ses-MotionLocBarsSML	(MotionLeft MotionRight) ×2, (BARS BARM BARL) ×2
ses-BarsLMSEyeCalib	RS, (BARL BARM BARS) ×2, RS, DeepMREye ×2
ses-Corners1	RS, (UpperLeft LowerLeft UpperRight LowerRight) ×2
ses-Corners2	RS, (LowerRight UpperRight LowerLeft UpperLeft) ×2
ses-Foveal	RS, (BAR WEDGE) ×4, RS

Each row lists the runs in acquisition order by their BIDS task- label.

Retinotopy

To map the visual field at high resolution across a wide range of eccentricities, three paradigms were combined. The standard paradigm swept a bar across the full display, 17.4 x 9.8 degrees, with central fixation, in three tasks with bar widths of 2.5, 1.0 and 0.4 degrees (BARL, BARM, BARS). The corners paradigm used the same display and the 1.0 degree bar but placed the fixation mark in one of the four corners of the screen, 1 degree from each of the two adjacent edges, so that the stimulus extended to 16.4 degrees horizontally and 8.8 degrees vertically from fixation into the opposite hemifield; each corner is its own task (UpperLeft, UpperRight, LowerLeft, LowerRight). The foveal paradigm restricted the stimulus to a central 4 x 4 degree window and mapped it with a 0.4 degree bar (BAR) and a 30 degree wedge rotating about fixation (WEDGE). Every task was acquired four times per participant, giving 36 mapping runs. Two shorter pilot runs with a single bar pass per direction (4bar) were acquired at the start of the screening session and are held out from the released fits. Gaze position was recorded during the foveal runs only.

Retinotopic mapping paradigms. Left, a representative stimulus frame per paradigm, with bar width (or wedge angle) and, for the foveal paradigms, the extent of the stimulated field; the fixation target is enlarged for legibility. Right, run structure, with block durations in seconds. Straight arrows give the direction of bar motion, curved arrows the direction of wedge rotation, and shaded blocks represent blank periods.

The standard, corner and foveal bar runs share one structure: an 18 s fixation period, four blocks of two orthogonal bar sweeps followed by an 18 s blank, and an 18 s fixation period to close. Sweep directions follow a palindromic sequence, so each cardinal direction occurs twice per run and the second half of the run reverses the first. Horizontal sweeps last 37.5 s and vertical sweeps 22.5 s on the full display, matching bar speed across the two screen dimensions. Both last 22.5 s inside the square foveal window. Standard and corner runs therefore last 330 s and foveal bar runs 270 s. Wedge runs last 300 s: after 15 s of fixation the wedge completes four 30 s clockwise rotations, pauses for a 30 s blank, completes four counter-clockwise rotations, and the run closes with 15 s of fixation. The pilot runs are shorter, at 192 s: a single sweep in each cardinal direction, the two pairs of sweeps separated by an 18 s blank, and 9 s of fixation followed by an 18 s blank at either end.

Background images

Instead of the flickering checkerboard used in classical retinotopy (Dumoulin & Wandell, 2008), the bar and wedge apertures revealed colourful cartoon images (Finzi et al., 2021), aiming to drive higher-level visual areas more effectively. A new image was drawn every 0.1 s, in random order from a bank of 63. The full-display runs stretched a 1244 x 700 px image across the whole screen. The foveal runs used the same images, scaled and cropped to 440 x 440 px to fit inside the 4 degree window.

Task

Participants were instructed to hold fixation throughout all runs and to press any button whenever the fixation mark changed. The mark is a small dark diagonal cross, 0.07 degrees (8 px) across, drawn over a larger, thin, low-contrast diagonal cross that serves as a fixation guide. In the standard bar runs the arms of the guide cross the entire display. In the corner runs they extend only 1 degree from fixation, so that the cross keeps four arms of equal length instead of being cut off by the nearby screen edges. In the foveal runs they extend 2 degrees, spanning the foveal window from corner to corner. At each target the small cross switches from dark to light against the mid-grey background for 300 ms and then back.

Target times were drawn independently per run, at a mean interval of 3.2 s jittered uniformly by ±0.9 s in the standard and corner runs and by ±0.5 s in the foveal and task-4bar runs, clipped to the range 2.5-4.5 s. Targets run from the end of the opening fixation period to the end of the run, so they occur during blank periods as well as during sweeps. Because the task is at fixation and is orthogonal to bar position, it holds attention and gaze without contributing a stimulus-locked component to the pRF fits.

The fitted pRF models, the recommended surface map, parameter definitions and quality-control caveats will be documented in an upcoming documentation update.

Localizers

Three functional localizers were acquired: a category localizer (fLoc) and a motion localizer within the precision mapping sessions, and an object localizer (task-oloc) in the eyetracking session (ses-31) of the main experiment, described here with the other two for convenience. All three are block designs with a fixation task and were run from precomputed, participant-independent trial sequences, so every participant saw identical stimulation. The eyetracking session was acquired at TR 1.9 s rather than 1.5 s; the preprocessed timeseries the localizer GLMs use are upsampled to TR 1 s in either case (Preprocessing).

Category localizer (fLoc)

The category localizer (task-floc, ses-4BarScreenfLoc) is the Stanford fLoc paradigm (Stigliani et al., 2015; stimuli and code at https://github.com/VPNL/fLoc), run in its two-stimulus-set mode. Each run shows five categories, each drawn from one of two subcategories of the fLoc image database:

Category	Stimulus set 1 (odd runs)	Stimulus set 2 (even runs)
face	adult faces	child faces
body	headless bodies	limbs
place	houses	corridors
object	cars	string instruments
character	pseudowords	numbers

Odd-numbered runs show stimulus set 1 and even-numbered runs stimulus set 2. The image sequence of every run is fixed and identical across participants: four sequences were generated in advance, runs 1-4 present sequences 1-4, and runs 5-8 present the same four sequences again, image for image, in the same order and with the same task probes. The four sequences use disjoint images, so each image appears in exactly one sequence and is seen twice per participant (once in runs 1-4, once in runs 5-8).

Block design. Images are shown for 400 ms followed by a 100 ms blank, 12 images per 6 s block. Each run has 6 blocks per category and 8 baseline blocks (fixation only), 38 blocks in total, in a pseudo-random order in which same-category blocks are occasionally adjacent. With a 9 s fixation period at the start and the end, a run lasts 246 s (164 volumes at the acquired TR of 1.5 s; 246 timepoints after upsampling to 1 s).

Task. One-back repetition detection with 18 repeats per run. A red fixation dot of 0.07 degrees diameter, the same extent as the retinotopy cross, is shown throughout.

Object localizer

The object localizer (task-oloc, ses-31) contrasts intact object images with scrambled versions of the same images, following Malach et al. (1995). The stimuli are 43 colour photographs of isolated everyday objects (a trophy, a vase, a locomotive, …) rendered on a grey grid background, and for each object a grid-scrambled counterpart in which the image tiles are shuffled over the same grid, so that intact and scrambled images share low-level content, colour and extent.

Block design. Images are presented at 750 px, about 6.8 degrees, and are shown for 400 ms followed by a 100 ms blank, 12 images per 6 s block. Each run has 8 object blocks, 8 scrambled blocks and 6 baseline blocks (fixation only), 22 blocks in total, in a pseudo-random order in which same-condition blocks are occasionally adjacent. With 9 s of fixation at the start and the end the design covers 150 s; 80 volumes were acquired at the TR of 1.9 s, i.e. 152 s, giving 152 timepoints after upsampling to 1 s. Every run draws on all 86 images, each shown about twice per run.

Task. One-back repetition detection with 20 repeats per run.

Run structure. Two fixed sequences exist. Runs 1 and 3 present sequence 1 and runs 2 and 4 present sequence 2, image for image and with the same probes, identically for all participants.

Motion localizer

The motion localizer (task-MotionLeft and task-MotionRight, ses-MotionLocBarsSML) is a random-dot kinematogram contrasting moving with static dots, presented to one visual hemifield at a time so that MT and MST can be separated by their different sensitivity to ipsilateral motion (Huk et al., 2002). In a MotionLeft run the fixation dot sits 7.7° right of the screen centre and the dot field is centred 4.2° left of it, i.e. 11.9° into the left hemifield; a MotionRight run mirrors this geometry. The dot field is a disc of 3.5° radius containing 200 grey dots of 0.2°, with a 0.5° dot-free zone around its centre. The stimulus therefore covers 8.4-15.4° of eccentricity along the horizontal meridian.

Block design. Each run alternates 10 moving and 10 static blocks of 12 s. During a moving block the dots move radially at 8°/s, alternating between expansion and contraction every 1 s, with a limited dot lifetime of 10 frames; during a static block the same dot field is shown stationary. With 9 s of fixation at the start and the end a run lasts 258 s (172 volumes at the acquired TR of 1.5 s; 258 timepoints after upsampling to 1 s).

Task. Fixation-colour-change detection: the fixation dot, 0.1 degrees in diameter inside a 0.2 degree surround, changes colour for 300 ms at random intervals of 4-9 s and participants press a button within 1.5 s.

Run structure. Block timing is fixed by the settings and identical across runs, hemifields and participants. Only the fixation-change times are random. Each participant has two runs per hemifield, four in total.

Behavioural data

In every retinotopy, category-localizer and motion-localizer run the presentation software logged the onset of each target or one-back probe together with the button presses and their reaction times, so hit rate and response latency can be derived per run. The object localizer was presented with the main experiment’s software in the eyetracking session; its behavioural records are covered by the Behavioral Data section of the main experiment above.

References

• Dumoulin SO and Wandell BA (2008) Population receptive field estimates in human visual cortex. NeuroImage 39:647-660. doi:10.1016/j.neuroimage.2007.09.034

• Finzi D, Gomez J, Nordt M, Rezai AA, Poltoratski S and Grill-Spector K (2021) Differential spatial computations in ventral and lateral face-selective regions are scaffolded by structural connections. Nature Communications 12:2278. doi:10.1038/s41467-021-22524-2

• Huk AC, Dougherty RF and Heeger DJ (2002) Retinotopy and functional subdivision of human areas MT and MST. Journal of Neuroscience 22:7195-7205. doi:10.1523/JNEUROSCI.22-16-07195.2002

• Malach R, Reppas JB, Benson RR, Kwong KK, Jiang H, Kennedy WA, Ledden PJ, Brady TJ, Rosen BR and Tootell RBH (1995) Object-related activity revealed by functional magnetic resonance imaging in human occipital cortex. PNAS 92:8135-8139. doi:10.1073/pnas.92.18.8135

• Stigliani A, Weiner KS and Grill-Spector K (2015) Temporal processing capacity in high-level visual cortex is domain specific. Journal of Neuroscience 35:12412-12424. doi:10.1523/JNEUROSCI.4822-14.2015