FaceX 03 / 25
Passive anti-spoof (liveness) in the browser
Passive liveness check that flags photos and screens held up to the camera — without extra user gestures.
Problem
Face login and attendance systems are easy to fool with a printed photo or a phone screen.
What it does
- Classifies live face vs spoof on single frames
- Runs next to detection/recognition in the same WASM pipeline
How it works
- Face crop → two MiniFASNet models (2 × 1.7 MB, Apache 2.0, MinivisionAI Silent-Face)
- Scores combined into a liveness decision in the browser
Stack & models
WebAssemblyMiniFASNet
Numbers
- 2 × 1.7 MB anti-spoof models
Figures come from project docs/README; no other benchmarks are claimed.