Duke team embeds neural computation directly inside camera pixels

Scobleizer · x · 2026-07-28

Duke researchers built RETINA, a vision sensor that runs AI inside the camera pixel, pushing GPU-like computation directly to where light lands.

The post frames it as a new kind of camera sensor: instead of capturing data first and processing later, the sensor itself performs neural-network-style work at the pixel level.

Related event: RETINA Sensor Embeds Neural Network Computing Directly Into Camera Pixels(3 posts)→

Original post →

More from Embodied

Embodied channel →