Aug 2026· International Conference on Digital Image Processing· Vol 14351, pp. 143510K - 143510K-11· 0 citations· 22 references
Engineering
TL;DR
HiDD-Net is proposed, a hierarchical dual-domain distillation framework that combines a teacher network that integrates multi-granularity spatial views via a cross-attention mechanism to explicitly align fine-grained anomalies with global context, and a compact student model that achieves a competitive cross-dataset average AUC.
Abstract
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.
Abstract. Deep-learning methods dominate remote-sensing change detection (CD), yet state-of-the-art models remain parameter-heavy and struggle with crisp boundaries, limiting their use on edge devices. We present LEDGNet, a Lightweight, Edge-knowledge- Distillation-Guided CD Network, that reconciles accuracy, boundary fidelity, and efficiency. LEDGNet integrates three purpose-built components: 1) an Edge Distillation Module that mines multi-scale boundary cues from a high-capacity teacher and transfers them to a compact student through an edge-aware loss; 2) StarLite, a depth-wise separable encoder that preserves fine spatial detail while minimizing floating-point operations; and 3) LiteDecoder, an inexpensive feature-fusion head that restores full resolution without bulky up-sampling. This design halves the parameters and inference time of mainstream fine-grained CD networks while enhancing edge sharpness. On the CDD and LEVIR-CD benchmarks, LEDGNet achieves competitive F1 performance while maintaining a compact footprint of 20.58 M parameters and 35.18 G FLOPs. With an inference time of 255 ms, it strikes a balance between resource consumption and detection efficiency, making it well-suited for high-efficiency remote sensing monitoring.
Tingyu Ji, Yixin Chen, Ruiqian Zhang et al.· ISPRS Annals of the Photogra...· 0 citations
Detecting audio-visual DeepFake (AVDeepFake) is becoming increasingly important as synthetic media tools become widely accessible and spread across consumer devices. In this study, we present an Detecting audio-visual DeepFakes (AV-DeepFakes) has become increasingly critical with the rapid proliferation of accessible synthetic media generation tools across consumer platforms. In this work, we propose a highperformance, deployment-efficient AV-DeepFake detection framework tailored for real-world consumer devices. The proposed model integrates a 3D convolutional visual encoder with a 2D convolutional audio encoder to learn synchronized multimodal representations, effectively capturing spatial, spectral, and prosodic inconsistencies inherent in manipulated content. To detect temporal forgeries, we introduce a bidirectional complementary boundary module that precisely localizes manipulation onsets and offsets. A cross-modal attention fusion mechanism aggregates modality-specific cues, while an uncertainty-aware gating strategy suppresses unreliable signals to improve robustness. Furthermore, a cross-modal discrepancy minimization loss encourages alignment for genuine samples while maximizing divergence for forged content, strengthening multimodal consistency learning. Extensive evaluations on FaceForensics++ and LAV-DF demonstrate the effectiveness of the proposed approach, achieving 97.1% AUC for clip-level detection and an 81.2% F1-score for temporal boundary localization, while reducing inference time by 5× compared with transformer-based methods.
Nasir Saleem, Adeel Hussain, Sami Bourouis et al.· International Journal of Int...· 0 citations
DINO-MAC is a simple yet effective model for robust deepfake detection under degraded conditions and is the first-place winner of the “NTIRE 2026 - Robust Deepfake Detection Challenge @ CVPR 2026”, significantly outperforms all competitors.
Chenfan Qu, Lianwen Jin, Junchi Li et al.· 3 citations
A dehazing framework named DKS-Net is proposed which fully utilizes the physics guiding features and extracting structural information in the spatial domain, and a Kernel Selective Feature Extraction Module (KSFE) is introduced to effectively captures structural patterns via large-kernel convolutions with dynamic selection capabilities and multi-scale semantic cues.
Zehao Shi, Han Wang, Xinyue Liu· International Conference on...· 0 citations
Today, the use of increasingly ubiquitous synthetic media, or ‘deepfakes’, has become a risk to online trust, information integrity and individual security and is being created by artificial intelligence (AI). The current approaches are mainly based on either spatial features of CNNs or high-level semantic representations of Vision Transformer; both have major drawbacks in effectively leveraging multi-domain forensic cues. This paper presents FAViT (Frequency-Aware Vision Transformer), a hybrid architecture capable of jointly utilizing spatial- and frequency-domain forensic information by the means of a bidirectional cross-attention fusion scheme. We use an 11-channel forensic tensor in each face image (including per-channel Fast Fourier Transform (FFT) magnitude maps, Discrete Wavelet Transform (DWT) sub-bands, channel noise residual maps, Sobel gradient magnitude and channels of Error Level Analysis (ELA)). A Frequency Branch CNN processes this multi-domain tensor and the original RGB image is encoded with a pretrained ViT-B/16 spatial branch. The two streams are combined through the bidirectional cross-attention which allows the model to localize both spatial and spectral manipulation artifacts. We also present an adversarial cleaning simulation pipeline which partitions the training process with five post-processing attack methods, namely GFPGAN neural face restoration, learned autoencoder cleaning, etc., to increase resistance to real-world forensic defenses. Tests of FaceForensics++ C23 (7926 images, consisting of four manipulation types) show that FAViT attains F1-score of 86.22, AUC-ROC of 94.26 and accuracy of 85.55 on the held-out test set. The strength analysis of 21 attack conditions shows that the max degradation in AUC is 30.3, with specific strengths in GFPGAN restoration (AUC = 98.51). Robustness is evaluated based on 21 post-processing attack cases that include JPEG compression, Gaussian blurring, down-sampling, and GFDGAN neural-based restoration; it should be noted that robustness against gradient-based adaptive attacks requires additional attention. Testing on the CIFAKE and Celeb-DF v2 datasets reveals some limitations of domain generalization.
Wasin Alkishri, Shahid Kamal, Jabar H. Yousif· Information· 0 citations
AI-generated image (AIGI) detection has become increasingly challenging due to the rapid advancement of generative models and the diminishing gap between synthetic and authentic content. Existing vision transformer-based detectors commonly rely on weighted-sum strategies to aggregate intermediate representations across transformer layers, often overlooking the inherently ordered semantic progression of hierarchical features from shallow texture cues to deep semantic representations. In this work, we propose \textbf{PE-Mamba}, a novel framework built upon a pre-trained PE-Core vision transformer with lightweight LoRA adaptation that introduces three complementary components for cross-layer feature aggregation and fusion. First, a bidirectional selective aggregator (BSA) processes layer-wise classification tokens through forward and backward selective scans, where the forward scan progressively accumulates shallow-to-deep forensic evidence, and the backward scan performs deep-to-shallow contextual refinement to reinterpret low-level cues in light of high-level semantic context. Second, a softmax-weighted aggregator (SWA) computes a learned global summary of all layer tokens as a complementary aggregation path. Third, a sigmoid-gated blend (SGA) adaptively fuses the BSA and SWA outputs via a learnable scalar gate, allowing the model to dynamically balance directional sequential evidence and global layer-wise aggregation. Extensive experiments on UniversalFakeDetect (96.6\% mACC, 99.5\% mAP) and AIGCDetect (95.3\% mACC, 98.1\% mAP) demonstrate that \methodname{} outperforms 18 detectors with superior generalization across diverse generative models, while training only 1.3\% of total parameters (0.13\% for LoRA alone).
Kutub Uddin, Nusrat Tasnim, K. Malik· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
