Yaniv Morgenstern

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

This paper compares how humans and generative AI models integrate visual cues in shape completion tasks, finding that humans use both low- and high-level cues while AI models rely primarily on low-level pattern continuation.

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

The paper shows humans can categorize novel natural shapes based on their biological origins without explicit knowledge of the underlying processes.

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

This study uses visual aftereffects to determine that high-level shape attributes, derived from natural shape statistics, drive perceptual distortions in human vision.

One shot generalization in humans revealed through a drawing task

The paper shows humans can generate novel variations of abstract shapes from a single exemplar by preserving key features and modifying parts while maintaining structure.

An image-computable model of human visual shape similarity

The paper develops a model called ShapeComp that predicts human shape similarity judgments with high accuracy using shape features from animal silhouettes.