Unified Face Attack Detection via Fine-Grained Semantic Guidance

2026-07-09

AI Summary

Existing face anti-spoofing methods often lack fine-grained semantic modeling of forgery cues, hindering the acquisition of generalizable and interpretable feature representations. To address this limitation, this work introduces MS-UFAD, a large-scale dataset comprising eight million spoofed facial images, each accompanied by fine-grained textual descriptions. Building upon this resource, we propose the Dual-Aligned Forgery Network (DAF-Net), which leverages a vision–language dual-alignment mechanism to jointly learn multimodal representations. Our approach is the first to incorporate fine-grained semantic guidance into face forgery detection, achieving significant performance gains over both purely visual baselines and those using coarse-grained textual descriptions across multiple benchmarks. These results demonstrate that semantic enrichment effectively enhances both the generalization capability and interpretability of forgery detection systems.

Abstract

The growing applications of facial recognition systems are accompanied by increasingly diverse security threats. Existing datasets lack detailed textual descriptions of forgery cues, leading most prior methods to treat face attack detection primarily as a visual recognition task. In this paper, building upon the large-scale MS-UFAD dataset which contains over 8 million attack images, we enrich each image with a fine-grained textual description of forgery cues. Furthermore, we propose a Dual Alignment Forgery Network(DAF-Net) to better leverage these textual information. Extensive experiments demonstrate that our approach extracts more generalizable and semantically meaningful forgery representations from attack images, outperforming both vision-only methods and approaches based on coarse-grained descriptions.

Problem

• Face Attack Detection
• Forgery Cues
• Fine-Grained Semantic Guidance
• Multimodal Learning
• Facial Recognition Security

Innovation

• fine-grained semantic guidance
• face attack detection
• textual description
• Dual Alignment Forgery Network
• MS-UFAD