To researchers’ surprise, deep learning vision algorithms often fail at classifying images because they mostly take cues from textures, not shapes.

The post Where We See Shapes, AI Sees Textures first appeared on Quanta Magazine