Developmental changes in visual clutter in infants' and toddlers' visual worlds

Talk 1, 2:30 pm, 34.11

Minju Kim¹ (minjukim023@gmail.com), Linda B. Smith¹; ¹Indiana University Bloomington

The everyday visual world is messy, noisy, and cluttered. Clutter is well-known to limit information extraction from a scene. Nonetheless, during the first two years of life, infants rapidly acquire visual categories of faces, objects, and scenes. Understanding the properties of clutter in the everyday environment where visual development occurs is essential to a comprehensive understanding of visual learning and category formation. Here, we quantify visual clutter in infants' daily experiences and its temporal structure. We analyzed 500,000 egocentric images from 40 infants across five age groups (1-3, 6-8, 10-12, 15-20, 22-26 months). Recordings captured naturalistic indoor and outdoor contexts, providing a dense sample of the scenes in front of infants’ faces. We used four converging measures: Feature Congestion as defined by Rosenholtz, the number of visual segments, semantic category diversity, and foregrounding. Distributions were highly skewed; therefore, we discretized scores into seven ordered categories based on percentiles ranging from Extremely Low to Extremely High. Temporal structure was examined by analyzing bouts of consecutive frames within the same clutter category. We observed evidence of a systematic restructuring of clutter, with convergent patterns across all four measures. For example, the Feature Congestion measure revealed that the youngest infants experienced the highest frequency of sustained, extremely low-clutter periods. In contrast, these long "pockets of simplicity" declined dramatically across development, becoming rare in the oldest age group. Concurrently, high-clutter exposures increased in both frequency and duration. Critically, this developmental trajectory—from sustained periods of uncluttered input to more frequent high-clutter input—was observed across measures, suggesting a systematic reorganization of the visual environment. The nature of the attentional tasks young perceivers need to solve changes with development. We will discuss the roles of infant visual development, early “sticky” attention, and foregrounded entities (faces and objects) as contributors to these developmental patterns.

Do you see what I see? Investigating the development of visual diet in children and teens using mobile eye-tracking in natural environments

Talk 2, 2:45 pm, 34.12

Elizabeth Jiwon Im^1,2 (ejim@stanford.edu), Kalanit Grill-Spector^1,2; ¹Stanford University, ²Stanford Wu Tsai Neuroscience Institute

Face recognition and word reading abilities improve during childhood and adolescence, in correlation with increases in selectivity for faces and words in the ventral temporal cortex. At the same time, selectivity for hands decreases. We hypothesize that these developments may be linked to changes in children’s visual diet across development. However, to date there has been no investigation of children’s and teenagers’ naturalistic visual diet, and how it develops remains unknown. For the first time, we investigate children’s (4-6yrs) and teenagers’ (14-17yrs) visual diet in natural environments (home: n=10; classroom: n=6) using mobile eye-tracking glasses (FOV=103°×77°). We introduce a novel validation procedure for natural viewing, estimating a measurement accuracy of 1.01°±1.69° in children (n=10) and 1.10°±1.52° in adults (n=10). We then leverage computer vision models to analyze egocentric scenes and gaze behavior. Our data show that the prevalence of faces and hands is similar between children and teenagers. However, fixation behaviors are different. While both age groups foveate more on faces than hands, there is a significantly higher proportion of faces in the central 5° of teenagers’ visual diet than that of children’s (14.9%±7 in teenagers vs 6.1%±2.3 in children, p=0.032, permutation test). Moreover, faces more often appear in the upper visual field for children (median y-position: 19.4°±5.2), while there is no vertical bias for faces in teenagers. In contrast, hands trend towards being in the lower visual field for teenagers (median y-position:﹣9.27°±10.89), while there is no vertical bias for hands in children. Together, our findings suggest that, despite shared environments, children and teenagers fixate and interact with the world differently, with visual field biases continuing to develop from early childhood into adolescence.

This work was supported with funds from NIH R01 EY 023915 awarded to KGS

Compromised Gestalt grouping in late-sighted children

Talk 3, 3:00 pm, 34.13

Lukas Vogelsang¹, Marin Vogelsang¹, Priti Gupta², Pragya Shah², Jeonghwan Cheon³, Purva Sethi², Stutee Narang², Suma Ganesh⁴, Pawan Sinha¹; ¹Massachusetts Institute of Technology, ²Project Prakash, Dr. Shroff’s Charity Eye Hospital, ³Independent Researcher, ⁴Pediatric Ophthalmology, Dr Shroff’s Charity Eye Hospital

Holistic perceptual organization and Gestalt-like grouping are critical for making sense of the complex visual environment. Past research has investigated the developmental timelines for acquiring these foundational capabilities, highlighting their importance in normal visual maturation. Here, we examined the consequences of curtailing early visual experience on perceptual grouping skills later in life. Specifically, we studied a unique population of visual observers – children from rural India who were born blind but gained sight through surgical intervention late in childhood. We adapted a paradigm previously used with normally-sighted infants. Participants viewed carefully designed square arrays of elements arranged in a uniform grid, where grouping was induced through similarity in luminance or figural form along alternating rows or columns, and reported whether they perceived vertical or horizontal organization. Longitudinal assessments of 16 patients from pre-operative to many months post-operative status revealed persistently poor performance. Despite notable gains in other visual abilities, participants exhibited no improvement over time, remaining substantially below matched controls. This deficit was consistent across all grouping cues. Complementary computational simulations suggest a candidate account for these findings: the deficits may arise from commencing visual experience with higher-fidelity input than is available to neonates. Deep networks trained on developmentally inspired degraded-to-undegraded input trajectories formed representations that more strongly encoded holistic organization in Gestalt stimuli. In contrast, networks trained without this developmental progression, akin to the experience of the late-sighted, failed to develop such organization. Together, these results provide evidence that early-onset blindness yields lasting impairments in Gestalt-like grouping, and suggest that not only the timing but also the degraded quality of early visual experience may play an important role in the emergence of holistic organization. Besides helping formulate clinical prognoses for late sight treatments, these findings have implications for understanding typical development and the design of developmentally-aligned computational training procedures.

NIH R01EY020517

Early Deprivation Selectively Impairs Cardinal Orientation Sensitivity and Reduces the Oblique Effect

Talk 4, 3:15 pm, 34.14

Md. Salman Sarkar¹ (optomsarkar@gmail.com), Bat-Sheva Hadad¹, Subiksha Saravanan², Shashikant Shetty², Amit Yashar¹; ¹University of Haifa, Israel, ²Aravind Eye Hospital, India

Background: Visual perception varies as a function of environmental regularities, demonstrating greater sensitivity to prevalent stimuli. In orientation processing, the visual system exhibits a well-known asymmetry in orientation sensitivity known as the oblique effect—greater perceptual sensitivity to cardinal (horizontal and vertical) orientations than to oblique ones. However, whether and how the visual system learns basic environmental regularities during early development—and how monocular or complete deprivation affects this process—remains unknown. Objective: We address this question by examining orientation sensitivity in individuals with a history of early visual deprivation due to dense cataract, either in both eyes (bilateral) or one eye (unilateral), and either from birth (congenital cataract) or acquired later in development (developmental cataract). Method: Participants aged 6–34 years (congenital-bilateral: n = 43; congenital-unilateral: n = 15; developmental-unilateral: n = 46; typically sighted: n = 61) completed a two-interval forced-choice (2IFC) orientation discrimination task across cardinal (0° and 90°) and oblique (45°) orientations. Thresholds were measured using a staircase procedure. Visual acuity and age were controlled for in the analysis. Results: Participants with early visual deprivation, either in both eyes or in one eye, showed a reduced standardized oblique effect, driven by poorer discrimination of cardinal orientations. The results remained consistent even after controlling for participants’ age and visual acuity as covariates. Finally, participants’ age showed a positive correlation specifically with cardinal orientation sensitivity in the typically sighted group. Conclusions: The results reveal that the visual system learns basic environmental regularities during development. Visual deprivation, in one or both eyes, is detrimental to the learning of these regularities. These findings suggest that, from birth and throughout development, the visual system adjusts its sensitivity based on the statistics of the environment.

Israel Science Foundation

The role of vision in infants’ multisensory localization and body remapping

Talk 5, 3:30 pm, 34.15

Monica Gori^1,2, Helene Vitale², Andrew Bremner³, Caudio Campus²; ¹IHMC, USA and IIT, Italy, ²IIT, Italy, ³University of Birmingham, UK

Background. The mechanisms underlying the development of multisensory integration in sensory deprivation remain unclear. In recent years, our work has highlighted the importance of sensory interactions in scaffolding multisensory development. We have shown that vision plays a critical role in shaping both auditory and tactile integration. In the absence of vision, auditory spatial representations and body representations are altered, making audio–tactile integration more challenging for blind individuals compared to sighted peers. Methods. We examined behavioral responses and event-related potentials (ERPs) in 12 sighted (S) and 12 severely visually impaired (SVI) infants. Participants were presented with unisensory (auditory, tactile) and multisensory (congruent, incongruent) stimuli delivered to their hands in both uncrossed and crossed positions. Neural activity was recorded using a high-density 128-channel EGI system. Results and Discussion. Together, the unisensory results suggest that SVI infants remain anchored to body-centered coordinates in early development, whereas sighted infants shift toward an external reference frame, revealing a two-step tactile processing mechanism. Indeed, SVI infants showed heightened contralateral somatosensory responses in an early time window, which correlated exclusively with their behavioral reactions and showed no evidence of tactile remapping. In contrast, sighted infants exhibited a later ipsilateral somatosensory response aligned with their behavioral performance. This highlights vision’s specific contribution to developing somatosensory remapping from early life. Regarding multisensory processing, congruent stimuli elicited activation of integrative centro-parietal (CP) and fronto-central (FC) regions within 180–220 ms in both groups, indicating preserved multisensory integration. However, in an earlier window (105–120 ms), sighted infants showed greater temporo-parietal activation (associated with auditory processing), whereas SVI infants showed stronger centro-parietal activation (associated with somatosensory processing). Consistent with our previous behavioral work (Gori et al., Curr.Biol. 2021), these results support the hypothesis that visual deprivation shapes early multisensory spatial representations.

Acknowledgement: This work is funded by EU H2020, ERC StG' MySpace (No. 948349).

Eccentricity Bias in the Infant Homolog of Adult Scene-Selective Cortex

Talk 6, 3:45 pm, 34.16

Sarah S. Tung¹ (sstung@stanford.edu), Emily M. Chen¹, Jason D. Yeatman², Cameron T. Ellis¹; ¹Department of Psychology, Stanford University, Stanford, CA, USA., ²Graduate School of Education, Stanford University, Stanford, CA, USA.

A central question in vision science is how human visual cortex becomes organized into distinct regions that selectively respond to content such as faces, scenes, and words. One hypothesis proposes that selectivity is scaffolded from ‘protomaps’—biases for features, like viewing location and spatial frequency, that covary with stimulus category. By hypothesis, protomaps are present at birth and shaped by experience, though their initial feature biases may persist. Indeed, in adults, category-selective regions are biased towards parts of the visual field where those categories typically appear. For example, scene-selective regions are more sensitive to peripherally presented stimuli than foveal stimuli. To test the protomap hypothesis, we conducted fMRI with awake behaving infants (age: 2-6 months, N=7) and collected independent measurements of category selectivity and eccentricity biases. Infants viewed blocks of scenes and words presented either foveally (6°) or peripherally (40°) in the visual field. We found evidence of eccentricity organization in early visual cortex in both infants and a validation sample of adults (N=7). In scene-selective visual cortex (anatomically defined in the collateral sulcus), we find scene-selective responses in adults but not in infants. Yet, consistent with the protomap hypothesis, infants (and adults) did show a significant eccentricity bias, responding more strongly to peripheral than foveal stimuli. Even voxels with the strongest evidence of scene-selectivity showed a significant eccentricity bias, suggesting that category selectivity and eccentricity are linked in infancy. Altogether, these results are consistent with the hypothesis that eccentricity serves as a scaffold for category-selectivity to emerge.

A multivariate approach to understanding the emergence of the visual word form area

Talk 7, 4:00 pm, 34.17

Kelly J. Hiersche¹ (hiersche.1@buckeyemail.osu.edu), Lauren A. Rydel¹, Zeynep M. Saygin¹; ¹The Ohio State University

The visual word form area (VWFA) is unique among ventral visual areas in that it shows protracted development, emerging only after a child can read. Prior work demonstrated connectivity to high-level language regions informs the eventual location of the VWFA, but the functional profile of the proto-VWFA is still unclear. Univariate word selectivity may emerge from initial multivariate representations that differentiate between visual words and other visual categories (e.g., faces, objects) or between linguistic vs. non-linguistic auditory stimuli. We scanned 111 children ages 3-12 years (24 longitudinal, including prereaders who transitioned into readers) on two runs of a high-level visual fMRI task and a subset on two runs of an auditory language task. We used functional regions of interest, multivariate representation analyses, and representational similarity analyses to understand the functional profile of emerging VWFA. We find that while pre-readers lack univariate word responses, they do show multivariate representations for written words. Representational similarity analyses showed that activation to words is dissimilar from other stimulus categories in the pre-VWFA and that this dissimilarity is almost identical to that of the VWFA in readers. The proto-VWFA in prereaders did not show multivariate (or univariate) representations for auditory language, much like results in readers. Finally, our longitudinal data supports these results, showing the proto-VWFA holds a multivariate representation for words, but not for auditory language, before the univariate VWFA emerges. Overall, our results suggest that multivariate representations may scaffold univariate selectivity of the VWFA as a child learns to read; interestingly, these do not seem to emerge from auditory representations to language, suggesting that any responses of this region to auditory language in older children or adults are likely a result of experience and challenge the theory that the proto-VWFA is part of the distributed language network.

Alfred P. Sloan FG-2018-10994, OSU College of Arts and Sciences, Women in Philanthropy Award, Chronic Brain Injury Pilot Award, R01 HD110401, (all to ZMS), NSF-GRFP DGE-1343012 (to K.J.H.)