Dhi-Technologies/amodal-counting-benchmark
Updated
Edge vision AI that refuses to guess: calibrated intervals, refusal gates, provenance, falsification ledgers. Synthetic benchmarks + demo.
Note A4 — visibility-corrected counting; calibrated intervals (synthetic).
Note E1 — precision-first cross-camera linking + bounded memory (synthetic).
Note E4 — predict-then-falsify alerting with counterfactuals (synthetic).
Note A3 — fixed camera to metric 3D via self-calibration (synthetic).
Note A5 — radiometric data engine + SSL harness (synthetic).
Note B1 — language-guided vision model factory; pipeline smoke fixtures.
Note Dhi Labs technical posts (thesis, E1, A4).
Explore visibility‑corrected object counting in synthetic scenes
Note Live demo: naive vs visibility-corrected count + calibrated interval.