Lab / computer-vision-attendance
Computer Vision Attendance
Face-embedding based attendance prototype where every bit of processing stays on-device — no biometric data leaves the machine.
What it tests
Roll-call attendance in a 40-student class burns teaching minutes. Cloud face recognition fixes that but ships biometric data to someone else's servers.
How it works
Enrolment captures a few faces per student and stores embeddings locally; a marked session compares live camera frames against those embeddings with a cosine threshold, all on-device.
The output is a simple CSV of present/absent with timestamps — the only record kept is attendance, never raw photos.
What came out of it
A working privacy-first prototype and a sharper instinct for where on-device ML beats cloud ML — and honest awareness of its limits (lighting, angle, twins of a classroom).