arXiv subjects

Zhenzhen Hu

Publications and source records attributed to Zhenzhen Hu.

At least 19 recordsLinked to original sources

arXiv2026

Visual Token Compression Enhances Robustness of MLLMs

In this paper, we show for the first time that visual token pruning enhances the robustness of Multimodal Large Language Models (MLLMs), mitigating vulnerabilities such as jailbreak attacks and hallucinations. Given that vision and language modalities cannot be perfectly aligned, the misaligned visual tokens might act as out-of-distribution (OOD) inputs, leading to unpredictable outputs and introducing potential vulnerabilities. Building on this insight, we aim to enhance model robustness against jailbreaks and hallucinations by reducing OOD visual tokens at robust-pruning layers, while also reducing inference cost as a side benefit. Specifically, we measure the distance between each visual token and the language feature space. Then, visual tokens with large distances are identified as OOD tokens, which can be iteratively pruned. To demonstrate the effectiveness of our method, we evaluate it on seven diverse popular benchmarks. Notably, our method yields an average improvement of 13.29\% in defending jailbreak attacks, consistently achieves competitive performance in mitigating hallucinations, and maintains strong results on general datasets like MME.

cs.CV↗

arXiv2026

Missing-Token Prompted Reliability-Aware Fusion for Robust Polyglot Speaker Identification

Accurate and robust multimodal speaker identification is essential for multimedia understanding and biometric authentication. However, real-world polyglot scenarios pose two key challenges: speaker-discriminative representations should generalize across languages, and the model should remain reliable when face information is unavailable. To address these challenges, we propose MRAF, a Missing-Token Prompted Reliability-Aware Fusion framework for polyglot speaker identification across complete-modality, missing-face, and cross-lingual scenarios. MRAF represents unavailable face inputs with a learnable missing token instead of fixed zero-valued features, providing a trainable representation of the missing visual state. This design reduces the distribution gap caused by missing inputs and allows subsequent reliability estimation and cross-modal fusion to operate within a unified token space. To adaptively integrate modalities with different reliability, MRAF further introduces a reliability-aware cross-attention fusion module, which estimates face and audio reliability scores, normalizes them into modality weights, and applies these weights to token representations before bidirectional cross-attention. In this way, the model can emphasize reliable modality cues while suppressing unreliable ones. During training, MRAF jointly optimizes multi-branch classification losses, audio-only knowledge distillation, and center loss to improve speaker discrimination and missing-modality robustness. Experiments on the official POLY-SIM 2026 test set demonstrate the effectiveness of the proposed framework. In the final evaluation, MRAF achieves 100% accuracy on P3 and P5, and obtains competitive results on the more challenging missing-face settings P4 and P6. The source code will be released at https://github.com/MSA-LMC/MRAF.

cs.SD↗

arXiv2026

Traits Run Deeper: Trait-Specific Asymmetric Fusion for Personality Assessment

Personality assessment aims to infer stable traits from dynamic behaviors across modalities like language, voice, and facial expressions. Existing approaches often adopt a uniform multimodal fusion strategy for all personality dimensions, overlooking trait-specific modality preferences and causing cross-modal interference. To address this, we propose Traits Run Deeper, a novel personality assessment framework consisting of three components. First, the Multimodal Foundation Representation (MFR) module constructs personality-oriented inputs and incorporates psychology-informed semantic templates as anchors, enabling foundation models to capture trait-relevant behaviors. Second, the Trait-Specific Modality Fusion (TSMF) module employs an asymmetric fusion mechanism, allowing each dimension to selectively exploit different modality pathways to capture heterogeneous preferences while reducing cross-modal contamination. Third, the Distribution-Calibrated Personality Regression (DCPR) module mitigates label imbalance and central tendency bias through target distribution calibration, improving robustness and stability. Experimental results on the AVI Challenge 2026 validation set show that our framework reduces mean squared error (MSE) by approximately 25% compared with the baseline. Consistent improvements on the official test set demonstrate that our method achieves the best performance and ranks first in the AVI Challenge 2026 Personality Assessment Track. The source code will be made available at [https://github.com/MSA-LMC/TraitsRunDeeper](https://github.com/MSA-LMC/TraitsRunDeeper).

cs.CV↗

arXiv2026

SDTalk: Structured Facial Priors and Dual-Branch Motion Fields for Generalizable Gaussian Talking Head Synthesis

High-quality, real-time talking head synthesis remains a fundamental challenge in computer vision. Existing reconstruction- and rendering-based methods typically rely on identity-specific models, limiting cross-identity generalization. To address this issue, we propose SDTalk, a one-shot 3D Gaussian Splatting (3DGS)-based framework that generalizes to unseen identities without personalized training or fine-tuning. Our framework comprises two modules with a two-stage training strategy. In the first stage, we incorporate structured facial priors into the reconstruction module and separately predict 3DGS parameters for visible and occluded regions, enabling complete head reconstruction from a single image. In the second stage, we introduce a dual-branch motion field to model coarse and fine facial dynamics, improving detail fidelity and lip synchronization. Experiments demonstrate that SDTalk surpasses existing methods in both visual quality and inference efficiency.

cs.CV↗

arXiv2026

To Fuse or to Drop? Dual-Path Learning for Resolving Modality Conflicts in Multimodal Emotion Recognition

Multimodal emotion recognition (MER) benefits from combining text, audio, and vision, yet standard fusion often fails when modalities conflict. Crucially, conflicts differ in resolvability: benign conflicts stem from missing, weak, or ambiguous cues and can be mitigated by cross-modal calibration, while severe conflicts arise from intrinsically contradictory (e.g., sarcasm) or misleading signals, for which forced fusion may amplify errors. Recognizing this, we propose Dual-Path Conflict Resolution (DCR), a unified framework that learns when to fuse and when to drop modalities. Path I (Affective Fusion Distiller, AFD) performs reverse distillation from audio/visual teachers to a textual student using temporally weighted class evidence, thereby enhancing representation-level calibration and improving fusion when alignment is beneficial. Path II (Affective Discernment Agent, ADA) formulates MER as a contextual bandit that selects among fusion and unimodal predictions based on a dual-view state and a calibration-aware reward, enabling decision-level arbitration under irreconcilable conflicts without requiring per-modality reliability labels. By taking into account the full multimodal context and coupling soft calibration with hard arbitration, DCR reconciles conflicts that can be aligned while bypassing misleading modalities when fusion is harmful. Across five benchmarks covering both dialogue-level and clip-level MER, DCR consistently outperforms competitive baselines or achieves highly competitive results. Further ablations, conflict-specific subset evaluation, and modality-selection analysis verify that AFD and ADA are complementary and jointly improve robust conflict-aware emotion recognition.

cs.MM↗

arXiv2026

Bidirectional Learning of Facial Action Units and Expressions via Structured Semantic Mapping across Heterogeneous Datasets

Facial action unit (AU) detection and facial expression (FE) recognition can be jointly viewed as affective facial behavior tasks, representing fine-grained muscular activations and coarse-grained holistic affective states, respectively. Despite their inherent semantic correlation, existing studies predominantly focus on knowledge transfer from AUs to FEs, while bidirectional learning remains insufficiently explored. In practice, this challenge is further compounded by heterogeneous data conditions, where AU and FE datasets differ in annotation paradigms (frame-level vs.\ clip-level), label granularity, and data availability and diversity, hindering effective joint learning. To address these issues, we propose a Structured Semantic Mapping (SSM) framework for bidirectional AU--FE learning under different data domains and heterogeneous supervision. SSM consists of three key components: (1) a shared visual backbone that learns unified facial representations from dynamic AU and FE videos; (2) semantic mediation via a Textual Semantic Prototype (TSP) module, which constructs structured semantic prototypes from fixed textual descriptions with learnable context prompts for supervision and cross-task alignment in a shared semantic space; and (3) a Dynamic Prior Mapping (DPM) module that incorporates FACS-derived prior knowledge and learns data-adaptive bidirectional association matrices in the textual semantic space for explicit knowledge transfer. Extensive experiments on popular AU detection and FE recognition benchmarks show that SSM consistently outperforms its single-task and multi-task baselines and achieves competitive performance against task-specific methods. The FE-to-AU results further show that holistic expression semantics provides useful supervision for fine-grained AU learning across heterogeneous datasets.

cs.CV↗

arXiv2025

Disentangling Foreground and Background for vision-Language Navigation via Online Augmentation

Following language instructions, vision-language navigation (VLN) agents are tasked with navigating unseen environments. While augmenting multifaceted visual representations has propelled advancements in VLN, the significance of foreground and background in visual observations remains underexplored. Intuitively, foreground regions provide semantic cues, whereas the background encompasses spatial connectivity information. Inspired on this insight, we propose a Consensus-driven Online Feature Augmentation strategy (COFA) with alternative foreground and background features to facilitate the navigable generalization. Specifically, we first leverage semantically-enhanced landmark identification to disentangle foreground and background as candidate augmented features. Subsequently, a consensus-driven online augmentation strategy encourages the agent to consolidate two-stage voting results on feature preferences according to diverse instructions and navigational locations. Experiments on REVERIE and R2R demonstrate that our online foreground-background augmentation boosts the generalization of baseline and attains state-of-the-art performance.

cs.CV↗

arXiv2025

CLAIP-Emo: Parameter-Efficient Adaptation of Language-supervised models for In-the-Wild Audiovisual Emotion Recognition

Audiovisual emotion recognition (AVER) in the wild is still hindered by pose variation, occlusion, and background noise. Prevailing methods primarily rely on large-scale domain-specific pre-training, which is costly and often mismatched to real-world affective data. To address this, we present CLAIP-Emo, a modular framework that reframes in-the-wild AVER as a parameter-efficient adaptation of language-supervised foundation models (CLIP/CLAP). Specifically, it (i) preserves language-supervised priors by freezing CLIP/CLAP backbones and performing emotion-oriented adaptation via LoRA (updating \ensuremath{\le}4.0\% of the total parameters), (ii) allocates temporal modeling asymmetrically, employing a lightweight Transformer for visual dynamics while applying mean pooling for audio prosody, and (iii) applies a simple fusion head for prediction. On DFEW and MAFW, CLAIP-Emo (ViT-L/14) achieves 80.14\% and 61.18\% weighted average recall with only 8M training parameters, setting a new state of the art. Our findings suggest that parameter-efficient adaptation of language-supervised foundation models provides a scalable alternative to domain-specific pre-training for real-world AVER. The code and models will be available at \href{https://github.com/MSA-LMC/CLAIP-Emo}{https://github.com/MSA-LMC/CLAIP-Emo}.

cs.MM↗

arXiv2025

Generalizable Engagement Estimation in Conversation via Domain Prompting and Parallel Attention

Accurate engagement estimation is essential for adaptive human-computer interaction systems, yet robust deployment is hindered by poor generalizability across diverse domains and challenges in modeling complex interaction dynamics.To tackle these issues, we propose DAPA (Domain-Adaptive Parallel Attention), a novel framework for generalizable conversational engagement modeling. DAPA introduces a Domain Prompting mechanism by prepending learnable domain-specific vectors to the input, explicitly conditioning the model on the data's origin to facilitate domain-aware adaptation while preserving generalizable engagement representations. To capture interactional synchrony, the framework also incorporates a Parallel Cross-Attention module that explicitly aligns reactive (forward BiLSTM) and anticipatory (backward BiLSTM) states between participants.Extensive experiments demonstrate that DAPA establishes a new state-of-the-art performance on several cross-cultural and cross-linguistic benchmarks, notably achieving an absolute improvement of 0.45 in Concordance Correlation Coefficient (CCC) over a strong baseline on the NoXi-J test set. The superiority of our method was also confirmed by winning the first place in the Multi-Domain Engagement Estimation Challenge at MultiMediate'25.

cs.CV↗

arXiv2025

Traits Run Deep: Enhancing Personality Assessment via Psychology-Guided LLM Representations and Multimodal Apparent Behaviors

Accurate and reliable personality assessment plays a vital role in many fields, such as emotional intelligence, mental health diagnostics, and personalized education. Unlike fleeting emotions, personality traits are stable, often subconsciously leaked through language, facial expressions, and body behaviors, with asynchronous patterns across modalities. It was hard to model personality semantics with traditional superficial features and seemed impossible to achieve effective cross-modal understanding. To address these challenges, we propose a novel personality assessment framework called \textit{\textbf{Traits Run Deep}}. It employs \textit{\textbf{psychology-informed prompts}} to elicit high-level personality-relevant semantic representations. Besides, it devises a \textit{\textbf{Text-Centric Trait Fusion Network}} that anchors rich text semantics to align and integrate asynchronous signals from other modalities. To be specific, such fusion module includes a Chunk-Wise Projector to decrease dimensionality, a Cross-Modal Connector and a Text Feature Enhancer for effective modality fusion and an ensemble regression head to improve generalization in data-scarce situations. To our knowledge, we are the first to apply personality-specific prompts to guide large language models (LLMs) in extracting personality-aware semantics for improved representation quality. Furthermore, extracting and fusing audio-visual apparent behavior features further improves the accuracy. Experimental results on the AVI validation set have demonstrated the effectiveness of the proposed components, i.e., approximately a 45\% reduction in mean squared error (MSE). Final evaluations on the test set of the AVI Challenge 2025 confirm our method's superiority, ranking first in the Personality Assessment track. The source code will be made available at https://github.com/MSA-LMC/TraitsRunDeep.

cs.CL↗

arXiv2025

Listening to the Unspoken: Exploring "365" Aspects of Multimodal Interview Performance Assessment

Interview performance assessment is essential for determining candidates' suitability for professional positions. To ensure holistic and fair evaluations, we propose a novel and comprehensive framework that explores ``365'' aspects of interview performance by integrating \textit{three} modalities (video, audio, and text), \textit{six} responses per candidate, and \textit{five} key evaluation dimensions. The framework employs modality-specific feature extractors to encode heterogeneous data streams and subsequently fused via a Shared Compression Multilayer Perceptron. This module compresses multimodal embeddings into a unified latent space, facilitating efficient feature interaction. To enhance prediction robustness, we incorporate a two-level ensemble learning strategy: (1) independent regression heads predict scores for each response, and (2) predictions are aggregated across responses using a mean-pooling mechanism to produce final scores for the five target dimensions. By listening to the unspoken, our approach captures both explicit and implicit cues from multimodal data, enabling comprehensive and unbiased assessments. Achieving a multi-dimensional average MSE of 0.1824, our framework secured first place in the AVI Challenge 2025, demonstrating its effectiveness and robustness in advancing automated and multimodal interview performance assessment. The full implementation is available at https://github.com/MSA-LMC/365Aspects.

cs.CL↗

arXiv2025

Rebalancing Contrastive Alignment with Bottlenecked Semantic Increments in Text-Video Retrieval

Recent progress in text-video retrieval has been largely driven by contrastive learning. However, existing methods often overlook the effect of the modality gap, which causes anchor representations to undergo in-place optimization (i.e., optimization tension) that limits their alignment capacity. Moreover, noisy hard negatives further distort the semantics of anchors. To address these issues, we propose GARE, a Gap-Aware Retrieval framework that introduces a learnable, pair-specific increment $\Delta_{ij}$ between text $t_i$ and video $v_j$, redistributing gradients to relieve optimization tension and absorb noise. We derive $\Delta_{ij}$ via a multivariate first-order Taylor expansion of the InfoNCE loss under a trust-region constraint, showing that it guides updates along locally consistent descent directions. A lightweight neural module conditioned on the semantic gap couples increments across batches for structure-aware correction. Furthermore, we regularize $\Delta$ through a variational information bottleneck with relaxed compression, enhancing stability and semantic consistency. Experiments on four benchmarks demonstrate that GARE consistently improves alignment accuracy and robustness, validating the effectiveness of gap-aware tension mitigation. Code is available at https://github.com/musicman217/GARE-text-video-retrieval.

cs.CV↗

arXiv2025

Concept Drift Guided LayerNorm Tuning for Efficient Multimodal Metaphor Identification

Metaphorical imagination, the ability to connect seemingly unrelated concepts, is fundamental to human cognition and communication. While understanding linguistic metaphors has advanced significantly, grasping multimodal metaphors, such as those found in internet memes, presents unique challenges due to their unconventional expressions and implied meanings. Existing methods for multimodal metaphor identification often struggle to bridge the gap between literal and figurative interpretations. Additionally, generative approaches that utilize large language models or text-to-image models, while promising, suffer from high computational costs. This paper introduces \textbf{C}oncept \textbf{D}rift \textbf{G}uided \textbf{L}ayerNorm \textbf{T}uning (\textbf{CDGLT}), a novel and training-efficient framework for multimodal metaphor identification. CDGLT incorporates two key innovations: (1) Concept Drift, a mechanism that leverages Spherical Linear Interpolation (SLERP) of cross-modal embeddings from a CLIP encoder to generate a new, divergent concept embedding. This drifted concept helps to alleviate the gap between literal features and the figurative task. (2) A prompt construction strategy, that adapts the method of feature extraction and fusion using pre-trained language models for the multimodal metaphor identification task. CDGLT achieves state-of-the-art performance on the MET-Meme benchmark while significantly reducing training costs compared to existing generative methods. Ablation studies demonstrate the effectiveness of both Concept Drift and our adapted LN Tuning approach. Our method represents a significant step towards efficient and accurate multimodal metaphor understanding. The code is available: \href{https://github.com/Qianvenh/CDGLT}{https://github.com/Qianvenh/CDGLT}.

cs.MM↗

arXiv2025

VAEmo: Efficient Representation Learning for Visual-Audio Emotion with Knowledge Injection

Audiovisual emotion recognition (AVER) aims to infer human emotions from nonverbal visual-audio (VA) cues, offering modality-complementary and language-agnostic advantages. However, AVER remains challenging due to the inherent ambiguity of emotional expressions, cross-modal expressive disparities, and the scarcity of reliably annotated data. Recent self-supervised AVER approaches have introduced strong multimodal representations, yet they predominantly rely on modality-specific encoders and coarse content-level alignment, limiting fine-grained emotional semantic modeling. To address these issues, we propose VAEmo, an efficient two-stage framework for emotion-centric joint VA representation learning with external knowledge injection. In Stage~1, a unified and lightweight representation network is pre-trained on large-scale speaker-centric VA corpora via masked reconstruction and contrastive objectives, mitigating the modality gap and learning expressive, complementary representations without emotion labels. In Stage~2, multimodal large language models automatically generate detailed affective descriptions according to our well-designed chain-of-thought prompting for only a small subset of VA samples; these rich textual semantics are then injected by aligning their corresponding embeddings with VA representations through dual-path contrastive learning, further bridging the emotion gap. Extensive experiments on multiple downstream AVER benchmarks show that VAEmo achieves state-of-the-art performance with a compact design, highlighting the benefit of unified cross-modal encoding and emotion-aware semantic guidance for efficient, generalizable VA emotion representations.

cs.CV↗

arXiv2025

PhysioSync: Temporal and Cross-Modal Contrastive Learning Inspired by Physiological Synchronization for EEG-Based Emotion Recognition

Electroencephalography (EEG) signals provide a promising and involuntary reflection of brain activity related to emotional states, offering significant advantages over behavioral cues like facial expressions. However, EEG signals are often noisy, affected by artifacts, and vary across individuals, complicating emotion recognition. While multimodal approaches have used Peripheral Physiological Signals (PPS) like GSR to complement EEG, they often overlook the dynamic synchronization and consistent semantics between the modalities. Additionally, the temporal dynamics of emotional fluctuations across different time resolutions in PPS remain underexplored. To address these challenges, we propose PhysioSync, a novel pre-training framework leveraging temporal and cross-modal contrastive learning, inspired by physiological synchronization phenomena. PhysioSync incorporates Cross-Modal Consistency Alignment (CM-CA) to model dynamic relationships between EEG and complementary PPS, enabling emotion-related synchronizations across modalities. Besides, it introduces Long- and Short-Term Temporal Contrastive Learning (LS-TCL) to capture emotional synchronization at different temporal resolutions within modalities. After pre-training, cross-resolution and cross-modal features are hierarchically fused and fine-tuned to enhance emotion recognition. Experiments on DEAP and DREAMER datasets demonstrate PhysioSync's advanced performance under uni-modal and cross-modal conditions, highlighting its effectiveness for EEG-centered emotion recognition.

cs.CV↗

arXiv2025

Video Flow as Time Series: Discovering Temporal Consistency and Variability for VideoQA

Video Question Answering (VideoQA) is a complex video-language task that demands a sophisticated understanding of both visual content and temporal dynamics. Traditional Transformer-style architectures, while effective in integrating multimodal data, often simplify temporal dynamics through positional encoding and fail to capture non-linear interactions within video sequences. In this paper, we introduce the Temporal Trio Transformer (T3T), a novel architecture that models time consistency and time variability. The T3T integrates three key components: Temporal Smoothing (TS), Temporal Difference (TD), and Temporal Fusion (TF). The TS module employs Brownian Bridge for capturing smooth, continuous temporal transitions, while the TD module identifies and encodes significant temporal variations and abrupt changes within the video content. Subsequently, the TF module synthesizes these temporal features with textual cues, facilitating a deeper contextual understanding and response accuracy. The efficacy of the T3T is demonstrated through extensive testing on multiple VideoQA benchmark datasets. Our results underscore the importance of a nuanced approach to temporal modeling in improving the accuracy and depth of video-based question answering.

cs.CV↗

arXiv2024

Agent Journey Beyond RGB: Hierarchical Semantic-Spatial Representation Enrichment for Vision-and-Language Navigation

Navigating unseen environments from natural language instructions remains challenging for egocentric agents in Vision-and-Language Navigation (VLN). Humans naturally ground concrete semantic knowledge within spatial layouts during indoor navigation. Although prior work has introduced diverse environment representations to improve reasoning, auxiliary modalities are often naively concatenated with RGB features, which underutilizes each modality's distinct contribution. We propose a hierarchical Semantic Understanding and Spatial Awareness (SUSA) architecture to enable agents to perceive and ground environments at multiple scales. Specifically, the Textual Semantic Understanding (TSU) module supports local action prediction by generating view-level descriptions, capturing fine-grained semantics and narrowing the modality gap between instructions and environments. Complementarily, the Depth Enhanced Spatial Perception (DSP) module incrementally builds a trajectory-level depth exploration map, providing a coarse-grained representation of global spatial layout. Extensive experiments show that the hierarchical representation enrichment of SUSA significantly improves navigation performance over the baseline on discrete VLN benchmarks (REVERIE, R2R, and SOON) and generalizes better to the continuous R2R-CE benchmark.

cs.CV↗

arXiv2024

Text Proxy: Decomposing Retrieval from a 1-to-N Relationship into N 1-to-1 Relationships for Text-Video Retrieval

Text-video retrieval (TVR) has seen substantial advancements in recent years, fueled by the utilization of pre-trained models and large language models (LLMs). Despite these advancements, achieving accurate matching in TVR remains challenging due to inherent disparities between video and textual modalities and irregularities in data representation. In this paper, we propose Text-Video-ProxyNet (TV-ProxyNet), a novel framework designed to decompose the conventional 1-to-N relationship of TVR into N distinct 1-to-1 relationships. By replacing a single text query with a series of text proxies, TV-ProxyNet not only broadens the query scope but also achieves a more precise expansion. Each text proxy is crafted through a refined iterative process, controlled by mechanisms we term as the director and dash, which regulate the proxy's direction and distance relative to the original text query. This setup not only facilitates more precise semantic alignment but also effectively manages the disparities and noise inherent in multimodal data. Our experiments on three representative video-text retrieval benchmarks, MSRVTT, DiDeMo, and ActivityNet Captions, demonstrate the effectiveness of TV-ProxyNet. The results show an improvement of 2.0% to 3.3% in R@1 over the baseline. TV-ProxyNet achieved state-of-the-art performance on MSRVTT and ActivityNet Captions, and a 2.0% improvement on DiDeMo compared to existing methods, validating our approach's ability to enhance semantic mapping and reduce error propensity.

cs.CV↗