Attendance recognition: three models, three distinct jobs
YuNet, MiniFASNetV2 and SFace separate face detection, liveness evidence and identity matching.
Locate the face before matching identity
Vispo Care uses publicly released models in its attendance image pipeline. YuNet locates faces. SFace aligns and crops the face region and extracts normalized features for identity matching. Detecting a face and deciding whose face it is are separate decisions.
This is face recognition, not text OCR. The models supply recognition components; Vispo integrates them with enrollment, identification and attendance workflows.
Liveness is a separate signal
MiniFASNetV2 analyzes a crop that includes context around the face and supplies liveness evidence. Its signal is distinct from identity similarity and detection confidence; those values should not be blended into an undefined accuracy percentage.
A liveness model is not an unconditional anti-spoofing guarantee. Lighting, pose and image quality can affect results, and uncertain outcomes need explicit handling.
Model versions and integration engineering
Care's model package records file versions and hashes to make its recognition components traceable. This describes the code and model integration; execution location and settings can vary by deployment.
Credit for the released models belongs to their authors. Naming their roles and sources distinguishes foundation components from Vispo's product engineering.