One-second smoothed peak 8.8. Independent detection accuracy unknown.
OBSERVED LICENSED SCENE
Crossing review.
Evidence and limits.
72-second licensed excerpt, 360 processed frames at target 5fps. Original analysis retained; this public report exposes aggregates only.
Independent crossing accuracy unknown.
Lost/end closures do not establish geometric exits.
Track IDs do not establish unique people.
Review interval 40–50s
50 sampled frames, mean 6.52 tracked detections, raw sampled peak 10. Highest observed mean among seven complete fixed 10-second bins; earliest tie wins. Recommendation: manually annotate detections and crossings before a staffing or capacity decision. Operational effect unvalidated.
Observed cost
Wall 25.87s; CPU 31.91s; peak memory 648.8MB. Monetary cost UNKNOWN. One local DirectML run does not establish production cost or throughput.
Forecast remains gated
No promoted forecast. Current gate requires at least 600s holdout and 120 origins per cell with uncertainty against persistence. This clip is insufficient.
Reproducible provenance
Analyzed source SHA256: 3cb900f0f351f371e6ace7ff71b51d30566badfb8310adafd8bd9475854d7b9d.
Aggregate JSON · Source and model versions · Proposed evaluation, privacy and support scope
ProtoplasmaKid / Wikimedia Commons / CC-BY-SA4.0 — original source / license. Display derivative: 72-second excerpt, resize, global blur, audio removal, re-encoding. Derivative/poster CC BY-SA 4.0; no endorsement.