Multi-behavior recommendation faces a critical challenge in practice: auxiliary behaviors (e.g., clicks, carts) are often noisy, weakly correlated, or semantically misaligned with the target behavior (e.g., purchase), which leads to biased preference learning and suboptimal performance. While existing methods attempt to fuse these heterogeneous signals, they inherently lack a principled mechanism to ensure robustness against such behavioral inconsistency. In this work, we propose Robust Multi-Behavior Recommendation towards Target Behaviors (RMBRec), a robust multi-behavior recommendation framework grounded in an information-theoretic robustness principle. We interpret robustness as a joint process of maximizing predictive information while minimizing its variance across heterogeneous behavioral environments. Under this perspective, the Representation Robustness Module (RRM) enhances local semantic consistency by maximizing the mutual information between users' auxiliary and target representations, whereas the Optimization Robustness Module (ORM) enforces global stability by minimizing the variance of predictive risks across behaviors, which is an efficient approximation to invariant risk minimization. This local-global collaboration bridges representation purification and optimization invariance in a theoretically coherent way. Extensive experiments on three real-world datasets demonstrate that RMBRec not only outperforms state-of-the-art methods in accuracy but also maintains remarkable stability under various noise perturbations. For reproducibility, our code is available at https://github.com/miaomiao-cai2/RMBRec/.

Popularity bias is a common challenge in recommender systems. It often causes unbalanced item recommendation performance and intensifies the Matthew effect. Due to limited user-item interactions, unpopular items are frequently constrained to the embedding neighborhoods of only a few users, leading to representation collapse and weakening the model's generalization. Although existing supervised alignment and reweighting methods can help mitigate this problem, they still face two major limitations: (1) they overlook the inherent variability among different Graph Convolutional Networks (GCNs) layers, which can result in negative gains in deeper layers; (2) they rely heavily on fixed hyperparameters to balance popular and unpopular items, limiting adaptability to diverse data distributions and increasing model complexity. To address these challenges, we propose Graph-Structured Dual Adaptation Framework (GSDA), a dual adaptive framework for mitigating popularity bias in recommendation. Our theoretical analysis shows that supervised alignment in GCNs is hindered by the over-smoothing effect, where the distinction between popular and unpopular items diminishes as layers deepen, reducing the effectiveness of alignment at deeper levels. To overcome this limitation, GSDA integrates a hierarchical adaptive alignment mechanism that counteracts entropy decay across layers together with a distribution-aware contrastive weighting strategy based on the Gini coefficient, enabling the model to adapt its debiasing strength dynamically without relying on fixed hyperparameters. Extensive experiments on three benchmark datasets demonstrate that GSDA effectively alleviates popularity bias while consistently outperforming state-of-the-art methods in recommendation performance.

Provider exposure fairness is crucial for sustaining a healthy content ecosystem and preventing monopolization in recommender systems. Yet, most existing methods either incorporate fairness constraints during model training, requiring expensive retraining when fairness objectives change, or rely on post-hoc reranking with fixed criteria, which lacks adaptability to diverse fairness requirements. To overcome these limitations, we propose Post-hoc Fairness Adaptation (PFA), a lightweight framework that equips a frozen recommender with a fairness adapter, enabling flexible fairness control without retraining the backbone model. Specifically, the fairness adapter learns personalized additive score adjustments from user-item embeddings, which are injected into the original ranking scores to steer provider exposure toward fairness. To train the adapter, we minimize the KL divergence between the actual and the target fair exposure distributions. However, this global objective implicitly treats all providers equally, ignoring structural disparities such as imbalanced provider group sizes and heterogeneous exposure within groups. Consequently, fairness may appear satisfied at an aggregate level while severe inter-group and intra-group exposure imbalances persist, undermining practical fairness. To address this, we design Hierarchical Exposure Fairness Alignment (HEFA), which explicitly balances inter- and intra-group provider exposure disparities, enabling flexible adaptation to diverse fairness requirements. To mitigate potential accuracy degradation, PFA jointly optimizes HEFA with a differentiable NDCG loss, enabling end-to-end fairness optimization while preserving ranking quality. Extensive experiments on three public datasets demonstrate that PFA achieves substantial fairness gains with negligible accuracy loss, consistently outperforming strong baselines.

Transformer-based Single Image Deraining (SID) methods have achieved remarkable success, primarily attributed to their robust capability in capturing long-range interactions. However, we've noticed that current methods handle rain-affected and unaffected regions concurrently, overlooking the disparities between these areas, resulting in confusion between rain streaks and background parts, and inabilities to obtain effective interactions, ultimately resulting in suboptimal deraining outcomes. To address the above issue, we introduce the Region Transformer (Regformer), a novel SID method that underlines the importance of independently processing rain-affected and unaffected regions while considering their combined impact for high-quality image reconstruction. The crux of our method is the innovative Region Transformer Block (RTB), which integrates a Region Masked Attention (RMA) mechanism and a Mixed Gate Forward Block (MGFB). Our RTB is used for attention selection of rain-affected and unaffected regions and local modeling of mixed scales. The RMA generates attention maps tailored to these two regions and their interactions, enabling our model to capture comprehensive features essential for rain removal. To better recover high-frequency textures and capture more local details, we develop the MGFB as a compensation module to complete local mixed scale modeling. Extensive experiments demonstrate that our model reaches state-of-the-art performance, significantly improving the image deraining quality. Our code and trained models are publicly available.

In this paper, we tackle the Aerial Vision-and-Dialog Navigation (AVDN) task in the training-free setting for resource-efficient high-altitude UAV navigation.Naively applying MLLMs leads to unreliable navigation due to weak directional grounding and the lack of explicit spatial memory.To address these issues, we propose PSC-AVDN, a training-free framework that tightly couples a three-stage Parsing-Search-Confirmation reasoning pipeline with a Structured Spatial Memory (SSM).The parsing stage uses an LLM to convert ambiguous dialogue instructions into stable geometric directional and destination cues.A Search Chain-of-Thought (S-CoT) then performs stepwise target exploration under high-altitude observations, and a Confirmation Chain-of-Thought (C-CoT) conducts fine-grained verification around candidate regions to resolve visual ambiguity.Meanwhile, SSM integrates three complementary sources of spatial cues, including multi-scale visual observation, spatial visual memory, and structured geometric memory to provide global spatial context and long-horizon consistency.Extensive experiments on ANDH and ANDH-Full show that PSC-AVDN establishes new state-of-the-art performance in the training-free setting, matching or surpassing several finetuned methods.Code will be publicly available at: https://github.com/QY6616/PSC-AVDN

Recent advances in generative models have sparked growing interest in moving beyond pure image generation toward transparent image generation, i.e., joint generation of image and its alpha mask. However, most existing approaches adopt a two-stage pipeline, where a diffusion-based model first generates an RGB image and a subsequent matting head predicts the alpha mask. This separation not only leads to error accumulation and inaccurate predictions but also overlooks the intrinsic correlation between the cross-modal data. In this work, we introduce Zippo, a unified diffusion framework, zipping color and transparency distributions into a single diffusion model, by learning joint distribution of RGB image and alpha mask. Zippo not only generates high-fidelity images but also produces plausible and sharp alpha masks. In practice, Zippo inflates the latent space into a unified representation that encodes cross-modal data, and builds upon it with a modality-aware diffusion process that flexibly switches between RGB and alpha domains. In this process, conditioning on one modality while denoising the other allows the model to generate RGB images from alpha masks and predict transparency from input images. In addition to single-modality prediction, we further design a modality-aware noise reassignment strategy to empower Zippo with the joint generation capability of RGB images and their corresponding alpha masks under text guidance. With these techniques, Zippo supports a wide range of transparent image generation tasks, including image-alpha joint generation, image matting, and alpha mask conditioned image generation. Extensive experiments demonstrate that Zippo not only delivers superior visual fidelity but also achieves competitive performance in visual downstream prediction, highlighting joint image-alpha modeling as a powerful alternative to traditional paradigms.

Recent advances in video reward models and post-training strategies have improved text-to-video (T2V) generation. While these models typically assess visual quality, motion quality, and text alignment, they often overlook key structural distortions, such as abnormal object appearances and interactions, which can degrade the overall quality of the generative video.To address this gap, we introduce REACT, a frame-level reward model designed specifically for structural distortions evaluation in generative videos. REACT assigns point-wise scores and attribution labels by reasoning over video frames, focusing on recognizing distortions. To support this, we construct a large-scale human preference dataset, annotated based on our proposed taxonomy of structural distortions, and generate additional data using a efficient Chain-of-Thought (CoT) synthesis pipeline. REACT is trained with a two-stage framework: ((1) supervised fine-tuning with masked loss for domain knowledge injection, followed by (2) reinforcement learning with Group Relative Policy Optimization (GRPO) and pairwise rewards to enhance reasoning capability and align output scores with human preferences. During inference, a dynamic sampling mechanism is introduced to focus on frames most likely to exhibit distortion.We also present REACT-Bench, a benchmark for generative video distortion evaluation. Experimental results demonstrate that REACT complements existing reward models in assessing structural distortion, achieving both accurate quantitative evaluations and interpretable attribution analysis.

Constantly discovering novel concepts is crucial in evolving environments. This paper explores the underexplored task of Continual Generalized Category Discovery (C-GCD), which aims to incrementally discover new classes from unlabeled data without forgetting old classes. In this work, we study a more practical setting, which includes more new classes over a longer period, without storing samples of past classes. In C-GCD, the model is initially trained on labeled data, followed by multiple incremental stages that incorporate both old and new classes in unlabeled data. The core challenge involves two conflicting objectives: discover new classes and prevent forgetting old ones. We identify two types of bias to which models are susceptible: prediction bias, where the model tends to predict new classes as old ones, and hardness bias, where forgetting on some particularly difficult old classes is exceptionally severe. To address these issues, we introduce a debiased learning framework, namely Happy, characterized by Hardness-aware prototype sampling and soft entropy regularization. For prediction bias, we first introduce clustering-guided initialization to provide robust features. In addition, we propose soft entropy regularization to assign appropriate probabilities to new classes, significantly enhancing their clustering performance. For hardness bias, we present hardness-aware prototype sampling, which effectively reduces the forgetting of previous classes. Building upon Happy, we further propose Happy++ by incorporating three training strategies, achieving more stable and generalizable performance. We further broaden the task scope by extending the model to detect out-of-distribution samples, which are commonly present in practical scenarios, thereby serving as a unified open-world classifier, called Happy++O. Experimental results demonstrate that our method proficiently manages the conflicts of C-GCD and achieves remarkable performance across various settings.

Micro-Actions (MAs) are subtle and spontaneous human behaviors that provide important non-verbal cues in social interaction and affective communication. However, their short duration, weak motion patterns, and fine-grained semantic differences make them difficult to annotate, model, and evaluate in a standardized manner. To promote academic research on micro-action analysis, we proposed and have annually organized the Micro-Action Analysis Grand Challenge (MAC) as a public benchmark platform for this emerging field. The first two editions of MAC established standardized evaluation settings for micro-action recognition and detection, providing publicly accessible datasets and protocols. Building upon these editions, this paper presents the 3rd MAC, held in conjunction with ACM Multimedia 2026. Under the theme of moving from recognition to fine-grained micro-action understanding, this edition further expands the scope of the challenge beyond conventional recognition and detection. In particular, we introduce a new task named fine-grained micro-action understanding, evaluated with the assistance of multimodal large language models, aiming to assess models' ability to capture fine-grained semantic cues and interpret subtle human micro-actions at a deeper level. We summarize the datasets, task settings, evaluation protocols, competition results, and representative solutions from top-performing teams. Finally, we discuss future directions for micro-action analysis and its broader role in human-centric video understanding.

Pose-guided person image synthesis aims to align a source image’s appearance with target pose features. Existing methods often fail to capture fine semantic details in complex scenes, causing texture distortion and misalignment under nonlinear deformations. We propose a diffusion-based framework with three components: (1) a Decoupled Learning Module that separates high-frequency textures, semantic structures, and colors to avoid feature interference; (2) a Multi-Scale Progressive Fusion Network that aligns appearance and pose via coarse-to-fine fusion with deformable convolutions, ensuring structural consistency under large pose changes; and (3) a Semantic-Aware Enhancement Module that introduces a Mamba-Like Linear Attention positional encoding to simulate a forgetting-gate effect, improving semantic modeling and reducing distortion. Experiments show clear gains in SSIM, FID, PSNR, and LPIPS, and visual results confirm superior preservation of fine details and robustness to large pose variations.

While existing time series foundation models primarily rely on large-scale unimodal pretraining, they lack complementary modalities to enhance time series understanding. Building multimodal foundation models is a natural next step, but it faces key challenges: 1) lack of a unified multimodal pretraining paradigm and large-scale multimodal corpora for time series analysis; 2) how to effectively integrate heterogeneous modalities and enhance model generalization. To address these challenges, we take an early step toward multimodal foundation models for time series analysis. We first propose a multimodal pretraining paradigm that leverages time series with endogenous modalities (derived images and text) and exogenous knowledge (real-world news), providing a comprehensive multi-view perspective for time series analysis. To support this, we develop an automated data construction pipeline to curate MM-TS, the first large-scale multimodal time series dataset spanning six domains, with up to one billion points. Then we propose HORAI, a frequency-enhanced multimodal foundation model. It integrates two core components: the Frequency-enhanced Cross-Modality Encoder and the Time-Frequency Decoder, designed to effectively fuse multimodal features and enhance model generalization across modalities and domains. After pretraining on MM-TS, HORAI achieves state-of-the-art zero-shot performance on time series forecasting and anomaly detection tasks, demonstrating strong generalization.

This work aims to tackle the all-in-one image restoration task, which seeks to handle multiple types of degradation with a single model. The primary challenge is to extract degradation representations from the input degraded images and use them to guide the model's adaptation to specific degradation types. Building on the insight that various degradations affect image content differently across frequency bands, we propose a new dual-transformer approach comprising two components: a frequency-aware degradation estimation transformer (Dformer) and a degradation-adaptive restoration transformer (Rformer). The Dformer captures the essential characteristics of various degradations by decomposing the input into different frequency components. By understanding how degradations affect these frequency components, the Dformer learns robust priors that effectively guide the restoration process. The Rformer then employs a degradation-adaptive self-attention (DA-SA) module to selectively focus on the most affected frequency components, guided by the learned degradation representations. Extensive experimental results demonstrate that our approach outperforms existing methods in five representative restoration tasks, including denoising, deraining, dehazing, deblurring, and low-light enhancement. In addition, our method offers benefits for handling real-world degradations, spatially variant degradations, and unseen degradation levels.

Binary neural networks (BNNs) have emerged as a promising solution for efficient low-level vision information processing, particularly in resource-constrained environments. However, applying BNNs to low-light image enhancement (LLIE) remains almost unexplored. A significant challenge in current BNNs is their limited ability to effectively address various degradation factors present in low-light images. To overcome this, we propose a framework named Binarized Low-Light Image Enhancement (BiLLIE), enabling smooth binarization of LLIE models. By introducing Adaptive Information Flow mechanism, the degradation factors in low-light images can be effectively controlled. In addition, we employ a two-step training strategy, aiming to gradually binarize LLIE models. We evaluated BiLLIE against seven BNN models on three binarized LLIE models. Experimental results demonstrate that BiLLIE achieves superior enhancement performance while significantly reducing computational and storage costs.

Existing video inpainting methods typically combine optical flow propagation with Transformer architectures, achieving promising inpainting results. However, they lack adaptive inpainting strategy optimization in diverse scenarios, and struggle to capture high-level temporal semantics, causing temporal inconsistencies and quality degradation. To address these challenges, we make one of the first attempts to introduce reinforcement learning into the video inpainting domain, establishing a closed-loop framework named CLIP-RL that enables adaptive strategy optimization. Specifically, video inpainting is reformulated as an agent-environment interaction, where the inpainting module functions as the agent's execution component, and a pre-trained inpainting detection module provides real-time quality feedback. Guided by a policy network and a composite reward function that incorporates a weighted temporal alignment loss, the agent dynamically selects actions to adjust the inpainting strategy and iteratively refines the inpainting results. Compared to ProPainter, CLIP-RL improves PSNR from 34.43 to 34.67 and SSIM from 0.974 to 0.986 on the YouTube-VOS dataset. Qualitative analysis demonstrates that CLIP-RL excels in detail preservation and artifact suppression, validating its superiority in video inpainting tasks.

Remote photoplethysmography (rPPG) enables contactless physiological monitoring by capturing subtle skin-color variations from facial videos. However, most existing methods predominantly rely on time-domain modeling, making them vulnerable to motion artifacts and illumination fluctuations, where weak physiological clues are easily overwhelmed by noise. To address these challenges, we propose FreqPhys, a frequency-guided rPPG framework that explicitly leverages physiological frequency priors for robust signal recovery. Specifically, FreqPhys first applies a Physiological Bandpass Filtering module to suppress out-of-band interference, and then performs Physiological Spectrum Modulation together with adaptive spectral selection to emphasize pulse-related frequency components while suppress residual in-band noise. A Cross-domain Representation Learning module further fuses these spectral priors with deep time-domain features to capture informative spatial–temporal dependencies. Finally, a frequency-aware conditional diffusion process progressively reconstructs high-fidelity rPPG signals. Extensive experiments on six benchmarks demonstrate that FreqPhys yields significant improvements over state-of-the-art approaches, particularly under challenging motion conditions. It highlights the importance of explicitly modeling physiological frequency priors. The source code will be released.

As large language models (LLMs) increasingly shape content generation, interaction, and decision-making across the Web, aligning them with human values has become a central objective in trustworthy AI. This challenge becomes even more pronounced when aligning multiple, potentially conflicting human values. Although recent approaches, such as reward reweighting, prompt-based supervised fine-tuning, and model merging, attempt to tackle multi-value alignment, they still face two major limitations: (1) training separate models for each value combination is prohibitively expensive; (2) value conflicts substantially degrade alignment performance. These limitations make it difficult to achieve favorable trade-offs across diverse human values. To address these challenges, we revisit multi-value alignment from the perspective of value consistency in data and propose VC-soup, a data filtering and parameter merging framework grounded in value-consistent learning. We first design a value consistency metric based on the cosine similarity between the reward-gap vector of each preference pair and an all-ones vector, which quantifies its cross-value coherence. We then filter out low-consistency preference pairs in each value dataset and train on the remaining data to obtain smooth, value-consistent policy models that better preserve linear mode connectivity. Finally, we linearly combine these policies and apply Pareto filtering across values to obtain solutions with balanced multi-value performance. Extensive experiments and theoretical analysis demonstrate that VC-soup effectively mitigates conflicts and consistently outperforms existing multi-value alignment methods.

Prompt tuning, a recently emerging paradigm, adapts vision-language pre-trained models to new tasks efficiently by learning "soft prompts" for frozen models. However, in few-shot scenarios, its effectiveness is limited by sensitivity to the initialization and the time-consuming search for optimal initialization, hindering rapid adaptation. Additionally, prompt tuning risks reducing the models' generalizability due to overfitting on scarce training samples. To overcome these challenges, we introduce a novel Gradient-RegulAted Meta-prompt learning (GRAM) framework that jointly meta-learns an efficient soft prompt initialization for better adaptation and a lightweight gradient regulating function for strong cross-domain generalizability in a meta-learning paradigm using only the weakly labeled image-text pre-training data. This is achieved through a Cross-Modal Hierarchical Clustering algorithm that organizes extensive image-text data into a structured hierarchy, facilitating robust meta-learning across diverse domains. Rather than designing a specific prompt tuning method, our GRAM can be easily incorporated into various prompt tuning methods in a model-agnostic way and bring about consistent improvement for them. Further, we consider a more practical but challenging setting: test-time prompt tuning with only unlabeled test samples and propose an improved structure-induced gradient regulating function to leverage the structured semantics of the meta-learning data for zero-shot generalization. This novel approach exploits the hierarchically clustered meta-learning data to model relationships between test-time data and meta-learning prototypes, facilitating the transfer of invariant knowledge without explicit annotations. Meanwhile, we introduce a structure complexity-informed strategy for adaptively constructing meta-training tasks and generating prototypes, which fully considers the diverse semantics within hierarchical clusters of different complexities. Comprehensive experiments demonstrate the state-of-the-art few- and zero-shot generalizability of our method.

Recent advancements in image synthesis, particularly with the advent of GAN and Diffusion models, have amplified public concerns regarding the dissemination of disinformation. To address such concerns, numerous AI-generated Image (AIGI) Detectors have been proposed and achieved promising performance in identifying fake images. However, there still lacks a systematic understanding of the adversarial robustness of AIGI detectors. In this paper, we examine the vulnerability of state-of-the-art AIGI detectors against adversarial attack under white-box and black-box settings, which has been rarely investigated so far. To this end, we propose a new method to attack AIGI detectors. First, inspired by the obvious difference between real images and fake images in the frequency domain, we add perturbations under the frequency domain to push the image away from its original frequency distribution. Second, we explore the full posterior distribution of the surrogate model to further narrow this gap between heterogeneous AIGI detectors, e.g. transferring adversarial examples across CNNs and ViTs. This is achieved by introducing a novel post-train Bayesian strategy that turns a single surrogate into a Bayesian one, capable of simulating diverse victim models using one pre-trained surrogate, without the need for re-training. We name our method as Frequency-based Post-train Bayesian Attack, or FPBA. Through FPBA, we show that adversarial attack is truly a real threat to AIGI detectors, because FPBA can deliver successful black-box attacks across models, generators, defense methods, and even evade cross-generator detection, which is a crucial real-world detection scenario. The code will be shared upon acceptance.

Mental health understanding in long-form videos requires nuanced reasoning over observable behavior, interpersonal context, and latent psychological states. Existing benchmarks largely reduce this task to coarse-grained classification, providing limited insight into whether models truly understand psychological phenomena or rely on superficial correlations. To address this limitation, we introduce MMHBench, a comprehensive multimodal benchmark for multi-perspective mental health understanding, comprising 268 long-form videos and 2,184 carefully curated questions. MMHBench organizes the evaluation into two complementary settings: (1) third-person assessment, consisting of 605 questions that focus on the interpretation of observable behaviors and multimodal evidence, and (2) first-person perspective-taking, comprising 1,579 questions that require perspective-conditioned reasoning to identify the interpretation of the mental state supported by the available multimodal evidence. We propose a Multi-Agent Question Generation (MAQG) framework that simulates diverse social roles to synthesize questions from multiple perspectives. The generated questions are refined through multi-role feedback and iterative optimization, followed by expert-guided verification to ensure quality and validity. Extensive evaluation of 22 representative multimodal large language models (MLLMs), spanning both open-source and leading closed-source models, demonstrates that long-form video mental health understanding remains highly challenging.