Face anti-spoofing in your browser

Is the face in front of the camera a real person, or a photo, a screen replay or a mask? A ResNet-18 presentation-attack detector trained on CelebA-Spoof, quantised to INT8 (11 MB) and run with onnxruntime-web next to the YuNet face detector. Camera frames and photos are processed on your device; nothing is uploaded or stored.

Learning project, not a security product: on the held-out test photos about one attack in four still passes as live.

Try your own face, then a photo of yourself on a phone screen.

What the model sees: a square 1.6x the face box, resized to 128 px.

 

P(attack)

Loading the runtime…