Event-based cameras capture visual information as asynchronous streams of per-pixel brightness changes, generating sparse, temporally precise data. Compared to conventional frame-based sensors, they offer significant advantages in capturing high-speed dynamics while consuming substantially less power. Predicting future event representations from past observations is an important problem, enabling downstream tasks such as future semantic segmentation or object tracking without requiring access to future sensor measurements. While recent state-of-the-art approaches achieve strong performance, they often rely on computationally heavy backbones and, in some cases, large-scale pretraining, limiting their applicability in resource-constrained scenarios. In this work, we introduce E-TIDE, a lightweight, end-to-end trainable architecture for event-tensor prediction that is designed to operate efficiently without large-scale pretraining. Our approach employs the TIDE module (Temporal Interaction for Dynamic Events), motivated by efficient spatiotemporal interaction design for sparse event tensors, to capture temporal dependencies via large-kernel mixing and activity-aware gating while maintaining low computational complexity. Experiments on standard event-based datasets demonstrate that our method achieves competitive performance with significantly reduced model size and training requirements, making it well-suited for real-time deployment under tight latency and memory budgets.

The rapid advancement of deepfake technology has significantly elevated the realism and accessibility of synthetic media. Emerging techniques, such as diffusion-based models and Neural Radiance Fields (NeRF), alongside enhancements in traditional Generative Adversarial Networks (GANs), have contributed to the sophisticated generation of deepfake videos. Concurrently, deepfake detection methods have seen notable progress, driven by innovations in Transformer architectures, contrastive learning, and other machine learning approaches. In this study, we conduct a comprehensive empirical analysis of state-of-the-art deepfake detection techniques, including human evaluation experiments against cutting-edge synthesis methods. Our findings highlight a concerning trend: many state-of-the-art detection models exhibit markedly poor performance when challenged with deepfakes produced by modern synthesis techniques, including poor performance by human participants against the best quality deepfakes. Through extensive experimentation, we provide evidence that underscores the urgent need for continued refinement of detection models to keep pace with the evolving capabilities of deepfake generation technologies. This research emphasizes the critical gap between current detection methodologies and the sophistication of new generation techniques, calling for intensified efforts in this crucial area of study.

Unauthorized reuse of digital artworks and creative content has become a widespread concern on social media and content-sharing platforms, where copyrighted images are copied, intentionally modified, and redistributed without permission. Billions of images are estimated to be stolen each day. To protect creators, copyright monitoring services use nearduplicate matching techniques, such as perceptual hashing, to detect unauthorized reuse despite modifications. However, existing approaches remain vulnerable to adversarial perturbations, allowing unauthorized reuse to evade copyright detection and undermining the reliability of monitoring services. A recent defense, CertPHash, aims to enhance adversarial robustness, but for DCT-based perceptual hashing it offers limited robustness gains while substantially reducing hash-bit utilization, which may increase collision risk when operating at scale. This limitation likely arises because CertPHash applies a generic learning-based approach across different perceptual hashing schemes, rather than accounting for the design-specific weaknesses of DCT-based perceptual hashing. In this work, we identify the root cause that makes DCT-based perceptual hashing susceptible to adversarial perturbations, and propose RD-PHash, a robustness enhancement tailored to DCT-based hashes. Our experiments show that, for DCT-based hashes such as pHash and PDQ, RD-PHash outperforms CertPHash in both hash-bit utilization and adversarial robustness, achieving up to $2.3 \times$ higher utilization while reducing the impact of existing adversarial attacks by up to 71%.

We introduce the MhAIM Dataset, enabling large-scale analysis and model training of human responses. Our study shows that people are better at identifying artificial intelligence content when posts include both text and visuals, and propose two composite metrics for downstream risk assessment and decision support.

Watermarks are increasingly embedded in AI-generated images to indicate their machine-generated origin. For reliable detection, watermarks are designed to remain robust against common image manipulations such as resizing and compression. In this paper, we uncover a previously overlooked vulnerability: the robustness property of watermarking schemes can inadvertently expose the image features that carry watermark signals. Exploiting this vulnerability, we introduce WRATH, the first watermark-agnostic black-box attack capable of both watermark removal and forgery. Our attack is practical and requires only a scheme's robustness information, which is typically public or can be obtained through simple robustness tests. When evaluated against state-of-theart watermarking schemes, including Amazon's, WRATH successfully attacks all evaluated schemes while preserving high perceptual image quality. We also discuss practical defenses to mitigate this vulnerability, and our findings call for a rethink of watermark security.

Publicly-acceesible face-manipulation tools have made deepfake creation accessible to non-expert users. Against these, existing defenses are mostly post-hoc, detecting only after forgery has occurred, and operating on still images rather than videos. Research is lacking in i) the proactive protection of published facial videos against black-box manipulation tools, and in (ii) understanding its perceptual effect on human viewers. We introduce TaintedPixels, a proactive video-protection method built around an asymmetric visibility trade-off: the embedded watermark should remain inconspicuous in the published video but become obvious once a downstream tool manipulates the video. TaintedPixels injects structured periodic perturbations into the blue channel of facial regions and refines them under stripe-visibility, color-cast, and video-level LPIPS budgets, with lightweight motion-adaptive deployment. We believe TaintedPixels is the first proactive defense designed specifically against black-box manipulation tools rather than image-level pipelines or specific surrogate generators. Across three publicly available off-the-shelf video manipulation tools and two off-the-shelf detectors, TaintedPixels attains the highest forgery fake rate while keeping perturbations small (LPIPS = 0.0042). Our non-expert human study, conducted on a diverse set of 300 video stimuli spanning different lighting conditions, backgrounds, and skin tones, shows that protected source videos draw a 3.26

Privacy concerns around ever increasing number of cameras are increasing in today's digital age. Although existing anonymization methods are able to obscure identity information, they often struggle to preserve the utility of the images. In this work, we introduce a training-free method for face anonymization that preserves key non-identity-related attributes. Our approach utilizes a pre-trained text-to-image diffusion model without requiring optimization or training. It begins by inverting the input image to recover its initial noise. The noise is then denoised through an identity-conditioned diffusion process, where modified identity embeddings ensure the anonymized face is distinct from the original identity. Our approach also supports localized anonymization, giving users control over which facial regions are anonymized or kept intact. Comprehensive evaluations against state-of-the-art methods show our approach excels in anonymization, attribute preservation, and image quality. Its flexibility, robustness, and practicality make it well-suited for real-world applications. Code and data can be found at https://github.com/hanweikung/nullface .

This article presents the Keystroke Verification Challenge- onGoing (KVC-onGoing)1, on which researchers can easily benchmark their systems in a common platform using large-scale public databases, the Aalto University Keystroke databases, and a standard experimental protocol. The keystroke data consist of tweet- long sequences of variable transcript text from over 185,000 subjects, acquired through desktop and mobile keyboards simulating real-life conditions. The results on the evaluation set of KVC-onGoing have proved the high discriminative power of keystroke dynamics, reaching values as low as 3.33% of Equal Error Rate (EER) and 11.96% of False Non-Match Rate (FNMR) @1% False Match Rate (FMR) in the desktop scenario, and 3.61% of EER and 17.44% of FNMR @1% at FMR in the mobile scenario, significantly improving previous state-of-the-art results. Concerning demographic fairness, the analyzed scores reflect the subjects' age and gender to various extents, not negligible in a few cases. The framework runs on CodaLab2.

As AI-generated content becomes widespread, so does the risk of misinformation. While prior research has primarily focused on identifying whether content is authentic, much less is known about how such content influences human perception and behavior. In domains like trading or the stock market, predicting how people react (e.g., whether a news post will go viral), can be more critical than verifying its factual accuracy. To address this, we take a human-centered approach and introduce the MhAIM Dataset, which contains 154,552 online posts (111,153 of them AI-generated), enabling large-scale analysis of how people respond to AI-generated content. Our human study reveals that people are better at identifying AI content when posts include both text and visuals, particularly when inconsistencies exist between the two. We propose three new metrics: trustworthiness, impact, and openness, to quantify how users judge and engage with online content. We present T-Lens, an LLM-based agent system designed to answer user queries by incorporating predicted human responses to multimodal information. At its core is HR-MCP (Human Response Model Context Protocol), built on the standardized Model Context Protocol (MCP), enabling seamless integration with any LLM. This integration allows T-Lens to better align with human reactions, enhancing both interpretability and interaction capabilities. Our work provides empirical insights and practical tools to equip LLMs with human-awareness capabilities. By highlighting the complex interplay among AI, human cognition, and information reception, our findings suggest actionable strategies for mitigating the risks of AI-driven misinformation.

Current face anonymization techniques often depend on identity loss calculated by face recognition models, which can be inaccurate and unreliable. Additionally, many methods require supplementary data such as facial landmarks and masks to guide the synthesis process. In contrast, our approach uses diffusion models with only a reconstruction loss, eliminating the need for facial landmarks or masks while still producing images with intricate, fine-grained details. We validated our results on two public benchmarks through both quantitative and qualitative evaluations. Our model achieves state-of-the-art performance in three key areas: identity anonymization, facial attribute preservation, and image quality. Beyond its primary function of anonymization, our model can also perform face swapping tasks by incorporating an additional facial image as input, demonstrating its versatility and potential for diverse applications. Our code and models are available at https://github.com/hanweikung/face_anon_simple .

Multi-face deepfake videos are becoming increasingly prevalent, often appearing in natural social settings that challenge existing detection methods. Most current approaches excel at single-face detection but struggle in multi-face scenarios, due to a lack of awareness of crucial contextual cues. In this work, we develop a novel approach that leverages human cognition to analyze and defend against multi-face deepfake videos. Through a series of human studies, we systematically examine how people detect deepfake faces in social settings. Our quantitative analysis reveals four key cues humans rely on: scene-motion coherence, inter-face appearance compatibility, interpersonal gaze alignment, and face-body consistency. Guided by these insights, we introduce , a novel framework designed to detect every fake face in multi-face scenarios. Extensive experiments on benchmark datasets show that improves average accuracy by 3.3% in in-dataset detection and 2.8% under real-world perturbations. Moreover, it outperforms existing methods by 5.8% on unseen datasets, demonstrating the generalization of human-inspired cues. further enhances interpretability by incorporating an LLM to provide human-readable explanations, making detection results more transparent and convincing. Our work sheds light on involving human factors to enhance defense against deepfakes.

In the treatment of burn wounds, an accurate estimation of the ratio of the wound’s area to the total body surface area ( https://github.com/nganntk/BurnWounds .

The emergence of generative artificial intelligence (GenAI) has exacerbated the challenges of Misinformation, Disinformation, and Mal-information (MDM) within digital ecosystems. These multifaceted challenges demand a re-evaluation of the digital information lifecycle and a deep understanding of its social impact. An interdisciplinary strategy integrating insights from technology, social sciences, and policy analysis is crucial to address these issues effectively. This paper introduces a three-tiered framework to scrutinize the lifecycle of GenAI-driven content from creation to consumption, emphasizing the consumer perspective. We examine the dynamics of consumer behavior that drive interactions with MDM, pinpoints vulnerabilities in the information dissemination process, and advocates for adaptive, evidence-based policies. Our interdisciplinary methodology aims to bolster information integrity and fortify public trust, equipping digital societies to manage the complexities of GenAI and proactively address the evolving challenges of digital misinformation. We conclude by discussing how GenAI can be leveraged to combat MDM, thereby creating a reflective cycle of technological advancement and mitigation.

Continuous Authentication (CA) using behavioral biometrics is a type of biometric identification that recognizes individuals based on their unique behavioral characteristics. Many behavioral biometrics can be captured through multiple sensors, each providing multichannel time-series data. Utilizing this multichannel data effectively can enhance the accuracy of behavioral biometrics-based CA. This paper extends BehaveFormer, a new framework that effectively combines time series data from multiple sensors to provide higher security in behavioral biometrics. BehaveFormer includes two Spatio-Temporal Dual Attention Transformers (STDAT), a novel transformer we introduce to extract more discriminative features from multichannel time-series data. Experimental results on two behavioral biometrics, Keystroke Dynamics and Swipe Dynamics with Inertial Measurement Unit (IMU), have shown State-of-the-art performance. For Keystroke, on three publicly available datasets (Aalto DB, HMOG DB, and HuMIdb), BehaveFormer outperforms the SOTA. For instance, BehaveFormer achieved an EER of 2.95% on the HuMIdb. For Swipe, on two publicly available datasets (HuMIdb and FETA) BehaveFormer outperforms the SOTA, for instance, BehaveFormer achieved an EER of 3.67% on the HuMIdb. Additionally, the BehaveFormer model shows superior performance in various CA-specific evaluation metrics. The proposed STDAT-based BehaveFormer architecture can also be effectively used for transfer learning. The model weights and reproducible experimental results are available at: https://github.com/nganntk/BehaveFormer

We present Foice, a novel deepfake attack against voice authentication systems. Foice generates a synthetic voice of the victim from just a single image of the victim's face, without requiring any voice sample. This synthetic voice is realistic enough to fool commercial authentication systems. Since face images are generally easier to obtain than voice samples, Foice effectively makes it easier for an attacker to mount large-scale attacks. The key idea lies in learning the partial correlation between face and voice features, and adding to that a face-independent voice feature sampled from a Gaussian distribution. We demonstrate the effectiveness of Foice with a comprehensive set of real-world experiments involving ten offline participants and an online dataset of 1,029 unique individuals. By evaluating eight state-of-the-art systems, including WeChat's Voiceprint and Microsoft Azure, we show that all these systems are vulnerable to Foice attack.

Institute, Warsaw, Poland This paper describes the results of the IEEE BigData 2023 Keystroke Verification Challenge 1 (KVC), that considers the biometric verification performance of Keystroke Dynamics (KD), captured as tweet-long sequences of variable transcript text from over 185,000 subjects. The data are obtained from two of the largest public databases of KD up to date, the Aalto Desktop and Mobile Keystroke Databases, guaranteeing a minimum amount of data per subject, age and gender annotations, absence of corrupted data, and avoiding excessively unbalanced subject distributions with respect to the considered demographic attributes. Several neural architectures were proposed by the participants, leading to global Equal Error Rates (EERs) as low as 3.33% and 3.61% achieved by the best team respectively in the desktop and mobile scenario, outperforming the current state of the art biometric verification performance for KD. Hosted on CodaLab 2 , the KVC will be made ongoing to represent a useful tool for the research community to compare different approaches under the same experimental conditions and to deepen the knowledge of the field.

The recent renaissance in generative models, driven primarily by the advent of diffusion models and iterative improvement in GAN methods, has enabled many creative applications. However, each advancement is also accompanied by a rise in the potential for misuse. In the arena of the deepfake generation, this is a key societal issue. In particular, the ability to modify segments of videos using such generative techniques creates a new paradigm of deepfakes which are mostly real videos altered slightly to distort the truth. This paradigm has been under-explored by the current deepfake detection methods in the academic literature. In this paper, we present a deepfake detection method that can address this issue by performing deepfake prediction at the frame and video levels. To facilitate testing our method, we prepared a new benchmark dataset where videos have both real and fake frame sequences with very subtle transitions. We provide a benchmark on the proposed dataset with our detection method which utilizes the Vision Transformer based on Scaling and Shifting [38] to learn spatial features, and a Timeseries Transformer to learn temporal features of the videos to help facilitate the interpretation of possible deepfakes. Extensive experiments on a variety of deepfake generation methods show excellent results by the proposed method on temporal segmentation and classical video-level predictions as well. In particular, the paradigm we address will form a powerful tool for the moderation of deepfakes, where human oversight can be better targeted to the parts of videos suspected of being deepfakes. All experiments can be reproduced at: github.com/rgb91/temporal-deepfake-segmentation.

Accelerated progress in voice cloning technology is making phone scams easier and exposing potential threats to politicians. Previous spoofing detection technology focused more on the fully faked speech. In this work, we create a Partial Synthetic Detection (Psynd) dataset and propose a fake segments localization system of the partially faked speech. Psynd dataset is a multi-speaker English corpus of approximately 13 hours in total at 24kHz sampling rate read English speech injected with synthetic speech. The fake segments are generated by state-of-art multi-speaker text-to-speech models with high similarity to the real speech to be injected. Our fake segments localization system consists of 3 parts: acoustic feature extraction, classification and post-processing. Frame level CQCC features are extracted and forwarded to a spoofing-discriminant ANN to predict real or fake label sequence. Assuming that the fake or real segments in the partially faked speech cannot be shorter than the duration of a phoneme, the labels of extreme short fake or real segments are flipped. We use 1-D IoU to evaluate the localization performance and get the result of 98.58% during the test, much higher than a random guess of $\frac{1}{3}$. We also explore extreme cases like fully faked, fully real and multi-fake-segments speech and degraded partially faked audio. Some benchmark results are presented on this dataset and show that a more robust detector is needed.

In supervised multi-task learning, the choice of auxiliary tasks is usually decided manually. This immediately raises the issue of how to choose suitable tasks, which in turn requires the need to measure the similarity between tasks. In this paper, we propose a task-similarity metric that depends solely on the task labels, not on the machine learning models. We show that auxiliary tasks may be synthetically generated from the main tasks with any desired similarity. Labels for these virtual tasks are also generated. Finally, we show that learning these virtual tasks along with the main task leads to real performance gains.

Continuous Authentication (CA) is proposed as an alter-native authentication scheme for modern personal devices. Gait is a suitable biometric for CA due to its availability and low resource requirements. However, the drastic change in the gait pattern with changes in actions (such as walking, running or climbing stairs) and changes in terrain (such as walking on a flat surface or down an incline) makes it chal-lenging to deploy in a real-world CA system. We show that standard gait features are influenced by different actions. The gait pattern of an action is also in-fluenced by the actions performed before and immediately after. Therefore, gait features are usually not robust to these action variations. We propose action invariant gait features to address this robustness issue. Our proposed method learns action invariant gait features utilizing a Siamese Net-work architecture with triplet loss and a unique triplet mining protocol. Our evaluations highlight that our action in-variant features are robust to pre and post action impacts and real world action variations. These features allows for a CA system to be enrolled using a single action (walk) and be used across multiple different actions encountered throughout the day.