TL;DR
Orthogonal Semantic Decoupling is proposed, a framework that decouples demographic semantics from forgery cues and out-performs state-of-the-art methods in both generalization and fairness, breaking the trade-off.
Cross-distribution generalization remains a critical bottleneck in deepfake detection. While recent efforts leverage the semantic priors of large-scale visual foundation models (VFMs), a noteworthy yet underexplored challenge remains: the susceptibility of these semantic priors to environmental interference from factors such as lighting and style. Crucially, this interference establishes spurious correlations between forgery cues and environmental patterns that severely limit generalization. To address this fundamental challenge, we propose an innovative Environment-Invariant Subspace Learning (EISL) framework. The core contribution of EISL is that it aims to disentangle features into orthogonal forgery-relevant invariant factors and environment-related residual factors via a learnable low-rank projection. To facilitate robust feature disentanglement, we also design an Environmental Intervention module that generates diverse and challenging intervention pairs, simulating out-of-distribution environmental shifts to guide the model toward discovering truly invariant forgery representations. Experiments across cross-dataset, cross-generator, whole-face synthesis, and corruption settings show consistent gains and competitive or leading performance against strong detectors, demonstrating improved robustness to unseen forgery types and environmental variations. This work provides a new perspective and a valuable exploration for understanding and tackling the generalization barriers of VFMs in deepfake detection.
A novel demographically balanced FDD benchmark is constructed to train and evaluate the fairness of detectors under both balanced and imbalanced population scenarios, and a novel expression and demographic perceptual vision-language model is introduced, termed FairForensics, for generalizable fair deepfake detection.
While Large Vision-Language Models (LVLMs), represented by LLaVA and GPT-4V, have demonstrated remarkable capabilities, their visual inputs remain vulnerable to adversarial attacks, posing significant security risks. Existing defense methods predominantly target single-task scenarios (e.g., zero-shot classification) and consequently lack generalizability across various multimodal tasks. To address this limitation, we propose a dual adversarial fine-tuning framework that jointly optimizes visual and semantic supervision signals from two modalities, enhancing model robustness while generalizing across multiple downstream tasks. The proposed framework comprises two core components, i.e., $\textbf{Visual}$ supervision branch and $\textbf{Semantic}$ supervision branch. The former branch leverages features from clean images, extracted via a frozen original vision encoder, to guide adversarial robustness while the latter incorporates caption-image alignment as a contextual signal to preserve semantic coherence under attack. Moreover, our method achieves cross-task robustness by simply replacing the CLIP vision encoder in the original model, with no need of separate task-specific retraining or architecture modifications.Extensive experiments demonstrate that our approach outperforms the state-of-the-art method in adversarial robustness evaluation across zero-shot classification, image captioning, and visual question answering (VQA) tasks.
Multimodal fake news detectors often generalize poorly across domains because they learn to trust unreliable evidence: domain-specific shortcuts amplified by imbalanced data and semantically inconsistent text-image pairs that make cross-modal evidence unreliable. We propose Expert-Guided Mutual Distillation (EGMD), which learns what evidence to trust across the prediction pipeline. At the input level, input-level calibration encodes pair-level coherence as a shared gain before fusion. At the representation level, an expert-guided teacher aligns domain statistics and encourages domain-specific patterns to concentrate in specialized experts. At the decision level, prototype-anchored domain-specific students use mutual learning and dual-channel distillation to inherit the teacher's feature geometry and calibrated predictions while discouraging local domain priors. We further construct Weibo_Balanced, a domain-balanced benchmark that isolates the effect of imbalance on generalization. Across four datasets in two languages, EGMD achieves state-of-the-art accuracy while reducing domain bias by up to 57.3%.
The rapid advancement of generative AI has made high-quality deepfakes increasingly accessible, raising serious concerns regarding misinformation and digital trust. Consequently, developing reliable deepfake detection systems has become a critical task in multimedia forensics. However, existing methods often face a severe trade-off between cross-dataset generalization and computational efficiency. They heavily rely on large-scale backbones, which limits their deployment on resource-constrained edge devices. To bridge this gap, we propose HiDD-Net, a hierarchical dual-domain distillation framework. Specifically, the teacher network integrates multi-granularity spatial views (global, local, and facial) via a cross-attention mechanism to explicitly align fine-grained anomalies with global context. Simultaneously, it captures subtle upsampling spectral artifacts using a three-level discrete wavelet transform (DWT) and Spatial Rich Model (SRM) features. Subsequently, we distill these fused spatial-frequency representations into a compact student model (MobileNetV3) by enforcing feature alignment in a shared semantic space. Extensive evaluations on DeepfakeBench demonstrate that the student achieves a competitive cross-dataset average AUC of 0.771 (vs. the teacher’s 0.796) with only 6.67M parameters. Furthermore, it supports real-time CPU inference (30+ FPS) and remains highly robust to JPEG compression and Gaussian blur, offering a practical solution for mobile forensic deployment.