Multimodal Large Language Models (MLLMs) show promise for Multimodal Emotion Recognition (MER) but often remain unreliable because sparse emotional cues could be easily overwhelmed and affected by redundant context. While fine-tuning is effective, it is usually costly when using large models. Training-free methods like chain-of-thought reasoning provide a practical alternative, but they mostly rely on heuristic prompting to influence the model behaviors and do not explicitly focus on emotion relevant tokens internally, which would allow decision-relevant emotional tokens to be diluted by environmental noise, resulting in unstable predictions. To address this limitation without training, we rethink MER from a world-model perspective that treats emotion as a latent state inferred from noisy and redundant multimodal observations. Under frozen parameters, this view suggests that robustness depends on constraining why and how tokens contribute to inference. Based on this insight, we propose WETR (World-Model inspired Emotion-aware Token Refinement), a training-free, plug-and-play regulator that reshapes token usage through two mechanisms: Noise-suppressed Token Selection (NTS), which suppresses redundant intra-modal noise, and State-strengthened Token Reweighting (STR), which amplifies decision-relevant emotional tokens. Experiments on multiple MER benchmarks demonstrate that WETR consistently improves accuracy and stability under frozen parameters, which also improves token-level interpretability.

Two-hand reconstruction from monocular images is hampered by complex poses and severe occlusions, which often cause interaction misalignment and two–hand penetration. We address this by decoupling the problem into 2D structural alignment and 3D spatial interaction alignment, each handled by a tailored component. For 2D alignment, we pioneer the attempt to unify heterogeneous structural priors (keypoints, segmentation, and depth) from vision foundation models as complementary structured guidance for two-hand recovery. Instead of extracting priors prediction as explicit inputs, we propose a fusion-alignment encoder that absorbs their structural knowledge implicitly, achieving foundation-level guidance without foundation-level cost. For 3D spatial alignment, we propose a two-hand diffusion model that learns a generative mapping from interpenetrated poses to realistic, collision-free configurations. Guided by collision gradients during denoising, the model converges toward the manifold of valid two-hand interactions, preserving geometric and kinematic coherence. This generative formulation approach enables physically credible reconstructions even under occlusion or ambiguous visual input. Extensive experiments on InterHand2.6M, HIC, and FreiHAND show state-of-the-art or leading performance in interaction alignment and penetration suppression. Code will be released publicly.

Foundation object detectors such as GLIP and Grounding DINO excel on general-domain data but often degrade in specialized and data-scarce settings like underwater imagery or industrial defects. Typical cross-domain few-shot approaches rely on fine-tuning scarce target data, incurring cost and overfitting risks. We instead ask: Can a frozen detector adapt with only one exemplar per class without training? To answer this, we introduce training-free one-shot domain generalization for object detection, where detectors must adapt to specialized domains with only one annotated exemplar per class and no weight updates. To tackle this task, we propose LAB-Det, which exploits Language As a domain-invariant Bridge. Instead of adapting visual features, we project each exemplar into a descriptive text that conditions and guides a frozen detector. This linguistic conditioning replaces gradient-based adaptation, enabling robust generalization in data-scarce domains. We evaluate on UODD (underwater) and NEU-DET (industrial defects), two widely adopted benchmarks for data-scarce detection, where object boundaries are often ambiguous, and LAB-Det achieves up to 5.4 mAP improvement over state-of-the-art fine-tuned baselines without updating a single parameter. These results establish linguistic adaptation as an efficient and interpretable alternative to fine-tuning in specialized detection settings.

Human emotion recognition is essential for emotionally intelligent systems in areas such as human-computer interaction, healthcare, and behavioral analysis. In real-world scenarios, emotional states are often ambiguous and fall outside predefined categories, making existing models unable to handle unseen emotion classes, thus limiting their use in sensitive domains like depression detection. To address this, open-set emotion recognition (O-ER) methods with pretrained models have been proposed to classify known emotions while detecting unknown ones. However, most focus on unimodal recognition, ignoring rich cues in multimodal data and limiting robustness. This motivates open-set multimodal emotion recognition (O-MER), which captures multimodal information to better recognize known emotions and detect unseen ones. However, directly fusing modalities often leads to semantically heterogeneous emotion representations, highlighting the need for a unified modality-interactive prompting mechanism to structurally align and fuse cross-modal affective cues. We propose Hierarchical Cross-modal Emotion-interactive Prompting (HCEP), a novel framework that adapts pretrained models for O-MER via a two-level prompt learning mechanism. HCEP includes three components: (1) Semantic-level Multimodal Emotion-aligning Prompting (SMEP), aligning and capturing multimodal emotion features via semantic emotion-unified prompts; (2) Class-level Unimodal Emotion-opposing Prompting (CUEP), refining decision boundaries per modality with positive and negative prompts; and (3) Dual-stream Prompt-driven Open-set Learning (DPOL), jointly optimizing known emotion classification and unknown emotion detection using a novel threshold-based discrimination strategy. Extensive experiments on five O-MER benchmarks demonstrate that HCEP achieves state-of-the-art results, outperforming the baseline with an average 16.79% AUROC and 28.49% OSCR relative gains, validating its strong capability to recognize known and detect unseen emotions.

Most referring object detection (ROD) models, especially the modern grounding detectors, are designed for data-rich conditions, yet many practical deployments, such as robotics, augmented reality, and other specialized domains, would face severe label scarcity. In such regimes, end-to-end grounding detectors need to learn spatial and semantic structure from scratch, wasting precious samples. We ask a simple question: Can explicit reasoning priors help models learn more efficiently when data is scarce? To explore this, we first introduce a Data-efficient Referring Object Detection (De-ROD) task, which is a benchmark protocol for measuring ROD performance in low-data and few-shot settings. We then propose the HeROD (Heuristic-inspired ROD), a lightweight, model-agnostic framework that injects explicit, heuristic-inspired spatial and semantic reasoning priors, which are interpretable signals derived based on the referring phrase, into 3 stages of a modern DETR-style pipeline: proposal ranking, prediction fusion, and Hungarian matching. By biasing both training and inference toward plausible candidates, these priors promise to improve label efficiency and convergence performance. On RefCOCO, RefCOCO+, and RefCOCOg, HeROD consistently outperforms strong grounding baselines in scarce-label regimes. More broadly, our results suggest that integrating simple, interpretable reasoning priors provides a practical and extensible path toward better data-efficient vision–language understanding.

Emotion Recognition (ER) is the process of analyzing and identifying human emotions from sensing data. Currently, the field heavily relies on facial expression recognition (FER) because visual channel conveys rich emotional cues. However, facial expressions are often used as social tools rather than manifestations of genuine inner emotions. To understand and bridge this gap between FER and ER, we introduce eye behaviors as an important emotional cue and construct an Eye-behavior-aided Multimodal Emotion Recognition (EMER) dataset. To collect data with genuine emotions, spontaneous emotion induction paradigm is exploited with stimulus material, during which non-invasive eye behavior data, like eye movement sequences and eye fixation maps, is captured together with facial expression videos. To better illustrate the gap between ER and FER, multi-view emotion labels for mutimodal ER and FER are separately annotated. Furthermore, based on the new dataset, we design a simple yet effective Eye-behavior-aided MER Transformer (EMERT) that enhances ER by bridging the emotion gap. EMERT leverages modality-adversarial feature decoupling and a multitask Transformer to model eye behaviors as a strong complement to facial expressions. In the experiment, we introduce seven multimodal benchmark protocols for a variety of comprehensive evaluations of the EMER dataset. The results show that the EMERT outperforms other state-of-the-art multimodal methods by a great margin, revealing the importance of modeling eye behaviors for robust ER. To sum up, we provide a comprehensive analysis of the importance of eye behaviors in ER, advancing the study on addressing the gap between FER and ER for more robust ER performance. Our EMER dataset and the trained EMERT models will be publicly available at https://github.com/kejun1/EMER.

Dynamic Facial Expression Recognition (DFER) models the temporal evolution of facial expressions in videos. In real-world deployments, changing scenarios distort expression trajectories over time, making it difficult for existing methods to maintain performance. While most current approaches address this issue through passive feature alignment across scenarios or domain-incremental learning techniques that preserve previously learned representations, they do not explicitly model scenario evolution over time, limiting their ability to robustly capture expression dynamics under scenario-incremental changes. To end this, we propose , an emotion world model for DFER that explicitly models latent emotion state evolution under scenario variations. Specifically, formulates scenario-incremental DFER as a progressive Bayesian inference problem over latent world states with dual temporal scales. Slow-timescale component () models scenario evolution using stochastic evolutionary priors, capturing long-term scenario effects and providing proactive guidance in new scenarios. Fast-timescale component () models frame-level expression dynamics with temporally consistent latent transitions, effectively decoupling expression dynamics from scenario influences. By jointly inferring latent states at both timescales, , shifts DFER from a passive feature discrimination process to a active probabilistic state inference under evolving scenarios. Experiments on FERV39k, DFEW, and MAFW demonstrate that , consistently outperforms state-of-the-art methods, achieving up to 3.84\% improvement while exhibiting strong cross-scenario stability and long-term robustness.

Self-supervised face representation learning (SFRL) shows strong potential for scalable face-related applications, yet remains vulnerable to adversarial attacks that cause dual facial semantic degradations, namely (1) structured distortions in key facial regions (e.g., subtle inter-ocular distance shifts) that disrupt identity-related features, and (2) unstructured additive noise (e.g., illumination artifacts) that entangles with face-related features in latent space. Existing defense methods struggle to deal with both facial semantic degradations in SFRL, resulting in limited robustness. To address this, inspired by existing reverse Diffusion approaches that effectively tackle the image denoising, we propose DAR-SFRL, a novel Degradation-adaptive Attack-Robust Self-supervised Face Representation Learning framework. DAR-SFRL models adversarial attacks as a degradation-based function composed of geometric distortions and additive noise, applying a multi-stage reverse Diffusion iterative process to recover facial semantics. At each stage of the process, DAR-SFRL employs: (1) an adaptive degraded-face restoration method that progressively reverses the degradation function and recovers fine-grained details from structured distortions, and (2) a noise-orthogonal contrastive learning mechanism to mitigate the impact of unstructured additive noise by maximizing the dissimilarity between noisy and clean image features in the latent space. Extensive experiments across tasks-including face recognition, facial expression recognition, and facial action unit detection-demonstrate that DAR-SFRL significantly outperforms state-of-the-art defenses under various adversarial attacks, highlighting its robustness and generalization in real-world face-aware applications. Our evaluation code is available at https://github.com/23wk/DAR-SFRL

Object Referring Analysis (ORA), commonly known as referring expression comprehension, requires the identification and localization of specific objects in an image based on natural descriptions. Unlike generic object detection, ORA requires both accurate language understanding and precise visual localization, making it inherently more complex. Although recent pre-trained large visual grounding detectors have achieved significant progress, they heavily rely on extensively labeled data and time-consuming learning. To address these, we introduce a novel, training-free framework for zero-shot ORA, termed FLORA (Formal Language for Object Referring and Analysis). FLORA harnesses the inherent reasoning capabilities of large language models (LLMs) and integrates a formal language model - a logical framework that regulates language within structured, rule-based descriptions - to provide effective zero-shot ORA. More specifically, our formal language model (FLM) enables an effective, logic-driven interpretation of object descriptions without necessitating any training processes. Built upon FLM-regulated LLM outputs, we further devise a Bayesian inference framework and employ appropriate off-the-shelf interpretive models to finalize the reasoning, delivering favorable robustness against LLM hallucinations and compelling ORA performance in a training-free manner. In practice, our FLORA boosts the zero-shot performance of existing pretrained grounding detectors by up to around 45 different challenging datasets also confirms that FLORA consistently surpasses current state-of-the-art zero-shot methods in both detection and segmentation tasks associated with zero-shot ORA. We believe our probabilistic parsing and reasoning of the LLM outputs elevate the reliability and interpretability of zero-shot ORA. We shall release codes upon publication.

Self-supervised facial representation learning (SFRL) methods, especially contrastive learning (CL) methods, have been increasingly popular due to their ability to perform face understanding without heavily relying on large-scale well-annotated datasets. However, analytically, current CL-based SFRL methods still perform unsatisfactorily in learning facial representations due to their tendency to learn pose-insensitive features, resulting in the loss of some useful pose details. This could be due to the inappropriate positive/negative pair selection within CL. To conquer this challenge, we propose a Pose-disentangled Contrastive Facial Representation Learning (PCFRL) framework to enhance pose awareness for SFRL. We achieve this by explicitly disentangling the pose-aware features from non-pose face-aware features and introducing appropriate sample calibration schemes for better CL with the disentangled features. In PCFRL, we first devise a pose-disentangled decoder with a delicately designed orthogonalizing regulation to perform the disentanglement; therefore, the learning on the pose-aware and non-pose face-aware features would not affect each other. Then, we introduce a false-negative pair calibration module to overcome the issue that the two types of disentangled features may not share the same negative pairs for CL. Our calibration employs a novel neighborhood-cohesive pair alignment method to identify pose and face false-negative pairs, respectively, and further help calibrate them to appropriate positive pairs. Lastly, we devise two calibrated CL losses, namely calibrated pose-aware and face-aware CL losses, for adaptively learning the calibrated pairs more effectively, ultimately enhancing the learning with the disentangled features and providing robust facial representations for various downstream tasks. In the experiments, we perform linear evaluations on four challenging downstream facial tasks with SFRL using our method, including facial expression recognition, face recognition, facial action unit detection, and head pose estimation. Experimental results show that PCFRL outperforms existing state-of-the-art methods by a substantial margin, demonstrating the importance of improving pose awareness for SFRL. Our evaluation code and model will be available at https://github.com/fulaoze/CV/tree/main .

Reinforcement learning (RL) is emerging as a powerful paradigm for enabling large language models (LLMs) to perform complex reasoning tasks. Recent advances indicate that integrating RL with retrieval-augmented generation (RAG) allows LLMs to dynamically incorporate external knowledge, leading to more informed and robust decision making. However, we identify a critical challenge during policy-driven trajectory sampling: LLMs are frequently trapped in unproductive reasoning paths, which we refer to as "dead ends", committing to overconfident yet incorrect conclusions. This severely hampers exploration and undermines effective policy optimization. To address this challenge, we propose REX-RAG (Reasoning Exploration with Policy Correction in Retrieval-Augmented Generation), a novel framework that explores alternative reasoning paths while maintaining rigorous policy learning through principled distributional corrections. Our approach introduces two key innovations: (1) Mixed Sampling Strategy, which combines a novel probe sampling method with exploratory prompts to escape dead ends; and (2) Policy Correction Mechanism, which employs importance sampling to correct distribution shifts induced by mixed sampling, thereby mitigating gradient estimation bias. We evaluate it on seven question-answering benchmarks, and the experimental results show that REX-RAG achieves average performance gains of 5.1

Despite advances in deep learning for remote sensing building extraction (RSBE), multitarget domain RSBE (MD-RSBE) remains challenging, as it requires transferring knowledge from a labeled source domain to multiple unlabeled target domains, with domain shifts in texture, style, and semantics. Existing domain adaptation (DA) and generalization (DG) methods face significant limitations: DA requires target-domain training, while DG needs multisource training, leading to high training costs and low generalization in practical MD-RSBE scenarios. To address this, we propose a multiview augmented single-source DG (MASDG) method, which effectively mitigates domain shifts across the RS source and target domains for robust MD-RSBE performance by enriching the diversity of the source domain through multiview augmentation and enforcing semantic consistency. Specifically, MASDG consists of three key components: texture-level domain augmentation (TDA) module, style-level domain augmentation (SDA) module, and semantic-invariant representation learning (SRL). To mitigate texture-level domain shift, TDA first introduces parameter-optimized multilayer random convolution to modify the texture of the source image, generating texture-augmented image pairs for simulating real-world texture diversity across various RS domains. Then, with each image pair from TDA, SDA employs two paralleled encoders, namely, the general feature encoder and the batch-guided style encoder, to formulate multiview building features, further mitigating style-level domain shift. Finally, SRL ensures SRL via a dual mechanism, including multiview segmentation loss and semantic consistency loss. The former generates predictions from diverse feature views (original, texture-augmented, and style-augmented), while the latter performs semantic alignment by minimizing distribution discrepancies among predictions, bridging semantic inconsistency to enable robust segmentation. Extensive experiments across three different MD-RSBE settings with seven different target domains demonstrate that our MASDG outperforms existing state-of-the-art methods by a significant margin.

Video-based facial expression recognition (VFER) is challenging due to variations caused by cultural background and expression camouflage. To tackle these problems, researchers introduced eye movement signals to complement visual information. However, existing methods either require expensive devices to capture high-quality eye movements or can only extract low-quality eye movements visually, making them ineffective in the real world. To address this, we propose an eye movement-instructed VFER (EM-VFER) that leverages high-quality eye movements to instruct the visual learning, obtaining robust performance without requiring costly devices during inference. Specifically, our EM-VFER operates in two stages: the high-quality eye movement pre-training stage and the eye movement-instructed video fine-tuning stage. In the pre-training, we compile an Eye-behavior-aided Multimodal Emotion Recognition (EMER) dataset and use it to train a multimodal Transformer. During the fine-tuning, we propose a novel progressive eye movement-instructed learning to take better advantage of the prior knowledge about high-quality eye movement signals from EMER. The instructed fine-tuning model could then make more robust predictions on downstream facial expression datasets. We evaluate our approach on three macro-expression datasets (DFEW, MAFW and Aff-wild2) and two micro-expression datasets (CASME III and CASME II). The results demonstrate that EM-VFER significantly outperforms existing methods.

Affective computing, which aims to recognize, interpret, and understand human emotions, provides benefits in healthcare, such as improving patient care and enhancing doctor-patient communication. However, there is a noticeable absence of a comprehensive summary of recent advancements in affective computing for healthcare, which could pose difficulties for researchers entering this field. To address this, our paper aims to provide an extensive literature review of related studies published in the last five years. We begin by analyzing trends, benefits, and limitations of recent datasets and affective computing methods devised for healthcare. Subsequently, we highlight several healthcare application hotspots of current technologies that could be promising for real-world deployment. Through our analysis, we identify and discuss some ongoing challenges in the field as evidenced by the literature. Concluding with a thorough review, we further offer potential future research directions and hope our findings and insights could guide related researchers to make better contributions to the evolution of affective computing in healthcare.

Multi-view camera-based 3D object detection has become popular due to its low cost, but accurately inferring 3D geometry solely from camera data remains challenging and may lead to inferior performance. Although distilling precise 3D geometry knowledge from LiDAR data could help tackle this challenge, the benefits of LiDAR information could be greatly hindered by the significant modality gap between different sensory modalities. To address this issue, we propose a Simulated multi-modal Distillation (SimDistill) method by carefully crafting the model architecture and distillation strategy. Specifically, we devise multi-modal architectures for both teacher and student models, including a LiDAR-camera fusion-based teacher and a simulated fusion-based student. Owing to the ``identical'' architecture design, the student can mimic the teacher to generate multi-modal features with merely multi-view images as input, where a geometry compensation module is introduced to bridge the modality gap. Furthermore, we propose a comprehensive multi-modal distillation scheme that supports intra-modal, cross-modal, and multi-modal fusion distillation simultaneously in the Bird's-eye-view space. Incorporating them together, our SimDistill can learn better feature representations for 3D object detection while maintaining a cost-effective camera-only deployment. Extensive experiments validate the effectiveness and superiority of SimDistill over state-of-the-art methods, achieving an improvement of 4.8% mAP and 4.1% NDS over the baseline detector. The source code will be released at https://github.com/ViTAE-Transformer/SimDistill.

Label-efficient segmentation aims to perform effective segmentation on input data using only sparse and limited ground-truth labels for training. This topic is widely studied in 3D point cloud segmentation due to the difficulty of annotating point clouds densely, while it is also essential for cost-effective segmentation on 2D images. Until recently, pseudo-labels have been widely employed to facilitate training with limited ground-truth labels, and promising progress has been witnessed in both the 2D and 3D segmentation. However, existing pseudo-labeling approaches could suffer heavily from the noises and variations in unlabelled data, which would result in significant discrepancies between generated pseudo-labels and current model predictions during training. We analyze that this can further confuse and affect the model learning process, which shows to be a shared problem in label-efficient learning across both 2D and 3D modalities. To address this issue, we propose a novel learning strategy to regularize the pseudo-labels generated for training, thus effectively narrowing the gaps between pseudo-labels and model predictions. More specifically, our method introduces an Entropy Regularization loss and a Distribution Alignment loss for label-efficient learning, resulting in an ERDA learning strategy. Interestingly, by using KL distance to formulate the distribution alignment loss, ERDA reduces to a deceptively simple cross-entropy-based loss which optimizes both the pseudo-label generation module and the segmentation model simultaneously. In addition, we innovate in the pseudo-label generation to make our ERDA consistently effective across both 2D and 3D data modalities for segmentation. Enjoying simplicity and more modality-agnostic pseudo-label generation, our method has shown outstanding performance in fully utilizing all unlabeled data points for training across ...

Understanding vehicles in images is important for various applications such as intelligent transportation and self-driving system. Existing vehicle-centric works typically pre-train models on large-scale classification datasets and then fine-tune them for specific downstream tasks. However, they neglect the specific characteristics of vehicle perception in different tasks and might thus lead to sub-optimal performance. To address this issue, we propose a novel vehicle-centric pre-training framework called VehicleMAE, which incorporates the structural information including the spatial structure from vehicle profile information and the semantic structure from informative high-level natural language descriptions for effective masked vehicle appearance reconstruction. To be specific, we explicitly extract the sketch lines of vehicles as a form of the spatial structure to guide vehicle reconstruction. The more comprehensive knowledge distilled from the CLIP big model based on the similarity between the paired/unpaired vehicle image-text sample is further taken into consideration to help achieve a better understanding of vehicles. A large-scale dataset is built to pre-train our model, termed Autobot1M, which contains about 1M vehicle images and 12693 text information. Extensive experiments on four vehicle-based downstream tasks fully validated the effectiveness of our VehicleMAE. The source code and pre-trained models will be released at https://github.com/Event-AHU/VehicleMAE.

Multimodal emotion recognition identifies human emotions from various data modalities like video, text, and audio. However, we found that this task can be easily affected by noisy information that does not contain useful semantics. To this end, we present a novel paradigm that attempts to extract noise-resistant features in its pipeline and introduces a noise-aware learning scheme to effectively improve the robustness of multimodal emotion understanding. Our new pipeline, namely Noise-Resistant Multimodal Transformer (NORM-TR), mainly introduces a Noise-Resistant Generic Feature (NRGF) extractor and a Transformer for the multimodal emotion recognition task. In particular, we make the NRGF extractor learn a generic and disturbance-insensitive representation so that consistent and meaningful semantics can be obtained. Furthermore, we apply a Transformer to incorporate Multimodal Features (MFs) of multimodal inputs based on their relations to the NRGF. Therefore, the possible insensitive but useful information of NRGF could be complemented by MFs that contain more details. To train the NORM-TR properly, our proposed noise-aware learning scheme complements normal emotion recognition losses by enhancing the learning against noises. Our learning scheme explicitly adds noises to either all the modalities or a specific modality at random locations of a multimodal input sequence. We correspondingly introduce two adversarial losses to encourage the NRGF extractor to learn to extract the NRGFs invariant to the added noises, thus facilitating the NORM-TR to achieve more favorable multimodal emotion recognition performance. In practice, on several popular multimodal datasets, our NORM-TR achieves state-of-the-art performance and outperforms existing methods by a large margin, which demonstrates that the ability to resist noisy information is important for effective emotion recognition.