RETINA puts a neural network inside each camera pixel for on-sensor vision

garrytan · x · 2026-07-29

Garry Tan reposts SynapseSemi’s RETINA, a vision sensor that runs a neural network inside the camera pixel itself.

The pitch is simple: move inference to the point where light lands, which could enable faster and more efficient vision processing than sending raw pixels to a separate model later.

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

Original post →

More from Embodied

Embodied channel →