Multimodal Large Language Models (MLLMs) are increasingly being deployed as automated content moderators. Within this landscape, we uncover a critical threat: Adversarial Smuggling Attacks. Unlike adversarial perturbations (for misclassification) and adversarial jailbreaks (for harmful output generation), adversarial smuggling exploits the Human-AI capability gap. It encodes harmful content into human-readable visual formats that remain AI-unreadable, thereby evading automated detection and enabling the dissemination of harmful content. We classify smuggling attacks into two pathways: (1) Perceptual Blindness, disrupting text recognition; and (2) Reasoning Blockade, inhibiting semantic understanding despite successful text recognition. To evaluate this threat, we constructed SmuggleBench, the first comprehensive benchmark comprising 1,700 adversarial smuggling attack instances. Evaluations on SmuggleBench reveal that both proprietary (e.g., GPT-5) and open-source (e.g., Qwen3-VL) state-of-the-art models are vulnerable to this threat, producing Attack Success Rates (ASR) exceeding 90

The past year has seen over 20 open-source document parsing models, yet thefield still benchmarks almost exclusively on OmniDocBench, a 1,355-pagemanually annotated dataset whose top scores have saturated above 90

This paper introduces the concept of Microscopic Spatial Intelligence (MiSI), the capability to perceive and reason about the spatial relationships of invisible microscopic entities, which is fundamental to scientific discovery. To assess the potential of Vision-Language Models (VLMs) in this domain, we propose a systematic benchmark framework MiSI-Bench. This framework features over 163,000 question-answer pairs and 587,000 images derived from approximately 4,000 molecular structures, covering nine complementary tasks that evaluate abilities ranging from elementary spatial transformations to complex relational identifications. Experimental results reveal that current state-of-the-art VLMs perform significantly below human level on this benchmark. However, a fine-tuned 7B model demonstrates substantial potential, even surpassing humans in spatial transformation tasks, while its poor performance in scientifically-grounded tasks like hydrogen bond recognition underscores the necessity of integrating explicit domain knowledge for progress toward scientific AGI. The datasets are available at https://huggingface.co/datasets/zongzhao/MiSI-bench.

Knowledge-based Visual Question Answering (KB-VQA) requires models to retrieve visual entities matching the query image from large-scale encyclopedic knowledge bases and answer related questions. Existing multimodal Retrieval Augmented Generation (mRAG) methods rely on global visual features to match candidate entities, yet when the knowledge base contains numerous visually similar entities, the retriever struggles to distinguish them, populating the candidate set with visually similar but factually mismatched distractors. Since subsequent processing steps such as noise filtering are also confined to this fixed candidate set, errors from failed retrieval inevitably propagate to the final answer. To address these challenges, we propose MMAgent-R^2, an agentic mRAG framework that integrates visual reranking and active rejection as its internal verification mechanism. Visual reranking directly compares query and candidate images, capturing discriminative details beyond textual descriptions to precisely identify the target entity among similar candidates; active rejection discards unreliable results and retrieves additional candidates when no confident match is found, moving beyond the fixed candidate pool. We design a composite reward function with step-level verification rewards and achieve joint optimization of external retrieval, internal verification, and answer generation via GRPO training. Experiments on InfoSeek, E-VQA, and MMhops demonstrate that achieves state-of-the-art performance, with particularly notable advantages in challenging retrieval scenarios and complex multi-image multi-hop reasoning tasks.

Multimodal tagging is essential for content understanding by assigning concise, semantically relevant tags to visual inputs. However, real-world tagging is inherently open-ended: user-generated content is noisy, long-tailed, and continuously evolving, challenging conventional closed-set or open-vocabulary classification methods. We propose Open-Tag, a generative framework for Open-world Multimodal Tagging that produces unordered, variable-length tag sequences in natural language without relying on predefined tag sets. Open-Tag introduces two key innovations: (1) an Order-Prompted Tag Sequence Generation that maps learnable, order-agnostic queries to latent tag semantics, enabling permutation-invariant tag generation, and (2) a Multi-Source Retrieval-Augmented Generation that fuses tag candidates from heterogeneous retrieval systems across visual, textual, and metadata modalities. A score normalization and aggregation strategy ensures robust fusion, enhancing the diversity and grounding of generated tags. To evaluate Open-Tag, we construct two large-scale datasets: CREATE-Tag (Chinese video) and PEXEL-Tag (English image), with over 3M videos and 160K images with tens of thousands of real-user tags. We propose a novel open-set evaluation metric, Tag Gain, to quantify the generation of relevant but previously unseen tags. Experiments show that Open-Tag outperforms state-of-the-art baselines on closed-set F1 and open-set Tag Gain, highlighting its generalization and novel tag discovery capabilities. Project webpage: https://createbenchmark.github.io/open-tag/ .

The ability to perform multi-modal multi-hop reasoning by iteratively integrating information across various modalities and external knowledge is critical for addressing complex real-world challenges. However, existing Multi-modal Large Language Models (MLLMs) are predominantly limited to single-step reasoning, as existing benchmarks lack the complexity needed to evaluate and drive multi-hop abilities. To bridge this gap, we introduce MMhops, a novel, large-scale benchmark designed to systematically evaluate and foster multi-modal multi-hop reasoning. MMhops dataset comprises two challenging task formats, Bridging and Comparison, which necessitate that models dynamically construct complex reasoning chains by integrating external knowledge. To tackle the challenges posed by MMhops, we propose MMhops-R1, a novel multi-modal Retrieval-Augmented Generation (mRAG) framework for dynamic reasoning. Our framework utilizes reinforcement learning to optimize the model for autonomously planning reasoning paths, formulating targeted queries, and synthesizing multi-level information. Comprehensive experiments demonstrate that MMhops-R1 significantly outperforms strong baselines on MMhops, highlighting that dynamic planning and multi-modal knowledge integration are crucial for complex reasoning. Moreover, MMhops-R1 demonstrates strong generalization to tasks requiring fixed-hop reasoning, underscoring the robustness of our dynamic planning approach.

Composed Image Retrieval (CIR) has demonstrated significant potential by enabling flexible, multimodal queries that combine a reference image and modification text.However, CIR inherently prioritizes semantic matching, struggling to reliably retrieve a user-specified instance across contexts. In practice, emphasizing concrete instance fidelity over broad semantics is often more consequential.In this work, we propose O bject- A nchored C omposed I mage R etrieval ( OACIR ), a novel fine-grained retrieval task that mandates strict instance-level consistency.To advance research on this task, we construct OACIRR ( OACIR on R eal-world images), the first large-scale, multi-domain benchmark comprising over 160K quadruples and four challenging candidate galleries enriched with hard-negative instance distractors.Each quadruple augments the compositional query with a bounding box that visually anchors the object in the reference image, providing a precise and flexible way to ensure instance preservation.To perform the OACIR task, we propose AdaFocal , a framework featuring a Context-Aware Attention Modulator that adaptively intensifies attention within the specified instance region, dynamically balancing focus between the anchored instance and the broader compositional context.Extensive experiments demonstrate that AdaFocal substantially outperforms existing compositional retrieval models, particularly in maintaining instance-level fidelity, thereby establishing a robust baseline for this challenging task while opening new directions for more flexible, instance-aware retrieval systems.

Advanced Multimodal Large Language Models (MLLMs) struggle with recent Knowledge-based Visual Question Answering (VQA) tasks, such as INFOSEEK and Encyclopedic-VQA, due to their limited and frozen knowledge scope, often leading to ambiguous and inaccurate responses. Thus, multimodal Retrieval-Augmented Generation (mRAG) is naturally introduced to provide MLLMs with comprehensive and up-to-date knowledge, effectively expanding the knowledge scope. However, current mRAG methods have inherent drawbacks, including: 1) Performing retrieval even when external knowledge is not needed. 2) Lacking of identification of evidence that supports the query. 3) Increasing model complexity due to additional information filtering modules or rules. To address these shortcomings, we propose a novel generalized framework called multimodal Retrieval-Reflection-Augmented Generation (mR$^{2}$AG), which achieves adaptive retrieval and useful information localization to enable answers through two easy-to-implement reflection operations, preventing high model complexity. In mR$^{2}$AG, Retrieval-Reflection is designed to distinguish different user queries and avoids redundant retrieval calls, and Relevance-Reflection is introduced to guide the MLLM in locating beneficial evidence of the retrieved content and generating answers accordingly. In addition, mR$^{2}$AG can be integrated into any well-trained MLLM with efficient fine-tuning on the proposed mR$^{2}$AG Instruction-Tuning dataset (mR$^{2}$AG-IT). mR$^{2}$AG significantly outperforms state-of-the-art MLLMs (e.g., GPT-4o) and mRAG-based MLLMs on INFOSEEK and Encyclopedic-VQA, while maintaining the exceptional capabilities of base MLLMs across a wide range of Visual-dependent tasks.

Multimodal Large Language Models (MLLMs) remain far from human-level performance in multi-view spatial reasoning, where models must establish object correspondences across view and infer coherent scene semantics. We analyze this limitation through the Transformation-Driven Visual Reasoning (TVR) task and find that Supervised Fine-Tuning (SFT) fails to capture cross-view consistency, whereas reinforcement learning (RL) fails to reliably identify key referential objects. To bridge this gap, we introduce multi-View Spatial TrAnsformation Reasoning (STAR-R1), a two-stage framework that combines process-supervised SFT with a referential-aware RL paradigm. STAR-R1 first learns structured spatial reasoning trajectories from high-quality CoTs and then uses fine-grained rewards on referential selection and answer correctness to encourage effective exploration and robust scene interpretation. Despite using only a small amount of high-quality training data, STAR-R1 surpasses state-of-the-art models with far more training data on the multi-view spatial understanding benchmarks TVR, MMSI-Bench, MindCube-Bench, and SPAR-Bench. Our study reveals the overlooked potential of RL in multi-view spatial understanding and points a way toward potentially achieving more human-like spatial reasoning in MLLMs.

Composed Image Retrieval (CIR) aims to retrieve target images from a gallery based on a reference image and modification text as a combined query. Recent approaches focus on balancing global information from two modalities and encode the query into a unified feature for retrieval. However, due to insufficient attention to fine-grained details, these coarse fusion methods often struggle with handling subtle visual alterations or intricate textual instructions. In this work, we propose DetailFusion, a novel dual-branch framework that effectively coordinates information across global and detailed granularities, thereby enabling detail-enhanced CIR. Our approach leverages atomic detail variation priors derived from an image editing dataset, supplemented by a detail-oriented optimization strategy to develop a Detail-oriented Inference Branch. Furthermore, we design an Adaptive Feature Compositor that dynamically fuses global and detailed features based on fine-grained information of each unique multimodal query. Extensive experiments and ablation analyses not only demonstrate that our method achieves state-of-the-art performance on both CIRR and FashionIQ datasets but also validate the effectiveness and cross-domain adaptability of detail enhancement for CIR.

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities across diverse tasks, yet they lag significantly behind humans in spatial reasoning. We investigate this gap through Transformation-Driven Visual Reasoning (TVR), a challenging task requiring identification of object transformations across images under varying viewpoints. While traditional Supervised Fine-Tuning (SFT) fails to generate coherent reasoning paths in cross-view settings, sparse-reward Reinforcement Learning (RL) suffers from inefficient exploration and slow convergence. To address these limitations, we propose STAR-R1, a novel framework that integrates a single-stage RL paradigm with a fine-grained reward mechanism tailored for TVR. Specifically, STAR-R1 rewards partial correctness while penalizing excessive enumeration and passive inaction, enabling efficient exploration and precise reasoning. Comprehensive evaluations demonstrate that STAR-R1 achieves state-of-the-art performance across all 11 metrics, outperforming SFT by 23

Large language models (LLMs) are trained on vast amounts of text from the Internet, but do they truly understand the viral content that rapidly spreads online—commonly known as memes? In this paper, we introduce CHIME, a dataset for CHinese Internet Meme Explanation. The dataset comprises popular phrase-based memes from the Chinese Internet, annotated with detailed information on their meaning, origin, example sentences, types, etc. To evaluate whether LLMs understand these memes, we designed two tasks. In the first task, we assessed the models’ ability to explain a given meme, identify its origin, and generate appropriate example sentences. The results show that while LLMs can explain the meanings of some memes, their performance declines significantly for culturally and linguistically nuanced meme types. Additionally, they consistently struggle to provide accurate origins for the memes. In the second task, we created a set of multiple-choice questions (MCQs) requiring LLMs to select the most appropriate meme to fill in a blank within a contextual sentence. While the evaluated models were able to provide correct answers, their performance remains noticeably below human levels. We have made CHIME public and hope it will facilitate future research on computational meme understanding.

Compared to traditional sentiment analysis, which only considers text, multimodal sentiment analysis needs to consider emotional signals from multimodal sources simultaneously and is therefore more consistent with the way how humans process sentiment in real-world scenarios. It involves processing emotional information from various sources such as natural language, images, videos, audio, physiological signals, etc. However, although other modalities also contain diverse emotional cues, natural language usually contains richer contextual information and therefore always occupies a crucial position in multimodal sentiment analysis. The emergence of ChatGPT has opened up immense potential for applying large language models (LLMs) to text-centric multimodal tasks. However, it is still unclear how existing LLMs can adapt better to text-centric multimodal sentiment analysis tasks. This survey aims to (1) present a comprehensive review of recent research in text-centric multimodal sentiment analysis tasks, (2) examine the potential of LLMs for text-centric multimodal sentiment analysis, outlining their approaches, advantages, and limitations, (3) summarize the application scenarios of LLM-based multimodal sentiment analysis technology, and (4) explore the challenges and potential research directions for multimodal sentiment analysis in the future.

Enhancing the fine-grained instance spatiotemporal motion perception capabilities of Video Large Language Models is crucial for improving their temporal and general video understanding. However, current models struggle to perceive detailed and complex instance motions. To address these challenges, we have made improvements from both data and model perspectives. In terms of data, we have meticulously curated iMOVE-IT, the first large-scale instance-motion-aware video instruction-tuning dataset. This dataset is enriched with comprehensive instance motion annotations and spatiotemporal mutual-supervision tasks, providing extensive training for the model's instance-motion-awareness. Building on this foundation, we introduce iMOVE, an instance-motion-aware video foundation model that utilizes Event-aware Spatiotemporal Efficient Modeling to retain informative instance spatiotemporal motion details while maintaining computational efficiency. It also incorporates Relative Spatiotemporal Position Tokens to ensure awareness of instance spatiotemporal positions. Evaluations indicate that iMOVE excels not only in video temporal understanding and general video understanding but also demonstrates significant advantages in long-term video understanding.

Mathematical problems in real-world scenarios are often presented in a purely vision-form, where textual problem statement and accompanying math figures, e.g., geometry figures and functional graphs, are integrated into a single image. This vision-form problem-solving task requires precise comprehension and reasoning on both textual and graphical elements in the images, posing significant challenge to current Multimodal Large Language Models (MLLMs), which process text and math figures in isolation. In this work, we propose VisionMath, the first exploration for vision-form mathematical problem-solving model, which employs a three-stage progressive multimodal reasoning alignment strategy to systematically enhance task-specific capabilities. Building upon a LLM proficient in unimodal mathematical reasoning, VisionMath first establishes foundational OCR capabilities through capturing rendered mathematical problem images. Subsequently, the model develops comprehensive understanding of figure structures and properties via learning from figure descriptions and mathematical educational videos. Finally, the model's reasoning capacity is activated using carefully constructed visual-form problemsolving datasets VisionMath-IT with chain-of-thought annotations. For comprehensive evaluation, we construct multilingual benchmarks covering diverse problem types, including geometry, algebra, function problems in both English and Chinese. Experimental results demonstrate that VisionMath significantly outperforms existing general-purpose and math-specialized models, showcasing superior multilingual and multi-type problem-solving capabilities. Our model weights, data and code will be made available at https://github.com/mengqiDyangge/VisionMath.

Understanding the content of events occurring in the video and their inherent temporal logic is crucial for video-text retrieval. However, web-crawled pre-training datasets often lack sufficient event information, and the widely adopted video-level cross-modal contrastive learning also struggles to capture detailed and complex video-text event alignment. To address these challenges, we make improvements from both data and model perspectives. In terms of pre-training data, we focus on supplementing the missing specific event content and event temporal transitions with the proposed event augmentation strategies. Based on the event-augmented data, we construct a novel Event-Aware Video-Text Retrieval model, i.e., EA-VTR, which achieves powerful video-text retrieval ability through superior video event awareness. EA-VTR can efficiently encode frame-level and video-level visual representations simultaneously, enabling detailed event content and complex event temporal cross-modal alignment, ultimately enhancing the comprehensive understanding of video events. Our method not only significantly outperforms existing approaches on multiple datasets for Text-to-Video Retrieval and Video Action Recognition tasks, but also demonstrates superior event content perceive ability on Multi-event Video-Text Retrieval and Video Moment Retrieval tasks, as well as outstanding event temporal logic understanding ability on Test of Time task.

Previous work for video captioning aims to objectively describe the video content but the captions lack human interest and attractiveness, limiting its practical application scenarios. The intention of video title generation (video titling) is to produce attractive titles, but there is a lack of benchmarks. This work offers CREATE, the first large-scale Chinese shoRt vidEo retrievAl and Title gEneration dataset, to assist research and applications in video titling, video captioning, and video retrieval in Chinese. CREATE comprises a high-quality labeled 210 K dataset and two web-scale 3 M and 10 M pre-training datasets, covering 51 categories, 50K+ tags, 537K+ manually annotated titles and captions, and 10M+ short videos with original video information. This work presents ACTEr, a unique Attractiveness-Consensus-based Title Evaluation, to objectively evaluate the quality of video title generation. This metric measures the semantic correlation between the candidate (model-generated title) and references (manual-labeled titles) and introduces attractive consensus weights to assess the attractiveness and relevance of the video title. Accordingly, this work proposes a novel multi-modal ALignment WIth Generation model, ALWIG, as one strong baseline to aid future model development. With the help of a tag-driven video-text alignment module and a GPT-based generation module, this model achieves video titling, captioning, and retrieval simultaneously. We believe that the release of the CREATE dataset, ACTEr metric, and ALWIG model will encourage in-depth research on the analysis and creation of Chinese short videos.

Dominant dual-encoder models enable efficient image-text retrieval but suffer from limited accuracy while the cross-encoder models offer higher accuracy at the expense of efficiency. Distilling cross-modality matching knowledge from cross-encoder to dual-encoder provides a natural approach to harness their strengths. Thus we investigate the following valuable question: how to make cross-encoder a good teacher for dual-encoder? Our findings are threefold:(1) Cross-modal similarity score distribution of cross-encoder is more concentrated while the result of dual-encoder is nearly normal making vanilla logit distillation less effective. However ranking distillation remains practical as it is not affected by the score distribution.(2) Only the relative order between hard negatives conveys valid knowledge while the order information between easy negatives has little significance.(3) Maintaining the coordination between distillation loss and dual-encoder training loss is beneficial for knowledge transfer. Based on these findings we propose a novel Contrastive Partial Ranking Distillation (CPRD) method which implements the objective of mimicking relative order between hard negative samples with contrastive learning. This approach coordinates with the training of the dual-encoder effectively transferring valid knowledge from the cross-encoder to the dual-encoder. Extensive experiments on image-text retrieval and ranking tasks show that our method surpasses other distillation methods and significantly improves the accuracy of dual-encoder.

Video Tagging intends to infer multiple tags spanning relevant content for a given video. Typically, video tags are freely defined and uploaded by a variety of users, so they have two characteristics: abundant in quantity and disordered intra-video. It is difficult for the existing multilabel classification and generation methods to adapt directly to this task. This paper proposes a novel generative model, Order-Prompted Tag Sequence Generation (OP-TSG), according to the above characteristics. It regards video tagging as a tag sequence generation problem guided by sample-dependent order prompts. These prompts are semantically aligned with tags and enable to decouple tag generation order, making the model focus on modeling the tag dependencies. Moreover, the word-based generation strategy enables the model to generate novel tags. To verify the effectiveness and generalization of the proposed method, a Chinese video tagging benchmark CREATE-tagging, and an English image tagging benchmark Pexel-tagging are established. Extensive results show that OP-TSG is significantly superior to other methods, especially the results on rare tags improve by 3.3% and 3% over SOTA methods on CREATE-tagging and Pexel-tagging, and novel tags generated on CREATE-tagging exhibit a tag gain of 7.04%.

Previous dual-encoder pre-training methods for video-text retrieval employ contrastive learning for cross-modal alignment in a latent space. However, such learned latent spaces often result in modality gap problem [26]. In this paper, we introduce a novel SemVTR framework designed to learn semantics-grounded video-text representations in a vocabulary space, in which each dimension corresponds to a semantic concept represented by a word. The representation is obtained by grounding video and text into semantically-related dimensions with high activation values. As video-text pairs share grounded dimensions, their vocabulary representations are expected to cluster together and thus alleviate modality gap problem. So, the crux of our method lies in grounding video and text into vocabulary space. Specifically, we propose a Multi-Granularity Video Semantics Grounding approach and a Textual Semantics Preserving training strategy. The visualization illustrates that SemVTR obtains semantics-gronded vocabulary representation and also alleviates the modality gap problem. SemVTR significantly outperforms existing methods on four video-text retrieval benchmarks.