Detection models are quantized with Kenosis and verified against COCO ground truth. The verification measures detection accuracy itself, not a proxy metric.
Every camera runs in its own isolated thread. At real-time tiers the pipeline drops zero frames, and telemetry counter-verifies it.
The pipeline is native Rust and memory-safe by construction, with no Python layer in the serving path.
The same build runs on CPU alone, and it picks up CUDA and DirectML acceleration when the host has it.
Prometheus and REST monitoring are built in, with per-stage latency histograms and overload alerting. Every performance claim on this page is reproducible from the daemon's own counters.
Measured — dual-camera headless daemon, re-validated 2026-07-13
2× 640×480@30 webcams · PP-YOLOE+ S INT8 @320 · Intel i5-13420H · ONNX Runtime CPU · ≥115 s steady-state windows, zero frame drops