Biology subjects

Morgenstern, Y.

Publications and source records attributed to Morgenstern, Y..

6 recordsLinked to original sources

bioRxiv2025

Human and generative AI integrate visual cues differently in a shape completion task

In amodal completion observers perceive complete objects despite partial occlusion. When two object parts are divided by an occluder, completion can result in perceiving one or two objects. This phenomenon involves both lower-level cues (e.g., symmetry, contour continuity) and higher-level cues (e.g., prior knowledge). Experiment 1 investigates how occluder size, familiarity, and symmetry affect human completions using a drawing task. Narrow occluders and asymmetry promote single-shape completions, while familiarity and (global) symmetry promote two-shape interpretations. Good continuation emerges as the strongest cue, with symmetry and familiarity playing increasingly important roles as occluder width increases. Experiment 2 compares human performance with three state-of-the-art generative AI models. Models often generated creative but non-compliant outputs, altering even unoccluded regions. We restricted analysis to instruction-following generations, identified through ratings by naive observers. Among compliant outputs, models showed some human-like biases (e.g., more two-shape completions for wide occluders), but failed with higher-level cues. They did not use symmetry to guide completions and showed reversed familiarity effects. Our findings highlight differences between human and AI completions. Humans integrate low- and high-level cues, whereas compliant outputs from the AI models rely primarily on low-level pattern continuation. Current AI models lack the flexible integration of multiple representational levels that characterize human perception. This work establishes an analytical framework for evaluating whether next-generation models achieve more human-like visual reasoning. HighlightsO_LIHumans see one or two objects behind occluders using geometric and semantic cues. C_LIO_LIHuman drawings and generative AI completions show how different cues modulate perception C_LIO_LIGood continuation dominates human and AI shape completions C_LIO_LISymmetry and familiarity guide humans, but not instruction-following AI outputs C_LIO_LIHumans integrate multiple levels; instruction-following AI engages lower-level-processing. C_LI

neuroscience↗

bioRxiv2024

Human shape perception spontaneously discovers the biological origin of novel, but natural, stimuli

Humans excel at categorizing objects by shape. This facility involves identifying shape features that objects have in common with other members of their class and relies--at least in part--on semantic/cognitive constructs. For example, plants sprout branches, fish grow fins, shoes are molded to our feet. Can humans parse shapes according to the processes that give shapes their key characteristics, even when such processes are hidden? To answer this, we investigated how humans perceive the shape of cells from the olfactory system of Xenopus laevis tadpoles. These objects are novel to most humans yet occur in nature and cluster into classes following their underlying biological function. We reconstructed 3D cell models through 3D-microscopy and photogrammetry, then conducted psychophysical experiments. Human participants performed two tasks: they arranged 3D-printed cell models by similarity and rated them along eight visual dimensions. Participants were highly consistent in their arrangements and ratings and spontaneously grouped stimuli to reflect the cell classes, unwittingly revealing the underlying processes shaping these forms. Our findings thus demonstrate that human perceptual organization mechanisms spontaneously parse the biological systematicities of never-before-seen, natural shapes. Integrating such human perceptual strategies into automated systems may enhance morphology-based analysis in biology and medicine.

neuroscience↗

bioRxiv2023

Contrast Sensitivity Function in Deep Networks

The contrast sensitivity function (CSF) is a fundamental signature of the visual system that has been measured extensively in several species. It is defined by the visibility threshold for sinusoidal gratings at all spatial fre-quencies. Here, we investigated the CSF in deep neural networks using the same 2AFC contrast detection paradigm as in human psychophysics. We examined 240 networks pretrained on several tasks. To obtain their corre-sponding CSFs, we trained a linear classifier on top of the extracted features from frozen pretrained networks. The linear classifier is exclusively trained on a contrast discrimination task with natural images. It has to find which of the two input images has higher contrast. The networks CSF is measured by detecting which one of two images contains a sinusoidal grating of varying orientation and spatial frequency. Our results demonstrate char-acteristics of the human CSF are manifested in deep networks both in the luminance channel (a band-limited inverted U-shaped function) and in the chromatic channels (two low-pass functions of similar properties). The exact shape of the networks CSF appears to be task-dependent. The human CSF is better captured by networks trained on low-level visual tasks such as image-denoising or autoencoding. However, human-like CSF also emerges in mid- and high-level tasks such as edge detection and object recognition. Our analysis shows that human-like CSF appears in all architectures but at different depths of processing, some at early layers, while others in intermediate and final layers. Overall, these results suggest that (i) deep networks model the human CSF faithfully, making them suitable candidates for applications of image quality and compression, (ii) efficient/purposeful processing of the natural world drives the CSF shape, and (iii) visual representation from all levels of visual hierarchy contribute to the tuning curve of the CSF, in turn implying a function which we intuitively think of as modulated by low-level visual features may arise as a consequence of pooling from a larger set of neurons at all levels of the visual system.

animal behavior and cognition↗

bioRxiv2023

The statistics of natural shapes predict high-level aftereffects in human vision

Shape perception is essential for numerous everyday behaviors from object recognition to grasping and handling objects. Yet how the brain encodes shape remains poorly understood. Here, we probed shape representations using visual aftereffects--perceptual distortions that occur following extended exposure to a stimulus--to resolve a long-standing debate about shape encoding. We implemented contrasting low-level and high-level computational models of neural adaptation, which made precise and distinct predictions about the illusory shape distortions the observers experience following adaptation. Directly pitting the predictions of the two models against one another revealed that the perceptual distortions are driven by high-level shape attributes derived from the statistics of natural shapes. Our findings suggest that the diverse shape attributes thought to underlie shape encoding (e.g., curvature distributions, skeletons, aspect ratio) are the result of a visual system that learns to encode natural shape geometries based on observing many objects.

neuroscience↗

bioRxiv2021

One shot generalization in humans revealed through a drawing task

Humans have the striking ability to learn and generalize new visual concepts from just a single exemplar. We suggest that when presented with a novel object, observers identify its significant features and infer a generative model of its shape, allowing them to mentally synthesize plausible variants. To test this, we showed participants abstract 2D shapes ( Exemplars) and asked them to draw new objects ( Variations) belonging to the same class. We show that this procedure created genuine novel categories. In line with our hypothesis, particular features of each Exemplar were preserved in its Variations and there was striking agreement between participants about which shape features were most distinctive. Also, we show that strategies to create Variations were strongly driven by part structure: new objects typically modified individual parts (e.g., limbs) of the Exemplar, often preserving part order, sometimes altering it. Together, our findings suggest that sophisticated internal generative models are key to how humans analyze and generalize from single exemplars.

neuroscience↗

bioRxiv2020

An image-computable model of human visual shape similarity

Shape is a defining feature of objects. Yet, no image-computable model accurately predicts how similar or different shapes appear to human observers. To address this, we developed a model ( ShapeComp), based on over 100 shape features (e.g., area, compactness, Fourier descriptors). When trained to capture the variance in a database of >25,000 animal silhouettes, ShapeComp predicts human shape similarity judgments almost perfectly (r2>0.99) without fitting any parameters to human data. To test the model, we created carefully selected arrays of complex novel shapes using a Generative Adversarial Network trained on the animal silhouettes, which we presented to observers in a wide range of tasks. Our findings show that human shape perception is inherently multidimensional and optimized for comparing natural shapes. ShapeComp outperforms conventional metrics, and can also be used to generate perceptually uniform stimulus sets, making it a powerful tool for investigating shape and object representations in the human brain.

neuroscience↗