Expand description
Deterministic CPU reference “vision” backend.
Honest scope: not a neural detector. Partitions the image into a 2×2 grid, scores each cell by channel mean and simple horizontal edge energy, and emits up to four class proposals. Used to:
- prove the ABI end-to-end on consumer machines without GPU weights;
- feed epistemic observation quins for rights/audit demos;
- regression-test semantic packing.
Replace with a P64-backed encoder when available — keep VisualModel.
Structs§
Constants§
- CLASS_
HIGH_ EDGE - CLASS_
LOW_ CONTRAST - CLASS_
MOSTLY_ BLUE - CLASS_
MOSTLY_ GREEN - CLASS_
MOSTLY_ RED - Stable class IRIs for the reference taxonomy.