Large Vision Language Models

When you read a paper about a new Vision-Language Model today, it can be easy to forget how strange this idea would have sounded not so long ago. Teaching machines to see was already hard. Teaching them to read and generate language was already hard. Asking them to do both at once - and then to reason, answer questions, follow instructions, and sometimes even surprise us - still carries a quiet trace of science fiction, even as it becomes routine. This book was born from a simple feeling: it is too easy to get lost. The field moves quickly, new model names appear constantly, and the gap between "I know the buzzwords" and "I actually understand how this works" can feel uncomfortably wide. I have felt that gap many times. If you are holding this book, you probably have too. My goal is not to provide an exhaustive catalog of every dataset, benchmark, and new model variant. Instead, I want to offer something more modest - and, I hope, more durable: a clear mental map of Vision-Language Models. Enough structure that you can read new papers with confidence; enough intuition that you can design your own systems without feeling as if you are assembling LEGO bricks blindly.

Recent advancements in visual context compression enable MLLMs to process ultra-long contexts efficiently by rendering text into images. However, we identify a critical vulnerability inherent to this paradigm: lowering image resolution inadvertently catalyzes jailbreaking. Our experiments reveal that the safety defenses of SOTA models deteriorate sharply as resolution degrades, surprisingly persisting even when text remains legible. We attribute this to Cognitive Overload'', hypothesizing that the effort required to decipher degraded inputs diverts attentional resources from safety auditing. This phenomenon is consistent across various visual perturbations, including noise and geometric distortion. To address this, we propose a simple Structured Cognitive Offloading'' strategy that mitigates these risks by enforcing a serialized pipeline to decouple visual transcription from safety assessment. Our work exposes a significant risk in vision-based compression and provides critical insights for the secure design of future MLLMs.

Vision Language Models (VLMs) offer the exciting possibility of processing text as rendered images, bypassing the need for tokenizing the text into long token sequences. Since VLM image encoders map fixed-size images to a fixed number of visual tokens, varying rendering resolution provides a fine-grained compression knob. However, accuracy deteriorates quickly as compression increases: characters shrink below the vision encoder's effective resolution, making them indistinguishable. To address this, we propose LensVLM, an inference framework and post-training recipe that enables VLMs to scan compressed images, then selectively expand only the relevant images to their uncompressed form via learned tools. Building on Qwen3.5-9B-Base, LensVLM maintains accuracy comparable to the full-text upper bound at 4.3x effective compression and outperforms retrieval-based, text- and visual-compression baselines up to 10.1x effective compression across seven text QA benchmarks. LensVLM also generalizes to multimodal document and code understanding tasks, with the accuracy gain over baselines growing as compression increases. Our analysis validates this approach: training makes visual compression robust to rendering choices, and as compression grows the model increasingly relies on expanded content rather than unreliable visual reading. The analysis also yields practical tool-choice guidance: text expansion is preferable for rendered text, while high-resolution image expansion suits native documents whose layout cues carry task-relevant information.

Vision-language-action (VLA) models inherit rich visual-semantic priors from pre-trained vision-language backbones, but adapting them to robotic control remains challenging. Full fine-tuning (FFT) is prone to overfitting on downstream robotic data and catastrophic forgetting of pretrained vision-language capabilities. Parameter-efficient fine-tuning (PEFT) better preserves pre-trained knowledge, yet existing PEFT methods still struggle to adapt effectively to robot control tasks. To address this gap, we propose VLA-GSE, a parameter-efficient VLA fine-tuning framework that improves control adaptation while retaining PEFT's knowledge preservation advantage. Specifically, VLA-GSE (Generalized and Specialized Experts) is initialized by spectrally decomposing the frozen backbone, assigning leading singular components to generalized experts (shared experts) and disjoint residual components to specialized experts (routed experts). This decomposition improves adaptation capacity under a fixed trainable-parameter budget. Under a comparable parameter budget, VLA-GSE updates only 2.51% of the full model parameters and consistently outperforms strong FFT and PEFT baselines. It achieves 81.2% average zero-shot success on LIBERO-Plus, preserves pre-trained VLM capability comparably to LoRA on multimodal understanding benchmarks, and improves real-world manipulation success under multiple distribution shifts. Code is available at: https://github.com/YuhuaJiang2002/VLA-GSE

Diabetic Retinopathy (DR) is a leading cause of preventable blindness among working-age adults worldwide, yet most automated screening systems are limited to image-level classification and lack clinically structured reporting. We propose Retina-RAG, a low-cost modular framework that jointly performs DR severity grading, macular edema (ME) detection, and report generation. The architecture decouples a high-performance retinal classifier and a parameter-efficient vision-language model (Qwen2.5-VL-7B-Instruct) adapted via Low-Rank Adaptation (LoRA), enabling flexible component integration. A retrieval-augmented generation (RAG) module injects curated ophthalmic knowledge together with structured classifier outputs at inference time to improve diagnostic consistency and reduce hallucinations. Retina-RAG achieves an F1-score of 0.731 for DR grading and 0.948 for ME detection, substantially outperforming zero-shot Qwen (0.096, 0.732) and MMed-RAG (0.541, 0.641) on a retinal disease detection dataset with captions. For report generation, Retina-RAG attains ROUGE-L 0.438 and SBERT similarity 0.884, exceeding all baselines. The full framework operates on a single consumer-grade GPU, demonstrating that clinically structured retinal AI can be achieved with modest computational resources.

Large-scale vision-language mixture-of-experts (VL-MoE) models provide strong multimodal capability, but efficient deployment on memory-constrained platforms remains difficult. Existing MoE offloading systems are largely designed for text-centric workloads and become much less effective for visual-heavy inputs, where large numbers of visual tokens induce broader and less predictable expert accesses. We present VisMMoE, a VL-MoE offloading system built on a single systems insight: pruning redundant visual tokens can improve offloading not only by reducing computation, but also by reshaping expert demand. We refer to this effect as \textit{visual-expert affinity}: token pruning makes expert accesses more concentrated within layers and more stable across layers, producing a smaller and more predictable expert working set. Guided by this insight, VisMMoE combines affinity-aware token compression, lookahead expert prediction, and cache/pipeline orchestration to improve expert locality and prefetch effectiveness under tight memory budgets. We implement VisMMoE on multiple frameworks and evaluate it on representative VL-MoE models and benchmarks. VisMMoE improves end-to-end inference performance by up to 2.68x and 1.61x, respectively, over strong baselines for today's VL-MoE deployments while maintaining competitive accuracy.

Despite the rapid evolution of training paradigms, the decoder backbone of large vision--language models (LVLMs) remains fundamentally rooted in the residual-connection Transformer architecture. Therefore, deciphering the distinct roles of internal modules is critical for understanding model mechanics and guiding architectural optimization. While prior statistical approaches have provided valuable attribution-based insights, they often lack a unified theoretical basis. To bridge this gap, we propose a unified framework grounded in information theory and geometry to quantify the geometric and entropic nature of residual updates. Applying this unified framework reveals a fundamental functional decoupling: Attention acts as a subspace-preserving operator focused on reconfiguration, whereas FFNs serve as subspace-expanding operators driving semantic innovation. Strikingly, further experiments demonstrate that replacing learned attention weights with predefined values (e.g., Gaussian noise) yields comparable or even superior performance across a majority of datasets relative to vanilla models. These results expose severe misallocation and redundancy in current mechanisms, suggesting that state-of-the-art LVLMs effectively ``get lost in attention'' rather than efficiently leveraging visual context.

The widespread adoption of camera-equipped mobile devices and wearables has enabled convenient capture of meal images, making food recognition a key component for real time dietary monitoring. However, real-world food images present challenges due to high intra-class similarity and the frequent presence of multiple food items within a single image. While deep learning models achieve strong performance in coarse grained classification, they often struggle to capture fine-grained attributes such as cooking style. Moreover, open-ended generation in modern vision-language models can produce non-canonical labels, limiting their practical deployment. We propose FoodCHA, a multimodal agentic framework that reformulates food recognition as a hierarchical decision-making process. By progressively anchoring predictions, FoodCHA guides subcategory identification using high-level categories and guides cooking style recognition using subcategories, improving semantic consistency and attribute-level discrimination. To ensure practical deployability, FoodCHA utilizes the compact Moondream-2B vision language model, which provides strong reasoning capability while maintaining lower computational and memory overhead. Experiments on FoodNExTDB show that FoodCHA outperforms Food-Llama-3.2-11B by 13.8% and 38.2% in category and subcategory recognition precision, respectively, and achieves a striking 153.2% improvement in cooking style classification precision.

Automated glaucoma detection is critical for preventing irreversible vision loss and reducing the burden on healthcare systems. However, ensuring fairness across diverse patient populations remains a significant challenge. In this paper, we propose FairEnc, a fair pretraining method for vision-language models (VLMs) that enables simultaneous debiasing across multiple sensitive attributes. FairEnc jointly mitigates biases in both textual and visual modalities with respect to multiple sensitive attributes, including race, gender, ethnicity, and language. Specifically, for the textual encoder, we leverage a large language model to generate synthetic clinical descriptions with varied sensitive attributes while preserving disease semantics, and employ a contrastive alignment objective to encourage demographic-invariant representations. For the visual encoder, we propose a dual-level fairness strategy that combines mutual information regularization to reduce statistical dependence between learned features and demographic groups, with multi-discriminator adversarial debiasing. Comprehensive experiments on the publicly available Harvard-FairVLMed dataset demonstrate that FairEnc effectively reduces demographic disparity as measured by DPD and DEOdds while achieving strong diagnostic performance under both zero-shot and linear probing evaluations. Additional experiments on the private FairFundus dataset show that FairEnc consistently preserves fairness advantages under cross-domain and cross-modality settings and maintains diagnostic performance within a competitive range. These results highlight FairEnc's ability to generalize fairness under distribution shifts, supporting its potential for more equitable deployment in real-world clinical settings. Our codebase and synthetic clinical notes are available at https://github.com/Mohamed-Elhabebe/FairEnc

Although Large Vision-Language Models (LVLMs) have demonstrated remarkable performance on downstream tasks, they frequently produce contents that deviate from visual information, leading to object hallucination. To tackle this, recent works mostly depend on expensive manual annotations and training cost, or decoding strategies which significantly increase inference time. In this work, we observe that LVLMs' attention to visual information is significantly enhanced when answering caption queries compared to non-caption queries. Inspired by this phenomenon, we propose Caption-guided Visual Attention Steering (CAST), a training-free, plug-and-play hallucination mitigation method that leverages the attention activation pattern corresponding to caption queries to enhance LVLMs' visual perception capability. Specifically, we use probing techniques to identify attention heads that are highly sensitive to caption queries and estimate optimized steering directions for their outputs. This steering strengthens LVLM's fine-grained visual perception capabilities, thereby effectively mitigating object hallucination. CAST reduced object hallucination by an average of 6.03% across five widely used LVLMs and five benchmarks including both discriminative and generative tasks, demonstrating state-of-the-art performance while adding little inference cost and preserving other foundational capabilities.

Vision-language models have demonstrated strong performance across robotic perception and instruction-following tasks. However, they still struggle with precise spatial reasoning, particularly in predicting object locations and fine-grained object states. We propose StateVLM, a vision-language model designed to learn fine-grained object representations, including object localization and grasp-relevant region prediction. We introduce a joint training objective that integrates an auxiliary regression loss (ARL) with the standard causal language modeling (CLM) objective to improve numerical reasoning and spatial understanding. To evaluate whether models can move beyond category-level grounding toward state-aware spatial understanding, we introduce an open-source benchmark, Object State Affordance Reasoning (OSAR), comprising 1,172 scenes with 7,746 individual objects and their corresponding bounding boxes. Empirical experiments on the Referring Expression Comprehension datasets demonstrate that integrating ARL improves model performance compared with CLM only. Experiments on the OSAR benchmark further demonstrate that StateVLM with ARL achieves an average improvement of 5.2% over models trained with CLM only. These results show that ARL is particularly beneficial for the complex affordance reasoning tasks that require object state understanding. The joint training objective in StateVLM demonstrates that integrating an auxiliary regression objective into VLM training improves numerical reasoning, and the OSAR benchmark provides a new testbed for understanding object state affordances in robotics.

While Large Vision-Language Models (LVLMs) offer powerful capabilities, they pose privacy risks by unintentionally memorizing sensitive personal information. Current unlearning benchmarks attempt to mitigate this using fictitious identities but overlook a critical stage 1 failure: models fail to effectively memorize target information initially, rendering subsequent unlearning evaluations unreliable. Diagnosing under-memorization and the multi-hop curse as root causes, we introduce ReMem, a Reliable Multi-hop and Multi-image Memorization Benchmark. ReMem ensures robust foundational learning through principled data scaling, reasoning-aware QA pairs, and diverse visual contexts. Additionally, we propose a novel Exposure metric to quantify the depth of information erasure from the model's internal probability distribution. Extensive experiments demonstrate that ReMem provides a rigorous and trustworthy framework for diagnosing both learning and unlearning behaviors in LVLMs.

Large Vision-Language Models (LVLMs), trained on web-scale data, risk memorizing and regenerating copyrighted visual content such as characters and logos, creating significant challenges. Machine unlearning offers a path to mitigate these risks by removing specific content post-training, but evaluating its effectiveness, especially in the complex multimodal setting of LVLMs, remains an open problem. Current evaluation methods often lack robustness or fail to capture the nuances of cross-modal concept erasure. To address this critical gap, we introduce the CoVUBench benchmark, the first framework specifically designed for evaluating copyright content unlearning in LVLMs. CoVUBench utilizes procedurally generated, legally safe synthetic data coupled with systematic visual variations spanning compositional changes and diverse domain manifestations to ensure realistic and robust evaluation of unlearning generalization. Our comprehensive multimodal evaluation protocol assesses both forgetting efficacy from the copyright holder perspective and the preservation of general model utility from the deployer viewpoint. By rigorously measuring this crucial trade-off, CoVUBench provides a standardized tool to advance the development of responsible and effective unlearning methods for LVLMs.

Multidimensional human understanding is essential for real-world applications such as film analysis and virtual digital humans, yet current LVLM benchmarks largely focus on single-task settings and lack fine-grained, human-centric evaluation. In this work, we introduce MHPR, a comprehensive benchmark for joint perception-reasoning over human-centric scenes spanning individual, multi-person, and human-object interaction dimensions. MHPR comprises a multi-level data design-Captioned Raw Data (C-RD), Supervised Fine-Tuning Data (SFT-D), Reinforcement Learning Data (RL-D), and Test Data (T-D)-together with an automated caption/VQA generation pipeline (ACVG) that performs category-wise attribute decomposition, attribute-specific rewriting, and multi-model voting to ensure high-quality, scalable annotations. We evaluate state-of-the-art vision-language models on fine-grained attributes (appearance, clothing, pose, parts) and high-level semantics (social relations, action semantics, spatial relations, intent and functionality). Our findings show that: 1) format-aligned SFT data substantially improves instruction following and stability; 2) challenge-focused RL data derived from bad-case analysis further enhances perception and reasoning on difficult instances; and 3) training Qwen2.5-VL-7B with MHPR yields significant gains, achieving near-parity with considerably larger models. We release ACVG and MHPR to facilitate reproducible, extensible research on human-centric perception and reasoning.

Vision-Language Models (VLMs) have broad potential in privacy-sensitive domains such as healthcare and finance, yet strict data-sharing constraints render centralized training infeasible. Federated Learning mitigates this issue by enabling decentralized training, but practical deployments face challenges due to client heterogeneity in computational resources, application requirements, and model architectures. Under extreme model and data heterogeneity, replacing parameter aggregation with preference-based collaboration offers a more suitable interface, as it eliminates the need for direct parameter or data exchange. Motivated by this, we propose MoR, a federated alignment framework that combines GRPO with Mixture-of-Rewards for heterogeneous VLMs. In MoR, each client locally trains a reward model from local preference annotations, capturing specific evaluation signals without exposing raw data. To combine these heterogeneous supervision signals, MoR introduces a Mixture-of-Rewards mechanism with learned routing, which adaptively fuses client reward models according to the input and alignment objective. The server then optimizes a base VLM using GRPO with a KL penalty to a reference model, enabling preference alignment without requiring client models to share architectures or parameters. Experiments on diverse public vision-language benchmarks demonstrate that MoR consistently outperforms federated alignment baselines in generalization and cross-client adaptability. Our approach provides a scalable solution for privacy-preserving alignment of heterogeneous VLMs under federated settings.

Image-based Joint-Embedding Predictive Architecture (I-JEPA) offers a promising approach to visual self-supervised learning through masked feature prediction. However with the inherent visual uncertainty at masked positions, feature prediction remains challenging and may fail to learn semantic representations. In this work, we propose Text-Conditional JEPA (TC-JEPA) that uses image captions to reduce the prediction uncertainty. Specifically, we modulate the predicted patch features using a fine-grained text conditioner that computes sparse cross-attention over input text tokens. With such conditioning, patch features become predictable as a function of text, thus are more semantically meaningful. We show TC-JEPA improves downstream performance and training stability, with promising scaling properties. TC-JEPA also offers a new vision-language pretraining paradigm based on feature prediction only, outperforming contrastive methods on diverse tasks, especially those requiring fine-grained visual understanding and reasoning.

Dual-system Vision-Language-Action (VLA) models achieve state-of-the-art robotic manipulation but are bottlenecked by the VLM backbone, which must execute at every control step while producing temporally redundant features. We propose Latent Bridge, a lightweight model that predicts VLM output deltas between timesteps, enabling the action head to operate on predicted outputs while the expensive VLM backbone is called only periodically. We instantiate Latent Bridge on two architecturally distinct VLAs: GR00T-N1.6 (feature-space bridge) and π0.5 (KV-cache bridge), demonstrating that the approach generalizes across VLA designs. Our task-agnostic DAgger training pipeline transfers across benchmarks without modification. Across four LIBERO suites, 24 RoboCasa kitchen tasks, and the ALOHA sim transfer-cube task, Latent Bridge achieves 95-100% performance retention while reducing VLM calls by 50-75%, yielding 1.65-1.73x net per-episode speedup.

Despite the success of Large-Vision Language Models (LVLMs), general optimization objectives (e.g., standard MLE) fail to constrain visual trajectories, leading to language bias and hallucination. To mitigate this, current methods introduce geometric priors from visual experts as additional supervision. However, we observe that such supervision is typically suboptimal: it is biased toward geometric precision and offers limited reasoning utility. To bridge this gap, we propose Perceptual Flow Network (PFlowNet), which eschews rigid alignment with the expert priors and achieves interpretable yet more effective visual reasoning. Specifically, PFlowNet decouples perception from reasoning to establish a self-conditioned generation process. Based on this, it integrates multi-dimensional rewards with vicinal geometric shaping via variational reinforcement learning, thereby facilitating reasoning-oriented perceptual behaviors while preserving visual reliability. PFlowNet delivers a provable performance guarantee and competitive empirical results, particularly setting new SOTA records on V* Bench (90.6%) and MME-RealWorld-lite (67.0%).

Vision-language models (VLMs) have demonstrated strong capabilities in multimodal perception and reasoning. However, deploying large VLMs on mobile devices remains challenging due to their substantial computational and memory demands. A practical alternative is device-edge co-inference, where a lightweight draft VLM on the mobile device collaborates with a larger target VLM on the edge server via speculative decoding. Nevertheless, directly extending speculative decoding to VLMs suffers from severe inefficiency due to excessive visual-token computation and high communication overhead. To address these challenges, we propose CoVSpec, an efficient collaborative speculative decoding framework for VLM inference. Specifically, we first develop a training-free visual token reduction framework that prunes redundant visual tokens on the mobile device by jointly considering query relevance, token activity, and low-rank dependency. Moreover, we design an adaptive drafting strategy that dynamically adjusts both the verification frequency and the draft length. In addition, we introduce a parallel branching mechanism with decoupled verification-correction to improve draft-side utilization during target-side verification and reduce correction-related transmission overhead. Experiments on multiple benchmarks show that CoVSpec achieves up to 2.21x higher throughput than target-only inference and reduces communication overhead by more than 96% compared with baselines, without compromising task accuracy.

Contrastive representation learning (CRL) underpins many modern foundation models. Despite recent theoretical progress, existing analyses suffer from several key limitations: (i) the statistical consistency of CRL remains poorly understood; (ii) available generalization bounds deteriorate as the number of negative samples increases, contradicting the empirical benefits of large negative sets; and (iii) the retrieval performance of CRL has received limited theoretical attention. In this paper, we develop a unified statistical learning theory for CRL. For downstream tasks, we evaluate retrieval quality using an AUC-type population criterion and show that the contrastive loss is \emph{statistically consistent} with optimal ranking. We further establish a \emph{calibration-style inequality} that quantitatively relates excess contrastive risk to excess retrieval suboptimality. For upstream training, we study both supervised and self-supervised contrastive objectives and derive generalization bounds of order $O(1/m + 1/\sqrt{n})$ and $O(1/\sqrt{m} + 1/\sqrt{n})$, respectively, where $m$ denotes the number of negative samples and $n$ the number of anchor points. These bounds not only explain the empirical advantages of large negative sets but also reveal an explicit trade-off between $m$ and $n$. Extensive experiments on large-scale vision--language models corroborate our theoretical predictions.

We introduce CADFS, a data-centric framework that enables large vision-language models to generate complex CAD design histories. Existing generative CAD systems are restricted to sketch-extrude operations due to simplified representations and limited datasets. We address this by introducing a FeatureScript-based representation and constructing a dataset of 450k real-world CAD models spanning 15 modeling operations. We obtain the dataset via a new pipeline that reconstructs clean, executable FeatureScript programs and provides multimodal annotations. Fine-tuning a VLM on this representation yields state-of-the-art results in text-conditioned CAD generation and image-based reconstruction, producing more accurate, diverse, and feature-rich designs than prior frameworks. Ablations show that each individual component of our framework, i.e., the FeatureScript representation, the extended operation set, and representation-aligned textual descriptions, significantly improves performance. Our framework substantially broadens the complexity and realism achievable in generative CAD. The CADFS framework and the new dataset are available at https://voyleg.github.io/cadfs/.

Multi-model learning has attracted great attention in visual-text tasks. However, visual-tabular data, which plays a pivotal role in high-stakes domains like healthcare and industry, remains underexplored. In this paper, we introduce \textit{VT-Bench}, the first unified benchmark for standardizing vision-tabular discriminative prediction and generative reasoning tasks. VT-Bench aggregates 14 datasets across 9 domains (medical-centric, while covering pets, media, and transportation) with over 756K samples. We evaluate 23 representative models, including unimodal experts, specialized visual-tabular models, general-purpose vision-language models (VLMs), and tool-augmented methods, highlighting substantial challenges of visual-tabular learning. We believe VT-Bench will stimulate the community to build more powerful multi-modal vision-tabular foundation models. Benchmark: https://github.com/Ziyi-Jia990/VT-Bench

Current vision language models face hallucination and robustness issues against ambiguous or corrupted modalities. We hypothesize that these issues can be addressed by exploiting the shared information between modalities to compensate for the impaired one. To this end, we analyze multimodal interactions -- redundant (shared), unique (exclusive), and synergistic (emergent) task-relevant information provided by the modalities -- to determine their impacts on model reliability. Specifically, amplifying redundant interactions would increase this exploitable shared information to resolve these issues; yet, modern instruction datasets often eliminate redundancies to prioritize visual grounding. We bridge this gap through a self-captioning workflow featuring a \textsc{Multimodal Interaction Gate}: a mechanism to convert unique interactions into redundant interactions. Our findings suggest that increasing redundancy can reduce visual induced errors by 38.3% and improve consistency by 16.8%.

Large vision-language models (VLMs) have advanced multimodal tasks such as video question answering (QA). However, VLMs face the challenge of selecting frames effectively and efficiently, as standard uniform sampling is expensive and performance may plateau. Inspired by active perception theory, which posits that models gain information by acquiring data that differs from their expectations, we introduce Video Active Perception (VAP), a training-free method to enhance long-form video QA using VLMs. Our approach treats keyframe selection as data acquisition in active perception and leverages a lightweight text-conditioned video generation model to represent prior world knowledge. Empirically, VAP achieves state-of-the-art zero-shot results on long-form or reasoning video QA datasets such as EgoSchema, NExT-QA, ActivityNet-QA, IntentQA, and CLEVRER, achieving an increase of up to 5.6 x frame efficiency by frames per question over standard GPT-4o, Gemini 1.5 Pro, and LLaVA-OV. Moreover, VAP shows stronger reasoning abilities than previous methods and effectively selects keyframes relevant to questions. These findings highlight the potential of leveraging active perception to improve the frame effectiveness and efficiency of long-form video QA.

Vision-Language Models (VLMs) frequently suffer from visual perception errors and hallucinations that compromise answer accuracy in complex reasoning tasks. Reinforcement Learning with Verifiable Rewards (RLVR) offers a promising solution by optimizing policies using answer correctness signals. Despite their effectiveness, prevailing RLVR methods face two critical limitations. First, much of the sampling budget is wasted on trajectories doomed to fail due to early visual description errors. Second, sparse rewards cannot distinguish whether failures stem from visual perception or reasoning stages. We introduce MIRL, a decoupled framework that addresses both limitations by leveraging mutual information (MI) between generated descriptions and visual inputs as a cheap pre-screening signal. This enables intelligent budget allocation toward high-potential trajectories via forking, while decoupled training provides independent MI-based rewards for visual perception optimization, resolving reward blindness. Experiments on six vision-language reasoning benchmarks demonstrate that MIRL achieves 70.22% average accuracy and successfully surpasses the performance of sampling 16 complete trajectories using only 10 pre-samples with top-6 selection (25% fewer complete trajectories). Our code is available at: https://anonymous.4open.science/r/mirl-main/.

A hierarchical cross-modal fusion model is proposed for vision-language question answering (VLQA) in industrial robotics, targeting the challenges of semantic ambiguity, complex environmental layouts, and domain-specific terminology common in modern manufacturing. The framework integrates advanced object detection, multi-scale visual encoding, syntactic parsing, and task-aware semantic attention to unite vision and language signals into a joint reasoning space. Region-based deep networks extract visual features, weighted embeddings aggregate, and recurrent neural parsing encodes sentence structures. Through fine-grained semantic alignment driven by adaptive fusion and cross-attention mechanisms, the system can handle operational queries, instruction steps, and anomaly detection with higher reliability. Compared to the existing VLQA benchmarks, validation experiments conducted on the IVQA and RIF benchmarks indicate improvements in semantic alignment, Top-1 accuracy, and robustness to ambiguous or procedural task queries. Ablation studies further quantify the impact of each architectural module, confirming the necessity of multi-level feature integration and context-driven gating for dependable industrial deployment. The technical advancements reported here provide core methodologies to improve the interpretability and operational effectiveness of industrial robots faced with diverse human-robot interaction tasks.

Visual perception in modern Vision-Language Models (VLMs) is constrained by a perceptual bandwidth bottleneck: a broad field of view preserves global context but sacrifices the fine-grained details required for complex reasoning. We argue that high-resolution visual reasoning is therefore not only semantic reasoning but also task-relevant evidence acquisition under limited perceptual bandwidth. Inspired by active vision and information foraging, we formalise this process as sequential Bayesian optimal experimental design (S-BOED), where an agent decides which visual evidence to acquire before answering. Since exact Bayesian inference is intractable in continuous gigapixel spaces, we derive a tractable coverage--resolution objective as a proxy for task-relevant information gain. We instantiate this framework with FOVEA, a training-free procedure that refines VLM crop proposals through evidence-oriented probing. Experiments on high-resolution benchmarks show consistent gains over direct and ReAct-style baselines, with particularly strong improvements in search-dominated remote-sensing settings.

Zero-shot vision-language models (VLMs) offer a promptable alternative to task-specific training for gross tumor volume (GTV) delineation in non-small-cell lung cancer (NSCLC), but the prompt dimensions that govern their spatial behavior remain poorly understood. We study this question by probing alignment directions in VoxTell on a held-out internal NSCLC tumor dataset through sub-prompt decomposition into diagnosis, demographic, staging, anatomical, generic, and irrelevant controls; attribute-wise perturbation robustness; specificity ladders; and cross-case prompt swaps, while benchmarking against fine-tuned and zero-shot baselines using the Dice Similarity Coefficient (DSC) with Wilcoxon signed-rank tests and Benjamini-Hochberg correction. Alignment analyses revealed that anatomical location is the dominant driver of VoxTell's spatial attention: 63.4 percent of location perturbations caused catastrophic drops, prompt specificity improved from generic to full descriptions except for diagnosis-only prompts, irrelevant prompts correctly yielded zero segmentation, and cross-case prompt swaps confirmed patient-specific conditioning (matched DSC 0.906 vs. mismatched 0.406). Histology and stage substitutions had minimal effect, indicating that the model prioritizes "where to look" over "what to look for." In this context, VoxTell, operating fully zero-shot, achieved a mean DSC of 0.613, statistically indistinguishable from nnUNet (0.690, adjusted p = 0.156) and Ahmed et al. (0.675, adjusted p = 0.679), while significantly outperforming all other zero-shot models. Together, these findings argue that segmentation VLMs should be evaluated not only by Dice, but also by the prompt dimensions to which they align.

While autoregressive Large Vision-Language Models (LVLMs) demonstrate remarkable proficiency in multimodal tasks, they face a "Visual Signal Dilution" phenomenon, where the accumulation of textual history expands the attention partition function, causing visual attention to decay inversely with generated sequence length. To counteract this, we propose Persistent Visual Memory (PVM), a lightweight learnable module designed to strengthen sustained, on-demand access to visual evidence. Integrated as a parallel branch alongside the Feed-Forward Network (FFN) in LVLMs, PVM establishes a distance-agnostic retrieval pathway that directly provides visual embeddings for enhanced visual perception, thereby structurally mitigating the signal suppression inherent to deep generation. Extensive experiments on Qwen3-VL models demonstrate that PVM brings notable improvements with negligible parameter overhead, delivering consistent average accuracy gains across both 4B and 8B scales, particularly in complex reasoning tasks that demand persistent visual perception. Furthermore, in-depth analysis reveals that PVM shows improved robustness in longer generations and accelerates internal prediction convergence.

In this paper, we present \textbf{Gen}erative \textbf{L}anguage-\textbf{I}mage \textbf{P}re-training (GenLIP), a minimalist generative pretraining framework for Vision Transformers (ViTs) designed for multimodal large language models (MLLMs). To better align vision encoders with the autoregressive nature of LLMs, GenLIP trains a ViT to predict language tokens directly from visual tokens using a standard language modeling objective, without contrastive batch construction or an additional text decoder. This design offers three key advantages: (1) \textbf{Simplicity}: a single transformer jointly models visual and textual tokens; (2) \textbf{Scalability}: it scales effectively with both data and model size; and (3) \textbf{Performance}: it achieves competitive or superior results across diverse multimodal benchmarks. Trained on 8B samples from Recap-DataComp-1B, GenLIP matches or surpasses strong baselines despite using substantially less pretraining data. After continued pretraining on multi-resolution images at native aspect ratios, GenLIP further improves on detail-sensitive tasks such as OCR and chart understanding, making it a strong foundation for vision encoders in MLLMs.

Key-Value (KV) cache has become a de facto component of modern Large Vision-Language Models (LVLMs) for inference. While it enhances decoding efficiency in Large Language Models (LLMs), its direct adoption in LVLMs introduces substantial GPU memory overhead due to the large number of vision tokens processed during the prefill stage. To tackle this problem, we propose LightKV, a novel approach that reduces KV cache size by exploiting the redundancy among vision-token embeddings. Guided by text prompts, LightKV employs cross-modality message passing to aggregate informative messages across vision tokens and progressively compress them during prefill. This prompt-aware guidance distinguishes our method from prior vision-only compression strategies. We evaluate LightKV on eight open-source LVLMs across eight public benchmark datasets, e.g., MME and SeedBench. Experimental results demonstrate that with only 55% of the original vision tokens, LightKV (a) halves the vision-token KV cache size, (b) reduces computation by up to 40%, and (c) preserves general-purpose performance while significantly outperforming existing baselines.

Single-prompt first-token probabilities from zero-shot vision-language model (VLM) safety classifiers are treated as decision scores, but we show they are unreliable under semantically equivalent prompt reformulation: even when the binary label is constrained to a fixed output position, equivalent prompts can induce materially different unsafe probabilities for the same sample. Across multimodal safety benchmarks and multiple VLM families, cross-prompt variance is strongly associated with prompt-level disagreement and higher error, making it a useful fragility diagnostic. A training-free mean ensemble improves NLL on all 14 dataset-model evaluation pairs and ECE on 12/14 relative to a train-selected single-prompt baseline, and wins more head-to-head NLL comparisons than labeled temperature scaling, Platt scaling, and isotonic regression applied to the same prompt. Ranking gains are consistent against the train-selected baseline on both AUROC and AUPRC, and against the full 15-prompt distribution remain consistent on AUPRC while softening on AUROC. Labeled calibration on top of the mean provides further gains when labels are available, identifying prompt averaging as a strong label-free first stage rather than a replacement for calibration. We frame this as a reliability stress test for zero-shot VLM first-token safety scores and recommend prompt-family evaluation with mean aggregation as a standard label-free reliability baseline.

Large Vision-Language Models (LVLMs) often suffer from hallucinations, generating descriptions that include visual details absent from the input image. Recent preference alignment methods typically rely on supervision distilled from stronger models such as GPT. However, this offline paradigm introduces a Supervision-Perception Mismatch: the student model is forced to align with fine-grained details beyond its perceptual capacity, learning to guess rather than to see. To obtain reliable self-supervision for online learning, we identify a Generative-Discriminative Gap within LVLMs, where models exhibit higher accuracy on discriminative verification than open-ended generation. Leveraging this capability, we propose \textbf{O}nline \textbf{S}elf-\textbf{CA}lib\textbf{R}ation (OSCAR), a framework that integrates Monte Carlo Tree Search with a Dual-Granularity Reward Mechanism to construct preference data and iteratively refines the model via Direct Preference Optimization. Extensive experiments demonstrate that OSCAR achieves state-of-the-art performance on hallucination benchmarks while improving general multimodal capabilities.

Despite the rapid progress of large vision-language models (LVLMs), fine-grained, state-conditioned GUI interaction remains challenging. Current evaluations offer limited coverage, imprecise target-state definitions, and an overreliance on final-task success, obscuring where and why agents fail. To address this gap, we introduce \textbf{FineState-Bench}, a benchmark that evaluates whether an agent can correctly ground an instruction to the intended UI control and reach the exact target state. FineState-Bench comprises 2,209 instances across desktop, web, and mobile platforms, spanning four interaction families and 23 UI component types, with each instance explicitly specifying an exact target state for fine-grained state setting. We further propose \textit{FineState-Metrics}, a four-stage diagnostic pipeline with stage-wise success rates: Localization Success Rate (SR@Loc), Interaction Success Rate (SR@Int), Exact State Success Rate at Locate (ES-SR@Loc), and Exact State Success Rate at Interact (ES-SR@Int), and a plug-and-play \textit{Visual Diagnostic Assistant} (VDA) that generates a Description and a bounding-box Localization Hint to diagnose visual grounding reason via controlled w/ vs.\ w/o comparisons. On FineState-Bench, exact goal-state success remains low: ES-SR@Int peaks at 32.8% on Web and 22.8% on average across platforms. With VDA localization hints, Gemini-2.5-Flash gains +14.9 ES-SR@Int points, suggesting substantial headroom from improved visual grounding, yet overall accuracy is still insufficient for reliable fine-grained state-conditioned interaction \href{https://github.com/FengxianJi/FineState-Bench}{Github.}

The computational cost of training a vision-language model (VLM) can be reduced by sampling the training data. Previous work on efficient VLM pre-training has pointed to the importance of semantic data balance, adjusting the distribution of topics in the data to improve VLM accuracy. However, existing efficient pre-training approaches may disproportionately remove rare concepts from the training corpus. As a result, long-tail concepts remain insufficiently represented in the training data and are not effectively captured during training. In this work, we introduce a dynamic cluster-based sampling approach (DynamiCS) that downsamples large clusters of data and upsamples small ones. We first demonstrate the advantage of our cluster-scaling approach, which maintains the relative order of semantic clusters in the data and emphasizes the long-tail. This approach contrasts with current work, which focuses only on flattening the semantic distribution of the data. Then, we show the importance of dynamic sampling, which applies sampling at each epoch to improve cross-epoch data diversity and make upsampling practical. Our experiments show that DynamiCS reduces the computational cost of VLM training and provides a performance advantage for long-tail concepts. Code available at https://github.com/MingliangLiang3/DynamiCS.

Deploying vision-language models (VLMs) in clinical settings demands auditable behavior under realistic failure conditions, yet the failure landscape of frontier VLMs on specialized medical inputs is poorly characterized. We audit five recent frontier and grounding-aware VLMs (Gemini2.5Pro, GPT-5, o3, GLM-4.5V, Qwen2.5VL) on Medical VQA along two trust-relevant axes. Perception: all models localize anatomical and pathological targets poorly -- the best model reaches only 0.23 mean IoU and 19.1% Acc@0.5 -- and exhibit clinically dangerous laterality confusion. Pipeline integration: a self-grounding pipeline, where the same model localizes then answers, degrades VQA accuracy for every model -- driven by both inaccurate localization and format-compliance failures under the two-step prompt (parse failure rises to 70%--99% for Gemini and GPT-5 on VQA-RAD). Replacing predicted boxes with ground-truth annotations recovers and improves VQA accuracy, consistent with the failure residing in the perception module rather than in the decomposition itself. These observational findings identify grounding quality as a primary trustworthiness bottleneck in our SLAKE bounding-box setting. As a complementary fine-tuning follow-up, supervised fine-tuning of Qwen2.5VL on combined Med-VQA training data attains the highest reported SLAKE open-ended recall (85.5%) among comparable methods, suggesting that the VQA-level gap is tractable with domain adaptation; whether this also closes the perception/trustworthiness bottleneck is left to future work.

When the visual style of text is considered, a wide variety can be observed in font, color, and size. However, when a word is read, its meaning is independent of the style in which it has been written or rendered. In this paper, we investigate whether, and how, the style in which a word is visualized in an image impacts the description that a Large Visual Language Model (LVLM) provides for the concept to which that word refers. Specifically, we investigate how functional text styles (readability-oriented, e.g., black sans-serif) versus decorative styles (display-oriented, e.g., colored cursive/script) affect LVLMs' descriptions of a concept in terms of the attributes of that concept. Our experiments study the situation in which the LVLM is able to correctly identify the concept referred to by a visual text, i.e., by a word or words rendered as an image, and in which the visual text style should not influence the attribute-based description that the LVLM produces. Our experimental results reveal that even when the concept is correctly identified, text style influences the model's attribute-based descriptions of the concept. Our findings demonstrate non-trivial style leakage from text style into semantic inference and motivate style-aware evaluation and mitigation for LVLM-based multimedia systems.

Vision-language models (VLMs) are increasingly used in autonomous driving because they combine visual perception with language-based reasoning, supporting more interpretable decision-making, yet their robustness to physical adversarial attacks, especially whether such attacks transfer across different VLM architectures, is not well understood and poses a practical risk when attackers do not know which model a vehicle uses. We address this gap with a systematic cross-architecture study of adversarial transferability in VLM-based driving, evaluating three representative architectures (Dolphins, OmniDrive, and LeapVAD) using physically realizable patches placed on roadside infrastructure in both crosswalk and highway scenarios. Our transfer-matrix evaluation shows high cross-architecture effectiveness, with transfer rates of 73-91% (mean TR = 0.815 for crosswalk and 0.833 for highway) and sustained frame-level manipulation over 64.7-79.4% of the critical decision window even when patches are not optimized for the target model.

Text-to-image person re-identification (TI-ReID) relies on natural-language text descriptions to retrieve top matching individuals from a gallery of reference images. While recent large vision-language models (VLMs) achieve strong retrieval performance, their decisions remain largely uninterpretable. Existing interpretability approaches in TI-ReID rely solely on slot-attention to highlight attended regions, but fail to reliably bind visual regions to semantically meaningful concepts, limiting interpretation to qualitative visualizations over a restricted vocabulary. This paper introduces InterPartAbility, an interpretable TI-ReID method that performs explicit part-wise matching and enables phrase-region grounding. Unlike parameter-heavy slot-attention methods that yield only qualitative interpretability, our open-vocabulary patch-phrase interaction module (PPIM) guides a standard TI-ReID model with concept-level phrases. Concept-based part phrases provide evidence that encourages the model to attend to the corresponding local image regions. InterPartAbility further leverages CLIP ViT self-attention to produce spatially concentrated patch activations aligned with each part-level phrase, yielding grounded explanation maps. Finally, a quantitative interpretability protocol for TI-ReID is introduced that extends current perturbation-based evaluation metrics into the TI-Reid domain. This includes a counterfactual region removal that measures retrieval degradation when top-ranked explanatory regions are removed. Empirical results on three challenging benchmarks show that InterPartAbility can achieve SOTA interpretability performance under these metrics, while sustaining competitive retrieval accuracy.

Breakthrough progress in vision-based navigation through unknown environments has been achieved by using multimodal large language models (MLLMs). These models can plan a sequence of motions by evaluating the current view at each time step against the task and goal given to the agent. However, current zero-shot Vision-and-Language Navigation (VLN) agents powered by MLLMs still tend to drift off course, halt prematurely, and achieve low overall success rates. We propose Three-Step Nav to counteract these failures with a three-view protocol: First, "look forward" to extract global landmarks and sketch a coarse plan. Then, "look now" to align the current visual observation with the next sub-goal for fine-grained guidance. Finally, "look backward" audits the entire trajectory to correct accumulated drift before stopping. Requiring no gradient updates or task-specific fine-tuning, our planner drops into existing VLN pipelines with minimal overhead. Three-Step Nav achieves state-of-the-art zero-shot performance on the R2R-CE and RxR-CE dataset. Our code is available at https://github.com/ZoeyZheng0/3-step-Nav.

Deploying Vision-Language Models (VLMs) on edge devices remains challenging due to their substantial computational and memory demands, which exceed the capabilities of resource-constrained embedded platforms. Conversely, fully offloading inference to the cloud is often impractical in bandwidth-limited environments, where transmitting raw visual data introduces substantial latency overhead. While recent edge-cloud collaborative architectures attempt to partition VLM workloads across devices, they typically rely on transmitting fixed-size representations, lacking adaptability to dynamic network conditions and failing to fully exploit semantic redundancy. In this paper, we propose a progressive semantic communication framework for edge-cloud VLM inference, using a Meta AutoEncoder that compresses visual tokens into adaptive, progressively refinable representations, enabling plug-and-play deployment with off-the-shelf VLMs without additional fine-tuning. This design allows flexible transmission at different information levels, providing a controllable trade-off between communication cost and semantic fidelity. We implement a full end-to-end edge-cloud system comprising an embedded NXP i.MX95 platform and a GPU server, communicating over bandwidth-constrained networks. Experimental results show that, at 1 Mbps uplink, the proposed progressive scheme significantly reduces network latency compared to full-edge and full-cloud solutions, while maintaining high semantic consistency even under high compression. The implementation code will be released upon publication at https://github.com/open-ep/ProSemComVLM.

Large Vision-Language Models (VLMs) have achieved remarkable multimodal performance yet remain prone to factual hallucinations, particularly in long-tail or specialized domains. Moreover, current models exhibit a weak capacity to refuse queries that exceed their parametric knowledge. In this paper, we propose a systematic framework to enhance the refusal capability of VLMs when facing such unknown questions. We first curate a model-specific "Visual-Idk" (Visual-I don't know) dataset, leveraging multi-sample consistency probing to distinguish between known and unknown facts. We then align the model using supervised fine-tuning followed by preference-aware optimization (e.g., DPO, ORPO) to effectively delineate its knowledge boundaries. Results on the Visual-Idk dataset show our method improves the Truthful Rate from 57.9% to 67.3%. Additionally, internal probing also demonstrates that the model genuinely recognizes its boundaries instead of just memorizing refusal patterns. Our framework further generalizes to out-of-distribution medical and perceptual domains, providing a robust path toward more trustworthy and prudent visual assistants.

Vision-language models have shown strong performance, but they often generalize poorly to specialized domains. While semi-supervised vision-language learning mitigates this limitation by leveraging a small set of labeled image-text pairs together with abundant unlabeled images, existing methods remain fundamentally pairwise and fail to model the global structure of multimodal representation manifolds. Existing topology-based alignment methods rely on persistence diagram matching, which neither guarantees geometric alignment nor utilizes the image-text pairing information central to vision-language learning. We propose Topology-Aware Multimodal Representation Alignment (ToMA), a framework that uses persistent homology to identify topologically salient edges and aligns them across modalities through available cross-modal correspondences. ToMA leverages both H_0-death edges and lightweight H_1-birth edges, allowing it to capture both connectivity and cycle structure without constructing 2-simplices. Experiments show that ToMA yields stable gains, with clear improvements on remote sensing and modest but consistent benefits on fashion retrieval. Additional analysis shows that ToMA is more stable than alternative topology-based objectives and that lightweight H_1-birth edges provide useful higher-order structural signals.

Chest X-ray interpretation is one of the most frequently performed diagnostic tasks in medicine and a primary target for AI development, yet current vision-language models are primarily trained on datasets of paired images and reports, not the cognitive processes and visual attention that underlie clinical reasoning. Here, we present CheXthought, a global, multimodal resource containing 103,592 chain-of-thought reasoning traces and 6,609,082 synchronized visual attention annotations across 50,312 multi-read chest X-rays from 501 radiologists in 71 countries. Our analysis reveals clinical reasoning patterns in how experts deploy distinct visual search strategies, integrate clinical context, and communicate uncertainty. We demonstrate the clinical utility of CheXthought across four dimensions. First, CheXthought reasoning significantly outperforms state-of-the-art vision-language model chain-of-thought in factual accuracy and spatial grounding. Second, visual attention data used as an inference-time hint recovers missed findings and significantly reduces hallucinations. Third, vision-language models trained on CheXthought data achieve significantly stronger pathology classification, visual faithfulness, temporal reasoning and uncertainty communication. Fourth, leveraging CheXthought's multi-reader annotations, we predict both human-human and human-AI disagreement directly from an image, enabling transparent communication of case difficulty, uncertainty and model reliability. These findings establish CheXthought as a resource for advancing multimodal clinical reasoning and the development of more transparent, interpretable vision-language models.

While Vision-Language Models (VLMs) have achieved state-of-the-art performance in general visual tasks, their perceptual robustness remains remarkably brittle when confronted with optical illusions. These failures are often attributed to shortcut heuristics, where models prioritize linguistic priors and memorized prototypes over direct visual evidence. In this work, we propose Structured Qualitative Inference (SQI), a training-free, data-centric framework designed to fortify visual grounding in frozen VLMs. SQI addresses perceptual anomalies through three systematic modules: (1) Axiomatic Constraint Injection, which suppresses erroneous metric estimations and quantitative hallucinations; (2) Hierarchical Scene Decomposition, which decouples target visual manifolds from complex background distractors; and (3) Counterfactual Self-Verification, an adversarial reasoning step that mitigates confirmation bias. By orchestrating these qualitative constraints at inference time, SQI effectively aligns high-level linguistic reasoning with low-level visual perception. Our framework was evaluated on the DataCV 2026 Challenge (Task I: Classic Illusion Understanding), where it ranked 2nd place overall. Experimental results demonstrate that SQI not only significantly enhances accuracy across diverse illusion categories but also provides superior diagnostic interpretability without any model fine-tuning. Our success underscores the potential of structured qualitative grounding as a robust paradigm for developing next-generation, illusion-resistant vision-language systems.

Vision-Language Models (VLMs) exhibit strong performance in instruction following and open-ended vision-language reasoning, yet they frequently generate fluent outputs that are weakly grounded in visual evidence. Prior works have shown that instruction prompting further worsens this issue by amplifying language priors, especially when the visual signal is uncertain or ambiguous. To address this challenge, we propose a decoding framework that explicitly balances linguistic informativeness and visual faithfulness during generation. Our method, Instruction-Evidence Contrastive Dual-Stream Decoding (IECD$^2$), maintains two parallel probability distribution of tokens at each decoding step: an instruction-driven stream that promotes expressive and informative responses, and an evidence-driven stream that enforces strict grounding in the image. These two streams are adaptively fused using a symmetric KL-based contrastive gate, which suppresses tokens favored by language priors but unsupported by visual evidence, while preserving them when both distributions agree. We evaluate IECD$^2$ on multiple datasets spanning various generative vision-language reasoning tasks such as captioning and visual question answering on multiple datasets such as, POPE, MME, VQAv2, AMBER, and MSCOCO. IECD$^2$ demonstrates consistent improvements in task accuracy and reasoning performance with substantial reduction in hallucination compared to state-of-the-art decoding approaches.

Large Vision-Language Models (LVLMs) have achieved remarkable progress in visual-textual understanding, yet their reliability is critically undermined by hallucinations, i.e., the generation of factually incorrect or inconsistent responses. While recent studies using steering vectors demonstrated promise in reducing hallucinations, a notable challenge remains: they inadvertently amplify the severity of residual hallucinations. We attribute this to their exclusive focus on the decoding stage, where errors accumulate autoregressively and progressively worsen subsequent hallucinatory outputs. To address this, we propose Prefill-Time Intervention (PTI), a novel steering paradigm that intervenes only once during the prefill stage, enhancing the initial Key-Value (KV) cache before error accumulation occurs. Specifically, PTI is modality-aware, deriving distinct directions for visual and textual representations. This intervention is decoupled to steer keys toward visually-grounded objects and values to filter background noise, correcting hallucination-prone representations at their source. Extensive experiments demonstrate PTI's significant performance in mitigating hallucinations and its generalizability across diverse decoding strategies, LVLMs, and benchmarks. Moreover, PTI is orthogonal to existing decoding-stage methods, enabling plug-and-play integration and further boosting performance. Code is available at: https://github.com/huaiyi66/PTI.

Diagram question answering (DQA) requires models to interpret structured visual representations such as charts, maps, infographics, circuit schematics, and scientific diagrams. Recent vision-language models (VLMs) often achieve high answer accuracy on these tasks, yet correct answers do not guarantee that models ground their reasoning in the diagram regions that support the prediction. Models may instead rely on textual correlations or dataset artifacts without identifying the visual evidence required to verify the answer. This limitation prevents reliable evaluation of diagram reasoning and reduces interpretability. We introduce DRAGON, a benchmark for evaluating evidence-grounded visual reasoning in diagrams. Given a diagram, a question, and the correct answer, a model must predict bounding boxes that correspond to the visual elements required to justify the answer. These evidence regions may include answer-bearing components, textual labels, legends, axes, connectors, and other supporting structures involved in the reasoning process. The DRAGON dataset contains 11,664 annotated question instances collected from six diagram QA datasets: ChartQA, Circuit-VQA, InfographicsVQA, MapIQ, MapWise, and AI2D. We release a 2,445-instance benchmark test set with human-verified reasoning evidence annotations and a standardized evaluation framework. We evaluate eight recent VLMs and analyze their ability to localize reasoning evidence across diverse diagram domains. DRAGON enables systematic evaluation of diagram reasoning and supports future research on models that ground their predictions in visual evidence.

Generative vision-language models can produce fluent medical image captions but remain prone to hallucination, over-specific diagnostic claims, and factual inconsistency-serious issues in pathology. We investigate retrieval-guided generation (RGG) as a safer alternative, where captions are formed by summarizing expert text from visually similar cases rather than generated de novo. On the ARCH histopathology dataset, RGG improves semantic alignment with ground truth, achieving cosine similarity of $\approx$0.60 versus $\approx$0.47 from MedGemma, with non-overlapping confidence intervals indicating a robust gain. A pathologist-led qualitative review shows better preservation of morphology-relevant terminology and fewer unsupported diagnoses, while revealing failure modes such as concept mixing and inherited over-specific labeling. Overall, retrieval-guided captioning offers a more transparent and reliable approach with clearer opportunities for auditing than fully generative methods.

In episodic memory with natural language queries (EM-NLQ), a user may ask a question (e.g., "Where did I place the mug?") that requires searching a long egocentric video, captured from the user's perspective, to find the moment that answers it. However, queries can be ambiguous or incomplete, leading to incorrect responses. Current methods ignore this key aspect and address EM-NLQ in a one-shot setup, limiting their applicability in real-world scenarios. In this work, we address this gap and introduce the Episodic Memory with Questions and Feedback task (EM-QnF). Here, the user can provide feedback on the model's initial prediction or add more information (e.g., "Before this. I'm looking for the big blue mug not the white one"), helping the model refine its predictions interactively. To this end, we collect datasets for feedback-based interaction and propose a lightweight training scheme that avoids expensive sequential optimization. We also introduce a plug-and-play Feedback ALignment Module (FALM) that enables existing EM-NLQ models to incorporate user feedback effectively. Our approach significantly improves over the state of the art on three challenging benchmarks and is better than or competitive with commercial large vision-language models while remaining efficient. Evaluation with human-generated feedback shows that it generalizes well to real-world scenarios.

Recent advancements in reinforcement learning with verifiable rewards (RLVR) have significantly improved the complex reasoning ability of vision-language models (VLMs). However, its outcome-level supervision is too coarse to diagnose and correct errors within the reasoning chain. To this end, we propose Perceval, a process reward model (PRM) that enables token-level error grounding, which can extract image-related claims from the response and compare them one by one with the visual evidence in the image, ultimately returning claims that contain perceptual errors. Perceval is trained with perception-intensive supervised training data. We then integrate Perceval into the RL training process to train the policy models. Specifically, compared to traditional GRPO, which applies sequence-level advantages, we apply token-level advantages by targeting penalties on hallucinated spans identified by Perceval, thus enabling fine-grained supervision signals. In addition to augmenting the training process, Perceval can also assist VLMs during the inference stage. Using Perceval, we can truncate the erroneous portions of the model's response, and then either have the model regenerate the response directly or induce the model to reflect on its previous output. This process can be repeated multiple times to achieve test-time scaling. Experiments show significant improvements on benchmarks from various domains across multiple reasoning VLMs trained with RL, highlighting the promise of perception-centric supervision as a general-purpose strategy. For test-time scaling, it also demonstrates consistent performance gains over other strategies, such as major voting. Our code and data will be publicly released at https://github.com/RUCAIBox/Perceval.

Vision-Language Models (VLMs) have demonstrated strong performance on textbook-style physics problems, yet they frequently fail when confronted with dynamic real-world scenarios that require temporal consistency and causal reasoning across frames. We identify two fundamental challenges underlying these failures: (1) spatio-temporal identity drift, where objects lose their physical identity across successive frames and break causal chains, and (2) volatility of inference-time insights, where a model may occasionally produce correct physical reasoning but never consolidates it for future reuse. To address these challenges, we propose PhysNote, an agentic framework that enables VLMs to externalize and refine physical knowledge through self-generated "Knowledge Notes." PhysNote stabilizes dynamic perception through spatio-temporal canonicalization, organizes self-generated insights into a hierarchical knowledge repository, and drives an iterative reasoning loop that grounds hypotheses in visual evidence before consolidating verified knowledge. Experiments on PhysBench demonstrate that PhysNote achieves 56.68% overall accuracy, a 4.96% improvement over the best multi-agent baseline, with consistent gains across all four physical reasoning domains.

Large Vision-Language Models (LVLMs) frequently suffer from hallucinations. Existing preference learning-based approaches largely rely on proprietary models to construct preference datasets. We identify that this reliance introduces a distributional mismatch between the proprietary and target models that hinders efficient alignment. To address this, we propose Alignment via VErified Self-correction DPO (AVES-DPO), a framework that aligns LVLMs using in-distribution data derived from the model's intrinsic knowledge. Our approach employs a consensus-based verification mechanism to diagnose diverse hallucinations and guides the model to self-correct, thereby generating preference pairs strictly compatible with its internal distribution. Extensive experiments demonstrate that AVES-DPO surpasses existing baselines in hallucination mitigation while requiring only 5.2k samples.

While Large Vision Language Models (LVLMs) demonstrate impressive capabilities, their substantial computational and memory requirements pose deployment challenges on resource-constrained edge devices. Current parameter reduction techniques primarily involve training LVLMs from small language models, but these methods offer limited flexibility and remain computationally intensive. We study a complementary route: compressing existing LVLMs by applying structured pruning to the language model backbone, followed by lightweight recovery training. Specifically, we investigate two structural pruning paradigms: layerwise and widthwise pruning, and pair them with supervised finetuning and knowledge distillation on logits and hidden states. Additionally, we assess the feasibility of conducting recovery training with only a small fraction of the available data. Our results show that widthwise pruning generally maintains better performance in low-resource scenarios, where computational resources are limited or there is insufficient finetuning data. As for the recovery training, finetuning only the multimodal projector is sufficient at small compression levels. Furthermore, a combination of supervised finetuning and hidden-state distillation yields optimal recovery across various pruning levels. Notably, effective recovery can be achieved using just 5% of the original data, while retaining over 95% of the original performance. Through empirical study on three representative LVLM families ranging from 3B to 7B parameters, this study offers actionable insights for practitioners to compress LVLMs without extensive computation resources or sufficient data. The code base is available at https://github.com/YiranHuangIrene/VLMCompression.git.

Recent advances in Vision-Language Models (VLMs) have benefited from Reinforcement Learning (RL) for enhanced reasoning. However, existing methods still face critical limitations, including the lack of low-level visual information and effective visual feedback. To address these problems, this paper proposes a unified multimodal interleaved reasoning framework \textbf{ForeSight}, which enables VLMs to \textbf{See Further} with low-level visual cues and \textbf{Think Deeper} with effective visual feedback. First, it introduces a set of low-level visual tools to integrate essential visual information into the reasoning chain, mitigating the neglect of fine-grained visual features. Second, a mask-based visual feedback mechanism is elaborated to incorporate visual reflection into the thinking process, enabling the model to dynamically re-examine and update its answers. Driven by RL, ForeSight learns to autonomously decide on tool invocation and answer verification, with the final answer accuracy as the reward signal. To evaluate the performance of the proposed framework, we construct a new dataset, Character and Grounding SalBench (CG-SalBench), based on the SalBench dataset. Experimental results demonstrate that the ForeSight-7B model significantly outperforms other models with the same parameter scale, and even surpasses the current SOTA closed-source models on certain metrics.

Current Vision-Language-Action (VLA) models predominantly rely on end-to-end fine-tuning. While effective, this paradigm compromises the inherent generalization capabilities of Vision-Language Models (VLMs) and incurs catastrophic forgetting. To address these limitations, we propose $M^2$-VLA, which demonstrates that a generalized VLM is able to serve as a powerful backbone for robotic manipulation directly. However, it remains a key challenge to bridge the gap between the high-level semantic understanding of VLMs and the precise requirements of robotic control. To overcome this, we introduce the Mixture of Layers (MoL) strategy that selectively extracts task-critical information from dense semantic features. Furthermore, to facilitate efficient trajectory learning under constrained model capacity, we propose a Meta Skill Module (MSM) that integrates strong inductive biases. Extensive experiments in both simulated and real-world environments demonstrate the effectiveness of our approach. Furthermore, generalization and ablation studies validate the architecture's zero-shot capabilities and confirm the contribution of each key component. Our code and pre-trained models will be made publicly available.

Detecting hate speech in memes is challenging due to their multimodal nature and subtle, culturally grounded cues such as sarcasm and context. While recent vision-language models (VLMs) enable joint reasoning over text and images, end-to-end prompting can be brittle, as a single prediction must resolve target, stance, implicitness, and irony. These challenges are amplified in multilingual settings. We propose a prompted weak supervision (PWS) approach that decomposes meme understanding into targeted, question-based labeling functions with constrained answer options for homophobia and transphobia detection in the LT-EDI 2026 shared task. Using a quantized Qwen3-VLM to extract features by answering targeted questions, our method outperforms direct VLM classification, with substantial gains for Chinese and Hindi, ranking 1st in English, 2nd in Chinese, and 3rd in Hindi. Iterative refinement via error-driven LF expansion and feature pruning reduces redundancy and improves generalization. Our results highlight the effectiveness of prompted weak supervision for multilingual multimodal hate speech detection.

Mixture-of-Experts (MoE) has become a prevalent backbone for large vision-language models (VLMs), yet how modality-specific signals should guide expert routing remains under-explored. Existing routing strategies are either hand-crafted or modality-agnostic, relying on idealized priors that ignore the layer-dependent modality fusion patterns in MoE-VLMs and provide little guidance for expert specialization. We propose Soft Modality-guided Expert Specialization (SMoES), which consists of dynamic soft modality scores that capture layer-dependent fusion patterns, an expert binning mechanism aligned with expert-parallel deployment, and an inter-bin mutual information regularization that encourages coherent modality specialization. Our method leverages attention-based or Gaussian-statistics modality scores to optimize mutual information regularization. Experiments across four MoE-based VLMs and 16 benchmarks demonstrate improvement on both effectiveness and efficiency: 0.9% and 4.2% average gain on multimodal and language-only tasks, 56.1% reduction in EP communication overhead, and 12.3% throughput improvement under realistic deployment. These results validate that aligning routing with modality-aware expert specialization unlocks MoE-VLM capacity and efficiency.

Graphical User Interface (GUI) element grounding (precisely locating elements on screenshots based on natural language instructions) is fundamental for agents interacting with GUIs. Deploying this capability directly on resource-constrained devices like mobile phones is increasingly critical for GUI agents requiring low latency. However, this goal faces a significant challenge, as current visual grounding methods typically employ large vision-language model (VLM) (more than 2.5B parameters), making them impractical for on-device execution due to memory and computational constraints. To address this, this paper introduces GoClick, a lightweight GUI element grounding VLM with only 230M parameters that achieves excellent visual grounding accuracy, even on par with significantly larger models. Simply downsizing existing decoder-only VLMs is a straightforward way to design a lightweight model, but our experiments reveal that this approach yields suboptimal results. Instead, we select an encoder-decoder architecture, which outperforms decoder-only alternatives at small parameter scales for GUI grounding tasks. Additionally, the limited capacity of small VLMs encourages us to develop a Progressive Data Refinement pipeline that utilizes task type filtering and data ratio adjustment to extract a high-quality 3.8M-sample core set from a 10.8M raw dataset. Training GoClick using this core set brings notable grounding accuracy gains. Our experiments show that GoClick excels on multiple GUI element grounding benchmarks while maintaining a small size and high inference speed. GoClick also enhances GUI agent performance when integrated into a device-cloud collaboration framework, where GoClick helps cloud-based task planners perform precise element localization and achieve higher success rates. We hope our method serves as a meaningful exploration within the GUI agent community.

Appreciating multi-figure paintings requires understanding how characters relate through subtle cues like gaze alignment, gesture, and spatial arrangement. We present MIRAGE, an evidence-centric framework designed to scaffold the exploration of these "micro-interactions" in multi-figure artworks. While such cues are essential for deep narrative appreciation, they are often distributed across complex scenes and difficult for viewers to systematically identify. Existing vision-language models (VLMs) frequently fail to provide reliable assistance, offering ungrounded interpretations that lack traceable visual evidence. MIRAGE addresses this by constructing a structured intermediate representation capturing identities, pose cues, and gaze hypotheses. However, the challenge extends beyond extracting these cues to coordinating them during interpretation. Without an explicit mechanism to organize and reconcile relational evidence, models often collapse multiple interaction hypotheses into a single unstable or weakly grounded narrative, even when low-level signals are available. This representation allows users to verify how high-level interpretations are anchored in low-level visual facts. By separating spatial grounding from narrative generation, MIRAGE enables users to inspect and reason about figure-to-figure relationships through a verifiable evidence layer. We evaluate MIRAGE against painting-only VLM baselines using a blind assessment protocol. Results show that MIRAGE significantly improves identity consistency, reduces relational hallucinations, and increases the coverage of subtle interactions. These findings suggest that structured grounding can serve as a critical interaction control layer, providing the necessary scaffolding for a more reliable, transparent, and human-led understanding of complex visual narratives.