D

Dinov2 Liveness Detection V2.2.3

Developed by nguyenkhoa
A liveness detection model based on the DINOv2 architecture, designed to distinguish real faces from spoofing attacks, achieving 99.32% accuracy on the evaluation dataset.
Downloads 346
Release Time : 1/23/2025

Model Overview

This model is a fine-tuned liveness detection model based on the DINOv2 architecture, specifically designed to identify real faces and various spoofing attacks (such as photos, video replays, etc.), applicable to identity verification and security systems.

Model Features

High-precision Detection
Achieves 99.32% accuracy on the evaluation dataset, with an F1 score of 0.9932, demonstrating excellent ability to distinguish real from fake faces.
Strong Anti-spoofing Capability
Exhibits an Attack Presentation Classification Error Rate (APCER) of 18.27% against various spoofing attacks (e.g., photos, video replays).
Low False Rejection Rate
Bona Fide Presentation Classification Error Rate (BPCER) is only 0.89%, effectively reducing misjudgments of genuine users.

Model Capabilities

Face Liveness Detection
Spoofing Attack Identification
Real-time Face Verification

Use Cases

Identity Authentication
Mobile Banking Identity Verification
Used in mobile banking applications to prevent photo or video replay attacks.
Effectively reduces financial fraud risks.
Access Control Systems
Used in smart access control systems to prevent unauthorized personnel from using photos or videos to deceive the system.
Enhances physical security protection levels.
Security Protection
Payment System Protection
Ensures the authenticity of users in mobile payment scenarios.
Reduces payment fraud.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase