Your visual system can average a whole crowd of objects almost instantly—but those averages are not independent. Put a larger crowd beside one group, and the other group can look smaller. The human visual system simultaneously computes summary statistics of object features, such as average size, across groups of objects.

Evidence suggests that summary statistics across groups of objects, or ensembles, are not independent but influence one another. The key question was whether one ensemble’s average size affects another ensemble’s perceived average size, and whether that change appears in visual-cortex activation.

The reported pattern was contrast-like: large distractor ensembles decreased target ensembles’ perceived average sizes, while smaller distractor ensembles increased them. The human visual system rapidly forms ensemble representations, efficiently summarising or averaging features across groups, including size, distance, and orientation.

Size averaging enables observers to rapidly, in less than fifty milliseconds, and accurately estimate the average size of a stimulus set, even without reporting individual items. Observers can also accurately report a group’s average size when the visibility of some objects is impaired by object-substitution masking.

These findings support the idea that individual objects and ensemble representations are represented differently in the visual system. Information about ensemble representations for different object groups is available relatively early and simultaneously across groups. What remained unclear was whether different summary statistics were independent descriptors, integrated at some processing level, and perceptually relevant when they interacted.

Earlier stimuli that induced the Ebbinghaus illusion altered the perceived average size of a set. Large Ebbinghaus inducers decreased perceived average size, whereas small inducers increased it, indicating contrast-driven modulation. This pattern is consistent with one object group acting as context and providing benchmarks for size judgments.

The contrast effect may arise from local feature interactions within individual objects or from higher-level operations over summary statistics. In the earlier experiment, Ebbinghaus inducers were homogeneous within each trial, either large or small, so an inducer’s size matched the group’s average size.

That confound made it difficult to determine whether contrast effects came from overall summary size statistics or from individual inducer features. The present experiment therefore used heterogeneous stimuli with variable sizes while keeping their average size constant, to disentangle individual object size from group average size.

Behavioural measures and fMRI tested whether one ensemble’s perceived average size influences another ensemble’s perceived average size in a contrast-like manner. Participants saw two colour-defined ensembles and reported the perceived average size of the target ensemble.

Target and distractor ensembles appeared in different visual-field quadrants, allowing their neural signatures to be separated in early retinotopic visual cortex. The design tested whether neural patterns would correspond with behavioural size-contrast results.

The stimuli consisted of two uniformly coloured ensembles: eighteen green circles and eighteen red circles. One colour-defined ensemble was the target, while the other was the task-irrelevant distractor. For fifteen participants, green was the target; for fourteen participants, red was the target, because full colour counterbalancing was not feasible.

Both ensembles had either a small average size of zero point seven degrees or a large average size of one point three degrees, producing four experimental conditions. Negative and positive size contrast referred to target averages smaller or larger than the distractor average, while matching conditions had equal average sizes.

The figure’s middle stimulus-display column shows the trial sequence most clearly: fourteen green target circles appear in one quadrant, while six red distractors occupy the diagonal quadrant. The later columns show the comparison circle, response options, and a brief noise mask, making clear that participants judged the target ensemble rather than the distractors.

The method of constant stimuli was used to detect the perceived average size of the target ensemble. Comparison-circle size varied around the target average in zero point one degree increments, with separate lists for small and large averages.

Ten comparison sizes were used, half smaller and half larger than the target ensemble’s average size. Each trial began with a one-thousand-millisecond fixation period, followed by target and distractor ensembles appearing in their quadrants for thirty-two milliseconds.

After another one-thousand-millisecond fixation period, a comparison circle appeared centrally for one thousand three hundred milliseconds; no response in that period counted as incorrect. In the fMRI session, one of five fixation durations was randomly assigned to introduce temporal jitter.

Participants completed four hundred eighty trials, lasting approximately thirty-seven minutes. Psychometric curves quantified the likelihood of judging the comparison circle as larger than the target circle. A logistic function modelled that probability, and the Point of Subjective Equality was the comparison size judged equal to the target at a probability of one half.

A two-by-two repeated-measures ANOVA tested target and distractor average-size effects on these points in the pre-test and fMRI session. The fMRI data were analysed using statistical parametric mapping software. Functional images were realigned to correct for inter-scan movement, normalised to the standard MNI single-subject template, and smoothed with a two-millimetre Gaussian kernel.

To investigate mutual size contrast, the neural signals from target and distractor ensembles first had to be disentangled. Subject-specific functional regions of interest were defined from localizer scans for voxels representing each of four visual-field quadrants.

This made it possible to examine the target ensemble in isolation and the influence of a distractor in the diagonally opposite quadrant. The quadrant-specific regions were later subdivided into visual areas V1, V2, and V3. The analysis compared the number of suprathreshold voxels for an ensemble of interest as a function of the context ensemble’s average size.

A context ensemble either matched the ensemble of interest, creating a size-match condition, or differed from it, creating a size-contrast condition. The results demonstrated a robust size-contrast effect. The target ensemble’s perceived average size was systematically modulated by the surrounding distractor ensemble, even though that ensemble was task-irrelevant.

This finding supports the hypothesis that a distractor ensemble’s average size influences the target ensemble’s perceived average size. Figure four shows that perceived average size, measured by the point of subjective equality, changes across combinations of target and distractor ensembles; asterisks mark significant paired comparisons.

Panels B and C show the same pattern in psychometric curves: solid lines are size-match conditions, while dashed lines represent negative and positive size contrast, with vertical lines marking the corresponding PSEs. Together, these results establish the pre-test size-contrast effect that the fMRI-session findings later replicated.

The fMRI-session behavioural results replicated the pre-test: the distractor ensemble modulated the target ensemble’s perceived average size. The perceived average was smaller for the small target condition, with a mean of zero point fifty-seven, than for the large target condition, with a mean of one point twenty-three.

Perceived average size was larger with a small distractor, mean zero point ninety-three, than with a large distractor, mean zero point eighty-seven. The ANOVA confirmed significant main effects of target average size and distractor average size.

The interaction between target and distractor average size was not significant. In panel A, the paired subject points and connecting lines show the same within-subject shift in both target-size groups, while the asterisks mark significant contrasts.

In panels B and C, the dashed and solid curves are horizontally separated within each color, with the midpoint guides making those PSE shifts visible at both single-subject and group levels. The analysis compared size-contrast conditions with size-match conditions to test how a context ensemble influenced the neural representation of an ensemble of interest.

Figure seven displayed percentage differences in activated voxels across V1, V2, and V3, averaged across visual-field quadrants. The contrasts were calculated separately for large and small ensembles of interest, and separately for target and distractor ensembles. In panels A and B, the bars for V1, V2, and V3 cluster near the baseline, with individual participants extending to roughly six percent.

Panels C and D show a similar near-baseline pattern, although panel C includes points reaching close to ten percent; the colored bars and error bars summarize the corresponding within-subject contrasts. The behavioural design could not directly measure perceived size in the task-irrelevant distractor, because participants judged only the target.

Even so, neural data suggested that distractor objects may also undergo size-contrast effects: retinotopic activation representing the distractor was modulated in a contrast-like manner. The distractor showed a significant negative size-contrast effect and a corresponding, although non-significant, positive pattern paralleling the target pattern.

The study provides strong evidence for mutual size contrast in ensemble representations, with each group acting as context for the other. Behaviourally, the target ensemble looked larger beside a smaller distractor and smaller beside a larger distractor.

The pattern resembles classical Ebbinghaus size contrast but shows that the phenomenon exists at the level of ensemble representations. Susceptibility to the Ebbinghaus illusion did not significantly correlate with the ensemble size-contrast effect. This suggests that size contrast for ensemble summary statistics and individual objects may involve distinct mechanisms despite similar perceptual outcomes.

The results revealed consistent underestimation of target-set average sizes across all conditions. That underestimation may reflect a general perceptual bias caused by the mere presence of a distractor ensemble, independent of its average size.

The design did not include a target-only control condition that could provide a direct baseline for behavioural judgments and neural responses to the target alone. Average size is not computed in isolation: two object groups mutually bias one another, with behavioural shifts echoed in early visual activation.

The result places size contrast at the level of ensemble summaries.