Facial recognition · isolated A/B lab
R&D only · no production integrations · no attendance submission · all test templates stay in page memory.
Open the separate A/B/C evaluation page · SFace is evaluation-only.
Synthetic comparison
Artificial 128-dimensional vectors test logic only. These results are not facial-recognition accuracy measurements.
Not run.
Consenting-participant camera experiment
Camera is off. No biometric test has run.
Enrollment uses one immutable frame for photograph and descriptor. The same validated reference is supplied to A and B to isolate recognition differences. No manual identity selection occurs during scanning.
No camera test identities enrolled.
System B research parameters
Illustrative, uncalibrated experiment values. Acceptance threshold remains 0.48; no production settings are changed.
Pose validation measures 2D eye-line roll only. Yaw/pitch and liveness are not inferred.
Live observations
Inference, quality work and confirmation span are separate. Accuracy/unknown rejection cannot be measured without labeled trials. A+B runs share one sequential workload and a 900 ms post-processing delay; they do not measure standalone throughput.
A queue: — · B queue: —
No live measurements.
Synthetic tests are available without models or camera permission.