In recent years, face anti-spoofing (FAS) has made notable progress in multimodal fusion, cross-domain generalization, and interpretability. With the development of large language models and reinforcement learning (RL), strategy-based training paradigms offer new opportunities for jointly modeling multimodality, generalization, and interpretability. However, compared to unimodal reasoning, multimodal reasoning introduces more complex logic, such as accurate feature representation and cross-modal verification, which significantly increases reasoning complexity and labeling difficulty. Due to the lack of high-quality annotations in existing multimodal FAS datasets, directly applying RL strategies is sub-optimal, hindering robust multimodal reasoning. In this paper, we find two key issues of supervised fine-tuning combined with reinforcement learning (SFT+RL) paradigms in multimodal FAS reasoning: 1) limited multimodal reasoning paths not only hinder the full utilization of multimodal information but also constrain the model’s exploration space after SFT, thereby affecting the effectiveness of subsequent RL; and 2) the mismatch between single-task supervision and the diversity of multimodal reasoning paths leads to reasoning confusion, where models may exploit shortcuts by directly mapping input images to answers, bypassing the intended reasoning process. These issues further increase the complexity of multimodal reasoning and hinder the effective application of RL strategies. To address these challenges, we propose the PA-FAS framework with a reasoning path enhancement strategy for high-quality extended reasoning sequences construction based on limited annotated data to enrich the reasoning paths and alleviate exploration constraints. Additionally, we introduce an answer shuffling mechanism during SFT for comprehensive multimodal analysis rather than mining superficial cues, thus encouraging deeper reasoning and avoiding shortcut learning. Our method significantly improves multimodal reasoning accuracy and generalization, and successfully unifies multimodal fusion, cross-domain generalization, and interpretability towards trustworthy multimodal FAS.

Face anti-spoofing (FAS) is crucial for protecting facial recognition systems from presentation attacks. Previous methods approached this task as a classification problem, lacking interpretability and reasoning behind the predicted results. Recently, multimodal large language models (MLLMs) have shown strong capabilities in perception, reasoning, and decision-making in visual tasks. However, there is currently no universal and comprehensive MLLM and dataset specifically designed for FAS task. To address this gap, we propose FaceShield, a MLLM for FAS, along with the corresponding pre-training and supervised fine-tuning (SFT) datasets, FaceShield-pre10K and FaceShield-sft45K. FaceShield is capable of determining the authenticity of faces, identifying types of spoofing attacks, providing reasoning for its judgments, and detecting attack areas. Specifically, we employ spoof-aware vision perception (SAVP) that incorporates both the original image and auxiliary information based on prior knowledge. We then use an prompt-guided vision token masking (PVTM) strategy to random mask vision tokens, thereby improving the model's generalization ability. We conducted extensive experiments on three benchmark datasets, demonstrating that FaceShield significantly outperforms previous deep learning models and general MLLMs on four FAS tasks, i.e., coarse-grained classification, fine-grained classification, reasoning, and attack localization.

Large Language Models (LLMs) hold rich implicit knowledge and powerful transferability. In this paper, we explore the combination of LLMs with the human skeleton to perform action classification and description. However, when treating LLM as a recognizer, two questions arise: 1) How can LLMs understand the skeleton? 2) How can LLMs distinguish among actions? To address these problems, we introduce a novel paradigm named learning Skeleton representation with visual-motion knowledge for Action Recognition (SUGAR). In our pipeline, we first utilize off-the-shelf large-scale video models as a knowledge base to generate visual, motion information related to actions. Then, we propose to supervise skeleton learning through this prior knowledge to yield discrete representations. Finally, we use the LLM with untouched pre-training weights to understand these representations and generate the desired action targets and descriptions. Notably, we present a Temporal Query Projection (TQP) module to continuously model the skeleton signals with long sequences. Experiments on several skeleton-based action classification benchmarks demonstrate the efficacy of our SUGAR. Moreover, experiments on zero-shot scenarios show that SUGAR is more versatile than linear-based methods.

Can Multimodal Large Language Models (MLLMs) discern confused objects that are visually present but audio-absent? To study this, we introduce a new benchmark, AV-ConfuseBench, which simulates an “Audio-Visual Confusion” scene by modifying the corresponding sound of an object in the video, e.g., mute the sounding object and ask MLLMs “Is there a/an {muted-object} sound”. Experimental results reveal that MLLMs, such as Qwen2.5-Omni and Gemini 2.5, struggle to discriminate non-existent audio due to visually dominated reasoning. Motivated by this observation, we introduce RL-CoMM, a Reinforcement Learning-based Collaborative Multi-MLLM that is built upon the Qwen2.5-Omni foundation. RL-CoMM includes two stages: 1) To alleviate visually dominated ambiguities, we introduce an external model, a Large Audio Language Model (LALM), as the reference model to generate audio-only reasoning. Then, we design a Step-wise Reasoning Reward function that enables MLLMs to self-improve audio-visual reasoning with the audio-only reference. 2) To ensure an accurate answer prediction, we introduce Answer-centered Confidence Optimization to reduce the uncertainty of potential heterogeneous reasoning differences. Extensive experiments on audio-visual question answering and audio-visual hallucination show that RL-CoMM improves accuracy by 10~30% over the baseline model with limited training data.

Unified video and action prediction models hold great potential for robotic manipulation, as future observations offer contextual cues for planning, while actions reveal how interactions shape the environment. However, most existing approaches treat observation and action generation in a monolithic and goal-agnostic manner, often leading to semantically misaligned predictions and incoherent behaviors. To this end, we propose H-GAR, a Hierarchical interaction framework via Goal-driven observation-Action Refinement. To anchor prediction to the task objective, H-GAR first produces a goal observation and a coarse action sketch that outline a high-level route toward the goal. To enable explicit interaction between observation and action under the guidance of the goal observation for more coherent decision-making, we devise two synergistic modules. (1) Goal-Conditioned Observation Synthesizer (GOS) synthesizes intermediate observations based on the coarse-grained actions and the predicted goal observation. (2) Interaction-Aware Action Refiner (IAAR) refines coarse actions into fine-grained, goal-consistent actions by leveraging feedback from the intermediate observations and a Historical Action Memory Bank that encodes prior actions to ensure temporal consistency. By integrating goal grounding with explicit action-observation interaction in a coarse-to-fine manner, H-GAR enables more accurate manipulation. Extensive experiments on both simulation and real-world robotic manipulation tasks demonstrate that H-GAR achieves state-of-the-art performance.

Recent Audio-Visual Question Answering (AVQA) methods have advanced significantly.However, most AVQA methods lack effective mechanisms for handling missing modalities, suffering from severe performance degradation in real-world scenarios with data interruptions.Furthermore, prevailing methods for handling missing modalities predominantly rely on generative imputation to synthesize missing features.While partially effective, these methods tend to capture inter-modal commonalities but struggle to acquire unique, modalityspecific knowledge within the missing data, leading to hallucinations and compromised reasoning accuracy.To tackle these challenges, we propose R 2 ScP, a novel framework that shifts the paradigm of missing modality handling from traditional generative imputation to retrieval-based recovery.Specifically, we leverage cross-modal retrieval via unified semantic embeddings to acquire missing domain-specific knowledge.To maximize semantic restoration, we introduce a context-aware adaptive purification mechanism that eliminates latent semantic noise within the retrieved data.Additionally, we employ a two-stage training strategy to explicitly model the semantic relationships between knowledge from different sources.Extensive experiments demonstrate that R 2 ScP significantly improves AVQA and enhances robustness in modal-incomplete scenarios.1

Abstract Facial expression recognition (FER) has emerged as an important research topic in recent years. However, current FER paradigms face challenges in generalization, lack semantic information aligned with natural language, and struggle to process both images and videos within a unified framework. Multimodal Large Language Models (MLLMs) have recently achieved success, offering advantages in addressing these issues and potentially overcoming the limitations of current FER paradigms. Nonetheless, directly applying pre-trained MLLMs to FER remains challenging due to insufficient instruction datasets and the inability of vision encoders to extract fine-grained facial information. Our zero-shot evaluations of existing open-source MLLMs on FER reveal a significant performance gap compared to GPT-4V and state-of-the-art supervised methods. In this paper, we aim to enhance MLLMs’ capabilities in understanding facial expressions. We first introduce a facial expression recognition instruction dataset ( FERID ), which has 376k category instructions and 339k conversational instructions. We then propose a novel MLLM, named EMO-LLaMA , which incorporates facial priors from a pretrained facial analysis network to enhance its understanding of human facial information. Specifically, we design a Face Info Mining module to extract both global and local facial information. Furthermore, we utilize a handcrafted prompt to introduce age-gender-race attributes, considering the emotional differences across diverse human groups. Extensive experiments show that EMO-LLaMA achieves results comparable to or competitive with SOTA on both static and dynamic FER datasets. The instruction dataset and code will be available at https://github.com/xxtars/EMO-LLaMA .

Federated learning (FL) enables collaborative model training on decentralized Internet of Things (IoT) data while keeping raw data local, thereby mitigating privacy risks. In practice, however, IoT deployments suffer from two coupled challenges: strict uplink bandwidth constraints and long-tailed label distributions where rare events are most critical. Existing work treats these issues separately—communication-efficient FL often neglects data imbalance, whereas federated long-tail methods typically incur heavy communication overheads and privacy risks. We propose FELT, short forFederated Ensemble Learning for Long-Tailed IoT Data, a framework that jointly addresses communication efficiency, long-tail robustness, and privacy through a communication-efficient private voting protocol. Instead of transmitting full model updates, FELT devices function as teacher models, transmitting only single-integer votes on public queries. This architecture reduces communication traffic by up to one order of magnitude (approx. 10×) compared to standard FL methods. On the server, FELT aggregates votes with calibrated noise to ensure (ε, δ)-differential privacy, followed by a post-hoc class-prior-based calibration that improves tail-class predictions with negligible extra computational overhead on the server side. Experiments on multiple real-world IoT datasets demonstrate that FELT substantially boosts tail-class F1 scores (up to 26% improvement) under the same privacy budget. These results highlight FELT as a practical solution for communication-constrained and privacy-sensitive IoT applications.

In this work, we focus on a special group of human body language — themicro-gesture (MG), which differs from the range of ordinary illustrative gestures in that they are not intentional behaviors performed to convey information to others, but rather unintentional behaviors driven by inner feelings. This characteristic introduces two novel challenges regarding micro-gestures that are worth rethinking. The first is whether strategies designed for other action recognition are entirely applicable to micro-gestures. The second is whether micro-gestures, as supplementary data, can provide additional insights for emotional understanding. In recognizing micro-gestures, we explore various augmentation strategies that take into account the subtle spatial and brief temporal characteristics of micro-gestures, often accompanied by repetitiveness, to determine more suitable augmentation methods. Considering the significance of temporal domain information for micro-gestures, we introduce a simple and efficient spatiotemporal balancing fusion method. We not only study our method on the considered micro-gesture dataset but also conduct experiments on mainstream gesture/action datasets. The results show that our approach performs well in micro-gesture recognition and on other datasets, achieving state-of-the-art performance compared to previous micro-gesture recognition methods. For emotional understanding based on micro-gestures, we construct complex emotional reasoning scenarios. Our evaluation, conducted with large language models, shows that micro-gestures play a significant and positive role in enhancing comprehensive emotional understanding. We confirm that our new insights contribute to advancing research in micro-gesture and emotional artificial intelligence.

The rapid advancement of AI-generated content (AIGC) presents significant challenges for digital forensics, necessitating robust and generalizable detection frameworks. Existing detection methods primarily rely on visual feature extraction, while vision-language model-based approaches are limited to class-label prompts, failing to capture quality-related artifacts introduced by different generative models. To address this limitation, we introduce QAFD, a novel Quality-Assisted Forgery Detection framework that incorporates image quality information into the detection process. Specifically, we design a quality queried attention block to effectively fuse class-based content prompts with quality-aware text prompts. This integration enhances the model’s ability to capture semantic artifacts related to degradation patterns commonly associated with AI-generated images. Furthermore, we introduce the Quality-Guided Forgery Adapter (QGFA) to incorporate quality-aware textual cues into the visual domain, improving feature extraction for both spatial and frequency-based forgery artifacts. This synergy allows frequency cues to enhance low-level artifact perception, while quality-aware guidance strengthens high-level discriminative representation. Extensive experiments demonstrate that QAFD achieves superior generalization to unseen generative models over three datasets and significantly maintains its robustness against common image post-processing operations.The codes will be released at github.

Facial expressions (FEs) and action units (AUs) are facial emotional representations at different levels of granularity. In the past, recognizing them has often been treated as two separate tasks. There are also some methods that use the knowledge of one to aid in recognizing the other, but currently, unified models capable of recognizing both FEs and AUs simultaneously remain rare. In this paper, we construct a unified model with strong generalization capability to jointly perform facial expression recognition (FER) and action unit detection (AUD). Considering the extremely limited training samples annotated with both FEs and AUs, we introduce a large amount of unlabeled facial data from the wild. We carefully design category-specific confidence margins and leverage the correspondences between FEs and AUs to assign credible pseudo-labels to the unlabeled facial data. Furthermore, we incorporate semantically richer textual descriptions as supervision and refine them through visual perception, leveraging the inherent correlations between AUs and between FEs and AUs to enhance their precision. Extensive experiments demonstrate the superiority of the proposed method from various perspectives, including a unified zero-shot benchmark for exploring the model's comprehensive generalization capability to recognize facial emotional representations across multiple datasets, as well as within-domain and cross-domain evaluations after fine-tuning. The code for the proposed method is available at https://github.com/yuankaishen2001/MGFER.

Emotional understanding and generation are often treated as separate tasks, yet they are inherently complementary and can mutually enhance each other. In this paper, we propose the UniEmo, a unified framework that seamlessly integrates these two tasks. The key challenge lies in the abstract nature of emotions, necessitating the extraction of visual representations beneficial for both tasks. To address this, we propose a hierarchical emotional understanding chain with learnable expert queries that progressively extracts multi-scale emotional features, thereby serving as a foundational step for unification. Simultaneously, we fuse these expert queries and emotional representations to guide the diffusion model in generating emotion-evoking images. To enhance the diversity and fidelity of the generated emotional images, we further introduce the emotional correlation coefficient and emotional condition loss into the fusion process. This step facilitates fusion and alignment for emotional generation guided by the understanding. In turn, we demonstrate that joint training allows the generation component to provide implicit feedback to the understanding part. Furthermore, we propose a novel data filtering algorithm to select high-quality and diverse emotional images generated by the well-trained model, which explicitly feedback into the understanding part. Together, these generation-driven dual feedback processes enhance the model's understanding capacity. Extensive experiments show that UniEmo significantly outperforms state-of-the-art methods in both emotional understanding and generation tasks. The code for the proposed method is available at https://github.com/JiuTian-VL/UniEmo.

Low-dose computed tomography plays a crucial role in reducing radiation exposure in clinical imaging, however, the resultant noise significantly impacts image quality and diagnostic precision. Recent transformer-based models have demonstrated strong denoising capabilities but are often constrained by high computational complexity. To overcome these limitations, we propose AMFA-Net, an adaptive multi-order feature aggregation network that provides a lightweight architecture for enhancing high-resolution feature representation in low-dose CT imaging. AMFA-Net effectively integrates local and global contexts within high-resolution feature maps while learning discriminative representations through multi-order context aggregation. We introduce an agent-based self-attention cross-shaped window transformer block that efficiently captures global context in high-resolution feature maps, which is subsequently fused with backbone features to preserve critical structural information. Our approach employs multi-order gated aggregation to adaptively guide the network in capturing expressive interactions that may be overlooked in fused features, thereby producing robust representations for denoised image reconstruction. Experiments on two challenging public datasets with 25% and 10% full-dose CT image quality demonstrate that our method surpasses state-of-the-art approaches in denoising performance with low computational cost, highlighting its potential for real-time medical applications.

Camera-based photoplethysmography (cbPPG) represents a non-invasive technique for capturing physiological parameters through facial videos, enabling the extraction of vital signs such as heart rate, respiration rate, and blood oxygen saturation without direct physical contact. Existing deep learning methods face two core challenges when dealing with cbPPG: firstly, extracting weak PPG signals from video segments with large spatial and temporal redundancy and understanding their periodic patterns in long contexts; secondly, accurately extracting PPG signals in complex lighting environments, especially in low-light conditions. To address these issues, this paper proposes an end-to-end method based on Mamba, named HRMamba. This method employs temporal difference mamba to process temporal signals and combines bidirectional state space to enable Mamba to robustly understand the scene and learn the periodic patterns of PPG. Furthermore, a luminance post-processing module is designed to extract luminance information from the video without enhancing lighting or altering the original video data, and embed it into the PPG signal. Experimental results demonstrate that HRMamba achieves state-of-the-art performance, and the designed luminance post-processing module can be applied in various lighting environments, significantly enhancing the performance in dark environments without degrading the performance in normal light scenes.

Audio-Visual Learning (AVL) aims at the audio-visual perception with both audio and vision modalities. AVL also suffers from data insufficiency in many applications as with other unimodal tasks. Concurrently, AVL often needs to continuously learn over time rather than all knowledge simultaneously. Considering the above two perspectives, our work mainly focuses on benchmarking the unexplored Few-Shot Audio-Visual Class-Incremental Learning (FS-AVCIL), i.e., continually perceiving novel categories described by a limited number of labeled examples with audio and visual modalities. Firstly, we provide the detailed task configuration together with a thorough analysis of the challenges in FS-AVCIL: (1) how to efficiently learn and fuse multimodal information with limited labeled examples; and (2) how to alleviate catastrophic forgetting cross-modal semantic correlations with limited data. Then, we propose an efficient framework based on Vision Transformer to solve FS-AVCIL. This framework contains two parts: temporal-residual prompting for audio-visual synergy adapter and temporal prompt regularization. Specifically, temporal-residual prompting is incorporated into the audio-visual adapter to efficiently finetune the pre-trained foundation model with limited data and capture audio-visual correlation by learning temporal-relevant prompts. Besides, we regularize temporal-relevant prompts to memorize previous knowledge by fully using the temporal knowledge from various perspectives. This framework is validated in audio-visual classification tasks under the FS-AVCIL scenario, and extensive experiments demonstrate its superior performance.

Multimodal sentiment analysis, which learns a model to process multiple modalities simultaneously and predict a sentiment value, is an important area of affective computing. Modeling sequential intra-modal information and enhancing cross-modal interactions are crucial to multimodal sentiment analysis. In this paper, we propose MSAmba, a novel hybrid Mamba-based architecture for multimodal sentiment analysis, consisting of two core blocks: Intra-Modal Sequential Mamba (ISM) block and Cross-Modal Hybrid Mamba (CHM) block, to comprehensively address the above-mentioned challenges with hybrid state space models. Firstly, the ISM block models the sequential information within each modality in a bi-directional manner with the assistance of global information. Subsequently, the CHM blocks explicitly model centralized cross-modal interaction with a hybrid combination of Mamba and attention mechanism to facilitate information fusion across modalities. Finally, joint learning of the intra-modal tokens and cross-modal tokens is utilized to predict the sentiment values. This paper serves as one of the pioneering works to unravel the outstanding performances and great research potential of Mamba-based methods in the task of multimodal sentiment analysis. Experiments on CMU-MOSI, CMU-MOSEI and CH-SIMS demonstrate the superior performance of the proposed MSAmba over prior Transformer-based and CNN-based methods.

For efficient and high-fidelity local facial attribute editing, most existing editing methods either require additional fine-tuning for different editing effects or tend to affect beyond the editing regions. Alternatively, inpainting methods can edit the target image region while preserving external areas. However, current inpainting methods still suffer from the generation misalignment with facial attributes description and the loss of facial skin details. To address these challenges, (i) a novel data utilization strategy is introduced to construct datasets consisting of attribute-text-image triples from a data-driven perspective, (ii) a Causality-Aware Condition Adapter is proposed to enhance the contextual causality modeling of specific details, which encodes the skin details from the original image while preventing conflicts between these cues and textual conditions. In addition, a Skin Transition Frequency Guidance technique is introduced for the local modeling of contextual causality via sampling guidance driven by low-frequency alignment. Extensive quantitative and qualitative experiments demonstrate the effectiveness of our method in boosting both fidelity and editability for localized attribute editing. Our codes will be made publicly available.

Figure skating, known as the "Art on Ice," is among the most artistic sports, challenging to understand due to its blend of technical elements (like jumps and spins) and overall artistic expression. Existing figure skating datasets mainly focus on single tasks, such as action recognition or scoring, lacking comprehensive annotations for both technical and artistic evaluation. Current sports research is largely centered on ball games, with limited relevance to artistic sports like figure skating. To address this, we introduce FSAnno, a large-scale dataset advancing artistic sports understanding through figure skating. FSAnno includes an open-access training and test dataset, alongside a benchmark dataset, FSBench, for fair model evaluation. FSBench consists of FSBench-Text, with multiple-choice questions and explanations, and FSBench-Motion, containing multimodal data and Question and Answer (QA) pairs, supporting tasks from technical analysis to performance commentary. Initial tests on FSBench reveal significant limitations in existing models’ understanding of artistic sports. We hope FSBench will become a key tool for evaluating and enhancing model comprehension of figure skating. All data, models, and more details are available at: https://github.com/Moomin-Fin/Ano.

Multi-object images are prevalent in various real-world scenarios, including augmented reality, advertisement design, and medical imaging. Efficient and precise editing of these images is critical for these applications. With the advent of Stable Diffusion (SD), high-quality image generation and editing have entered a new era. However, existing methods often struggle to consider each object both individually and part of the whole image editing, both of which are crucial for ensuring consistent quantity perception, resulting in suboptimal perceptual performance. To address these challenges, we propose MoEdit, an auxiliaryfree multi-object image editing framework. MoEdit facilitates high-quality multi-object image editing in terms of style transfer, object reinvention, and background regeneration, while ensuring consistent quantity perception between inputs and outputs, even with a large number of objects. To achieve this, we introduce the Feature Compensation (FeCom) module, which ensures the distinction and separability of each object attribute by minimizing the in-between interlacing. Additionally, we present the Quantity Attention (QTTN) module, which perceives and preserves quantity consistency by effective control in editing, without relying on auxiliary tools. By leveraging the SD model, MoEdit enables customized preservation and modification of specific concepts in inputs with high quality. Experimental results demonstrate that our MoEdit achieves State-Of-The-Art (SOTA) performance in multi-object image editing. Data and codes are available at https://github.com/Tear-kitty/MoEdit.

This paper investigates the enhancement of reasoning capabilities in language models through token-level multi-model collaboration.Our approach selects the optimal tokens from the next token distributions provided by multiple models to perform autoregressive reasoning.Contrary to the assumption that more models yield better results, we introduce a distribution distance-based dynamic selection strategy (DDS) to optimize the multi-model collaboration process.To address the critical challenge of vocabulary misalignment in multi-model collaboration, we propose the concept of minimal complete semantic units (MCSU), which is simple yet enables multiple language models to achieve natural alignment within the linguistic space.Experimental results across various benchmarks demonstrate the superiority of our method.The code will be available at https://github.com/Fanye12/DDS.

Missing modalities are a common challenge in real-world multimodal learning scenarios, occurring during both training and testing. Existing methods for managing missing modalities often require the design of separate prompts for each modality or missing case, leading to complex designs and a substantial increase in the number of parameters to be learned. As the number of modalities grows, these methods become increasingly inefficient due to parameter redundancy. To address these issues, we propose Evidence-based Parameter-Efficient Prompting (EPE-P), a novel and parameter-efficient method for pretrained multimodal networks. Our approach introduces a streamlined design that integrates prompting information across different modalities, reducing complexity and mitigating redundant parameters. Furthermore, we propose an Evidence-based Loss function to better handle the uncertainty associated with missing modalities, improving the model’s decision-making. Our experiments demonstrate that EPE-P outperforms existing prompting-based methods in terms of both effectiveness and efficiency. The code is released at https://github.com/Boris-Jobs/EPE-P_MLLMs-Robustness.

Imperceptible adversarial attacks have recently attracted increasing research interests. Existing methods typically incorporate external modules or loss terms other than a simple lp-norm into the attack process to achieve imperceptibility, while we argue that such additional designs may not be necessary. In this paper, we rethink the essence of imperceptible attacks and propose two simple yet effective strategies to unleash the potential of PGD, the common and classical attack, for imperceptibility from an optimization perspective. Specifically, the Dynamic Step Size is introduced to find the optimal solution with minimal attack cost towards the decision boundary of the attacked model, and the Adaptive Early Stop strategy is adopted to reduce the redundant strength of adversarial perturbations to the minimum level. The proposed PGD-Imperceptible (PGD-Imp) attack achieves state-of-the-art results in imperceptible adversarial attacks for both untargeted and targeted scenarios. When performing untargeted attacks against ResNet-50, PGD-Imp attains 100% (+0.3%) ASR, 0.89 (-1.76) l2distance, and 52.93 (+9.2) PSNR with 57s (-371s) running time, significantly outperforming existing methods.

In the domain of facial recognition security, multimodal Face Anti-Spoofing (FAS) is essential for countering presentation attacks. However, existing technologies encounter challenges due to modality biases and imbalances, as well as domain shifts. Our research introduces a Mixture of Experts (MoE) model to address these issues effectively. We identified three limitations in traditional MoE approaches to multimodal FAS: (1) Coarse-grained experts’ inability to capture nuanced spoofing indicators; (2) Gated networks’ susceptibility to input noise affecting decision-making; (3) MoE’s sensitivity to prompt tokens leading to overfitting with conventional learning methods. To mitigate these, we propose the Bypass Isolated Gating MoE (BIG-MoE) framework, featuring: (1) Fine-grained experts for enhanced detection of subtle spoofing cues; (2) An isolation gating mechanism to counteract input noise; (3) A novel differential convolutional prompt bypass enriching the gating network with critical local features, thereby improving perceptual capabilities. Extensive experiments on four benchmark datasets demonstrate significant generalization performance improvement in multimodal FAS task. The code is released at https://github.com/murInJ/BIG-MoE.

Adversarial attacks present a critical challenge to deep neural networks' robustness, particularly in transfer scenarios across different model architectures. However, the transferability of adversarial attacks faces a fundamental dilemma between Exploitation (maximizing attack potency) and Exploration (enhancing cross-model generalization). Traditional momentum-based methods over-prioritize Exploitation, i.e., higher loss maxima for attack potency but weakened generalization (narrow loss surface). Conversely, recent methods with inner-iteration sampling over-prioritize Exploration, i.e., flatter loss surfaces for cross-model generalization but weakened attack potency (suboptimal local maxima). To resolve this dilemma, we propose a simple yet effective Gradient-Guided Sampling (GGS), which harmonizes both objectives through guiding sampling along the gradient ascent direction to improve both sampling efficiency and stability. Specifically, based on MI-FGSM, GGS introduces inner-iteration random sampling and guides the sampling direction using the gradient from the previous inner-iteration (the sampling's magnitude is determined by a random distribution). This mechanism encourages adversarial examples to reside in balanced regions with both flatness for cross-model generalization and higher local maxima for strong attack potency. Comprehensive experiments across multiple DNN architectures and multimodal large language models (MLLMs) demonstrate the superiority of our method over state-of-the-art transfer attacks. Code is made available at https://github.com/anuin-cat/GGS.

With the availability of diverse sensor modalities (i.e., RGB, Depth, Infrared) and the success of multi-modal learning, multi-modal face anti-spoofing (FAS) has emerged as a prominent research focus. The intuition behind it is that leveraging multiple modalities can uncover more intrinsic spoofing traces. However, this approach presents more risk of misalignment. We identify two main types of misalignment: (1) \textbf{Intra-domain modality misalignment}, where the importance of each modality varies across different attacks. For instance, certain modalities (e.g., Depth) may be non-defensive against specific attacks (e.g., 3D mask), indicating that each modality has unique strengths and weaknesses in countering particular attacks. Consequently, simple fusion strategies may fall short. (2) \textbf{Inter-domain modality misalignment}, where the introduction of additional modalities exacerbates domain shifts, potentially overshadowing the benefits of complementary fusion. To tackle (1), we propose a alignment module between modalities based on mutual information, which adaptively enhances favorable modalities while suppressing unfavorable ones. To address (2), we employ a dual alignment optimization method that aligns both sub-domain hyperplanes and modality angle margins, thereby mitigating domain gaps. Our method, dubbed \textbf{D}ual \textbf{A}lignment of \textbf{D}omain and \textbf{M}odality (DADM), achieves state-of-the-art performance in extensive experiments across four challenging protocols demonstrating its robustness in multi-modal domain generalization scenarios. The codes will be released soon.

Contrastive language-image pretraining (CLIP) has significantly advanced image-based vision learning. A pressing topic subsequently arises: how can we effectively adapt CLIP to the video domain? Recent studies have focused on adjusting either the textual or visual branch of CLIP for action recognition. However, we argue that adaptations of both branches are crucial. In this paper, we propose a **C**ontrastive **L**anguage-**A**ction **V**ideo Learn**er** (**CLAVER**), designed to shift CLIP's focus from the alignment of static visual objects and concrete nouns to the alignment of dynamic action behaviors and abstract verbs. Specifically, we introduce a novel Kronecker mask attention for temporal modeling. Our tailored Kronecker mask offers three benefits 1) it expands the temporal receptive field for each token, 2) it serves as an effective spatiotemporal heterogeneity inductive bias, mitigating the issue of spatiotemporal homogenization, and 3) it can be seamlessly plugged into transformer-based models. Regarding the textual branch, we leverage large language models to generate diverse, sentence-level and semantically rich interpretive prompts of actions, which shift the model's focus towards the verb comprehension. Extensive experiments on various benchmarks and learning scenarios demonstrate the superiority and generality of our approach. The code will be available soon.

Recapture detection is critical in forensic tasks, especially for authentication of face and document images in electronic Know Your Customer (e-KYC) processes. While deep learning has advanced face anti-spoofing (FAS) and document presentation attack detection (DPAD), weak forensic cues still hinder reliability. We propose the Forensicability Assessment Network (FANet) to assess sample forensicability and reject low-forensicability samples before recapture detection. This enhances the overall performance and reliability of e-KYC systems. FANet operates independently, without relying on real training data or actual forensic features, ensuring strong generalization across various scenarios. It combines image quality and forensic task cues, defining three forensicability classes based on domain knowledge. FANet is trained with cross-entropy loss, updating centers using a momentum-based approach. Experimental results demonstrate significant improvements in reliability by filtering out low-forensicability samples. This work introduces the first comprehensive approach to assessing forensicability, ensuring more reliable recapture detection in face and document images. The source codes are available at https://github.com/chenlewis/FANet.

Recently, 3D Gaussian Splatting (3DGS) has emerged as a prominent framework for novel view synthesis, providing high fidelity and rapid rendering speed. However, the substantial data volume of 3DGS and its attributes impede its practical utility, requiring compression techniques for reducing memory cost. Nevertheless, the unorganized shape of 3DGS leads to difficulties in compression. To formulate unstructured attributes into normative distribution, we propose a well-structured tri-plane to encode Gaussian attributes, leveraging the distribution of attributes for compression. To exploit the correlations among adjacent Gaussians, K-Nearest Neighbors (KNN) is used when decoding Gaussian distribution from the Tri-plane. We also introduce Gaussian position information as a prior of the position-sensitive decoder. Additionally, we incorporate an adaptive wavelet loss, aiming to focus on the high-frequency details as iterations increase. Our approach has achieved results that are comparable to or surpass that of SOTA 3D Gaussians Splatting compression work in extensive experiments across multiple datasets. The codes is available at TC-GS.

Facial Action Units (AUs) detection is a cornerstone of objective facial expression analysis and a critical focus in affective computing. Despite its importance, AU detection faces significant challenges, such as the high cost of AU annotation and the limited availability of datasets. These constraints often lead to overfitting in existing methods, resulting in substantial performance degradation when applied across diverse datasets. Addressing these issues is essential for improving the reliability and generalizability of AU detection methods. Moreover, many current approaches leverage Transformers for their effectiveness in long-context modeling, but they are hindered by the quadratic complexity of self-attention. Recently, Test-Time Training (TTT) layers have emerged as a promising solution for long-sequence modeling. Additionally, TTT applies self-supervised learning for iterative updates during both training and inference, offering a potential pathway to mitigate the generalization challenges inherent in AU detection tasks. In this paper, we propose a novel vision backbone tailored for AU detection, incorporating bidirectional TTT blocks, named AU-TTT. Our approach introduces TTT Linear to the AU detection task and optimizes image scanning mechanisms for enhanced performance. Additionally, we design an AU-specific Region of Interest (RoI) scanning mechanism to capture fine-grained facial features critical for AU detection. Experimental results demonstrate that our method achieves competitive performance in both within-domain and cross-domain scenarios.

The integration of pathologic images and genomic data for survival analysis has gained increasing attention with advances in multimodal learning. However, current methods often ignore biological characteristics, such as heterogeneity and sparsity, both within and across modalities, ultimately limiting their adaptability to clinical practice. To address these challenges, we propose AdaMHF: Adaptive Multimodal Hierarchical Fusion, a framework designed for efficient, comprehensive, and tailored feature extraction and fusion. AdaMHF is specifically adapted to the uniqueness of medical data, enabling accurate predictions with minimal resource consumption, even under challenging scenarios with missing modalities. Initially, AdaMHF employs an experts expansion and residual structure to activate specialized experts for extracting heterogeneous and sparse features. Extracted tokens undergo refinement via selection and aggregation, reducing the weight of non-dominant features while preserving comprehensive information. Subsequently, the encoded features are hierarchically fused, allowing multi-grained interactions across modalities to be captured. Furthermore, we introduce a survival prediction benchmark designed to resolve scenarios with missing modalities, mirroring real-world clinical conditions. Extensive experiments on TCGA datasets demonstrate that AdaMHF surpasses current state-of-the-art (SOTA) methods, showcasing exceptional performance in both complete and incomplete modality settings. Code is available in AdaMHF.

Multimodal news contains a wealth of information and is easily affected by deepfake modeling attacks. To combat the latest image and text generation methods, we present a new Multimodal Fake News Detection dataset (MFND) containing 11 manipulated types, designed to detect and localize highly authentic fake news. Furthermore, we propose a Shallow-Deep Multitask Learning (SDML) model for fake news, which fully uses unimodal and mutual modal features to mine the intrinsic semantics of news. Under shallow inference, we propose the momentum distillation-based light punishment contrastive learning for fine-grained uniform spatial image and text semantic alignment, and an adaptive cross-modal fusion module to enhance mutual modal features. Under deep inference, we design a two-branch framework to augment the image and text unimodal features, respectively merging with mutual modalities features, for four predictions via dedicated detection and localization projections. Experiments on both mainstream and our proposed datasets demonstrate the superiority of the model. Codes and dataset are released at https://github.com/yunan-wang33/sdml.

In dyadic interactions, facial reactions are crucial for conveying an individuals' responses to their conversational partners. Individuals may exhibit varied but appropriate facial reactions (AFRs) when perceiving the same behavioral expression. Although some recent methods can already respond multiple appropriate facial reactions to the given human speaker behaviors, the AFRs generated by these methods often fail to adequately preserve crucial head motions, leading to visual jitter and unnatural transitions between generated AFR segments. In this paper, we propose a novel and generic PFLPosNet framework which addresses the aforementioned problems at both pre-processing and post-processing stages, where a new pose-aware face behavior localization method PFL is introduced to retain the head pose displacement information from the source data. In addition, the framework proposes a real-time head pose adjustment method, PosNet, to ensure continuity and smoothness in the visual output of the model when using data with correct head pose displacement. Experimental results demonstrate that our approach not only generates more coherent and natural facial reaction sequences but also significantly outperforms existing online MAFRG methods in terms of continuity and smoothness. Our code is made available at https://github.com/rainforcetime/PFLPosNet.

Generating multiple appropriate facial reactions (MAFR) is essential for effective human-agent interaction. However, existing methods typically do not jointly model both local and global emotional cues, and neglect the temporal dynamics of facial expressions, leading to emotionally inconsistent and less natural reactions. In this work, we combine local and global emotional features to form a more comprehensive emotional representation. Our method further introduces motion-aware visual features that capture the dynamic evolution of facial expressions beyond static frames. By integrating both appearance and motion information within a structured generative framework, our approach enables more context-aware and temporally natural listener reactions. Experimental results demonstrate that our method outperforms existing approaches in both reaction diversity and appropriateness, which ranked first in the React 2025 challenge offline track.The implementation code can be accessed at:https://github.com/mtv-2025-react/mtv-2025-2025.git.

Recent advances in face forgery detection have shown strong in-domain performance but often fail to generalize to out-of-distribution data, especially when confronted with unseen manipulation techniques or domain shifts (e.g., lighting conditions, camera noise). We propose a novel Mixture-of-Experts framework, termed GM-DF, that decouples domain-specific and domain-invariant features to tackle cross-domain face forgery detection. Our method builds upon a foundation model (CLIP) and incorporates three key modules: (1) Dataset-Embedding Generator that leverages lightweight expert layers and database-aware feature normalization to adaptively modulate features at a per-domain level, capturing idiosyncratic cues without overfitting; (2) Multi-Dataset Representation mechanism that fuses these expert embeddings using scaled dot-product attention and integrates a mask image modeling (MIM) task to amplify local forgery artifacts; (3) Meta-Domain-Embedding Optimizer, inspired by MAML, which alternates between domain-specific (inner-loop) and domain-invariant (outer-loop) updates to facilitate rapid adaptation on new domains. Additionally, inspired by [13] (Yossi Gandelsman, Alexei A Efros, and Jacob Steinhardt. 2024. Interpreting the second-order effects of neurons in clip. arXiv preprint arXiv:2406.04341 (2024)) we introduce second-order feature propagation in the intermediate layers of CLIP to enhance fine-grained artifact cues and propose domain-class disentangled prompts to flexibly encode multi-domain text representations. Together, these strategies enable GM-DF to learn robust, shared forgery cues while preserving essential domain nuances. Our extensive experiments on multiple cross-domain benchmarks demonstrate that GM-DF significantly outperforms state-of-the-art approaches in both detection accuracy and domain transferability, reducing reliance on superficial artifacts and improving generalization to unseen forgeries. Importantly, our design requires minimal overhead beyond standard CLIP, making GM-DF both effective and computationally efficient for real-world face forgery detection.

In current web environment, fake news spreads rapidly across online social networks, posing serious threats to society. Existing multimodal fake news detection methods can generally be classified into knowledge-based and semantic-based approaches. However, these methods are heavily rely on human expertise and feedback, lacking flexibility. To address this challenge, we propose a Dynamic Analysis and Adaptive Discriminator (DAAD) approach for fake news detection. For knowledge-based methods, we introduce the Monte Carlo Tree Search algorithm to leverage the self-reflective capabilities of large language models (LLMs) for prompt optimization, providing richer, domain-specific details and guidance to the LLMs, while enabling more flexible integration of LLM comment on news content. For semantic-based methods, we define four typical deceit patterns: emotional exaggeration, logical inconsistency, image manipulation, and semantic inconsistency, to reveal the mechanisms behind fake news creation. To detect these patterns, we carefully design four discriminators and expand them in depth and breadth, using the soft-routing mechanism to explore optimal detection models. Experimental results on three real-world datasets demonstrate the superiority of our approach.

Current affective computing paradigms often treat emotional understanding and generation as separate tasks, yet they inherently possess symbiotic potential for mutual enhancement. In this paper, we aim to bridge the gap by developing a unified framework. The primary challenge lies in the extraction of precise and semantically rich representations of abstract emotions, which are crucial for both tasks. To address this, we harness the Chain-of-Thought reasoning at the latent space of multimodal large language models and propose EmoSym, a unified framework built upon this advanced foundation. Our framework is executed through three key steps: 1) Emotional reasoning knowledge compression. To enable efficient transfer of emotional reasoning priors, we design specialized reasoning tokens to compact emotion-aware contexts from external reasoning knowledge bases into latent representations. 2) Verifiable reinforcement reasoning optimization. To ensure more reliable and consistent emotional reasoning, we develop a verifiable reinforcement learning paradigm to further enhance the reasoning token by emotion-specific verifiable reward signals. Processed through the above two steps, the reasoning token simultaneously enhances emotional understanding while enriching semantic representations, benefiting subsequent emotional generation tasks. 3) Reasoning-augmented generation and online feedback. We then fuse it with emotional representations and feed them into a diffusion model to generate emotion-evoking images. Additionally, to create a generative-to-understanding enhancement feedback, we propose an Online Emotional Memory Bank (OEMB). It leverages newly generated images to progressively update the training dataset in the training process to reinforce understanding. Extensive experiments demonstrate the superior capabilities of our framework in both emotional understanding and generation tasks.

In this paper, we propose BeamLLM, a vision-empowered millimeter-wave (mmWave) beam prediction framework leveraging large language models (LLMs) to enhance the accuracy and robustness of beam prediction. By integrating computer vision (CV) with LLMs’ cross-modal reasoning capabilities, the framework extracts user equipment (UE) positional features from RGB images and aligns visual-temporal features with LLMs’ semantic space through reprogramming techniques. Evaluated on a realistic vehicle-to-infrastructure (V2I) scenario, BeamLLM achieves 61.01% top-1 accuracy and 97.39% top-3 accuracy in standard prediction tasks, outperforming traditional deep learning models. In few-shot prediction scenarios, performance degradation is limited to 12.56% (top-1) and 5.55% (top3) from time sample 1 to 10, demonstrating superior prediction capability.

Multimodal omics provide deeper insight into the biological processes and cellular functions, especially transcriptomics and proteomics. Computational methods have been proposed for the integration of single-cell multimodal omics of transcriptomics and proteomics. However, existing methods primarily concentrate on the alignment of different omics, overlooking the unique information inherent in each omics type. Moreover, as the majority of single-cell cohorts only encompass one omics, it becomes critical to transfer the knowledge learnt from multimodal omics to enhance unimodal omics analysis. Therefore, we proposed a novel framework that leverages masked autoencoder with cross-attention mechanism, called scMMAE (single-cell multimodal masked autoencoder), to fuse multimodal omics and enhance unimodal omics analysis. scMMAE simultaneously captures both the shared features and the distinctive information of two single-cell omics modalities and transfers the knowledge to enhance single-cell transcriptome data. Comparative evaluations against benchmarking methods across various cohorts revealed a notable improvement, with an increase of up to 21% in the adjusted Rand index and up to 12% in normalized mutual information in the context of multimodal fusion. In the realm of unimodal omics, scMMAE demonstrated an overall enhancement of approximately 20% in the adjusted Rand index and nearly 10% in normalized mutual information. Other nine metrics, including the Fowlkes-Mallows index and silhouette coefficient, further underscored the high performance of scMMAE. Significantly, scMMAE exhibits an elevated level of proficiency in distinguishing between different cell types, particularly on CD4 and CD8 T cells. Availability and implementation: scMMAE source code at https://github.com/DM0815/scMMAE/.

Abstract Personalized content synthesis technologies based on diffusion models have achieved significant breakthroughs, allowing for the creation of specific images from just a few reference photos. However, when these technologies are used to create fake news or unsettling content targeting individuals, they pose a substantial risk to society. To address this issue, current methods generate adversarial samples by adversarially maximizing the training loss, thus disrupting any personalized generation model trained with these samples. However, these methods are not efficient and do not fully consider the intrinsic mechanisms of successful personalization attacks. In this paper, we introduce an innovative Disruptive Text-Image Alignment (DTIA) framework. Based on the analysis that existing methods succeed in their attacks due to an overfitting of text to noise, which results in inaccurate face information matching, we have designed a Text-Image Mis-Match Attack framework. This framework aims to disrupt the model’s learning of associations between input faces and specific texts, thereby reducing unnecessary computational load. We also studied how the choice of timestep in diffusion models affects adversarial attacks and proposed a step schedule strategy to enhance algorithm efficiency. Extensive experiments on facial benchmarks have demonstrated that our DTIA framework not only disrupts personalized generation models but also significantly improves model efficiency.

Face Anti-Spoofing (FAS) is constantly challenged by new attack types and mediums, and thus it is crucial for a FAS model to not only mitigate Catastrophic Forgetting (CF) of previously learned spoofing knowledge on the training data during continual learning but also enhance the model's generalization ability to potential spoofing attacks. In this paper, we first highlight that current strategies for catastrophic forgetting are not well-suited to the imperceptible nature of spoofing information in FAS and lack the focus on improving generalization capability. Then, the instance-wise dynamic central difference convolutional adapter module with the weighted ensemble strategy for Vision Transformer (ViT) is proposed for efficiently fine-tuning with low-shot data by extracting generalized spoofing texture information. Furthermore, we find that catastrophic forgetting in FAS can be reflected through the inconsistent attention matrices of ViT between different continual sessions, as the attention matrices embody relationships of spoofing clues between different patch tokens. Hence, we introduce attention consistency regularization by learning and reusing attention matrices to alleviate catastrophic forgetting. Finally, we devise new protocols and conduct extensive experiments to validate the superior performance of alleviating catastrophic forgetting and generalization on unseen domains.

Face Anti-Spoofing (FAS) is essential for securing face recognition systems against presentation attacks. Recent advances in sensor technology and multimodal learning have enabled the development of multimodal FAS systems. However, existing methods often struggle to generalize to unseen attacks and diverse environments due to two key challenges: (1) Modality unreliability, where sensors such as depth and infrared suffer from severe domain shifts, impairing the reliability of cross-modal fusion; and (2) Modality imbalance, where over-reliance on a dominant modality weakens the model's robustness against attacks that affect other modalities. To overcome these issues, we propose MMDG++, a multimodal domain-generalized FAS framework built upon the vision-language model CLIP. In MMDG++, we design the Uncertainty-Guided Cross-Adapter++ (U-Adapter++) to filter out unreliable regions within each modality, enabling more reliable multimodal interactions. Additionally, we introduce Rebalanced Modality Gradient Modulation (ReGrad) for adaptive gradient modulation to balance modality convergence. To further enhance generalization, propose Asymmetric Domain Prompts (ADPs) that leverage CLIP's language priors to learn generalized decision boundaries across modalities. We also develop a novel multimodal FAS benchmark to evaluate generalizability under various deployment conditions. Extensive experiments across this benchmark show our method outperforms state-of-the-art FAS methods, demonstrating superior generalization capability.

Multimodal Large Language Models (MLLMs) have gained significant attention due to their rich internal implicit knowledge for cross-modal learning. Although advances in bringing audio-visuals into LLMs have resulted in boosts for a variety of Audio-Visual Question Answering (AVQA) tasks, they still face two crucial challenges: 1) audio-visual ambiguity, and 2) audio-visual hallucination. Existing MLLMs can respond to audio-visual content, yet sometimes fail to describe specific objects due to the ambiguity or hallucination of responses. To overcome the two aforementioned issues, we introduce the CAT+, which enhances MLLM to ensure more robust multimodal understanding. We first propose the Sequential Question-guided Module (SQM), which combines tiny transformer layers and cascades Q-Formers to realize a solid audio-visual grounding. After feature alignment and high-quality instruction tuning, we introduce Ambiguity Scoring Direct Preference Optimization (AS-DPO) to correct the problem of CAT+ bias toward ambiguous descriptions. To explore the hallucinatory deficits of MLLMs in dynamic audio-visual scenes, we build a new Audio-visual Hallucination Benchmark, named AVHbench. This benchmark detects the extent of MLLM's hallucinations across three different protocols in the perceptual object, counting, and holistic description tasks. Extensive experiments across video-based understanding, open-ended, and close-ended AVQA demonstrate the superior performance of our method. The AVHbench is released at https://github.com/rikeilong/Bay-CAT.

Facial Action Unit (AU) detection is a crucial task in affective computing and social robotics as it helps to identify emotions expressed through facial expressions. Anatomically, there are innumerable correlations between AUs, which contain rich information and are vital for AU detection. Previous methods used fixed AU correlations based on expert experience or statistical rules on specific benchmarks, but it is challenging to comprehensively reflect complex correlations between AUs via hand-crafted settings. There are alternative methods that employ a fully connected graph to learn these dependencies exhaustively. However, these approaches can result in a computational explosion and high dependency with a large dataset. To address these challenges, this paper proposes a novel self-adjusting AU-correlation learning (SACL) method with less computation for AU detection. This method adaptively learns and updates AU correlation graphs by efficiently leveraging the characteristics of different levels of AU motion and emotion representation information extracted in different stages of the network. Moreover, this paper explores the role of multi-scale learning in correlation information extraction, and design a simple yet effective multi-scale feature learning (MSFL) method to promote better performance in AU detection. By integrating AU correlation information with multi-scale features, the proposed method obtains a more robust feature representation for the final AU detection. Extensive experiments show that the proposed method outperforms the state-of-the-art methods on widely used AU detection benchmark datasets, with only 28.7% and 12.0% of the parameters and FLOPs of the best method, respectively.

Recently, vision transformer based multimodal learning methods have been proposed to improve the robustness of face anti-spoofing (FAS) systems. However, multimodal face data collected from the real world is often imperfect due to missing modalities from various imaging sensors. Recently, flexible-modal FAS (Yu et al. 2023) has attracted more attention, which aims to develop a unified multimodal FAS model using complete multimodal face data but is insensitive to test-time missing modalities. In this paper, we tackle one main challenge in flexible-modal FAS, i.e., when missing modality occurs either during training or testing in real-world situations. Inspired by the recent success of the prompt learning in language models, we proposeVisualPrompt flexible-modalFAS(VP-FAS), which learns the modal-relevant prompts to adapt the frozen pre-trained foundation model to downstream flexible-modal FAS task. Specifically, both vanilla visual prompts and residual contextual prompts are plugged into multimodal transformers to handle general missing-modality cases, while only requiring less than 4% learnable parameters compared to training the entire model. Furthermore, missing-modality regularization is proposed to force models to learn consistent multimodal feature embeddings when missing partial modalities. Extensive experiments conducted on two multimodal FAS benchmark datasets demonstrate the effectiveness of our VP-FAS framework that improves the performance under various missing-modality cases while alleviating the requirement of heavy model re-training.

Radar sensing is gaining increasing attention due to its unique advantages, including being device-free, privacy-preserving, and capable of penetrating obstacles. It has been extensively studied in various applications such as human activity recognition, vital sign monitoring, and person identification. However, most existing research focuses on a single specific application, and there remains a lack of studies or datasets dedicated to multi-task radar sensing. In this paper, we collected a dataset for two sensing tasks, including gesture recognition and person identification, via a miniature mm-wave radar. The raw radar signals were processed using micro-Doppler and range-Doppler techniques to extract spectral and spatial representations. We propose an improved multi-task radar sensing framework (MT-DualFormer) that incorporates attention-based cross-task feature distillation and contrastive learning to maximize task performance. MT-DualFormer consists of dual branches with CNN and Transformer modules, capturing both spatial and temporal dependencies in radar data. Attention-based cross-task feature distillation enables knowledge transfer between gesture recognition and person identification tasks. Meanwhile, contrastive learning ensures embedding space separability, facilitating robust task-specific classification. In the evaluation, MT-DualFormer achieves accuracy rates of 98.87% for gesture recognition and 97.96% for person identification, surpassing five representative multi-task approaches and ten state-of-the-art models. This study underscores the importance of leveraging task correlations to enhance the performance of radar-based sensing systems.

The rapid development of photo-realistic face generation methods has raised significant concerns in society and academia, highlighting the urgent need for robust and generalizable face forgery detection (FFD) techniques. Although existing approaches mainly capture face forgery patterns using image modality, other modalities like fine-grained noises and texts are not fully explored, which limits the generalization capability of the model. In addition, most FFD methods tend to identify facial images generated by GAN, but struggle to detect unseen diffusion-synthesized ones. To address the limitations, we aim to leverage the cutting-edge foundation model, contrastive language-image pre-training (CLIP), to achieve generalizable diffusion face forgery detection (DFFD). In this paper, we propose a novel multi-modal fine-grained CLIP (MFCLIP) model, which mines comprehensive and fine-grained forgery traces across image-noise modalities via language-guided face forgery representation learning, to facilitate the advancement of DFFD. Specifically, we devise a fine-grained language encoder (FLE) that extracts fine global language features from hierarchical text prompts. We design a multi-modal vision encoder (MVE) to capture global image forgery embeddings as well as fine-grained noise forgery patterns extracted from the richest patch, and integrate them to mine general visual forgery traces. Moreover, we build an innovative plug-and-play sample pair attention (SPA) method to emphasize relevant negative pairs and suppress irrelevant ones, allowing cross-modality sample pairs to conduct more flexible alignment. Extensive experiments and visualizations show that our model outperforms the state of the arts on different settings like cross-generator, cross-forgery, and cross-dataset evaluations. Our code will be available at https://github.com/Jenine-321/MFCLIP.

Camouflaged object detection (COD) and salient object detection (SOD) are two distinct yet closely-related computer vision tasks widely studied during the past decades. Though sharing the same purpose of segmenting an image into binary foreground and background regions, their distinction lies in the fact that COD focuses on concealed objects hidden in the image, while SOD concentrates on the most prominent objects in the image. Building universal segmentation models is currently a hot topic in the community. Previous works achieved good performance on certain task by stacking various hand-designed modules and multi-scale features. However, these careful task-specific designs also make them lose their potential as general-purpose architectures. Therefore, we hope to build general architectures that can be applied to both tasks. In this work, we propose a simple yet effective network (SENet) based on vision Transformer (ViT), by employing a simple design of an asymmetric ViT-based encoder-decoder structure, we yield competitive results on both tasks, exhibiting greater versatility than meticulously crafted ones. To enhance the performance of universal architectures on both tasks, we propose some general methods targeting some common difficulties of the two tasks. First, we use image reconstruction as an auxiliary task during training to increase the difficulty of training, forcing the network to have a better perception of the image as a whole to help with segmentation tasks. In addition, we propose a local information capture module (LICM) to make up for the limitations of the patch-level attention mechanism in pixel-level COD and SOD tasks and a dynamic weighted loss (DW loss) to solve the problem that small target samples are more difficult to locate and segment in both tasks. Finally, we also conduct a preliminary exploration of joint training, trying to use one model to complete two tasks simultaneously. Extensive experiments on multiple benchmark datasets demonstrate the effectiveness of our method. The code is available at https://github.com/linuxsino/SENet.

Attenuation compensation (AC), while being beneficial for visual-interpretation tasks in myocardial perfusion imaging (MPI) by SPECT, typically requires the availability of a separate X-ray CT component, leading to additional radiation dose, higher costs, and potentially inaccurate diagnosis due to SPECT/CT misalignment. To address these issues, we developed a method for cardiac SPECT AC using deep learning and emission scatter-window photons without a separate transmission scan (CTLESS). In this method, an estimated attenuation map reconstructed from scatter-energy window projections is segmented into different regions using a multi-channel input multi-decoder network trained on CT scans. Pre-defined attenuation coefficients are assigned to these regions, yielding the attenuation map used for AC. We objectively evaluated this method in a retrospective study with anonymized clinical SPECT/CT stress MPI images on the clinical task of detecting defects with an anthropomorphic model observer. CTLESS yielded statistically non-inferior performance compared to a CT-based AC (CTAC) method and significantly outperformed a non-AC (NAC) method on this clinical task. Similar results were observed in stratified analyses with different sexes, defect extents and severities. The method was observed to generalize across two SPECT scanners, each with a different camera. In addition, CTLESS yielded similar performance as CTAC and outperformed NAC method on the metrics of root mean squared error and structural similarity index measure. Moreover, as we reduced the training dataset size, CTLESS yielded relatively stable AUC values and generally outperformed another DL-based AC method that directly estimated the attenuation coefficient within each voxel. These results demonstrate the capability of the CTLESS method for transmission-less AC in SPECT and motivate further clinical evaluation.

Visible-infrared person re-identification is a challenging task in video surveillance. Most existing works achieve performance gains by aligning feature distributions or image styles across modalities, whereas the multi-granularity information and domain knowledge are usually neglected. Motivated by these issues, we propose a novel modality-aware domain alignment network (MDANet) for visible-infrared person re-identification (VI-ReID), which utilizes global-local context cues and the generalized domain alignment strategy to solve modal differences and poor generalization. Firstly, modality-aware global-local context attention (MGLCA) is proposed to obtain multi-granularity context features and identity-aware patterns. Secondly, we present a generalized domain alignment learning head (GDALH) to relieve the modality discrepancy and enhance the generalization of MDANet, whose core idea is to enrich feature diversity in the domain alignment procedure. Finally, the entire network model is trained by proposing cross-modality circle, classification, and domain alignment losses in an end-to-end fashion. We conduct comprehensive experiments on two standards and their corrupted VI-ReID datasets to validate the robustness and generalization of our approach. MDANet is obviously superior to the most state-of-the-art methods. Specifically, the proposed method can gain 8.86% and 2.50% in Rank-1 accuracy on SYSU-MM01 (all-search and single-shot mode) and RegDB (infrared to visible mode) datasets, respectively. The source code will be made available soon.

The goal of Few-Shot Continual Learning (FSCL) is to incrementally learn novel tasks with limited labeled samples and preserve previous capabilities simultaneously. However, current FSCL works lack research on domain increment and domain generalization ability, which cannot cope with changes in the visual perception environment. In this paper, we set up a Generalized FSCL (GFSCL) protocol involving both class- and domain-incremental scenarios together with domain generalization assessment. Firstly, two benchmark datasets and protocols are newly arranged, and detailed baselines are provided for this unexplored configuration. Furthermore, we find that common continual learning methods have poor generalization ability on unseen domains and cannot better tackle catastrophic forgetting issue in cross-incremental tasks. Hence, we propose a rehearsal-free framework based on Vision Transformer (ViT) named Contrastive Mixture of Adapters (CMoA). It contains two non-conflicting parts: (1) By applying the fast-adaptation characteristic of adapter-embedded ViT, the mixture of Adapters (MoA) module is incorporated into ViT. For stability purpose, cosine similarity regularization and dynamic weighting are designed to make each adapter learn specific knowledge and concentrate on particular classes. (2) To further enhance domain generalization ability, we alleviate the intra-class variation by prototype-calibrated contrastive learning to improve domain-invariant representation learning. Finally, six evaluation indicators showing the overall performance and forgetting are compared by comprehensive experiments on two benchmark datasets to validate the efficacy of CMoA, and the results illustrate that CMoA can achieve comparative performance with rehearsal-based continual learning methods.

Visible-infrared person re-identification (VI-ReID) presents significant challenges on account of the substantial cross-modality gap and intra-class variations. Most existing methods primarily concentrate on aligning cross-modality at the feature or image levels and training with an equal number of samples from different modalities. However, in the real world, there exists an issue of modality imbalance between visible and infrared data. Besides, imbalanced samples between train and test impact the robustness and generalization of the VI-ReID. To alleviate this problem, we propose a dual-path imbalanced feature compensation network (DICNet) for VI-ReID, which provides equal opportunities for each modality to learn inconsistent information from different identities of others, enhancing identity discrimination performance and generalization. First, a modality consistency perception (MCP) module is designed to assist the backbone focus on spatial and channel information, extracting diverse and salient features to enhance feature representation. Second, we propose a cross-modality features re-assignment strategy to simulate modality imbalance by grouping and re-organizing the cross-modality features. Third, we perform bidirectional heterogeneous cooperative compensation with cross-modality imbalanced feature interaction modules (CIFIMs), allowing our network to explore the identity-aware patterns from imbalanced features of multiple groups for cross-modality interaction and fusion. Further, we design a feature re-construction difference loss to reduce cross-modality discrepancy and enrich feature diversity within each modality. Extensive experiments on three mainstream datasets show the superiority of the DICNet. Additionally, competitive results in corrupted scenarios verify its generalization and robustness.

For Multimodal Sentiment Analysis (MSA), previous methods concentrate on designing sophisticated fusion strategies and performing representation learning across heterogeneous modalities, aiming to leverage multimodal signals to detect human sentiment. However, these approaches fail to address the long-standing issue of corrupted modal details in videos, which may be caused by the challenge of the excessive loss of emotionally relevant semantics resulted from the degradation of detailed information. In this work, we aim to improve the robustness capacity of resisting corruption in MSA, by introducing a Hierarchical Frequency Restoration and Adaptive Modality Enforcement (HFR-AME) approach. The HFR-AME progressively recovers blurred detailed cues in each modality while enhancing the discriminative power of modal representations. Specifically, to reconstruct distinct frequency band features, we propose to equip the HFR module with a key component called the Frequency Multimodal UNet (FM-UNet), so as to utilize complementary modal features as conditions. This meticulous restoration process, performed from low to high frequency, facilitates the comprehensive recovery of intricate details. Meanwhile, to adaptively integrate these diverse frequency features, we introduce the AME module to enhance the beneficial modal frequencies while suppressing irrelevant ones, with the goal of strengthening the restored modal representations. Extensive experiments show our HFR-AME outperforms state-of-the-art methods on the CMU-MOSI and CMU-MOSEI datasets, improving 7-class accuracy by 0.5% and 0.6%, respectively. Further analysis also confirms its cross-lingual generalization and competitive computational efficiency. Our code is made available at https://github.com/nianhua20/HFR-AME .

Face Anti-Spoofing (FAS) is crucial for securing face recognition systems against presentation attacks. With ad-vancements in sensor manufacture and multi-modal learning techniques, many multi-modal FAS approaches have emerged. However, they face challenges in generalizing to unseen attacks and deployment conditions. These chal-lenges arise from (1) modality unreliability, where some modality sensors like depth and infrared undergo signifi-cant domain shifts in varying environments, leading to the spread of unreliable information during cross-modal feature fusion, and (2) modality imbalance, where training overly relies on a dominant modality hinders the conver-gence of others, reducing effectiveness against attack types that are indistinguishable by sorely using the dominant modality. To address modality unreliability, we propose the Uncertainty-Guided Cross-Adapter (U-Adapter) to recognize unreliably detected regions within each modality and suppress the impact of unreliable regions on other modal-ities. For modality imbalance, we propose a Rebalanced Modality Gradient Modulation (ReGrad) strategy to rebal-ance the convergence speed of all modalities by adaptively adjusting their gradients. Besides, we provide the first large-scale benchmark for evaluating multi-modal FAS per-formance under domain generalization scenarios. Exten-sive experiments demonstrate that our method outperforms state-of-the-art methods. Source codes and protocols are released on https://github.com/OMGGGGG/mmdg.

Face anti-spoofing techniques based on domain generalization have recently been studied widely. Adversarial learning and meta-learning techniques have been adopted to learn domain-invariant representations. However, prior approaches often consider the dataset gap as the primary factor behind domain shifts. This perspective is not fine-grained enough to reflect the intrinsic gap among the data accurately. In our work, we redefine domains based on identities rather than datasets, aiming to disentangle liveness and identity attributes. We emphasize ignoring the adverse effect of identity shift, focusing on learning identity-invariant liveness representations through orthogonalizing liveness and identity features. To cope with style shifts, we propose Style Cross module to expand the stylistic diversity and Channel-wise Style Attention module to weaken the sensitivity to style shifts, aiming to learn robust liveness representations. Furthermore, acknowledging the asymmetry between live and spoof samples, we introduce a novel contrastive loss, Asymmetric Augmented Instance Contrast. Extensive experiments on four public datasets demonstrate that our method achieves state-of-the-art performance under cross-dataset and limited source dataset scenarios. Additionally, our method has good scalability when expanding diversity of identities. Code is available at https://github.com/yjyddq/DLIF.

Deception detection within audio-visual modalities is vital across diverse sectors, notably in customs security and multimedia anti-fraud. However, this notable efficacy is lost by the necessity to train and deploy separate models for each conceivable modality scenario, leading to redundancy and inefficiency. Moreover, real-world environments where multi-modal models are deployed often fail to meet these idealized conditions. To overcome these challenges and further elevate performance levels, we propose an advanced Transformer-based framework complemented by an Audio-Visual Adapter (AVA) integrating temporal features from both audio and visual modalities. In addition, we introduce an innovative multi-modal contrastive learning method that is designed to enhance the correlation between uni-modal features and their integrated counterparts within a consistent feature space. Our designed method can deal with the flexible-model scenario instead of deploying different models for various modalities. Empirical evaluations conducted on two benchmark datasets have validated the superiority of our proposed model over other multi-modal fusion techniques, particularly in scenarios characterized by varying and missing modalities. This strongly affirms the effectiveness of our approach in significantly boosting the accuracy of deception detection in complex, real-world multi-modal scenarios. The codes will be released soon.

Deep neural networks (DNNs) have been applied in many computer vision tasks and achieved state-of-the-art (SOTA) performance. However, misclassification will occur when DNNs predict adversarial examples which are created by adding human-imperceptible adversarial noise to natural examples. This limits the application of DNN in security-critical fields. In order to enhance the robustness of models, previous research has primarily focused on the unimodal domain, such as image recognition and video understanding. Although multi-modal learning has achieved advanced performance in various tasks, such as action recognition, research on the robustness of RGB-skeleton action recognition models is scarce. In this paper, we systematically investigate how to improve the robustness of RGB-skeleton action recognition models. We initially conducted empirical analysis on the robustness of different modalities and observed that the skeleton modality is more robust than the RGB modality. Motivated by this observation, we propose the Attention-based Modality Reweighter (AMR), which utilizes an attention layer to re-weight the two modalities, enabling the model to learn more robust features. Our AMR is plug-and-play, allowing easy integration with multimodal models. To demonstrate the effectiveness of AMR, we conducted extensive experiments on various datasets. For example, compared to the SOTA methods, AMR exhibits a 43.77% improvement against PGD20 attacks on the NTURGB+D 60 dataset. Furthermore, it effectively balances the differences in robustness between different modalities.

Diffusion-based personalized visual content generation technologies have achieved significant breakthroughs, allowing for the creation of specific objects by just learning from a few reference photos. However, when misused to fabricate fake news or unsettling content targeting individuals, these technologies could cause considerable societal harm. To address this problem, current methods generate adversarial samples by adversarially maximizing the training loss, thereby disrupting the output of any personalized generation model trained with these samples. However, the existing methods fail to achieve effective defense and maintain stealthiness, as they overlook the intrinsic properties of diffusion models. In this paper, we introduce a novel Dual-Domain Anti-Personalization framework (DDAP). Specifically, we have developed Spatial Perturbation Learning (SPL) by exploiting the fixed and perturbation-sensitive nature of the image encoder in personalized generation. Subsequently, we have designed a Frequency Perturbation Learning (FPL) method that utilizes the characteristics of diffusion models in the frequency domain. The SPL disrupts the overall texture of the generated images, while the FPL focuses on image details. By alternating between these two methods, we construct the DDAP framework, effectively harnessing the strengths of both domains. To further enhance the visual quality of the adversarial samples, we design a localization module to accurately capture attentive areas while ensuring the effectiveness of the attack and avoiding unnecessary disturbances in the background. Extensive experiments on facial benchmarks have shown that the proposed DDAP enhances the disruption of personalized generation models while also maintaining high quality in adversarial samples, making it more effective in protecting privacy in practical applications.

Document Presentation Attack Detection (DPAD) is an important measure in protecting the authenticity of a document image. However, recent DPAD methods demand additional resources, such as manual effort in collecting additional data or knowing the parameters of acquisition devices. This work proposes a DPAD method based on multi-modal disentangled traces (MMDT) without the above drawbacks. We first disentangle the recaptured traces by a self-supervised disentanglement and synthesis network to enhance the generalization capacity in document images with different contents and layouts. Then, unlike the existing DPAD approaches that rely only on data in the RGB domain, we propose to explicitly employ the disentangled recaptured traces as new modalities in the transformer backbone through adaptive multi-modal adapters to fuse RGB/trace features efficiently. Visualization of the disentangled traces confirms the effectiveness of the proposed method in different document contents. Extensive experiments on three benchmark datasets demonstrate the superiority of our MMDT method on representing forensic traces of recapturing distortion.

Abstract Inference using skeleton to steer RGB videos is applicable to fine‐grained activities in indoor human action recognition (IHAR). However, existing methods that explore only spatial alignment are prone to bias, resulting in limited performance. The authors propose a Three‐stage Guidance (3sG) framework, leveraging skeleton knowledge to promote RGB in three stages. First, a soft shading image is proposed for alleviating background noise in videos, allowing the network to directly focus more on the motion region. Second, the authors propose to extract RGB frames of interest to reduce the computational effort. Furthermore, to explore more fully the complementary information between skeletons and RGB, the skeleton is coupled to the frame representation in a different spatial–temporal sharing pattern. Third, the global skeleton and skeleton‐guided RGB features are fed into the shared classifiers, which approximate the logit distributions of the two to enhance the performance in RGB unimodal. Finally, a fusion strategy that utilizes two learnable parameters to adaptively integrate the skeleton with the RGB is proposed. 3sG outperforms the state‐of‐the‐art results on the Toyota Smarthome dataset while it is more efficient than similar methods on the NTU RGB+D dataset.

Abstract Recently, vision transformer (ViT) based multimodal learning methods have been proposed to improve the robustness of face anti-spoofing (FAS) systems. However, there are still no works to explore the fundamental natures (e.g., modality-aware inputs, suitable multimodal pre-training, and efficient finetuning) in vanilla ViT for multimodal FAS. In this paper, we investigate three key factors (i.e., inputs, pre-training, and finetuning) in ViT for multimodal FAS with RGB, Infrared (IR), and Depth. First, in terms of the ViT inputs, we find that leveraging local feature descriptors (such as histograms of oriented gradients) benefits the ViT on IR modality but not RGB or Depth modalities. Second, in consideration of the task (FAS vs. generic object classification) and modality (multimodal vs. unimodal) gaps, ImageNet pre-trained models might be sub-optimal for the multimodal FAS task. Finally, in observation of the inefficiency on direct finetuning the whole or partial ViT, we design an adaptive multimodal adapter (AMA), which can efficiently aggregate local multimodal features while freezing majority of ViT parameters. To bridge these gaps, we propose the modality-asymmetric masked autoencoder (M $$^{2}$$ 2 A $$^{2}$$ 2 E) for multimodal FAS self-supervised pre-training without costly annotated labels. Compared with the previous modality-symmetric autoencoder, the proposed M $$^{2}$$ 2 A $$^{2}$$ 2 E is able to learn more intrinsic task-aware representation and compatible with modality-agnostic (e.g., unimodal, bimodal, and trimodal) downstream settings. Extensive experiments with both unimodal (RGB, Depth, IR) and multimodal (RGB+Depth, RGB+IR, Depth+IR, RGB+Depth+IR) settings conducted on multimodal FAS benchmarks demonstrate the superior performance of the proposed methods. One highlight is that the proposed method is robust under various missing-modality cases where previous multimodal FAS models suffer serious performance drops. We hope these findings and solutions can facilitate the future research for ViT-based multimodal FAS.

Poses are effective in interpreting fine-grained human activities, especially when encountering complex visual information. Unimodal methods for action recognition unsatisfactorily to daily activities due to the lack of a more comprehensive perspective. Multimodal methods to combine pose and visual are still not exhaustive enough in mining complementary information. Therefore, we propose a Pose-promote (Ppromo) framework that utilizes a priori knowledge of pose joints to perceive visual information progressively. We first introduce a temporal promote module to activate each video segment using temporally synchronized joint weights. Then a spatial promote module is proposed to capture the key regions in visuals using the learned pose attentions. To further refine the bimodal associations, the global inter-promote module is proposed to align global pose-visual semantics at the feature granularity. Finally, a learnable late fusion strategy between visual and pose is applied for accurate inference. Ppromo achieves state-of-the-art performance on three publicly available datasets.

Dynamic facial expression recognition (DFER) plays a vital role in understanding human emotions and behaviors. Existing efforts tend to fall into a single modality self-supervised pretraining learning paradigm, which limits the representation ability of models. Besides, coarse-grained temporal modeling struggles to capture subtle facial expression representations from various inputs. In this letter, we propose a novel method for DFER, termed fine-grained temporal-enhanced transformer (FTET-DFER), which consists of two stages. First, we employ the inherent correlation between visual and auditory modalities in real videos, to capture temporally dense representations such as facial movements and expressions, in a self-supervised audio-visual learning manner. Second, we utilize the learned embeddings as targets, to achieve the DFER. In addition, we design the FTET block to study fine-grained temporal-enhanced facial expression features based on intra-clip locally-enhanced relations as well as inter-clip locally-enhanced global relationships in videos. Extensive experiments show that FTET-DFER outperforms the state-of-the-arts through within-dataset and cross-dataset evaluation.

Face anti-spoofing (FAS) and face forgery detection play vital roles in securing face biometric systems from presentation attacks (PAs) and vicious digital manipulation (e.g., deepfakes). Despite satisfactory performance upon large-scale data and powerful deep models, recent advances in face spoofing and forgery detection approaches usually focus on 1) unimodal visual appearance or physiological (i.e., remote photoplethysmography (rPPG)) cues; and 2) separated feature representation for FAS or face forgery detection. On one side, unimodal appearance and rPPG features are respectively vulnerable to high-fidelity face 3D mask and video replay attacks, inspiring us to design reliable multi-modal fusion mechanisms for generalized FAS. On the other side, there are rich common features across FAS and face forgery detection tasks (e.g., periodic rPPG rhythms and vanilla appearance for bonafides), providing solid evidence to design a joint FAS and face forgery detection system in a multi-task learning fashion. In this paper, we establish the first joint face spoofing and forgery detection benchmark using both visual appearance and physiological rPPG cues. To enhance the rPPG periodicity discrimination, we design a two-branch physiological network using both facial spatio-temporal rPPG signal map and its continuous wavelet transformed counterpart as inputs. To mitigate the modality bias and improve the fusion efficacy, we conduct a weighted batch and layer normalization for both appearance and rPPG features before multi-modal fusion. We also investigate prevalent deep models, feature fusion strategies and multi-task learning configurations for joint face spoofing and forgery detection. We find that the generalization capacities of both unimodal (appearance or rPPG) and multi-modal (appearance+rPPG) models can be obviously improved via joint training on these two tasks. We hope this new benchmark will facilitate the future research of both FAS and deepfake detection communities. The codes will be released athttps://github.com/ZitongYu/Benchmarking.

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.

In view of inconsistent face acquisition procedure in face anti-spoofing, the detection performance on the target domain generally suffers severe degradation under source-specific gradient optimization. Existing domain adaptation face anti-spoofing methods focus on improving model generalization capability through feature matching, which do not consider the gradient discrepancy between the source and target domains. To this end, this work develops a category-conditional gradient alignment guided face anti-spoofing algorithm (CCGA-FAS) from a novel perspective of gradient discrepancy elimination. Technically, the category-conditional gradient alignment mechanism maximizes the cosine similarity of the gradient vectors generated by source and target samples within the live and spoof categories separately, which promotes the source and target domains to follow similar gradient descent directions during optimization. Considering that the gradient vector generation and alignment is computationally dependent on reliable category information, a temporal knowledge and flexible threshold based dynamic category measurer is devised to provide pseudo category information for unlabelled target samples in an easy-to-hard manner. The optimization for CCGA-FAS is implemented under the teacher-student structure, where the student model serves as the gradient optimization backbone, and the category prediction simultaneously benefits from the teacher and student models to consolidate the alignment stability. Experimental results and analysis demonstrate that the proposed method outperforms the state-of-the-art methods in both unsupervised and K-shot semi-supervised domain adaptive face anti-spoofing scenarios.

The rapid advancement of photorealistic generators has reached a critical juncture where the discrepancy between authentic and manipulated images is increasingly indistinguishable. Thus, benchmarking and advancing techniques detecting digital manipulation become an urgent issue. Although there have been a number of publicly available face forgery datasets, the forgery faces are mostly generated using GAN-based synthesis technology, which does not involve the most recent technologies like diffusion. The diversity and quality of images generated by diffusion models have been significantly improved and thus a much more challenging face forgery dataset shall be used to evaluate SOTA forgery detection literature. In this paper, we propose a large-scale, diverse, and fine-grained high-fidelity dataset, namely GenFace, to facilitate the advancement of deepfake detection, which contains a large number of forgery faces generated by advanced generators such as the diffusion-based model and more detailed labels about the manipulation approaches and adopted generators. In addition to evaluating SOTA approaches on our benchmark, we design an innovative Cross Appearance-Edge Learning (CAEL) detector to capture multi-grained appearance and edge global representations, and detect discriminative and general forgery traces. Moreover, we devise an Appearance-Edge Cross-Attention (AECA) module to explore the various integrations across two domains. Extensive experiment results and visualizations show that our detection model outperforms the state of the arts on different settings like cross-generator, cross-forgery, and cross-dataset evaluations. Code and datasets will be available athttps://github.com/Jenine-321/GenFace.

By training first with a large base dataset, Few-Shot Class-Incremental Learning (FSCIL) aims at continually learning a sequence of few-shot learning tasks with novel classes. There are mainly two challenges in FSCIL: the overfitting issue of novel classes with limited labeled samples and the catastrophic forgetting of previously seen classes. The current protocol of FSCIL is built by mimicking the general class-incremental learning setting by building a unified framework, while the existing frameworks for FSCIL on this protocol always bias to the classes in the base dataset because the dominant performance of the deep model is decided by the size of the training dataset. Moreover, it is difficult to handle the stability-plasticity constraint in a unified FSCIL framework. To solve these issues, we rethink the configuration of FSCIL with the open-set hypothesis by reserving the possibility in the first session for incoming categories. To find a better decision boundary of close space and open space, Hyperbolic Reciprocal Point Learning module (Hyper-RPL) is built on Reciprocal Point Learning with hyperbolic neural networks. Besides, when learning novel categories from limited labeled data, we incorporate a hyperbolic metric learning (Hyper-Metric) module into the distillation-based framework to alleviate the overfitting issue and better handle the trade-off issue between the preservation of old knowledge and the acquisition of new knowledge. Finally, the comprehensive assessments of the proposed configuration and modules on three benchmark datasets are executed to validate the effectiveness, and state-of-the-art results are achieved.

Remote photoplethysmography (rPPG) is an important technique for detecting human vital signs and has received extensive attention. For a long time, researchers have focused attention on supervised methods that rely on large amounts of labeled data. These methods are limited by their need for large amounts of data and the difficulty of acquiring ground truth physiological signals. To address these issues, several self-supervised methods based on contrastive learning have been proposed. However, they focus on contrastive learning between samples, which neglects inherent self-similar priors in physiological signals and seems to have a limited ability to cope with noise. In this paper, a linear self-supervised reconstruction task was designed for extracting the inherent self-similar priors in physiological signals. In addition, a specific noise-insensitive strategy was explored for reducing the interference of motion and illumination. The framework proposed in this paper, rPPG-MAE, demonstrates excellent performance even on the challenging VIPL-HR dataset. We also evaluate the proposed method on two public datasets, namely, PURE and UBFC-rPPG. The results show that our method not only outperforms existing self-supervised methods but also outperforms state-of-the-art (SOTA) supervised methods. One important observation is that the quality of the dataset appears to be more important than the size of the dataset used in self-supervised pretraining of the rPPG. The source code is available athttps://github.com/linuxsino/rPPG-MAE.

Recently, deep learning technique has been widely employed to deal with face super-resolution (FSR) problem. It aims to predict the nonlinear relationship between the low-resolution (LR) face images and corresponding high-resolution (HR) ones, which could recover the high-frequency details from the LR degraded textures. However, either CNN-based or Transformer-based approaches mostly enhance the details by exploiting the relationship of local pixels or patches on LR features, the nonlocal features are not fully taken into account for producing high-frequency textures. To improve the above problem, we design a novel dual-branch module which consists of Transformer and CNN respectively. The Transformer branch extracts multiple scale feature embeddings and explores local and nonlocal self-attention simultaneously. Thus, the parallel self-attention mechanism has superior capabilities to capture the local and nonlocal dependencies on face image in the face reconstruction. Furthermore, the traditional CNNs usually extract features by combining pixels in a local convolutional kernel, it may be not effective to recover lost high-frequency details since the variations of local pixels are not well measured, which is important in recovering vivid edges and contours. To this end, we propose the local variation based attention block on the CNN branch, which could enhance the capabilities by directly extracting features from the variation of neighboring pixels. Finally, the Transformer-branch and CNN-branch are combined together by the modulation block to fuse both nonlocal and local advantages from two branches. Experimental results demonstrate the effectiveness of the proposed method when compared with state-of-the-art approaches.

The action anticipation task refers to predicting what action will happen based on observed videos, which requires the model to have a strong ability to summarize the present and then reason about the future. Experience and common sense suggest that there is a significant correlation between different actions, which provides valuable prior knowledge for the action anticipation task. However, previous methods have not effectively modeled this underlying statistical relationship. To address this issue, we propose a novel end-to-end video modeling architecture that utilizes attention mechanisms, named Anticipation via Recognition and Reasoning (ARR). ARR decomposes the action anticipation task into action recognition and sequence reasoning tasks and effectively learns the statistical relationship between actions by next action prediction (NAP). In comparison to existing temporal aggregation strategies, ARR is able to extract more effective features from observable videos to make more reasonable predictions. In addition, to address the challenge of relationship modeling that requires extensive training data, we propose an innovative approach for the unsupervised pre-training of the decoder, which leverages the inherent temporal dynamics of video to enhance the reasoning capabilities of the network. Extensive experiments on the Epic-kitchen-100, EGTEA Gaze+, and 50salads datasets demonstrate the efficacy of the proposed methods. The code is available at https://github.com/linuxsino/ARR .

Remote photoplethysmography (rPPG) enables non-contact heart rate (HR) estimation from facial videos which gives significant convenience compared with traditional contact-based measurements. In the real-world long-term health monitoring scenario, the distance of the participants and their head movements usually vary by time, resulting in the inaccurate rPPG measurement due to the varying face resolution and complex motion artifacts. Different from the previous rPPG models designed for a constant distance between camera and participants, in this paper, we propose two plug-and-play blocks (i.e., physiological signal feature extraction block (PFE) and temporal face alignment block (TFA)) to alleviate the degradation of changing distance and head motion. On one side, guided with representative-area information, PFE adaptively encodes the arbitrary resolution facial frames to the fixed-resolution facial structure features. On the other side, leveraging the estimated optical flow, TFA is able to counteract the rPPG signal confusion caused by the head movement thus benefit the motion-robust rPPG signal recovery. Besides, we also train the model with a cross-resolution constraint using a two-stream dual-resolution framework, which further helps PFE learn resolution-robust facial rPPG features. Extensive experiments on three benchmark datasets (UBFC-rPPG, COHFACE and PURE) demonstrate the superior performance of the proposed method. One highlight is that with PFE and TFA, the off-the-shelf spatio-temporal rPPG models can predict more robust rPPG signals under both varying face resolution and severe head movement scenarios. The codes are available at https://github.com/LJWGIT/Arbitrary_Resolution_rPPG.

Remote photoplethysmography (rPPG) technology has drawn increasing attention in recent years. It can extract Blood Volume Pulse (BVP) from facial videos, making many applications like health monitoring and emotional analysis more accessible. However, as the BVP signal is easily affected by environmental changes, existing methods struggle to generalize well for unseen domains. In this paper, we systematically address the domain shift problem in the rPPG measurement task. We show that most domain generalization methods do not work well in this problem, as domain labels are ambiguous in complicated environmental changes. In light of this, we propose a domain-label-free approach called NEuron STructure modeling (NEST). NEST improves the generalization capacity by maximizing the coverage of feature space during training, which reduces the chance for under-optimized feature activation during inference. Besides, NEST can also enrich and enhance domain invariant features across multi-domain. We create and benchmark a large-scale domain generalization protocol for the rPPG measurement task. Extensive experiments show that our approach outperforms the state-of-the-art methods on both cross-dataset and intra-dataset settings. The codes are available at https://github.com/LuPaoPao/NEST.

Face Anti-Spoofing (FAS) is recently studied under the continual learning setting, where the FAS models are expected to evolve after encountering data from new domains. However, existing methods need extra replay buffers to store previous data for rehearsal, which becomes infeasible when previous data is unavailable because of privacy issues. In this paper, we propose the first rehearsal-free method for Domain Continual Learning (DCL) of FAS, which deals with catastrophic forgetting and unseen domain generalization problems simultaneously. For better generalization to unseen domains, we design the Dynamic Central Difference Convolutional Adapter (DCDCA) to adapt Vision Transformer (ViT) models during the continual learning sessions. To alleviate the forgetting of previous domains without using previous data, we propose the Proxy Prototype Contrastive Regularization (PPCR) to constrain the continual learning with previous domain knowledge from the proxy prototypes. Simulating practical DCL scenarios, we devise two new protocols which evaluate both generalization and anti-forgetting performance. Extensive experimental results show that our proposed method can improve the generalization performance in unseen domains and alleviate the catastrophic forgetting of previous knowledge. The code and protocol files are released on https://github.com/RizhaoCai/DCL-FAS-ICCV2023.

Deception detection in conversations is a challenging yet important task, having pivotal applications in many fields such as credibility assessment in business, multimedia anti-frauds, and custom security. Despite this, deception detection research is hindered by the lack of high-quality deception datasets, as well as the difficulties of learning multimodal features effectively. To address this issue, we introduce DOLOS1, the largest gameshow deception detection dataset with rich deceptive conversations. DOLOS includes 1, 675 video clips featuring 213 subjects, and it has been labeled with audio-visual feature annotations. We provide train-test, duration, and gender protocols to investigate the impact of different factors. We benchmark our dataset on previously proposed deception detection approaches. To further improve the performance by fine-tuning fewer parameters, we propose Parameter-Efficient Crossmodal Learning (PECL), where a Uniform Temporal Adapter (UT-Adapter) explores temporal attention in transformer-based architectures, and a crossmodal fusion module, Plug-in Audio-Visual Fusion (PAVF), combines crossmodal information from audio-visual features. Based on the rich fine-grained audio-visual annotations on DOLOS, we also exploit multi-task learning to enhance performance by concurrently predicting deception and audiovisual features. Experimental results demonstrate the desired quality of the DOLOS dataset and the effectiveness of the PECL. The DOLOS dataset and the source codes are available at here.

Abstract Remote photoplethysmography (rPPG), which aims at measuring heart activities and physiological signals from facial video without any contact, has great potential in many applications (e.g., remote healthcare and affective computing). Recent deep learning approaches focus on mining subtle rPPG clues using convolutional neural networks with limited spatio-temporal receptive fields, which neglect the long-range spatio-temporal perception and interaction for rPPG modeling. In this paper, we propose two end-to-end video transformer based architectures, namely PhysFormer and PhysFormer++, to adaptively aggregate both local and global spatio-temporal features for rPPG representation enhancement. As key modules in PhysFormer, the temporal difference transformers first enhance the quasi-periodic rPPG features with temporal difference guided global attention, and then refine the local spatio-temporal representation against interference. To better exploit the temporal contextual and periodic rPPG clues, we also extend the PhysFormer to the two-pathway SlowFast based PhysFormer++ with temporal difference periodic and cross-attention transformers. Furthermore, we propose the label distribution learning and a curriculum learning inspired dynamic constraint in frequency domain, which provide elaborate supervisions for PhysFormer and PhysFormer++ and alleviate overfitting. Comprehensive experiments are performed on four benchmark datasets to show our superior performance on both intra- and cross-dataset testings. Unlike most transformer networks needed pretraining from large-scale datasets, the proposed PhysFormer family can be easily trained from scratch on rPPG datasets, which makes it promising as a novel transformer baseline for the rPPG community.

Face anti-spoofing (FAS) has lately attracted increasing attention due to its vital role in securing face recognition systems from presentation attacks (PAs). As more and more realistic PAs with novel types spring up, early-stage FAS methods based on handcrafted features become unreliable due to their limited representation capacity. With the emergence of large-scale academic datasets in the recent decade, deep learning based FAS achieves remarkable performance and dominates this area. However, existing reviews in this field mainly focus on the handcrafted features, which are outdated and uninspiring for the progress of FAS community. In this paper, to stimulate future research, we present the first comprehensive review of recent advances in deep learning based FAS. It covers several novel and insightful components: 1) besides supervision with binary label (e.g., '0' for bonafide versus '1' for PAs), we also investigate recent methods with pixel-wise supervision (e.g., pseudo depth map); 2) in addition to traditional intra-dataset evaluation, we collect and analyze the latest methods specially designed for domain generalization and open-set FAS; and 3) besides commercial RGB camera, we summarize the deep learning applications under multi-modal (e.g., depth and infrared) or specialized (e.g., light field and flash) sensors. We conclude this survey by emphasizing current open issues and highlighting potential prospects.

The availability of handy multi-modal (i.e., RGB-D) sensors has brought about a surge of face anti-spoofing research. However, the current multi-modal face presentation attack detection (PAD) has two defects: (1) The framework based on multi-modal fusion requires providing modalities consistent with the training input, which seriously limits the deployment scenario. (2) The performance of ConvNet-based model on high fidelity datasets is increasingly limited. In this work, we present a pure transformer-based framework, dubbed the Flexible Modal Vision Transformer (FM-ViT), for face anti-spoofing to flexibly target any single-modal (i.e., RGB) attack scenarios with the help of available multi-modal data. Specifically, FM-ViT retains a specific branch for each modality to capture different modal information and introduces the Cross-Modal Transformer Block (CMTB), which consists of two cascaded attentions named Multi-headed Mutual-Attention (MMA) and Fusion-Attention (MFA) to guide each modal branch to mine potential features from informative patch tokens, and to learn modality-agnostic liveness features by enriching the modal information of own CLS token, respectively. Experiments demonstrate that the single model trained based on FM-ViT can not only flexibly evaluate different modal samples, but also outperforms existing single-modal frameworks by a large margin, and approaches the multi-modal frameworks introduced with smaller FLOPs and model parameters.

Face anti-spoofing (FAS) plays a crucial role in securing face recognition systems. Empirically, given an image, a model with more consistent output on different views (i.e., augmentations) of this image usually performs better. Motivated by this exciting observation, we conjecture that encouraging feature consistency of different views may be a promising way to boost FAS models. In this paper, we explore this way thoroughly by enhancing both Embedding-level and Prediction-level Consistency Regularization (EPCR) in FAS. Specifically, at the embedding level, we design a dense similarity loss to maximize the similarities between all positions of two intermediate feature maps in a self-supervised fashion; while at the prediction level, we optimize the mean square error between the predictions of two views. Notably, our EPCR is free of annotations and can directly integrate into semi-supervised learning schemes. Considering different application scenarios, we further design five diverse semi-supervised protocols to measure semi-supervised FAS techniques. We conduct extensive experiments to show that EPCR can significantly improve the performance of several supervised and semi-supervised tasks on benchmark datasets. The codes and protocols are available athttps://github.com/clks-wzz/EPCR.

We present a customized 3D mesh Transformer model for the pose transfer task. As the 3D pose transfer essentially is a deformation procedure dependent on the given meshes, the intuition of this work is to perceive the geometric inconsistency between the given meshes with the powerful self-attention mechanism. Specifically, we propose a novel geometry-contrastive Transformer that has an efficient 3D structured perceiving ability to the global geometric inconsistencies across the given meshes. Moreover, locally, a simple yet efficient central geodesic contrastive loss is further proposed to improve the regional geometric-inconsistency learning. At last, we present a latent isometric regularization module together with a novel semi-synthesized dataset for the cross-dataset 3D pose transfer task towards unknown spaces. The massive experimental results prove the efficacy of our approach by showing state-of-the-art quantitative performances on SMPL-NPT, FAUST and our new proposed dataset SMG-3D datasets, as well as promising qualitative results on MG-cloth and SMAL datasets. It's demonstrated that our method can achieve robust 3D pose transfer and be generalized to challenging meshes from unknown spaces on cross-dataset tasks. The code and dataset are made available. Code is available: https://github.com/mikecheninoulu/CGT.

With diverse presentation attacks emerging continually, generalizable face anti-spoofing (FAS) has drawn growing attention. Most existing methods implement domain generalization (DG) on the complete representations. However, different image statistics may have unique properties for the FAS tasks. In this work, we separate the complete representation into content and style ones. A novel Shuffled Style Assembly Network (SSAN) is proposed to extract and reassemble different content and style features for a stylized feature space. Then, to obtain a generalized representation, a contrastive learning strategy is developed to emphasize liveness-related style information while suppress the domain-specific one. Finally, the representations of the correct assemblies are used to distinguish between living and spoofing during the inferring. On the other hand, despite the decent performance, there still exists a gap between academia and industry, due to the difference in data quantity and distribution. Thus, a new large-scale benchmark for FAS is built up to further evaluate the performance of algorithms in reality. Both qualitative and quantitative results on existing and proposed benchmarks demonstrate the effectiveness of our methods. The codes will be available at https://github.com/wangzhuo2019/SSAN.

Remote photoplethysmography (rPPG), which aims at measuring heart activities and physiological signals from facial video without any contact, has great potential in many applications. Recent deep learning approaches focus on mining subtle rPPG clues using convolutional neural networks with limited spatio-temporal receptive fields, which neglect the long-range spatio-temporal perception and interaction for rPPG modeling. In this paper, we propose the PhysFormer, an end-to-end video transformer based architecture, to adaptively aggregate both local and global spatio-temporal features for rPPG representation enhancement. As key modules in PhysFormer, the temporal difference transformers first enhance the quasi-periodic rPPG features with temporal difference guided global attention, and then refine the local spatio-temporal representation against interference. Furthermore, we also propose the label distribution learning and a curriculum learning inspired dynamic constraint in frequency domain, which provide elaborate supervisions for PhysFormer and alleviate overfitting. Comprehensive experiments are performed on four benchmark datasets to show our superior performance on both intra- and cross-dataset testings. One highlight is that, unlike most transformer networks needed pretraining from large-scale datasets, the proposed PhysFormer can be easily trained from scratch on rPPG datasets, which makes it promising as a novel transformer baseline for the rPPG community. The codes are available at https://github.com/ZitongYu/PhysFormer.

Privacy with the use of face images is becoming a major concern in civilians’ applications. Recent studies have exploited privacy protection methods by means of facial attributes editing or de-identifying face images. Altering attributes causes loss of information for facial analysis while most de-identification studies did not quantitatively evaluate how well facial attributes are preserved. Moreover, state-of-the-art face analysis utilized 3D information for better performance. Existing face privacy studies only focusing in 2D domain is a key limitation towards the compatibility of more advanced 3D face analysis. This paper presents the first study on the possibility of 3D face de-identification with preserving facial attributes. We systematically evaluate the performance of 2D/3D face/facial attribute recognition and develop 2D/3D de-identification methods with preserving facial attributes using Auto Encoder and Generative Adversarial Networks approaches. We present comprehensive and reproducible experimental results using a publicly available 3D face database with facial attribute annotations for benchmarking and further research. https://github.com/kevinhmcheng/3d-face-de-id

Face Presentation Attack Detection (PAD) is an important measure to prevent spoof attacks for face biometric systems. Many works based on Convolution Neural Networks (CNNs) for face PAD formulate the problem as an image-level binary classification task without considering the context. Alternatively, Vision Transformers (ViT) using self-attention to attend the context of an image become the mainstreams in face PAD. Inspired by ViT, we propose a Video-based Transformer for face PAD (ViTransPAD) with short/long-range spatio-temporal attention which can not only focus on local details with short-range attention within a frame but also capture long-range dependencies over frames. Instead of using coarse image patches with single-scale as in ViT, we pro-pose the Multi-scale Multi-Head Self-Attention (MsMHSA) module to accommodate multi-scale patch partitions of Q, K, V feature maps to different heads on a single transformer in a coarse-to-fine manner, which enables to learn a fine-grained representation to perform pixel-level discrimination for face PAD. Due to lack inductive biases of convolutions in pure transformers, we also introduce convolutions to our ViTransPAD to integrate the desirable properties of CNNs. The extensive experiments show the effectiveness of our proposed ViTransPAD with a preferable accuracy-computation balance, which can serve as a new backbone for face PAD.

Face Super-resolution (FSR) task works for generating high-resolution (HR) face images from the corresponding low-resolution (LR) inputs, which has received a lot of attentions because of the wide application prospects. However, due to the diversity of facial texture and the difficulty of reconstructing detailed content from degraded images, FSR technology is still far away from being solved. In this paper, we propose a novel and effective face super-resolution framework based on Transformer, namely Interconnected Dual Pyramid Transformer (IDPT). Instead of straightly stacking cascaded feature reconstruction blocks, the proposed IDPT designs the pyramid encoder/decoder Transformer architecture to extract coarse and detailed facial textures respectively, while the relationship between the dual pyramid Transformers is further explored by a bottom pyramid feature extractor. The pyramid encoder/decoder structure is devised to adapt various characteristics of textures in different spatial spaces hierarchically. A novel fusing modulation module is inserted in each spatial layer to guide the refinement of detailed texture by the corresponding coarse texture, while fusing the shallow-layer coarse feature and corresponding deep-layer detailed feature simultaneously. Extensive experiments and visualizations on various datasets demonstrate the superiority of the proposed method for face super-resolution tasks.

Face anti-spoofing (FAS) secures face recognition from presentation attacks (PAs). Existing FAS methods usually supervise PA detectors with handcrafted binary or pixel-wise labels. However, handcrafted labels may are not the most adequate way to supervise PA detectors learning sufficient and intrinsic spoofing cues. Instead of using the handcrafted labels, we propose a novel Meta-Teacher FAS (MT-FAS) method to train a meta-teacher for supervising PA detectors more effectively. The meta-teacher is trained in a bi-level optimization manner to learn the ability to supervise the PA detectors learning rich spoofing cues. The bi-level optimization contains two key components: 1) a lower-level training in which the meta-teacher supervises the detector's learning process on the training set; and 2) a higher-level training in which the meta-teacher's teaching performance is optimized by minimizing the detector's validation loss. Our meta-teacher differs significantly from existing teacher-student models because the meta-teacher is explicitly trained for better teaching the detector (student), whereas existing teachers are trained for outstanding accuracy neglecting teaching ability. Extensive experiments on five FAS benchmarks show that with the proposed MT-FAS, the trained meta-teacher 1) provides better-suited supervision than both handcrafted labels and existing teacher-student models; and 2) significantly improves the performances of PA detectors.

Conventional 2D face biometric systems are vulnerable to presentation attacks performed with different face artefacts, e.g., printouts, video-replays and wearable 3D masks. The research focus in face presentation attack detection (PAD) has been recently shifting towards end-to-end learning of deep representations directly from annotated data rather than designing hand-crafted (low-level) features. However, even the state-of-the-art deep learning based face PAD models have shown unsatisfying generalization performance when facing unknown attacks or acquisition conditions due to lack of representative training and tuning data available in the existing public benchmarks. To alleviate this issue, we propose a video pre-processing technique called Temporal Sequence Sampling (TSS) for 2D face PAD by removing the estimated inter-frame 2D affine motion in the view and encoding the appearance and dynamics of the resulting smoothed video sequence into a single RGB image. Furthermore, we leverage the features of a Convolutional Neural Network (CNN) by introducing a self-supervised representation learning scheme, where the labels are automatically generated by the TSS method as the stabilized frames accumulated over video clips of different temporal lengths provide the supervision. The learnt feature representations are then fine-tuned for the downstream task using labelled face PAD data. Our extensive experiments on four public benchmarks, namely Replay-Attack, MSU-MFSD, CASIA-FASD and OULU-NPU, demonstrate that the proposed framework provides promising generalization capability and encourage further study in this domain.

Face presentation attack detection (PAD) is essential to secure face recognition systems primarily from high-fidelity mask attacks. Most existing 3D mask PAD benchmarks suffer from several drawbacks: 1) a limited number of mask identities, types of sensors, and a total number of videos; 2) low-fidelity quality of facial masks. Basic deep models and remote photoplethysmography (rPPG) methods achieved acceptable performance on these benchmarks but still far from the needs of practical scenarios. To bridge the gap to real-world applications, we introduce a large-scale High-Fidelity Mask dataset, namely HiFiMask. Specifically, a total amount of 54,600 videos are recorded from 75 subjects with 225 realistic masks by 7 new kinds of sensors. Along with the dataset, we propose a novel Contrastive Context-aware Learning (CCL) framework. CCL is a new training methodology for supervised PAD tasks, which is able to learn by leveraging rich contexts accurately (e.g., subjects, mask material and lighting) among pairs of live faces and high-fidelity mask attacks. Extensive experimental evaluations on HiFiMask and three additional 3D mask datasets demonstrate the effectiveness of our method. The codes and dataset will be released soon.

Pain is a significant indicator that shows people are suffering from an unwell experience and its automatic estimation has attracted much interest in recent years. Of late, most estimation methods are designed to capture the dynamic pain information from visual signals while a few physiological-signal based methods can provide extra potential cues to analyze the pain more accurately. However, it is still challenging to capture the physiological data from patients as it requires contact devices and patients’ cooperation. In this paper, we propose to leverage the pseudo physiological information by generating new modal data from the original visual videos and jointly estimating the pain by an end-to-end network. To extract the representations from bi-modal data, we design a spatio-temporal pain estimation network, which employs a dual-branch framework for extracting pain-aware visual and pseudo physiological features separately and fuses the features in a probabilistic way. The inherent vital sign, i.e., heart rate gain (HRG), from pseudo physiological information can be utilized as an auxiliary signal and integrated with the visual pain estimation framework. Moreover, specially-designed 3D convolution filters and attention structures are employed to extract spatio-temporal features for both branches. To use the HRG as an auxiliary way for pain estimation, we propose a probabilistic inference model by jointly considering the visual branch and physiological branch, which makes our model estimate the pain comprehensively. Experiments on two publicly-available datasets show the effectiveness of introducing the pseudo modality, and the proposed method can outperform the state-of-the-art methods.

We introduce a new dataset for the emotional artificial intelligence research: identity-free video dataset for Micro-Gesture Understanding and Emotion analysis (iMiGUE). Different from existing public datasets, iMiGUE focuses on nonverbal body gestures without using any identity information, while the predominant researches of emotion analysis concern sensitive biometric data, like face and speech. Most importantly, iMiGUE focuses on micro-gestures, i.e., unintentional behaviors driven by inner feelings, which are different from ordinary scope of gestures from other gesture datasets which are mostly intentionally performed for illustrative purposes. Furthermore, iMiGUE is designed to evaluate the ability of models to analyze the emotional states by integrating information of recognized micro-gesture, rather than just recognizing prototypes in the sequences separately (or isolatedly). This is because the real need for emotion AI is to understand the emotional states behind gestures in a holistic way. Moreover, to counter for the challenge of imbalanced sample distribution of this dataset, an unsupervised learning method is proposed to capture latent representations from the micro-gesture sequences themselves. We systematically investigate representative methods on this dataset, and comprehensive experimental results reveal several interesting insights from the iMiGUE, e.g., micro-gesture-based analysis can promote emotion understanding. We confirm that the new iMiGUE dataset could advance studies of micro-gesture and emotion AI.

Recently, deep Convolutional Neural Networks (CNNs) can achieve human-level performance in edge detection with the rich and abstract edge representation capacities. However, the high performance of CNN based edge detection is achieved with a large pretrained CNN backbone, which is memory and energy consuming. In addition, it is surprising that the previous wisdom from the traditional edge detectors, such as Canny, Sobel, and LBP are rarely investigated in the rapid-developing deep learning era. To address these issues, we propose a simple, lightweight yet effective architecture named Pixel Difference Network (PiDiNet) for efficient edge detection. PiDiNet adopts novel pixel difference convolutions that integrate the traditional edge detection operators into the popular convolutional operations in modern CNNs for enhanced performance on the task, which enjoys the best of both worlds. Extensive experiments on BSDS500, NYUD, and Multicue are provided to demonstrate its effectiveness, and its high training and inference efficiency. Surprisingly, when training from scratch with only the BSDS500 and VOC datasets, PiDiNet can surpass the recorded result of human perception (0.807 vs. 0.803 in ODS F-measure) on the BSDS500 dataset with 100 FPS and less than 1M parameters. A faster version of PiDiNet with less than 0.1M parameters can still achieve comparable performance among state of the arts with 200 FPS. Results on the NYUD and Multicue datasets show similar observations. The codes are available at https://github.com/zhuoinoulu/pidinet.

Automatic pain recognition is paramount for medical diagnosis and treatment. The existing works fall into three categories: assessing facial appearance changes, exploiting physiological cues, or fusing them in a multi-modal manner. However, (1) appearance changes are easily affected by subjective factors which impedes objective pain recognition. Besides, the appearance-based approaches ignore long-range spatial-temporal dependencies that are important for modeling expressions over time; (2) the physiological cues are obtained by attaching sensors on human body, which is inconvenient and uncomfortable. In this paper, we present a novel multi-task learning framework which encodes both appearance changes and physiological cues in a non-contact manner for pain recognition. The framework is able to capture both local and long-range dependencies via the proposed attention mechanism for the learned appearance representations, which are further enriched by temporally attended physiological cues (remote photoplethysmography, rPPG) that are recovered from videos in the auxiliary task. This framework is dubbed rPPG-enriched Spatio-Temporal Attention Network (rSTAN) and allows us to establish the state-of-the-art performance of non-contact pain recognition on publicly available pain databases. It demonstrates that rPPG predictions can be used as an auxiliary task to facilitate non-contact automatic pain recognition.

Face anti-spoofing (FAS) plays a vital role in securing face recognition systems. Recently, central difference convolution (CDC) has shown its excellent representation capacity for the FAS task via leveraging local gradient features. However, aggregating central difference clues from all neighbors/directions simultaneously makes the CDC redundant and sub-optimized in the training phase. In this paper, we propose two Cross Central Difference Convolutions (C-CDC), which exploit the difference of the center and surround sparse local features from the horizontal/vertical and diagonal directions, respectively. It is interesting to find that, with only five ninth parameters and less computational cost, C-CDC even outperforms the full directional CDC. Based on these two decoupled C-CDC, a powerful Dual-Cross Central Difference Network (DC-CDN) is established with Cross Feature Interaction Modules (CFIM) for mutual relation mining and local detailed representation enhancement. Furthermore, a novel Patch Exchange (PE) augmentation strategy for FAS is proposed via simply exchanging the face patches as well as their dense labels from random samples. Thus, the augmented samples contain richer live/spoof patterns and diverse domain distributions, which benefits the intrinsic and robust feature learning. Comprehensive experiments are performed on four benchmark datasets with three testing protocols to demonstrate our state-of-the-art performance.

Face anti-spoofing (FAS) plays a vital role in securing face recognition systems. Existing methods heavily rely on the expert-designed networks, which may lead to a sub-optimal solution for FAS task. Here we propose the first FAS method based on neural architecture search (NAS), called NAS-FAS, to discover the well-suited task-aware networks. Unlike previous NAS works mainly focus on developing efficient search strategies in generic object classification, we pay more attention to study the search spaces for FAS task. The challenges of utilizing NAS for FAS are in two folds: the networks searched on 1) a specific acquisition condition might perform poorly in unseen conditions, and 2) particular spoofing attacks might generalize badly for unseen attacks. To overcome these two issues, we develop a novel search space consisting of central difference convolution and pooling operators. Moreover, an efficient static-dynamic representation is exploited for fully mining the FAS-aware spatio-temporal discrepancy. Besides, we propose Domain/Type-aware Meta-NAS, which leverages cross-domain/type knowledge for robust searching. Finally, in order to evaluate the NAS transferability for cross datasets and unknown attack types, we release a large-scale 3D mask dataset, namely CASIA-SURF 3DMask, for supporting the new 'cross-dataset cross-type' testing protocol. Experiments demonstrate that the proposed NAS-FAS achieves state-of-the-art performance on nine FAS benchmark datasets with four testing protocols.

3D mask face presentation attack detection (PAD) plays a vital role in securing face recognition systems from emergent 3D mask attacks. Recently, remote photoplethysmography (rPPG) has been developed as an intrinsic liveness clue for 3D mask PAD without relying on the mask appearance. However, the rPPG features for 3D mask PAD are still needed expert knowledge to design manually, which limits its further progress in the deep learning and big data era. In this letter, we propose a pure rPPG transformer (TransRPPG) framework for learning intrinsic liveness representation efficiently. At first, rPPG-based multi-scale spatial-temporal maps (MSTmap) are constructed from facial skin and background regions. Then the transformer fully mines the global relationship within MSTmaps for liveness representation, and gives a binary prediction for 3D mask detection. Comprehensive experiments are conducted on two benchmark datasets to demonstrate the efficacy of the TransRPPG on both intra- and cross-dataset testings. Our TransRPPG is lightweight and efficient (with only 547 K parameters and 763 M FLOPs), which is promising for mobile-level applications.

Gesture recognition has attracted considerable attention owing to its great potential in applications. Although the great progress has been made recently in multi-modal learning methods, existing methods still lack effective integration to fully explore synergies among spatio-temporal modalities effectively for gesture recognition. The problems are partially due to the fact that the existing manually designed network architectures have low efficiency in the joint learning of multi-modalities. In this paper, we propose the first neural architecture search (NAS)-based method for RGB-D gesture recognition. The proposed method includes two key components: 1) enhanced temporal representation via the proposed 3D Central Difference Convolution (3D-CDC) family, which is able to capture rich temporal context via aggregating temporal difference information; and 2) optimized backbones for multi-sampling-rate branches and lateral connections among varied modalities. The resultant multi-modal multi-rate network provides a new perspective to understand the relationship between RGB and depth modalities and their temporal dynamics. Comprehensive experiments are performed on three benchmark datasets (IsoGD, NvGesture, and EgoGesture), demonstrating the state-of-the-art performance in both single- and multi-modality settings. The code is available at https://github.com/ZitongYu/3DCDC-NAS.

Face anti-spoofing is crucial to the security of face recognition systems. Most previous methods formulate face anti-spoofing as a supervised learning problem to detect various predefined presentation attacks, which need large scale training data to cover as many attacks as possible. However, the trained model is easy to overfit several common attacks and is still vulnerable to unseen attacks. To overcome this challenge, the detector should: 1) learn discriminative features that can generalize to unseen spoofing types from predefined presentation attacks; 2) quickly adapt to new spoofing types by learning from both the predefined attacks and a few examples of the new spoofing types. Therefore, we define face anti-spoofing as a zero- and few-shot learning problem. In this paper, we propose a novel Adaptive Inner-update Meta Face Anti-Spoofing (AIM-FAS) method to tackle this problem through meta-learning. Specifically, AIM-FAS trains a meta-learner focusing on the task of detecting unseen spoofing types by learning from predefined living and spoofing faces and a few examples of new attacks. To assess the proposed approach, we propose several benchmarks for zero- and few-shot FAS. Experiments show its superior performances on the presented benchmarks to existing methods in existing zero-shot FAS protocols.

Face anti-spoofing is critical to the security of face recognition systems. Depth supervised learning has been proven as one of the most effective methods for face anti-spoofing. Despite the great success, most previous works still formulate the problem as a single-frame multi-task one by simply augmenting the loss with depth, while neglecting the detailed fine-grained information and the interplay between facial depths and moving patterns. In contrast, we design a new approach to detect presentation attacks from multiple frames based on two insights: 1) detailed discriminative clues (e.g., spatial gradient magnitude) between living and spoofing face may be discarded through stacked vanilla convolutions, and 2) the dynamics of 3D moving faces provide important clues in detecting the spoofing faces. The proposed method is able to capture discriminative details via Residual Spatial Gradient Block (RSGB) and encode spatio-temporal information from Spatio-Temporal Propagation Module (STPM) efficiently. Moreover, a novel Contrastive Depth Loss is presented for more accurate depth supervision. To assess the efficacy of our method, we also collect a Double-modal Anti-spoofing Dataset (DMAD) which provides actual depth for each sample. The experiments demonstrate that the proposed approach achieves state-of-the-art results on five benchmark datasets including OULU-NPU, SiW, CASIA-MFSD, Replay-Attack, and the new DMAD. Codes will be available at https://github.com/clks-wzz/FAS-SGTD.

Face anti-spoofing (FAS) plays a vital role in face recognition systems. Most state-of-the-art FAS methods 1) rely on stacked convolutions and expert-designed network, which is weak in describing detailed fine-grained information and easily being ineffective when the environment varies (e.g., different illumination), and 2) prefer to use long sequence as input to extract dynamic features, making them difficult to deploy into scenarios which need quick response. Here we propose a novel frame level FAS method based on Central Difference Convolution (CDC), which is able to capture intrinsic detailed patterns via aggregating both intensity and gradient information. A network built with CDC, called the Central Difference Convolutional Network (CDCN), is able to provide more robust modeling capacity than its counterpart built with vanilla convolution. Furthermore, over a specifically designed CDC search space, Neural Architecture Search (NAS) is utilized to discover a more powerful network structure (CDCN++), which can be assembled with Multiscale Attention Fusion Module (MAFM) for further boosting performance. Comprehensive experiments are performed on six benchmark datasets to show that 1) the proposed method not only achieves superior performance on intra-dataset testing (especially 0.2% ACER in Protocol-1 of OULU-NPU dataset), 2) it also generalizes well on cross-dataset testing (particularly 6.5% HTER from CASIA-MFSD to Replay-Attack datasets). The codes are available at https://github.com/ZitongYu/CDCN.

With the development of mobile devices, it is hopeful and pressing to deploy face recognition and face anti-spoofing (FAS) model on cell phone or portable devices. Most of existing face anti-spoofing methods focus on building computational costly detector for better spoofing face detection performance. However, these detectors are unfriendly to be deployed on the mobile device for real-time FAS applications. In this paper, we propose a neural architecture search (NAS) based method called Auto-FAS, intending to discover well-suitable lightweight networks for mobile-level face anti-spoofing. In Auto-FAS, a special search space is designed to restrict the model's size, and pixel-wise binary supervision is used to improve the model's performance. We demonstrate both the effectiveness and efficiency of the proposed approach on three public benchmark datasets, which shows the potential real-time FAS application for mobile devices.

Remote photoplethysmography (rPPG), which aims at measuring heart activities without any contact, has great potential in many applications (e.g., remote healthcare). Existing end-to-end rPPG and heart rate (HR) measurement methods from facial videos are vulnerable to the less-constrained scenarios (e.g., with head movement and bad illumination). In this letter, we explore the reason why existing end-to-end networks perform poorly in challenging conditions and establish a strong end-to-end baseline (AutoHR) for remote HR measurement with neural architecture search (NAS). The proposed method includes three parts: 1) a powerful searched backbone with novel Temporal Difference Convolution (TDC), intending to capture intrinsic rPPG-aware clues between frames; 2) a hybrid loss function considering constraints from both time and frequency domains; and 3) spatio-temporal data augmentation strategies for better representation learning. Comprehensive experiments are performed on three benchmark datasets, and we achieved superior performance on both intra- and cross-dataset testings.

Atrial fibrillation (AF) is one of the most common cardiac arrhythmias, which particularly occurs in the elderly individuals with heart disease. Though AF is often asymptomatic during normal activities, it has huge potential risks for stroke and other severe diseases. Thus, early detection of AF has great importance in the field of public health. Currently, electrocardiography (ECG) is the commonly used measure for the diagnosis of AF, which presents the irregular rhythm of waveform for AF patients. However, the measurement of the ECG signal requires special medical acquisition devices, which are not comfortable for practical monitoring in daily life. In this paper, we explore a very promising algorithm to detect AF from remote face videos by analyzing the color variations of face skin. The main challenge is that the current remote photoplethysmography (rPPG) technique is rather immature, which causes difficulty in extracting accurate pulse signals for describing the cardiac rhythm. To solve this problem, we first utilize various rPPG algorithms to capture pulse rhythms from different regions on the face video. We then investigate biomedical statistical methods to extract suitable features from each pulse signal. Due to the imprecision of video-extracted pulse signals, some traditional physiological features may lose their utility since they were originally proposed for ECG signals. Furthermore, some of them are very susceptible to the influence of noise. Thus, we propose a feature fusion algorithm to select and combine reasonable information from multiple physiological features, which aims to preserve the discriminability of detecting AF in the presence of the noise and outlier disturbances. The experimental results on a real-world database demonstrate the effectiveness of the proposed method in providing useful information for AF detection.

Remote photoplethysmography (rPPG), which aims at measuring heart activities without any contact, has great potential in many applications (e.g., remote healthcare). Existing rPPG approaches rely on analyzing very fine details of facial videos, which are prone to be affected by video compression. Here we propose a two-stage, end-to-end method using hidden rPPG information enhancement and attention networks, which is the first attempt to counter video compression loss and recover rPPG signals from highly compressed videos. The method includes two parts: 1) a Spatio-Temporal Video Enhancement Network (STVEN) for video enhancement, and 2) an rPPG network (rPPGNet) for rPPG signal recovery. The rPPGNet can work on its own for robust rPPG measurement, and the STVEN network can be added and jointly trained to further boost the performance especially on highly compressed videos. Comprehensive experiments are performed on two benchmark datasets to show that, 1) the proposed method not only achieves superior performance on compressed videos with high-quality videos pair, 2) it also generalizes well on novel data with only compressed videos available, which implies the promising potential for real-world applications.

Deep part-based methods in recent literature have revealed the great potential of learning local part-level representation for pedestrian image in the task of person re-identification. However, global features that capture discriminative holistic information of human body are usually ignored or not well exploited. This motivates us to investigate joint learning global and local features from pedestrian images. Specifically, in this work, we propose a novel framework termed tree branch network (TBN) for person re-identification. Given a pedestrain image, the feature maps generated by the backbone CNN, are partitioned recursively into several pieces, each of which is followed by a bottleneck structure that learns finer-grained features for each level in the hierarchical tree-like framework. In this way, representations are learned in a coarse-to-fine manner and finally assembled to produce more discriminative image descriptions. Experimental results demonstrate the effectiveness of the global and local feature learning method in the proposed TBN framework. We also show significant improvement in performance over state-of-the-art methods on three public benchmarks: Market-1501, CUHK-03 and DukeMTMC.