In a groundbreaking study that challenges long-held assumptions about the development of visual processing in the infant brain, researchers have revealed that high-level visual features are already present in the ventral stream by as early as two months of age. This discovery provides compelling evidence that the neural architecture supporting complex visual perception emerges far earlier than previously believed, and undergoes a refinement process rather than a simple hierarchical build-up from rudimentary to advanced processing stages.
The ventral stream — a crucial pathway in the brain responsible for object recognition and categorization — has traditionally been thought to develop through a bottom-up progression. Early visual areas were assumed to initially process simple, low-level features such as edges and colors. Over time, the progressively higher-order areas were believed to integrate these simple signals into more complex representations, facilitating recognition of objects, categories, and even animacy. However, this new research upends that paradigm, revealing a surprisingly rich repertoire of high-level visual encoding present in infants’ brains as young as two months.
Using a sophisticated technique known as representational similarity analysis (RSA), the research team meticulously compared neural activity patterns within various regions of the ventral stream to computational models of visual features — spanning simple perceptual cues to categorical distinctions captured by deep neural networks. This approach enabled them to map out the evolving functional specializations of visual cortical regions over the course of infancy and early childhood, with unprecedented resolution.
Their findings indicate that the early visual cortex (EVC) in infants does not merely encode simple low-level attributes but instead represents a diverse range of features at birth. These features correspond to both basic perceptual qualities and intricate categorical distinctions, indicating a complex amalgam of neural responses right from the outset. Over time, however, these early visual regions become increasingly specialized toward processing low-level visual features, fine-tuning their role to handle foundational aspects of visual input more effectively.
In the ventral visual cortex (VVC), which is situated further along the processing hierarchy and typically associated with higher-order object recognition, a similar richness of feature representation is observed at two months. Indeed, the VVC initially exhibits a bias towards categorical features, such as distinguishing animate from inanimate objects, suggestive of an innate predisposition for these significant conceptual distinctions. As development progresses, these high-level categorical representations become more functionally distinct and refined, likely reflecting the impact of visual experience and neural maturation.
Even the lateral occipital (LO) cortex, regarded as a mid-level visual processing area, displays a notably protracted developmental timeline. Unlike the VVC and EVC, LO does not show early robust feature representation and instead appears to develop object-selective processing capabilities at a slower pace. This finding highlights that the ventral stream’s developmental trajectory is not uniform but instead marked by region-specific timelines, further challenging simplified hierarchical models.
Crucially, the study observes no evidence for the previously hypothesized bottom-up progression from simple to complex feature representations along the ventral stream. Instead, the data reveals an alternative developmental cascade: high-level feature representations are present across the hierarchy from a young age and are subsequently refined and segregated as the brain matures and accumulates visual experience. This non-hierarchical progression suggests that the visual system is primed early on with robust category-level processing, which is then sculpted by developmental and experiential factors.
The decline in the influence of complex features captured by higher layers of deep neural networks on early visual representations further supports the notion of specialized refinement with age. Although deep neural network models offer a valuable analogy for understanding cortical processing, the brain’s developmental dynamics appear to optimize different processing nodes distinctly, decreasing reliance on complex feature inputs in early visual regions while consolidating them in higher-level domains.
By employing supervised deep convolutional neural networks, such as AlexNet, as computational benchmarks, the researchers were able to tease apart the intricacies of visual feature representation in the infant brain. Correlational analyses between brain response similarity patterns and network layer representations exemplify how computational neuroscience can illuminate developmental processes, bridging the gap between machine learning models and human neurobiology.
This study’s implications extend beyond basic science, offering potential insights into developmental disorders affecting visual perception and object recognition. Understanding that complex visual representations emerge early and are honed rather than built anew has critical ramifications for early diagnosis and intervention strategies in conditions such as autism spectrum disorder, where visual processing abnormalities are often observed.
Furthermore, the finding of early categorical organization by animacy in the ventral visual cortex aligns with evolutionary perspectives that prioritize the detection of animate entities for survival. This early bias for animacy categorization highlights the innate foundations upon which experiential learning elaborates, suggesting that certain visual category preferences are hardwired rather than solely learned.
The detailed visualization of these findings is encapsulated in Fig. 7 of the original publication, where scatter plots illustrate regional correlations between neural representational similarity matrices and models of feature complexity. The size and opacity of plotted points effectively convey the strength and reliability of these correlations, emphasizing the dynamic tuning of functional specialization across development.
Altogether, this compelling body of evidence redefines our understanding of the visual system’s ontogeny. Rather than a gradual build-up from simple edge detection to complex object recognition, the infant brain boasts a sophisticated, albeit initially less differentiated, network of feature representations. These early features are then meticulously sculpted into adult-like specialization through experience-dependent plasticity and neural fine-tuning.
As the field continues to explore the developmental trajectories of cortical processing, the integration of cutting-edge computational models with longitudinal neuroimaging holds promise for uncovering the nuanced interplay between innate neural architecture and environmental shaping. This research marks a pivotal step forward, inviting reconsideration of foundational theories about sensory system maturation and underlining the remarkable capabilities of the infant brain.
In sum, the revelation that infants possess rich high-level visual categorizations within the ventral temporal cortex as early as two months of age challenges prevailing dogma and opens new avenues for research into brain development, cognitive neuroscience, and artificial intelligence. The sophisticated experimental design and analytical rigor exhibited by O’Doherty and colleagues illuminate the path for future inquiries into how humans come to perceive and interpret their visual world from the very beginning of life.
Subject of Research: Development of visual feature representations in the infant ventral stream.
Article Title: Infants have rich visual categories in ventrotemporal cortex at 2 months of age.
Article References: O’Doherty, C., Dineen, Á.T., Truzzi, A. et al. Infants have rich visual categories in ventrotemporal cortex at 2 months of age. Nat Neurosci (2026). https://doi.org/10.1038/s41593-025-02187-8
Image Credits: AI Generated
DOI: https://doi.org/10.1038/s41593-025-02187-8
Tags: cognitive neuroscience of infantsearly brain developmenthigh-level encoding in ventral cortexhigh-level visual featuresinfant visual processingneural architecture in infantsobject recognition in infantsrefinement of visual processingrepresentational similarity analysisventral stream developmentvisual categorization in babiesvisual perception in infants
