Large Vision-Language Models (LVLMs) recently achieve significant breakthroughs in understanding complex visual-textual contexts. However, hallucination issues still limit their real-world applicability. Although previous mitigation methods effectively reduce hallucinations in photographic images, they largely overlook the potential risks posed by stylized images, which play crucial roles in critical scenarios such as game scene understanding, art education, and medical analysis. In this work, we first construct a dataset comprising photographic images and their corresponding stylized versions with carefully annotated caption labels. We then conduct head-to-head comparisons on both discriminative and generative tasks by benchmarking 13 advanced LVLMs on the collected datasets. Our findings reveal that stylized images tend to induce significantly more hallucinations than their photographic counterparts. To address this issue, we propose Style-Aware Visual Early Revision (SAVER), a novel mechanism that dynamically adjusts LVLMs' final outputs based on the token-level visual attention patterns, leveraging early-layer feedback to mitigate hallucinations caused by stylized images. Extensive experiments demonstrate that SAVER achieves state-of-the-art performance in hallucination mitigation across various models, datasets, and tasks.

Open-set face forgery detection poses significant security threats and presents substantial challenges for existing detection models. These detectors primarily have two limitations: they cannot generalize across unknown forgery domains or inefficiently adapt to new data. To address these issues, we introduce an approach that is both general and parameter-efficient for face forgery detection. Our method builds on the assumption that different forgery source domains exhibit distinct style statistics. Specifically, we design a forgery-style-mixture formulation that augments the diversity of forgery source domains, enhancing the model’s generalizability across unseen domains. In addition, previous methods typically require fully fine-tuning pretrained networks, consuming substantial time and computational resources. Drawing on recent advancements in vision transformers (ViT) for face forgery detection, we develop a parameter-efficient ViT-based detection model that includes lightweight forgery feature extraction modules and enables the model to extract global and local forgery clues simultaneously. We only optimize the inserted lightweight modules during training, maintaining the original ViT structure with its pre-trained weights. This training strategy effectively preserves the informative pre-trained knowledge while flexibly adapting the model to the task of Deepfake detection. Extensive experimental results demonstrate that the designed model achieves state-of-the-art generalizability with significantly reduced trainable parameters, representing an important step toward open-set Deepfake detection in the wild.

Deepfakes have recently raised significant trust issues and security concerns among the public. Compared to CNN-based face forgery detectors, ViT-based methods take advantage of the expressivity of transformers, achieving superior detection performance. However, these approaches still exhibit the following limitations: (1) Fully fine-tuning ViT-based models from ImageNet weights demands substantial computational and storage resources; (2) ViT-based methods struggle to capture local forgery clues, leading to model bias; (3) These methods limit their scope on only one or few face forgery features, resulting in limited generalizability. To tackle these challenges, this work introduces Mixture-of-Experts modules for Face Forgery Detection (MoE-FFD), a generalized yet parameter-efficient ViT-based approach. MoE-FFD only updates lightweight Low-Rank Adaptation (LoRA) and Adapter layers while keeping the ViT backbone frozen, thereby achieving parameter-efficient training. Moreover, MoE-FFD leverages the expressivity of transformers and local priors of CNNs to simultaneously extract global and local forgery clues. Additionally, novel MoE modules are designed to scale the model's capacity and smartly select optimal forgery experts, further enhancing forgery detection performance. Our proposed learning scheme can be seamlessly adapted to various transformer backbones in a plug-and-play manner. Extensive experimental results demonstrate that the proposed method achieves state-of-the-art face forgery detection performance with significantly reduced parameter overhead in cross-dataset, cross-manipulation, and robustness evaluations. Our ablation studies further validate the effectiveness of the designed components and the proposed learning scheme. The code is available at: https://github.com/LoveSiameseCat/MoE-FFD.

The increasing sophistication of image manipulation techniques demands robust forensic solutions that can both reliably detect alterations and precisely localize tampered regions. Recent Multimodal Large Language Models (MLLMs) show promise by leveraging world knowledge and semantic understanding for context-aware detection, yet they struggle with perceiving subtle, low-level forensic artifacts crucial for accurate manipulation localization. This paper presents a novel Propose-Rectify framework that effectively bridges semantic reasoning with forensic-specific analysis. In the proposal stage, our approach utilizes a forensic-adapted LLaVA model to generate initial manipulation analysis and preliminary localization of suspicious regions based on semantic understanding and contextual reasoning. In the rectification stage, we introduce a Forensics Rectification Module that systematically validates and refines these initial proposals through multi-scale forensic feature analysis, integrating technical evidence from several specialized filters. Additionally, we present an Enhanced Segmentation Module that incorporates critical forensic cues into SAM’s encoded image embeddings, thereby overcoming inherent semantic biases to achieve precise delineation of manipulated regions. By synergistically combining advanced multimodal reasoning with established forensic methodologies, our framework ensures that initial semantic proposals are systematically validated and enhanced through concrete technical evidence, resulting in comprehensive detection accuracy and localization precision. Extensive experimental validation demonstrates state-of-the-art performance across diverse datasets with exceptional robustness and generalization capabilities.

Hui Liu, Wenya Wang, Hao Sun, Chris Xing Tian, Chenqi Kong, Xin Dong, Haoliang Li. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

With the rise of social media, vast amounts of user-uploaded videos (e.g., YouTube) are utilized as training data for Visual Object Tracking (VOT). However, the VOT community has largely overlooked video data-privacy issues, as many private videos have been collected and used for training commercial models without authorization. To alleviate these issues, this paper presents the first investigation on preventing personal video data from unauthorized exploitation by deep trackers. Existing methods for preventing unauthorized data use primarily focus on image-based tasks (e.g., image classification), directly applying them to videos reveals several limitations, including inefficiency, limited effectiveness, and poor generalizability. To address these issues, we propose a novel generative framework for generating Temporal Unlearnable Examples (TUEs), and whose efficient computation makes it scalable for usage on large-scale video datasets. The trackers trained w/ TUEs heavily rely on unlearnable noises for temporal matching, ignoring the original data structure and thus ensuring training video data-privacy. To enhance the effectiveness of TUEs, we introduce a temporal contrastive loss, which further corrupts the learning of existing trackers when using our TUEs for training. Extensive experiments demonstrate that our approach achieves state-of-the-art performance in video data-privacy protection, with strong transferability across VOT models, datasets, and temporal matching tasks.

Most existing unlearnable strategies focus on preventing unauthorized users from training single-task learning (STL) models with personal data. Nevertheless, the paradigm has recently shifted towards multi-task data and multi-task learning (MTL), targeting generalist and foundation models that can handle multiple tasks simultaneously. Despite their growing importance, MTL data and models have been largely neglected while pursuing unlearnable strategies. This paper presents MTL-UE, the first unified framework for generating unlearnable examples for multi-task data and MTL models. Instead of optimizing perturbations for each sample, we design a generator-based structure that introduces label priors and class-wise feature embeddings which leads to much better attacking performance. In addition, MTL-UE incorporates intra-task and inter-task embedding regularization to increase inter-class separation and suppress intra-class variance which enhances the attack robustness greatly. Furthermore, MTL-UE is versatile with good supports for dense prediction tasks in MTL. It is also plug-and-play allowing integrating existing surrogate-dependent unlearnable methods with little adaptation. Extensive experiments show that MTL-UE achieves superior attacking performance consistently across 4 MTL datasets, 3 base UE methods, 5 model backbones, and 5 MTL task-weighting strategies. Code is available at https://github.com/yuyi-sd/MTL-UE.

Digital image forensics plays a crucial role in image authentication and manipulation localization. Despite the progress powered by deep neural networks, existing forgery localization methodologies exhibit limitations when deployed to unseen datasets and perturbed images (i.e., lack of generalization and robustness to real-world applications). To circumvent these problems and aid image integrity, this paper presents a generalized and robust manipulation localization model through the analysis of pixel inconsistency artifacts. The rationale is grounded on the observation that most image signal processors (ISP) involve the demosaicing process, which introduces pixel correlations in pristine images. Moreover, manipulating operations, including splicing, copy-move, and inpainting, directly affect such pixel regularity. We, therefore, first split the input image into several blocks and design masked self-attention mechanisms to model the global pixel dependency in input images. Simultaneously, we optimize another local pixel dependency stream to mine local manipulation clues within input forgery images. In addition, we design novel Learning-to-Weight Modules (LWM) to combine features from the two streams, thereby enhancing the final forgery localization performance. To improve the training process, we propose a novel Pixel-Inconsistency Data Augmentation (PIDA) strategy, driving the model to focus on capturing inherent pixel-level artifacts instead of mining semantic forgery traces. This work establishes a comprehensive benchmark integrating 16 representative detection models across 12 datasets. Extensive experiments show that our method successfully extracts inherent pixel-inconsistency forgery fingerprints and achieve state-of-the-art generalization and robustness performances in image manipulation localization.

With the rapid progress of generative models, the current challenge in face forgery detection is how to effectively detect realistic manipulated faces from different unseen domains. Though previous studies show that pre-trained Vision Transformer (ViT) based models can achieve some promising results after fully fine-tuning on the Deepfake dataset, their generalization performances are still unsatisfactory. To this end, we present a Forgery-aware Adaptive Vision Transformer (FA-ViT) under the adaptive learning paradigm for generalized face forgery detection, where the parameters in the pre-trained ViT are kept fixed while the designed adaptive modules are optimized to capture forgery features. Specifically, a global adaptive module is designed to model long-range interactions among input tokens, which takes advantage of self-attention mechanism to mine global forgery clues. To further explore essential local forgery clues, a local adaptive module is proposed to expose local inconsistencies by enhancing the local contextual association. In addition, we introduce a fine-grained adaptive learning module that emphasizes the common compact representation of genuine faces through relationship learning in fine-grained pairs, driving these proposed adaptive modules to be aware of fine-grained forgery-aware information. Extensive experiments demonstrate that our FA-ViT achieves state-of-the-arts results in the cross-dataset evaluation, and enhances the robustness against unseen perturbations. Particularly, FA-ViT achieves 93.83% and 78.32% AUC scores on Celeb-DF and DFDC datasets in the cross-dataset evaluation. The code and trained model have been released at:https://github.com/LoveSiameseCat/FAViT.

Adversarial examples, characterized by imperceptible perturbations, pose significant threats to deep neural networks by misleading their predictions. A critical aspect of these examples is their transferability, allowing them to deceive unseen models in closed-box scenarios. Despite the widespread exploration of defense methods, including those on transferability, they show limitations: inefficient deployment, ineffective defense, and degraded performance on clean images. In this work, we introduce a novel training paradigm aimed at enhancing robustness against transferable adversarial examples (TAEs) in a more efficient and effective way. We propose a model that exhibits random guessing behavior when presented with clean data$\boldsymbol {x}$as input, and generates accurate predictions when with triggered data$\boldsymbol {x}+\boldsymbol {\tau }$. Importantly, the trigger$\boldsymbol {\tau }$remains constant for all data instances. We refer to these models as models with trigger activation. We are surprised to find that these models exhibit certain robustness against TAEs. Through the consideration of first-order gradients, we provide a theoretical analysis of this robustness. Moreover, through the joint optimization of the learnable trigger and the model, we achieve improved robustness to transferable attacks. Extensive experiments conducted across diverse datasets, evaluating a variety of attacking methods, underscore the effectiveness and superiority of our approach.

Recent advances in AI-powered image editing tools have significantly lowered the barrier to image modification, raising pressing security concerns those related to spreading misinformation and disinformation on social platforms. Image provenance analysis is crucial in this context, as it identifies relevant images within a database and constructs a relationship graph by mining hidden manipulation and transformation cues, thereby providing concrete evidence chains. This paper introduces a novel end-to-end deep learning framework designed to explore the structural information of provenance graphs. Our proposed method distinguishes from previous approaches in two main ways. First, unlike earlier methods that rely on prior knowledge and have limited generalizability, our framework relies upon a patch attention mechanism to capture image provenance clues for local manipulations and global transformations, thereby enhancing graph construction performance. Second, while previous methods primarily focus on identifying tampering traces only between image pairs, they often overlook the hidden information embedded in the topology of the provenance graph. Our approach aligns the model training objectives with the final graph construction task, incorporating the overall structural information of the graph into the training process. We integrate graph structure information with the attention mechanism, enabling precise determination of the direction of transformation. Experimental results show the superiority of the proposed method over previous approaches, underscoring its effectiveness in addressing the challenges of image provenance analysis.

Face presentation attacks (FPA), also known as face spoofing, have brought increasing concerns to the public through various malicious applications, such as financial fraud and privacy leakage. Therefore, safeguarding face recognition systems against FPA is of utmost importance. Although existing learning-based face anti-spoofing (FAS) models can achieve outstanding detection performance, they lack generalization capability and suffer significant performance drops in unforeseen environments. Many methodologies seek to use auxiliary modality data (e.g., depth and infrared maps) during the presentation attack detection (PAD) to address this limitation. However, these methods can be limited since (1) they require specific sensors such as depth and infrared cameras for data capture, which are rarely available on commodity mobile devices, and (2) they cannot work properly in practical scenarios when either modality is missing or of poor quality. In this paper, we devise an accurate and robustMultiModalMobileFaceAnti-Spoofing system namedM$^{3}$FASto overcome the issues above. The primary innovation of this work lies in the following aspects: (1) To achieve robust PAD, our system combines visual and auditory modalities using three commonly available sensors: camera, speaker, and microphone; (2) We design a novel two-branch neural network with three hierarchical feature aggregation modules to perform cross-modal feature fusion; (3). We propose a multi-head training strategy, allowing the model to output predictions from the vision, acoustic, and fusion heads, resulting in a more flexible PAD. Extensive experiments have demonstrated the accuracy, robustness, and flexibility of M$^{3}$FAS under various challenging experimental settings. The source code and dataset are available at:https://github.com/ChenqiKONG/M3FAS/.

Face Anti-Spoofing (FAS) aims to detect malicious attempts to invade a face recognition system by presenting spoofed faces. State-of-the-art FAS techniques predominantly rely on deep learning models but their cross-domain generalization capabilities are often hindered by the domain shift problem, which arises due to different distributions between training and testing data. In this study, we develop a generalized FAS method under the Efficient Parameter Transfer Learning (EPTL) paradigm, where we adapt the pre-trained Vision Transformer models for the FAS task. During training, the adapter modules are inserted into the pre-trained ViT model, and the adapters are updated while other pre-trained parameters remain fixed. We find the limitations of previous vanilla adapters in that they are based on linear layers, which lack a spoofing-aware inductive bias and thus restrict the cross-domain generalization. To address this limitation and achieve cross-domain generalized FAS, we propose a novel Statistical Adapter (S-Adapter) that gathers local discriminative and statistical information from localized token histograms. To further improve the generalization of the statistical tokens, we propose a novel Token Style Regularization (TSR), which aims to reduce domain style variance by regularizing Gram matrices extracted from tokens across different domains. Our experimental results demonstrate that our proposed S-Adapter and TSR provide significant benefits in both zero-shot and few-shot cross-domain testing, outperforming state-of-the-art methods on several benchmark tests. We will release the source code upon acceptance.

Face forgery detection is essential in combating malicious digital face attacks. Previous methods mainly rely on prior expert knowledge to capture specific forgery clues, such as noise patterns, blending boundaries, and frequency artifacts. However, these methods tend to get trapped in local optima, resulting in limited robustness and generalization capability. To address these issues, we propose a novel Critical Forgery Mining (CFM) framework, which can be flexibly assembled with various backbones to boost their generalization and robustness performance. Specifically, we first build a fine-grained triplet and suppress specific forgery traces through prior knowledge-agnostic data augmentation. Subsequently, we propose a fine-grained relation learning prototype to mine critical information in forgeries through instance and local similarity-aware losses. Moreover, we design a novel progressive learning controller to guide the model to focus on principal feature components, enabling it to learn critical forgery features in a coarse-to-fine manner. The proposed method achieves state-of-the-art forgery detection performance under various challenging evaluation settings. The source code is available at:https://github.com/LoveSiameseCat/CFM.

The image recapture attack is an effective image manipulation method to erase certain forensic traces, and when targeting on personal document images, it poses a great threat to the security of e-commerce and other web applications. Considering the current learning-based methods suffer from serious over-fitting problem, in this paper, we propose a novel two-branch deep neural network by mining better generalized recapture artifacts with a designed frequency filter bank and multi-scale cross-attention fusion module. In the extensive experiment, we show that our method can achieve better generalization capability compared with state-of-the-art techniques on different scenarios.

Attention mechanisms are dominating the explainability of deep models. They produce probability distributions over the input, which are widely deemed as feature-importance indicators. However, in this paper, we find one critical limitation in attention explanations: weakness in identifying the polarity of feature impact. This would be somehow misleading – features with higher attention weights may not faithfully contribute to model predictions; instead, they can impose suppression effects. With this finding, we reflect on the explainability of current attention-based techniques, such as Attention $\bigodot$ Gradient and LRP-based attention explanations. We first propose an actionable diagnostic methodology (henceforth faithfulness violation test) to measure the consistency between explanation weights and the impact polarity. Through the extensive experiments, we then show that most tested explanation methods are unexpectedly hindered by the faithfulness violation issue, especially the raw attention. Empirical analyses on the factors affecting violation issues further provide useful observations for adopting explanation methods in attention models.

Recently, there has been an exponential increase in the security concerns raised by faking face (e.g., deepfake), which automatically changes the identity with a specifically learned deep generative model. With numerous approaches proposed to identify the fake content, much less work has been dedicated to automatically revealing the authentic one that is originally acquired. Here, we propose a new paradigm that seeks to reveal the authentic face hidden behind the fake one by leveraging the joint information of face and audio. More specifically, given the fake face as well as the audio segment, the cross-modality transferable capability is exploited by learning to generate the feature of the authentic face, based on the underlying clues from the audio as well as the fake face appearance. The effectiveness of the proposed scheme is validated through a series of evaluations, and experimental results show that the proposed model achieves promising face reconstruction performance in revealing the hidden faces, in terms of reconstruction quality, as well as identity and face attribute inference accuracy.

The technological advancements of deep learning have enabled sophisticated face manipulation schemes, raising severe trust issues and security concerns in modern society. Generally speaking, detecting manipulated faces and locating the potentially altered regions are challenging tasks. Herein, we propose a conceptually simple but effective method to efficiently detect forged faces in an image while simultaneously locating the manipulated regions. The proposed scheme relies on a segmentation map that delivers meaningful high-level semantic information clues about the image. Furthermore, a noise map is estimated, playing a complementary role in capturing low-level clues and subsequently empowering decision-making. Finally, the features from these two modules are combined to distinguish fake faces. Extensive experiments show that the proposed model achieves state-of-the-art detection accuracy and remarkable localization performance.

2D face presentation attacks are one of the most notorious and pervasive face spoofing types, which have caused pressing security issues to facial authentication systems. While RGB-based face anti-spoofing (FAS) models have proven to counter the face spoofing attack effectively, most existing FAS models suffer from the overfitting problem (i.e., lack generalization capability to data collected from an unseen environment). Recently, many models have been devoted to capturing auxiliary information (e.g., depth and infrared maps) to achieve a more robust face liveness detection performance. However, these methods require expensive sensors and cost extra hardware to capture the specific modality information, limiting their applications in practical scenarios. To tackle these problems, we devise a novel and cost-effective FAS system based on the acoustic modality, named Echo-FAS, which employs the crafted acoustic signal as the probe to perform face liveness detection. We first propose to build a large-scale, high-diversity, and acoustic-based FAS database, Echo-Spoof. Then, based upon Echo-Spoof, we propose designing a novel two-branch framework that combines the global and local frequency clues of input signals to distinguish inputs, live vs. spoofing faces accurately. The devised Echo-FAS comprises the following three merits: (1) It only needs one available speaker and microphone as sensors while not requiring any expensive hardware; (2) It can successfully capture the 3D geometrical information of input queries and achieve a remarkable face anti-spoofing performance; and (3) It can be handily allied with other RGB-based FAS models to mitigate the overfitting problem in the RGB modality and make the FAS model more accurate and robust. Our proposed Echo-FAS provides new insights regarding the development of FAS systems for mobile devices.

In this paper, we propose a no-reference (NR) image quality assessment (IQA) method via feature level pseudo-reference (PR) hallucination. The proposed quality assessment framework is rooted in the view that the perceptually meaningful features could be well exploited to characterize the visual quality, and the natural image statistical behaviors are exploited in an effort to deliver the accurate predictions. Herein, the PR features from the distorted images are learned by a mutual learning scheme with the pristine reference as the supervision, and the discriminative characteristics of PR features are further ensured with the triplet constraints. Given a distorted image for quality inference, the feature level disentanglement is performed with an invertible neural layer for final quality prediction, leading to the PR and the corresponding distortion features for comparison. The effectiveness of our proposed method is demonstrated on four popular IQA databases, and superior performance on cross-database evaluation also reveals the high generalization capability of our method. The implementation of our method is publicly available on https://github.com/Baoliang93/FPR.