Advances in Face Anti-Spoofing and Backdoor Attack Detection

The field of face anti-spoofing and backdoor attack detection is rapidly evolving, with a growing focus on developing more robust and generalizable models. Recent research has highlighted the importance of moving beyond traditional memorization-based approaches and instead leveraging reinforcement learning and multimodal large language models to improve the interpretability and decision-making capabilities of face anti-spoofing systems. Notable papers in this area include:

  • A face anti-spoofing method that uses reinforcement fine-tuning to stimulate the capabilities of multimodal large language models, achieving state-of-the-art cross-domain generalization performance.
  • A backdoor attack framework specifically designed for Vision Mamba, which leverages a Resonant Frequency Trigger to create stealthy, distributed triggers and achieves superior attack success rates while maintaining clean data accuracy.

Sources

Exploring Task-Solving Paradigm for Generalized Cross-Domain Face Anti-Spoofing via Reinforcement Fine-Tuning

Trident: Detecting Face Forgeries with Adversarial Triplet Learning

PNAct: Crafting Backdoor Attacks in Safe Reinforcement Learning

BadViM: Backdoor Attack against Vision Mamba

Reasoner for Real-World Event Detection: Scaling Reinforcement Learning via Adaptive Perplexity-Aware Sampling Strategy

Survivability of Backdoor Attacks on Unconstrained Face Recognition Systems

Built with on top of