< Back to all clusters
[TECHNOLOGY] · United States · 43 sources

started · updated

NYU researchers find AI struggles to recognize shapes and silhouettes

A research team led by Biyu He at the New York University Grossman School of Medicine has identified significant weaknesses in artificial intelligence regarding shape and silhouette recognition. According to a study published in the journal ‘iScience’, modern deep neural networks struggle to identify objects when presented as simple silhouettes or outlines, a task that humans perform with high reliability.

The study compared human visual performance against more than 200 different AI models. Researchers systematically manipulated 240 images across 48 categories, including black silhouettes without internal structure and patterns like fur textures without clear contours. The findings suggest that while the human brain prioritizes overall shape, AI algorithms tend to focus on local details, surface patterns, and textures. Consequently, minor distortions or the absence of specific details can cause these systems to fail in identifying the intended object.

Entities

Biyu He · Mugihiko Kato · New York University Grossman School of Medicine · iScience

Claims

What the coverage asserts, and how many sources carry each claim.

Sources