Machine perception · Research prototype
AEyes
Photonic eyes.
What if a machine could recognize an object from its visual signature instead of asking a giant model to guess?
SYSTEM / 01
What it is.
InputImages transformed into photon and retinal-inspired measurements
MethodDeterministic features, signatures, geometry, and matching
OutputAn identity verdict with an inspectable evidence trail
AmbitionUseful visual intelligence without a large generative model in the loop
WHAT EXISTS
Evidence.
- Implemented photon capture, retinal transform, physical photoreceptor, pattern engine, and graph modules.
- A July 2026 evaluation recorded 80% raw accuracy across a small 15-fixture set and 63% on the stricter rubric.
- The physical vision stage uses deterministic codec and retinal extractors with a Naka–Rushton response.
WHAT REMAINS
Limits.
- The current prototype is not human-grade and does not claim to be.
- Earlier 95.1% figures measured near-duplicate index recall, not general object recognition.
- Hard negatives, shape, geometry, context, and end-to-end physical vision still need deeper validation.