When Deepfakes Look Real: Detecting AI-Generated Faces with Unlabeled Data due to Annotation Challenges

2025-08-12

AI Summary

The increasing photorealism of AI-generated faces undermines human annotation reliability, while existing supervised deepfake detection methods suffer severe performance degradation on unlabeled social media data due to distribution shift. Method: This paper proposes an unsupervised deepfake detection framework based on a dual-path network that jointly integrates text-guided cross-domain vision–semantics alignment, curriculum-based pseudo-label optimization, and cross-domain knowledge distillation—thereby mitigating both distribution shift and catastrophic forgetting. Crucially, learnable prompts are employed to enable robust multimodal embedding alignment. Contribution/Results: To our knowledge, this is the first work achieving robust unsupervised detection under highly overlapping real/fake face distributions. Evaluated on 11 mainstream benchmarks, our method achieves an average accuracy gain of +6.3% over state-of-the-art approaches, significantly improving unlabeled-data utilization and generalization across domains.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Multimodal Learning

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Existing deepfake detection methods heavily depend on labeled training data. However, as AI-generated content becomes increasingly realistic, even extbf{human annotators struggle to distinguish} between deepfakes and authentic images. This makes the labeling process both time-consuming and less reliable. Specifically, there is a growing demand for approaches that can effectively utilize large-scale unlabeled data from online social networks. Unlike typical unsupervised learning tasks, where categories are distinct, AI-generated faces closely mimic real image distributions and share strong similarities, causing performance drop in conventional strategies. In this paper, we introduce the Dual-Path Guidance Network (DPGNet), to tackle two key challenges: (1) bridging the domain gap between faces from different generation models, and (2) utilizing unlabeled image samples. The method features two core modules: text-guided cross-domain alignment, which uses learnable prompts to unify visual and textual embeddings into a domain-invariant feature space, and curriculum-driven pseudo label generation, which dynamically exploit more informative unlabeled samples. To prevent catastrophic forgetting, we also facilitate bridging between domains via cross-domain knowledge distillation. Extensive experiments on extbf{11 popular datasets}, show that DPGNet outperforms SoTA approaches by extbf{6.3%}, highlighting its effectiveness in leveraging unlabeled data to address the annotation challenges posed by the increasing realism of deepfakes.

Problem

• Detect AI-generated faces without labeled training data
• Bridge domain gap between different generation models
• Utilize unlabeled data to improve detection accuracy

Innovation

• Dual-Path Guidance Network for deepfake detection
• Text-guided cross-domain alignment for invariant features
• Curriculum-driven pseudo label generation for unlabeled data