Large Vision Language Models

Despite the remarkable progress of large vision language models (LVLMs), object hallucination remains a fundamental challenge that hinders their trustworthy deployment. A key finding motivates our work: real and hallucinated object tokens are clearly separable in hidden representations, yet this separability is largely lost at the language-modeling (LM) head. We propose TruthLens, a self-evaluation framework that teaches the LM head to expose a per-object truthfulness signal without any auxiliary model or additional inference cost. Concretely, a rarely-used special token is repurposed as a reference token. For each object-token position, we extract the log-probability assigned to this special token by the LM head, and define its difference from a predefined constant as the truthfulness score. The model is then fine-tuned with an MSE objective that drives scores toward 1 for real objects and 0 for hallucinated ones, while a divergence constraint preserves the original generation capability. Despite being trained on only a limited set of object categories, TruthLens generalizes effectively to benchmarks with substantially larger label spaces. Extensive experiments across multiple LVLMs demonstrate state-of-the-art performance; notably, on Qwen2.5-VL-7B, TruthLens outperforms the previous best method on MS-COCO by over 17% in AUROC. Our code is available at https://github.com/wyqstan/TruthLens.

Recent vision-language models (VLMs) can generate executable CAD programs from images, but existing methods mainly target coarse, general-purpose 3D objects and rarely address the fine-grained geometry and millimeter-level tolerances required in industrial mechanical design. We introduce OmniMech, the first million-scale benchmark for evaluating VLMs on executable CAD generation from industrial manufacturing data. OmniMech contains more than 251,000 fully dimensioned and toleranced 2D orthographic drawings, paired with native CAD models, multi-view renderings, mesh, STEP and B-rep representations, and rich semantic annotations. The benchmark includes four tasks: (1) parametric CAD program synthesis from engineering drawings; (2) diagram-to-3D reasoning for geometrically and structurally consistent reconstruction; (3) annotation-grounded reasoning over dimensions, symbols, feature callouts, and manufacturing constraints; and (4) tool-augmented agentic reasoning using visualization, measurement, CAD execution, and verification tools. Experiments show that current VLMs and CAD-specialized models still struggle with executable program synthesis, fine-grained 3D reconstruction, and reliable enforcement of dimensions and tolerances. We will release the benchmark data, evaluation code, and tool interfaces to support future research.

Deep vision models exploit shortcuts, relying on cues that correlate with supervision signals. Prior work has focused on visible biases, such as object-background or texture correlations. We identify a different source of shortcut learning: invisible metadata traces embedded at the pixel level, for metadata such as image processing and photo acquisition. We hypothesize that large-scale semantic supervision, whether through categorical labels (ImageNet) or billion-scale captions (LAION), naturally induces metadata-semantics correlations during pretraining, leading models to convert low-level signals into predictive features. By introducing controlled metadata-semantics correlations, we show that stronger ones produce systematically higher sensitivity to metadata traces and larger performance degradation under metadata distribution shifts. We further explore mitigation strategies applied during and after pretraining that reduce sensitivity not only to targeted metadata but also to unseen ones, without sacrificing performance on downstream tasks. Metadata sensitivity also has a positive side: it partly explains the strong generated-image detection ability of some encoders, while its mitigation can improve out-of-distribution generalization. Code: https://github.com/ryan-caesar-ramos/visual-encoder-traces

Background: Accurate glioma subregion delineation is important for radiotherapy planning and longitudinal monitoring, but manual contour correction is time-consuming. Models such as nnU-Net may generalize imperfectly and lack clinician-directed text correction. Purpose: We investigated adapting a three-dimensional (3D) vision-language foundation model for text-guided brain tumor segmentation refinement. Methods: We developed a lightweight VoxTell-based framework. Pretrained VoxTell generated initial masks. Oracle prompts derived from segmentation errors encoded target, action, location, imaging evidence, edit size, and preservation constraints. Frozen Qwen/VoxTell prompt embeddings were injected through trainable projections into its multiscale decoder conditioning; other weights remained frozen. Training, validation, and testing used 901, 100, and 250 BraTS-GLI cases. Cross-dataset transfer was evaluated on 100 meningioma, metastasis, pediatric tumor, and UPENN-GBM cases. Results: On the internal test set using post-contrast T1-weighted input, correct instructions improved subregion Dice similarity coefficient (DSC; enhancing tumor, edema, and necrotic/non-enhancing core) from $0.774\pm0.158$ to $0.796\pm0.137$. They outperformed blank prompts ($0.762\pm0.155$; Holm-adjusted $p<0.001$, $d_z=0.71$) and contradictory prompts ($0.770\pm0.163$; $p<0.001$, $d_z=0.48$). In cross-dataset testing, correct instructions improved DSC from $0.527\pm0.287$ to $0.550\pm0.278$ and outperformed contradictory instructions ($0.504\pm0.275$; $p<0.001$, $d_z=0.43$). Conclusion: A 3D vision-language foundation model can perform instruction-guided refinement of glioma subregion segmentations. Sensitivity to correct, blank, and contradictory prompts suggests text-dependent contour editing rather than nonspecific post-processing, supporting further evaluation as a clinician-in-the-loop tool.

Standard accuracy metrics for VLMs often mask significant reliability failures in sensitive domains. In this work, we utilize a histopathology-validated brain MRI dataset to systematically assess the diagnostic robustness of four VLM families under evidence-preserving perturbations. By reordering anatomical slices and swapping target label positions, we evaluate whether models maintain consistent predictions when clinical evidence remains invariant. Our results reveal significant vulnerabilities in presentation-order stability, with models exhibiting prediction flips in up to 48.9% of cases under simple sequence reversals. We further identify a textual selection bias, where label reordering triggers inconsistent diagnoses in up to 67.8% of cases despite identical visual inputs. Negative-control tests further reveal diagnostic overcommitment: models generate categorical diagnoses in up to 76.1% of cases after expert-annotated lesion slices are removed. These results demonstrate that high accuracy can overestimate clinical reliability, masking sensitivity to sequential presentation and textual framing that is not captured by aggregate accuracy. Our findings highlight the necessity of stability-based metrics for the deployment of VLMs in safety-critical clinical applications. Our evaluation data and code will be made public upon acceptance.

Dreams can be emotionally intense but difficult to communicate. We describe the Dream Scene Visualiser (DSV) system which turns written dream descriptions into a temporal sequence of four panel images visualising the dream. This starts with a large language model prompted to split a dream description into four chronological parts. Then a text-to-image model produces images for each part with visual coherence maintained across the sequence, and DSV regenerates any image not suitably matching the text. We evaluate DSV over 50 visualisations from dream descriptions in DreamBank, and report quality, fidelity and coherence results via objective measures employing the CLIP, DINOv2 and Qwen2-VL vision-language models.

Vision-language-action (VLA) models provide a unified paradigm for connecting visual perception, language understanding, and robotic control. However, existing VLA models still face major challenges in long-horizon tasks: sparse expert demonstrations constrain cross-task compositional generalization; the non-Markovian nature of long-horizon tasks makes it difficult for policies conditioned only on current observations to maintain temporal consistency; limited closed-loop error correction allows execution errors to accumulate; and end-to-end action fine-tuning may weaken the high-level semantic representations of vision-language model (VLM) backbones. To address these issues, we propose a hierarchical long-horizon VLA architecture with an explicit language-memory module. The central idea is to convert discrete temporal observations into a coherent textual memory sequence with temporal logic. The system is decoupled into a high-level VLM and a low-level VLA: the high-level VLM performs semantic reasoning through a visual question answering training paradigm, while the low-level VLA executes precise continuous control conditioned on subtask instructions and visual observations. The high-level VLM recursively updates both language memory and subtask instructions using the previous memory as a contextual anchor, enabling persistent temporal tracking and dynamic correction during long-horizon execution. We evaluate the proposed method in multiple simulation environments and conduct sim-to-real experiments on a real robotic platform. The results demonstrate that explicit language memory improves the success rate and robustness of VLA models on complex long-horizon tasks while providing an interpretable semantic account of the decision process.

Learning manipulation skills from human videos is promising for scalable robot learning. However, the embodiment mismatch between humans and robots makes this challenging. One promising solution is to learn object-centric actionable affordances that are embodiment-agnostic. In this work, we propose a framework that leverages egocentric human videos with state-of-the-art 3D Structure-from-Motion and hand mesh reconstruction to extract actionable affordances such as visual, grasp, and trajectory affordances that explicitly encode where to interact, how to grasp, and how to move. We construct EgoAffordance, a large-scale dataset comprising 204K episodes with 5.6M visual affordances and 11.6M grasp and trajectory affordances. Building on this, we introduce VLAff, a large vision-language model-based unified foundation model that learns cross-modal correlations across all actionable affordances. Given a visual observation and instruction, VLAff generates visual affordance heatmaps, grasp poses, and trajectories, which are then converted into directly executable actions by utilizing 3D scene information. Through extensive experiments, we demonstrate that VLAff not only achieves state-of-the-art performance on visual affordance prediction, but can also be effectively applied to real robot applications such as zero-shot manipulation and affordance-guided robot learning.

Vision-language models (VLMs) have made rapid progress in visual perception and increasingly support real-world tasks that depend on images. Many such tasks, however, require more than rec- ognizing what an image contains: a model must use visual evidence to make a complete decision whose parts jointly satisfy global constraints. We introduce COMPLEXITYWORLD, a benchmark of 390 tasks across 39 domain-inspired visual worlds and 29 decision categories. Each task is generated from a hidden structured specification, rendered as a visual scene, and scored by an exe- cutable verifier that accepts any feasible solution. Under direct inference, all evaluated models ex- cept GPT-5.6-Sol remain below 40% verifier ac- ceptance rate (VAR), while GPT-5.6-Sol reaches 75.6%. Performance improves substantially when the same decision information is made explicit in structured form, yet varies sharply across equiva- lent visual presentations. Agent scaffolds provide smaller, model-dependent gains. Together, these results reveal a persistent visual-to-decision bot- tleneck that additional inference alone does not remove.

Picture books and comics have long been used to disseminate hateful narratives because they are easily understood even by children, as exemplified by the notorious Nazi propaganda picture book \emph{Der Giftpilz}. Recently, frontier text-to-image (T2I) systems such as Gemini and GPT-Image have enabled conversational generation with consistent characters and scenes across turns, making hateful visual stories, namely ordered image groups that collectively convey hateful narratives, cheap and scalable to produce. Although prior work has studied hateful content generation by T2I systems, it focuses on individual images, leaving group-level hateful meaning largely unexplored. We aim to address the gap. Concretely, we introduce \texttt{HatefulStoryPrompts}, comprising 330 multi-turn configurations from 55 hateful stories across two languages and three visual styles, and evaluate five frontier models over 4,950 attempts. Every model completes over 80% of the stories, with the strongest reaching 99.0%. We further evaluate existing moderation systems on \texttt{HatefulVisualStory}, a human-labeled dataset of 969 hateful image sets and 990 benign controls, and find that they frequently miss group-level hateful meaning: dedicated safety models achieve at most 34.9% recall, while a strong vision-language model reaches 67.5%. Finally, we propose complementary proactive and post-generation defenses. An interaction-aware monitor achieves 97.3% recall for prompt-only sessions and 92.6% when the user supplies the first image, while post-generation methods jointly analyzing completed image groups reach 80.2%. Our work shows that, as image generation evolves from isolated outputs to coherent visual narratives, safety must evolve accordingly, from per-image moderation to stateful reasoning over interactions and image relationships.

Predicting item difficulty from content can provide an initial estimate for newly developed questions before sufficient student responses are available. Existing approaches typically represent the question stem and answer choices as text. When mathematics items contain visual components, a common pipeline first textualizes that evidence and then applies a text predictor. We ask: how should visual evidence be represented for item difficulty prediction? We compare question text alone, visual textualization, which expresses visual evidence in language, and image-native modeling, which retains the original image. Using Eedi items with difficulty calibrated from student responses, we train large language models (LLMs) and vision-language models (VLMs) directly for difficulty regression. Both visual interfaces achieve the lowest point estimates, although the leading systems cannot be reliably ordered. Open-VLM textualization yields lower RMSE point estimates for all evaluated LLMs, while broader adaptation does so for all image-native VLMs. Test-time interventions show dependence on the paired full-item image, but do not isolate the additional visual component. The two visual interfaces also make partially complementary item-level errors and differ substantially in computational workflow. Thus, textualization should not be treated as the only practical interface: image-native modeling is a competitive alternative whose effectiveness depends on how the VLM is adapted.

Frozen vision-language-action (VLA) policies generate temporally extended action chunks, but long-horizon manipulation remains vulnerable to accumulated execution error and visual aliasing across task stages. Successful rollouts provide useful corrective evidence, yet current frame retrieval can return progress-misaligned actions,while direct replay or time-domain fusion can overwrite the reactive structure of the policy proposal. We introduce Retrieve in Time, Correct in Frequency (RTCF), a training-free test-time correction framework that improves frozen VLA performance with low model-side overhead.RTCF separates which experience to retrieve from which part of its action to transfer. Progressive Memory Alignment (PMA) causally aligns the growing visual execution history with complete successful trajectories through incrementally updated monotonic frontiers, jointly identifying a relevant memory and the current aligned memory position without stage labels. From the aligned action chunk,RTCF transfers a coefficient-wise-clipped low-frequency residual on motion channels. Higher-frequency components and gripper decisions remain inherited from the frozen policy. Across four LIBERO suites and 2,000 episodes per condition, RTCF raises aggregate success from 86.4% to 88.4% and improves LIBERO-Long from 61.6% to 68.6%.These gains require no parameter updates, repeated VLA inference, or additional GPU resources: correction can be performed on the client CPU after a single policy invocation, and the median latencies sum to only 10.99 ms per action chunk

Visual inputs in vision-language models (VLMs) are often encoded into substantially longer token sequences than text, making visual tokens a major bottleneck for efficient inference. Abundant recent methods address this bottleneck by scoring token importance and pruning low-scoring tokens in a single pass. However, one-shot scoring is insufficient because a token's prompt-relevant usefulness depends on the evidence already retained. Motivated by this insight, we introduce DIVE (Dynamic Iterative Visual Evidence Construction), a training-free framework that recasts visual-token pruning as dynamic evidence construction. DIVE repeatedly selects the remaining token with the highest residual-conditioned score, updates the visual and prompt residuals to discount the evidence already explained, and re-evaluates the remaining tokens. This select-update-re-evaluate process builds a retained set of complementary, prompt-relevant evidence. Experiments across eight image-understanding benchmarks show that DIVE consistently preserves performance across token budgets. With an 88.9% reduction in visual tokens, DIVE retains 98.2% of the uncompressed model's average performance. Code is available at https://github.com/Zhong-Chenchen/DIVE.git.

Mixture-of-experts vision-language models (MoE-VLMs) increase model capacity with sparse expert activation, yet deployment requires storing the full expert pool. Training-free expert merging reduces this burden, and many routing-based methods aggregate routing statistics across all tokens to determine merge compatibility. However, MoE-VLM inference is phase-structured: image-context tokens carry visual content, question tokens specify the query, and answer tokens produce the output, with different counts and routing distributions. Because image-context tokens are far more numerous, global aggregation can overemphasize image-context processing and obscure phase-conditioned expert roles, making experts serving different phases appear interchangeable and degrading model performance. We therefore argue that MoE-VLM expert merging should preserve phase-conditioned expert roles, judging compatibility by how experts serve different phases rather than globally aggregated routing statistics. Based on this view, we propose RoleMerge, a training-free method that constructs each expert's Routing Role Profile (RRP) from phase-normalized routing statistics, capturing its relative phase preference. Guided by expert-phase information loss, RoleMerge merges experts with compatible profiles and their corresponding router entries while preserving answer-decoding expert distinctions. Experiments on three models and multiple benchmarks show that RoleMerge preserves more of the full model's performance than alternative expert-merging methods at matched expert-retention ratios, with relative improvements of up to 9.6 percent in six-task macro-average performance. These results validate phase-conditioned expert roles as a more effective basis than global routing aggregation for MoE-VLM expert merging.

Group-based reinforcement learning methods for multimodal large language models typically rely on trajectory-level credit assignment that applies a single advantage to all tokens in a response. However, multimodal reasoning involves substantially higher perceptual uncertainty than text-only settings, where the model must repeatedly re-examine visual information to verify intermediate interpretations, and different visual groundings can lead to divergent reasoning paths, making such uniform credit assignment particularly inadequate and causing relative advantages to progressively degenerate toward zero. To address these challenges, we propose Multi-Branch Policy Optimization (MBPO), a tree-based framework that constructs reasoning trees at vision-language decision boundaries, enabling sibling branches to explore diverse visual hypotheses and assigning segment-level credit through branch-relative advantages. We further introduce a temporal replay buffer to reuse informative segments while controlling policy staleness. Experiments on several multimodal reasoning benchmarks show that MBPO outperforms representative baselines, improving both learning signal quality and optimization efficiency. The code is publicly available at https://github.com/ShuaiLyu0110/MBPO.

Edge platforms used for aerial observation must interpret aircraft imagery under limited memory, limited compute, and intermittent connectivity. This setting is difficult for standard RGB-only recognition models and general-purpose vision-language models, especially when calibrated thermal and LiDAR aircraft data are unavailable. We present TriCLE, an application-oriented tri-modal vision-language system for aircraft taxonomic grouping under edge constraints. From a single RGB aircraft image, TriCLE generates a structure-preserving FLIR-style thermal view and a pseudo-LiDAR depth projection, then fuses the aligned views with task instructions in a compact Qwen3-VL backbone. The model is aligned to an expert aircraft taxonomy based on propulsion, airframe family, size, design era, and configuration, so its outputs reflect engineering-relevant similarity rather than only surface appearance. We evaluate supervised fine-tuning, rotation-preserving SFT, and three policy-alignment strategies: GRPO, GSPO, and DAPO. Sequence-level GSPO gives the strongest validation performance, reaching 88.33% validation accuracy and 0.91 weighted F1 on valid aircraft outputs. On a held-out aircraft test partition, GSPO achieves 78.00% accuracy and 0.793 weighted F1 while preserving 94.00% parseable output formatting. After 4-bit quantization and attention-memory optimization, the aligned 4B model fits an 8GB deployment target and processes each tri-modal triplet in 1.48 seconds. These results support TriCLE as a practical prototype for interpretable, edge-feasible aircraft grouping, while emphasizing the need for further validation on real aligned thermal and LiDAR sensor streams.

Existing scaling strategies for Multimodal Large Language Models (MLLMs) typically expand either model parameters or sequential inference computation, incurring substantial memory or latency overhead. More importantly, most existing methods fail to alter the rigid, fixed computation allocation between the Vision Transformer and the Large Language Model components, limiting task-specific optimization. To address this, we introduce the Parallel Vision-Language (ParVL) scaling framework for MLLMs, which scales parallel computation by reusing the existing ViT and LLM backbone parameters across multiple vision and language branches. This framework raises a central question: given a fixed backbone parameter budget, how should additional shared-backbone computation be allocated between the vision and language modalities? We instantiate each parallel computational stream with branch-specific prefix parameters over a shared backbone, and train the entire model end-to-end via full-parameter supervised fine-tuning on roughly 13B tokens. We systematically study the computation-allocation trade-off between the ViT encoder and LLM decoder. ParVL improves overall multimodal performance over same-recipe single-branch baselines, and the best evaluated vision--language allocation varies across tasks. Code is available at https://github.com/YangYangGirl/ParVL.

Designing effective reward functions remains a major bottleneck in Reinforcement Learning (RL). Recent work uses large foundation Vision-Language Models (VLMs) as reward models, computing text-observation similarity to bypass manual reward engineering. Although promising, these rewards are often noisy and unreliable, limiting their direct utility during deployment. We present Structure-Aware Fine-Tuning (SAFT), a simple, self-supervised method that refines these imperfect reward signals online without access to ground-truth supervision. SAFT leverages intrinsic structural priors to regularize the VLM's latent space via LoRA adapters. We rigorously evaluate SAFT across a spectrum of base model capabilities to demonstrate its versatility. Our results show that SAFT consistently denoises the reward landscape, yielding faster policy convergence and substantially improved alignment (EPIC distance) relative to the underlying base model, suggesting that failures can often be attributed to structural brittleness rather than semantic misunderstanding. By replacing extensive human preference annotation with structural inductive biases inherent to the task, SAFT offers a scalable path for stabilizing text-conditioned RL and underscores the broader value of incorporating task structure as a general inductive bias.

Large vision--language models (LVLMs) demonstrate strong multimodal reasoning capabilities but remain prone to hallucination, where model predictions are not grounded in visual evidence. Existing black-box hallucination detection methods estimate uncertainty through a single consistency metric, implicitly assuming that model uncertainty can be adequately characterized by a single measure. However, hallucinations exhibit diverse manifestations of uncertainty across different behavioral probes, making a single measure insufficient to characterize their underlying behavior. We propose \emph{Unique Hallucination Pattern (UHP) Detection}, a fully black-box framework that models hallucination as a structured uncertainty pattern defined by two axes: perturbation modality (image vs.\ text) and logical polarity (a statement vs.\ its negation). Their intersection produces four complementary consistency groups that capture distinct manifestations of model uncertainty, from which both within-group and between-group features are extracted to train a lightweight classifier. Through comprehensive experiments on AMBER and PhD across three LVLMs, UHP Detection consistently outperforms prior black-box and white-box baselines, with improvements of up to $+18.72\%$ AUC-ROC and $+20.07\%$ AUC-PR over the strongest black-box methods. Extensive ablation studies demonstrate that each consistency group contributes complementary information and that their combination forms a structured hallucination pattern. Furthermore, cross-dataset evaluation shows that this learned pattern generalizes across benchmarks, indicating that hallucination behavior reflects a model-specific consistency pattern. \textbf{Code is publicly available at} https://github.com/amirezzati/uhpdet.

While Vision-Language Models (VLMs) have demonstrated remarkable performance in processing and understanding both text and images, their large parameter sizes lead to significant computational overhead, limiting their deployment on resource-constrained devices. While pruning has been effective for compressing Large Language Models (LLMs), directly applying it to VLMs leads to significant performance drops, largely due to redundant visual tokens interfering with importance estimation. To this end, we propose SlimVLM, a structured pruning framework designed to compress VLMs while preserving their task performance. We introduce an adaptive visual token selection strategy for VLMs that leverages average text-to-visual attention scores to assess the importance of visual tokens, removing redundant ones during pruning based on a set threshold, thereby optimizing the importance calculation. Recognizing the varying tolerance to sparsity across different modules, we also propose a Sensitivity-aware dynamic pruning mechanism that determines the appropriate pruning ratio for each module by calculating the linear reconstruction error between the outputs of the pruned and unpruned modules, ensuring overall performance stability. Experimental results show that SlimVLM outperforms existing methods across multiple multimodal benchmarks, achieving state-of-the-art performance.

Generative Vision-Language Models (VLMs) commonly treat bounding-box coordinates as independent output symbols, leaving numerical order and axis semantics implicit. We identify this representation as an important source of error in visual grounding. Hi-Token encodes each coordinate with axis-specific tokens for the hundreds, tens, and ones digits, which adds coarse-to-fine structure and increases token reuse while retaining the existing VLM architecture. Hi-GAR complements this representation with a geometry-based reward for Group Relative Policy Optimization (GRPO), using box overlap and coordinate accuracy at multiple scales. Controlled comparisons under matched training conditions show that Hi-Token improves localization throughout the evaluated IoU range. Hi-GAR further reduces low-overlap predictions and is used only during training. Experiments on three VLM backbones and the RefCOCO family show consistent gains across models and benchmarks. Hi-R1 achieves higher values than strong specialist baselines on most reported metrics. Analyses of token frequency, digit boundaries, object scale, and IoU distributions explain the effects of coordinate representation and reward training. The results show that structured coordinate generation provides an effective approach to generative visual grounding.

Medical visual grounding connects free-form clinical queries to spatial evidence in medical images and is an important component of interpretable medical artificial intelligence. However, general-purpose grounding models are predominantly trained on natural images, while existing medical localization resources remain fragmented across imaging modalities, datasets, and task formulations. To address this gap, we construct LocAnyMed-200K, a multimodal medical visual grounding dataset containing approximately 200K image-query-answer examples across computed tomography, optical medical imaging, ultrasound, and X-ray. We harmonize heterogeneous detection and localization resources into a unified free-form instruction format that supports one or multiple bounding boxes, point coordinates, and no-target outputs for negative queries. Full-parameter fine-tuning of LocateAnything-3B on LocAnyMed-200K improves F1@IoU 0.50 from 10.64 to 85.59 on a held-out evaluation split, demonstrating that large-scale domain-specific supervision can equip a general grounding model with effective medical localization capabilities. Beyond spatial coordinates, a clinically interpretable grounding system should also communicate the evidence supporting its prediction. We therefore derive LocAnyMed-CoT-20K, a rationale-augmented subset that connects anatomical context, visual observations, and spatial conclusions through structured reasoning and further improves cross-source generalization through fine-tuning. Together, these resources provide a unified foundation for studying both localization accuracy and rationale quality across heterogeneous medical imaging modalities. The code is publicly available at https://github.com/MiliLab/LocAnyMed.

Post-training reinforcement learning (RL) algorithms are commonly used to align large vision-language models (LVLMs) with human intent and the requirements of visual reasoning tasks. However, existing RL-based alignment methods are often resource-intensive and encounter mismatches between training objectives and inference-time distributions. To bridge this gap, we propose a novel test-time alignment approach that leverages trajectory-guided structured sampling for dynamic inference-time refinement, achieving better alignment with visual grounding and ensuring logical consistency. Our approach begins with curating a reasoning memory bank via a trajectory learning algorithm, which decomposes complex question solving into ordered sequences of predefined reasoning patterns. It subsequently accomplishes inference-time alignment by first collecting trajectories from reasoning memory bank to establish a global structural reasoning prior, and then using an iterative Markov Chain Monte Carlo (MCMC) algorithm for localized multi-objective refinement of the reasoning trace. Experiments across multiple multimodal reasoning datasets demonstrate that our approach significantly improves accuracy without incurring prohibitive inference overhead. These results establish trajectory-guided test-time sampling as a scalable and effective alternative to traditional post-training alignment, particularly for complex visual reasoning tasks.

Explainable aesthetic image cropping requires not only localizing a visually pleasing crop but also explaining why it is preferred. Existing crop-and-explain methods largely treat explanation as post-hoc text generation and overlook composition, a key aesthetic factor that links crop decisions with interpretable reasoning. In this paper, we reformulate explainable aesthetic image cropping as a structured crop-composition-explanation problem. To support this setting, we introduce COMEX, a new benchmark built through image expansion and an IO-reversal pipeline. COMEX contains 33,161 quadruples, each consisting of an expanded image, a crop box, a composition category, and a composition-grounded explanation, enabling joint learning of crop localization, composition understanding, and explanation generation. We further propose a two-stage SFT+GRPO framework, where supervised fine-tuning establishes the structured output protocol and basic cropping ability, and GRPO further improves crop quality, composition prediction, and explanation faithfulness. We benchmark 15 large vision-language models and existing cropping methods on COMEX, establishing a comprehensive testbed for composition-grounded explainable aesthetic cropping. Experiments on both COMEX and prior benchmarks demonstrate the effectiveness and transferability of our framework, with strong performance across evaluation metrics.

Prior work has shown that LLMs encode the truth of factual propositions along linear directions in activation space. It's unclear how these representations extend to contextual truth: propositions whose truth is determined by in-context evidence rather than world knowledge. We show that LLMs maintain a linear representation of contextual truth that persists across structurally different output policies, even when the output doesn't require the model to determine a proposition's truth, and show causal evidence via steering experiments. Using the transcripts from a collaborative vision-language task that requires two LLMs to maintain a shared common ground, we show that truth representations of a proposition are significantly swayed by partner assertions about that proposition, even when the LLM has enough evidence to determine its truth. We find evidence that propositions near the decision boundary are more susceptible to having their truth shifted through partner assertions. Separating representation from output distinguish two forms of sycophancy that output behavior alone cannot: the model may accommodate a false proposition while continuing to represent it as false, or shift its representation across the boundary. The latter is 2.59x more common when the model agrees by restating the false claim explicitly than when it agrees implicitly.

Reachability analysis for visuomotor policies is difficult because large visual encoders make end-to-end set propagation computationally expensive and excessively conservative. We therefore freeze the visual encoder and confine set propagation to a low-dimensional interface between it and the downstream policy, with the interface set calibrated from held-out camera-pose perturbations. Propagating this set through the policy with zonotopes yields a terminal output-enclosure width that set-based training optimizes directly. During evaluation, camera-pose perturbations are sampled from the prescribed distribution, and rollout-level split conformal calibration converts the resulting action-deviation scores into a probabilistic reachable-action radius with finite-sample coverage. In controlled manipulation experiments, set-based training reduces this radius while preserving closed-loop task capability, and matched behavior-only, observational-consistency, and pointwise-adversarial controls all leave a larger radius.

Vision-language models (VLMs) are increasingly deployed as reasoning agents in real-world visual assessment pipelines, yet their spatial grounding remains unreliable for fine-grained, visually ambiguous targets. We study this gap in the context of automated vehicle damage assessment, where fine-grained defects such as scratches and hairline cracks occupy few pixels, produce weak gradient signal, and are easily confused with reflections and surface texture. We show that a state-of-the-art VLM (Qwen-VL) achieves strong semantic classification accuracy (87.3%) on this task but is systematically ungrounded at the spatial level: it hallucinates damage in reflective regions, misses elongated scratches entirely, and produces spatially inconsistent outputs when prompted for localization. We propose TinyDamage, a hybrid architecture that delegates spatial grounding to a dedicated multi-task segmentation model while reserving the VLM for semantic reasoning and report generation. On the segmentation side, we find that the choice of loss function has an outsized and underexplored effect on tiny-object grounding: focal loss, widely used for class imbalance, collapses tiny-damage detection to zero, while a supervised contrastive objective measurably improves damage/background separability. We integrate the segmentation model into a 7-node LangGraph agent pipeline that grounds every VLM generation step in the segmentation output, and show that this grounding reduces the report hallucination rate from 92% (text-only) and 78% (image-only) to 31% in a controlled evaluation on 100 human-verified reports. We introduce DET_l, a permissive per-category detection metric for evaluating tiny-object grounding under class imbalance, and report latency and reliability characteristics of the deployed pipeline.

Deploying vision-language models (VLMs) for safety-critical spatial reasoning on resource-constrained autonomous driving platforms requires both compact model size and reliable metric grounding. We present MoRAL (Multimodal Reasoning for Autonomous Language Models), a two-stage fine-tuning pipeline that teaches Cosmos-Reason2-2B to first read a physics-encoded Bird's Eye View (BEV) representation and then reason over it for driving decisions. The BEV image encodes LiDAR metric distance as color bands, object class as cluster morphology, and radar Doppler velocity as directional wedge overlays, externalizing spatial perception into the input image so that no learned 3D backbone is required at inference. Stage 1 fine-tunes the vision encoder on 60,000 grounding records; zero-shot baselines produce no parseable BEV outputs, confirming the vocabulary requires explicit training. Stage 2 fine-tunes the full model (52M parameters, 2.4% of total) on 57,696 chain-of-thought records generated by Cosmos-Reason2-8B as teacher, spanning eight driving question types. On 2,304 held-out nuScenes frames evaluated by Gemma 4 (31B) calibrated against human review, MoRAL wins seven of eight question types over a zero-shot 8B baseline despite using four times fewer parameters, with the largest margins on question types requiring structured multi-step physics reasoning. Emergency braking recall improves from 10.8% to 47.8%, output degeneration falls from 94.1% to 20.8%, and the full pipeline fits a consumer 8 GB GPU at 42 tok/s without quantization. These results establish a reproducible foundation for compact, physics-grounded VLM reasoning on mobile edge platforms.

Forecasting future states from video sequences is a critical challenge for autonomous robotic systems and a fundamental objective of world modeling. Prior generative methods operating at the pixel level inevitably overemphasize task-irrelevant details, leading to prohibitive computational overhead. While latent-based approaches attempt to mitigate this by predicting features directly, the persistent reliance on heavy decoders for state-to-task mapping remains a computational bottleneck. In this work, we propose Decoder-Free Feature Forecasting (DF$^3$), a novel framework that models world evolution entirely within the latent space and directly derives task outputs, completely eliminating the need for a decoder. Specifically, DF$^3$ injects learnable spatial queries into the terminal blocks of a frozen vision foundation model to extract future state representations directly. By employing a lightweight, unified Motion-Aware Context Fusion (MACF) mechanism that seamlessly integrates coarse flow warping with fine-grained latent cross-correlation, these queries interact with historical token representations to explicitly align and forecast the feature of the next frame. Subsequently, a specialized set of task queries probes these forecasted features for the downstream task. Extensive experiments on public benchmarks and zero-shot deployment in a robotic simulator demonstrate that DF$^3$ achieves performance comparable to state-of-the-art methods while offering superior efficiency and flexibility for integrated perception and control.

Existing astronomy foundation models provide strong galaxy representations, but adapting them to new survey conditions and survey-specific morphology recognition tasks still requires substantial human supervision. We show that VLM-based VQA systems contain meaningful visual-semantic priors that can serve as weak supervision for downstream morphology classifiers and improve morphology classification under limited human-label budgets. We first introduce a survey-oriented VQA benchmark spanning two representative imaging regimes and evaluate state-of-the-art VLMs on galaxy morphology questions. The results show that these models capture useful morphology signals and informative uncertainty, but are not sufficiently reliable to replace human annotators. Motivated by this finding, we use a general-purpose VLM as a morphology teacher for Zoobot, an astronomy foundation model pretrained on large-scale Galaxy Zoo annotations. Across two survey domains and multiple annotation budgets, the VLM teacher consistently improves Zoobot's downstream morphology classification. These results demonstrate that a general-purpose VLM provides knowledge complementary to an astronomy foundation model and can teach it to better recognize galaxy morphology under limited human supervision. The resulting pipeline is designed for label-efficient adaptation to forthcoming large-scale surveys, including the Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST) and the Nancy Grace Roman Space Telescope. The benchmark and code are publicly available at https://github.com/fw-ic/VLM-morphology-teacher.

Visual representations of VLA models remain unreliable for spatially precise robotic manipulation. We uncover that vision encoders in VLAs also exhibit attention artifacts previously documented in generic Vision Transformers, and further show that, in embodied policies, these artifacts are closely associated with spatial perception capabilities acquired during post-training. As the encoder learns task-relevant information such as object location, depth ordering, and local geometry, limited global-token capacity causes part of this information to spill into low-information patch tokens. We introduce AtVLA, a framework that inserts learnable register tokens into the visual encoder. Trained end-to-end using only embodied data and the original action objective, these registers emerge as dedicated carriers of embodied spatial information, while the remaining patch tokens recover clean and spatially faithful attention distributions crucial for precise target localization and fine-grained contact. Clean attention restores reliable localization, but cannot recover geometric details lost in low-resolution observations. AtVLA therefore couples attention rectification with uncertainty-gated local refinement. The action expert samples multiple action chunks and estimates uncertainty from their disagreement; only for uncertain predictions, action-conditioned attention rollout identifies the task-relevant region, which is cropped, re-encoded at high resolution, and appended to the cached prefix for refined action generation. Across LIBERO, SimplerEnv, and a challenging single-view real-world benchmark, AtVLA improves the average LIBERO success rate from 94.2% to 98.4% and real-world success from 46.5% to 69.0%. The cropping is triggered on approximately 30% of replanning steps, resulting in only 1.4-1.6x the total computation of the base model under the representative deployment setting.

Vision-language models (VLMs) exhibit strong generalization across multimodal tasks but remain vulnerable to adversarial perturbations. Existing attacks typically follow single-trajectory gradient optimization or task-specific objectives, limiting search-space exploration and cross-task transferability. We propose an evolutionary-computation-guided cross-modal attack framework for unified VLMs. The framework adaptively searches both textual and visual spaces. On the textual side, it evolves hard negative semantic embeddings around the source-category representation to provide diverse cross-modal repulsion. On the visual side, it maintains a population of object-region perturbations and combines momentum-based gradient updates with evolutionary selection, mutation, and crossover to more reliably explore multiple feasible trajectories. Jointly optimizing semantic negative guidance and localized perturbations generates adversarial examples that consistently shift source-object semantics toward target categories across vision-language tasks. Theoretical analyses show that the co-evolutionary search preserves perturbation feasibility, prevents degradation of the best observed fitness, and increases the probability of reaching high-margin adversarial regions compared with single-trajectory optimization. Experiments on Florence-2, OFA, and UnifiedIO-2 demonstrate strong overall attack performance across image captioning, object detection, region categorization, and object localization. Ablation studies further verify the complementary effectiveness of text-side semantic evolution and image-side perturbation evolution, as well as the framework's efficiency and cross-task transferability.

Vision-language models (VLMs) remain unreliable when predictions require fine-grained visual evidence. We identify a previously overlooked cause: spectral response rigidity. Despite substantial frequency variation across images and tasks, pretrained vision encoders exhibit persistent, encoder-specific layerwise spectral profiles that change only marginally under downstream fine-tuning. Since pretrained vision encoders only receive images, they cannot adapt spectral extraction to the evidence required by the current query. We therefore propose HAFI-VLM, which introduces a task-conditioned frequency pathway while preserving the pretrained semantic representation. Hierarchical Adaptive Frequency Injection (HAFI) retrieves complementary low-, mid-, and high-frequency evidence at multiple encoder depths using text-modulated, spatially aligned cross-attention. A Visual Enrichment Layer Adapter further recalibrates shallow LLM attention to effectively utilize the enriched visual tokens. Experiments on LLaVA-1.5 and Qwen2.5-VL demonstrate consistent improvements in general VQA, text-rich understanding, and hallucination robustness, outperforming representation-level enhancement methods and most resolution- or cropping-based approaches without additional high-resolution encoding. Mechanistic analyses show that HAFI restores task-dependent spectral allocation while retaining semantic attention, establishing frequency enrichment as a distinct and effective route for improving VLM perception.

Remote-sensing videos enable real-time observation of changes in target attributes, short-term activities, and scene evolution. They record motion, actions, interactions, and scene changes that cannot be captured by isolated images. Existing models primarily target single images or discrete temporal observations spanning a long time range. However, a unified evaluation setting for assessing vision-language models on continuous remote-sensing video understanding remains lacking. We introduce RSVideo-10K, a remote-sensing video dataset comprising 10,773 instances, 1.47 million frames, and 17.02 hours of footage, containing both unmanned aerial vehicles and satellite platforms. Its fixed evaluation benchmark, RSVideo-Bench, contains 2,731 test instances and evaluates two complementary aspects of remote-sensing video understanding: L1 Perception and L2 Reasoning, spanning seven capability groups and 17 tasks. Evaluations show that current vision-language models still struggle to recover small local evidence, track short-lived states, and use scene-constrained spatial relations. Based on this analysis, we further propose RSVideo, a reinforcement learning framework for small-target spatiotemporal focusing that selects question-relevant regions across frames and suppresses redundant background tokens. RSVideo achieves a maximum absolute improvement of 9.01% with InternVL3.5-14B and attains the highest accuracy of 40.63% with Qwen3.6-27B across 26 open-source vision-language backbones. Codes will be available at https://github.com/HongjieZhou0329/RSVideo.

The deployment of Vision-Language Models (VLMs) in autonomous driving (AD) systems is constrained by on-board computing power, restricting vehicles to small VLMs (SVLMs) with limited perception and reasoning capabilities. Infrastructure-assisted AD alleviates this resource constraint by enabling collaboration with large VLMs (LVLMs) at edge servers. However, in dynamic vehicular environments, the utility of sensory data for downstream tasks decays rapidly, making timeliness of information a critical concern. To balance the accuracy gains of LVLMs with their latency-induced timeliness degradation, we develop a Timeliness-Aware Large-Small VLM Collaboration (TALSC) framework. Specifically, we first model the Age of Information (AoI) evolution for VLM inference and characterize the coupling among AoI, token length, and task performance to formulate a general timeliness metric. Building on this, we propose the TALSC online scheduling algorithm. Since scheduling decisions have a delayed impact on future timeliness metric and the output token number is unknown at scheduling time, we design a Lyapunov drift-plus-estimated-penalty algorithm and provides a guaranteed performance. In simulation, we first conduct a case study to derive a fitted timeliness metric based on nuScenes dataset, and further show that TALSC outperforms baselines under various communication and computing settings, achieving up to a 12.6% normalized improvement in Micro-F1 score compared with the best-performing baseline.

Rapid and accurate post-disaster building damage assessment is essential, yet remains a challenging task. Unmanned Aerial Vehicle (UAV) imagery offers a timely and high-resolution view of affected areas, but existing Computer Vision (CV) models often demand large annotated datasets, generalize poorly across geographic regions and their assessment policies, and are confined to the specific tasks they were trained for. Large Vision-Language Models (LVLMs) offer a promising alternative through their strong reasoning and generalization capabilities, but fall short on precise, low-level perception tasks such as object detection and accurate bounding box generation. Furthermore, they often require a substantial amount of data for effective fine-tuning on domain-specific tasks. In this paper, we propose a hybrid framework that decouples detection from damage assessment, combining the precision of CV models with the reasoning power of LVLMs. A CV model first detects buildings and generates bounding boxes on the image that are then passed to an LVLM for damage classification and contextual interpretation. We evaluated our framework on two real-world benchmarks: RescueNet and FloodNet. In particular, the best combination under this framework accurately counts intact, partially damaged and completely destroyed buildings, surpassing isolated baselines by up to 2.1 R^2 points, while requiring only limited annotated data for the detection stage. Beyond reporting aggregate gains, we provide a detailed analysis of failure scenarios and edge cases, offering practical insights for practitioners and concrete directions for future work. Our source code and data are publicly available to the research community via the following repository: https://github.com/ungquanghuy-kddi/VLM_GDINO.git

Vision-Language Models (VLMs) perform well on commonsense reasoning tasks but struggle with visual spatial reasoning. Most existing solutions introduce extra 3D prior inputs or external spatial encoders, which increase complexity and degrade the underlying VLMs' general-purpose capabilities after spatial fine-tuning. To this end, we propose a parameter-efficient \textit{\textbf{Spatio}-vision \textbf{L}anguage \textbf{M}odels (SpatioLM)}, that enhances spatial intelligence without extra 3D prior inputs or third-party spatial encoders. Concretely, we design a plug-and-play and non-invasive spatio-vision module that elicits the spatial knowledge inherent in VLMs. Furthermore, we innovatively leverage pseudo depth and camera information as supervision to guide the model in learning physically coherent representations. Extensive experiments show that SpatioLM achieves significant improvements in diverse tasks, including spatial perception and understanding while effectively limiting the degradation of general capabilities. Notably, the model achieves an impressive score of 71.6 on the VSI-Bench (the first model to surpass 70). In addition, it attains competitive performance when transferred to embodied manipulation tasks. Code is available at \href{https://github.com/xiaomi-research/spatio-lm}{\faGithub~spatio-lm}.

Addressing urban blight has seen increased focus in the past 15 years. Assessing urban blight is essential for guiding urban planning, targeting rehabilitation, and safeguarding public health, yet traditional residential blight surveys are difficult to maintain at scale due to the labor-intensive cost and long-term cycle. This study introduced a scalable framework for estimating residential blight using open-source large vision-language models on multiple views. Structured prompts guided models to evaluate housing attributes, including roof integrity, wall damage, and broken or boarded openings, producing both binary assessments and probabilistic estimates of disrepair. To evaluate the performance of these visual assessments, we compared professional human annotations of these features across several models, including an ensemble stacking approach based on XGBoost and a weighted scoring system. Results showed that (i) multiple street views can contribute to the improvement of accuracy, (ii) large vision-language models have different strengths of inference, (iii) the ensemble learner outperforms individual base models, enhancing robustness across all residential conditions and blight assessment. The practical application of the method allows low-cost tracking and management of housing stock conditions, providing a regularly updatable complement to traditional blight surveys.

We present our ImageCLEF 2026 Multimodal Reasoning system for the Visual Multiple Choice Question Answering (Visual MCQ) and Visual Open Question Answering (Visual OpenQA) subtasks. The challenge requires reliable reasoning over multilingual educational and scientific images with dense text, diagrams, charts, tables, formulas, and units, while enforcing strict answer formats. Our central finding is that robust output control is as important as model choice. For Visual MCQ, we replace fragile free-form generation with direct candidate label scoring from vision-language model logits, then combine complementary runs through score fusion and voting. For Visual OpenQA, we use image enhancement, concise final answer prompting, deterministic decoding, and targeted post-processing to remove reasoning traces and formatting artifacts. Without task-specific model training, our official submissions achieved third place in Visual MCQ with 0.7108 accuracy and first place in Visual OpenQA with 0.6488 COMET, 0.1391 BLEU, 0.2762 ROUGE L, and 0.2383 METEOR. The results highlight the practical value of inference engineering: careful scoring, ensembling, prompting, and cleanup can turn strong VLMs into reliable competition systems.

The challenge of fair deepfake detection (FDD) has attracted increasing attention. Existing fairness-enhanced detectors often suffer from suboptimal generalization to unseen manipulations and fairness across demographic groups. They are typically developed and evaluated on demographically imbalanced distributions, resulting in biased predictions toward minority groups. In this paper, we construct a novel demographically balanced FDD benchmark to train and evaluate the fairness of detectors under both balanced and imbalanced population scenarios. Additionally, we introduce a novel expression and demographic perceptual vision-language model, termed FairForensics, for generalizable fair deepfake detection. FairForensics conducts face forgery generalization enhancement and demographic-aware fairness regularization. During face forgery generalization enhancement, built upon the novel observation of significant distribution differences between pristine and forged expression vectors, we design an expression encoder to capture high-level expression-guided forgery patterns, and an expression-perceptual visual encoder that integrates global appearance and expression forgery features while mitigating identity bias using an identity-aware patch perturbation module. Under demographic-aware fairness regularization, we propose a demographic-guided language encoder to extract population-aware global language embeddings, which boosts the decoupling of forgery features from demographic information via vision-language alignment. We devise a population-aware prototype fairness objective to enforce both inter-class separability and intra-class alignment across demographic subgroups. Extensive experiments on our balanced demographic benchmark show that our method achieves the state-of-the-art in terms of generalization and fairness.

Vision-Language Models (VLMs) face a critical computational bottleneck when processing high-resolution imagery due to the $O(N^2)$ memory complexity of Softmax Multi-Head Attention (MHA). While substituting MHA with independent Multi-Layer Perceptrons (MLPs) achieves $O(N)$ scaling, it strips the architecture of spatial sequence routing, severely degrading global scene understanding and object permanence. In this paper, we propose the Linear Multi-Timescale Retention (LIA-MTR) module, a memory-efficient cross-modal bridge. By integrating an ELU-based positive feature mapping with adaptive write-gating and log-linearly distributed recurrent decays, LIA-MTR mathematically compresses continuous visual sequences into bounded memory states. Theoretical analysis proves the architecture operates with strict $O(N)$ sequence-interaction complexity. Empirically, synthetic retrieval evaluations demonstrate that LIA-MTR flawlessly routes context across 16,000 tokens, eliminating the "Lost in the Middle" degradation typical of naive linear attention. Hardware benchmarking reveals infinite-context scaling capabilities, natively processing 262,144 visual patches within an 11.2 GB VRAM footprint, whereas standard MHA suffers out-of-memory failure at 16,384 patches. Furthermore, following instruction tuning on 665K conversational samples, LIA-MTR significantly outperforms an industry-standard MLP baseline on the MME benchmark (71.00% vs. 68.11%), driven by a 10% absolute improvement in object permanence and superior global semantic extraction. This work establishes a mathematically rigorous, computationally flat foundation for infinite-context Vision-Language integration.

Vision-language models (VLMs) are emerging as a key component of embodied intelligence, with growing applications in auto-labeling and end-to-end autonomous driving. However, existing approaches for improving spatiotemporal reasoning in VLMs often rely on complex preprocessing pipelines, expensive human annotations, or synthetic data, which limit scalability and introduce potential sim-to-real gaps. Moreover, although these methods have improved spatiotemporal understanding, they still lack strong metric reasoning capabilities for dynamic scenes, such as estimating object motion in real-world units. Prior work has explored LiDAR-based metric depth supervision to enhance spatial perception, but it does not directly address temporal reasoning. We introduce STAR-VLM, an automotive radar-supervised framework that enhances spatiotemporal VLMs with motion reasoning and metric velocity estimation for autonomous driving. Automotive radar is a low-cost and widely deployed sensor that provides complementary spatiotemporal supervision through range and Doppler measurements. By leveraging these measurements as label-free ground truth during training, STAR-VLM improves the metric spatiotemporal reasoning ability of VLMs. Through experiments on driving scenarios, we show that STAR-VLM achieves state-of-the-art performance on both motion classification and metric velocity estimation, outperforming even task-specific methods designed for each task. These results highlight automotive radar as a scalable and cost-effective source of supervision for building metric-aware spatiotemporal VLMs for real-world autonomous driving.

Hard symbolic-reasoning tasks such as Sudoku, maze pathfinding, and ARC remain challenging for LLMs due to their fixed-depth autoregressive reasoning, which limits systematic search, refinement, and backtracking. While recursive models such as Hierarchical Reasoning Model (HRM) and Tiny Recursive Model (TRM) address this limitation through iterative latent-state refinement, they are typically task-specific and do not leverage pretrained language priors. We propose R-Qwen, a recursive reasoning framework built upon a pretrained Qwen backbone. R-Qwen repeatedly refines a candidate solution through programmatic self-recursion and deep supervision, combining the structured iterative computation of recursive models with the linguistic and reasoning priors of pretrained LLMs. We further adapt Hierarchical Supervision Weighting (HSW) to autoregressive models by exponentially weighting losses across recursive steps. HSW reduces gradient variance by at least 50%, improves the signal-to-noise ratio of stochastic gradients, and accelerates convergence. Across eight challenging benchmarks, R-Qwen consistently outperforms prior recursive reasoning models and substantially larger LLMs while using a comparable number of trainable parameters. Notably, on ARC-AGI dataset, our model achieves a 27.6% improvement over the baseline, highlighting the effectiveness of recursive refinement for general symbolic reasoning. These results suggest that recursive reasoning mechanisms and pretrained language model priors are complementary approaches for improving symbolic puzzle-solving. Code and models will be released after acceptance.

Vision Language Models (VLMs) have demonstrated exceptional performance across various tasks. However, they have not yet been thoroughly evaluated on more complex tasks. The Persuasion Model, conceived by Aristotle, resembles a triangle shape, which highlights its inherent challenges related to personal biases. To assess the progress of VLMs on these complex tasks, we use the ImageArg datasets, focusing on the Logos, Ethos, and Pathos detection tasks. Our findings indicate that models from the Qwen family achieve improved F1 scores, with Qwen3 performing exceptionally well on the Logos and Pathos tasks, while Qwen2 exhibits competitive performance on the more complex Ethos detection task. We release the code to foster research in this direction.

Referring expression segmentation requires language conditioned localization and pixel-accurate masks, but monolithic models can be costly to deploy. We present VespaSeg, a modular pipeline that grounds a text query with a compact vision-language model and converts the predicted box to a mask with MobileSAM. We study Florence-2-base, Florence-2-large, and Moondream2 grounders together with targeted adaptation of the grounding and segmentation stages. Under a repository-specific RefCOCO validation protocol containing the first expression for each of 3,811 referenced-object records, the adapted Florence-2-base pipeline obtains 73.64 mean intersection over union (mIoU) and 84.60 precision at IoU 0.5. On an NVIDIA RTX 6000 Ada GPU it processes 22.8 cached-image queries per second with 2.20 GB mean allocated GPU memory. A matched 500-query comparison gives 73.73 mIoU for Florence-2-base and 72.82 for Florence-2-large, while the base model is 1.70 times faster and uses 1.17 GB less allocated memory. Ablations show that ground-truth-box adaptation raises MobileSAM mIoU from 82.22 to 86.61 and that reducing the Florence-2 output-token budget from 64 to 32 preserves accuracy. These results support compact, modular grounding and segmentation, while also exposing the need for evaluation on the complete standard RefCOCO expression splits and deployment hardware.

Fine-grained visual representations are essential for medical image analysis, particularly when diagnostically relevant evidence is subtle and spatially localized. Modern transformer-based medical vision encoders must therefore learn patch-level representations that are both clinically meaningful and spatially consistent. Without these properties, large vision-language models (LVLMs) operate on an ambiguous visual foundation, limiting their ability to generate clinically reliable and spatially grounded responses. However, existing training strategies for medical vision encoders rarely achieve both objectives. Image-text alignment provides clinically meaningful supervision primarily at the image level, leaving the spatial localization of diagnostic evidence weakly constrained. In contrast, self-supervised learning promotes spatial consistency but lacks the semantic supervision needed to distinguish visually similar yet clinically distinct regions. To address this gap, we present LoFi, a medical vision foundation model built on location-aware fine-grained representation learning. LoFi trains a vision encoder with a lightweight large language model under grounding and grounded captioning objectives. Because these objectives require predicting location from clinical text and vice versa, spatial consistency emerges without any explicit patch-level regularization. To enable training at scale, we construct MedG, a large-scale medical grounding dataset of 4.48M image-text-box triplets curated from 84 datasets spanning 7 modalities. Across phrase grounding, visual question answering, and region-based organ classification under perturbations, LoFi consistently outperforms general-purpose and medical vision foundation models as well as state-of-the-art LVLMs. Code is available at https://github.com/myeongkyunkang/lofi-medg.

Large vision--language models (VLMs) can reason step by step about complex visual scenes, but this open-ended, autoregressive chain-of-thought (CoT) approach is poorly suited to safety-critical, rule-governed settings such as industrial surveillance, where decisions must be bounded, deterministic, and low-latency. Because CoT inference cost scales jointly with reasoning length and the number of concurrent streams, it creates a throughput bottleneck that precludes the real-time, multistream monitoring required for industrial accountability. Here we present MonitorVLM-v2, a deployment-oriented framework that recasts VLM-based safety assessment as probabilistic inference over a finite regulatory decision space, compressing multimodal reasoning into single-step rule-ID predictions and reducing decoding from a variable-length sequence to a single token. We introduce symbolic policy optimization (SymPO), a novel contrastive policy optimization algorithm that sharpens decision boundaries within this finite symbolic space, together with an entropy-driven triage mechanism that routes uncertain predictions to human reviewers for expert confirmation. In a four-month prospective deployment across 10 concurrent camera feeds in an operational underground mining facility, MonitorVLM-v2 achieved a 19.45-fold increase in inference speed and identified 2.78 times as many confirmed violations as the site's routine manual inspection workflow, demonstrating the practical value of compressed symbolic decision-making for real-time, auditable industrial monitoring.

Many open educational resources are lacking in accessibility, especially in-depth image descriptions. In subjects like Science and Mathematics, however, it can be particularly difficult to write image descriptions since there can be many complicated expressions and names depending upon the course level. To help fill that gap in a small way, we introduce Math Image Descriptions for Accessible Learning (MIDAL), a math image-description dataset of 2,020 mathematical images spanning multiple educational levels, to aid in training vision language models to create image descriptions following accessibility best practices. We hope MIDAL is a valuable resource in enhancing the conversation and innovation regarding accessibility of STEM content in higher education. This dataset is however not just limited in math description generation but can also be used to fine-tune language models that can have improved mathematical reasoning and answers.

Multimodal large language models have advanced visual understanding, yet perception-intensive reasoning remains challenging. Recent latent visual reasoning methods introduce hidden-space computation before answering, but they often rely on costly intermediate supervision, such as bounding boxes, sketches, or interleaved rationales. These strategies focus on how latent states should be shaped, but do not explicitly assess whether the latent is useful for the final answer. We propose LUT, a latent reasoning framework trained with only standard VQA pairs. LUT centers training on Latent Utility at two levels. At the trajectory level, we propose Utility-Aware Latent Distillation SFT, which explores answer-relevant latent trajectories, selects qualified trajectories by their information gain, and distills more reliable and learnable supervision through curriculum learning. At the step level, we propose Latent Attribution Policy Optimization, which uses answer-to-latent attribution to differentially optimize latent steps during reinforcement learning. Experiments on perception-intensive visual reasoning benchmarks show that LUT outperforms previous latent reasoning methods and remains competitive with latent-text interleaved methods with lower annotation cost.

Vision-language models return structured chest-radiograph findings through interfaces exposing no confidence score, so a receiving institution cannot read off how far to trust an individual judgment. Whether agreement with an institution's reference standard transfers across sites, findings, prediction directions and question formats is largely unmeasured. We evaluated three generative vision-language models on three institutional chest-radiograph corpora and six findings under two elicitation protocols, comprising more than 345,000 finding-level predictions, and estimated finding-by-direction reference agreement at a receiving institution from a small budget of local labels. Estimation strategies were then stress-tested under repeated strict institution-held-out evaluation. Under evaluation excluding the receiving institution from development entirely, adaptive selection among the seven estimators that design admits did not improve on simple fixed alternatives: it achieved a mean Brier score of 0.1083, against 0.0853 for always using a Beta-Binomial empirical-Bayes estimator and 0.0855 for a target-only logistic model. Those two differ by 0.0003, less than this family's own sensitivity to a change of solver version, and each leads in about half the settings, so no default can be recommended. Their advantage over estimators pooling across institutions was concentrated at one site and not confirmatory once clustered by institution, and a plug-in empirical-Bayes posterior-predictive count interval at a nominal 95% level covered 87.0%, less at the hardest institution. Reference agreement therefore has to be re-evaluated per site and per interface; these results concern agreement with institutional labels, not clinical correctness.

Frozen vision foundation models are commonly evaluated through a single global image embedding, but this interface can conflate missing information with information lost at readout time. We study this distinction by keeping a pretrained vision encoder frozen and varying only the readout applied to its final patch tokens. We compare standard global readouts against a lightweight foveated readout, which attention-pools patch tokens using a learned or question-conditioned query, and against an oracle readout with access to the annotated target region. We evaluate these interfaces on three localized binding problems: a controlled synthetic color--shape binding task under clutter, a color-free crowded shape-detection variant, and a GQA-derived natural-image task where paired questions ask for the colors of different same-category objects in the same image. Global readouts perform near perfectly when the synthetic target appears alone, but collapse under clutter and counterfactual target edits, whereas the foveated readout recovers most of the oracle-accessible signal. On the GQA-derived task, question-independent global image vectors improve only modestly over question-only priors, while question-conditioned foveation substantially improves paired localized color accuracy. A counterfactual nuisance-to-signal ratio explains the synthetic failures: global pooling dilutes localized label-changing evidence while exposing the probe to nuisance variation from irrelevant objects. These results indicate that apparent spatial blindness in frozen vision models can arise from the global embedding interface rather than from an absence of spatial information in the frozen patch tokens.

Recent advances in large vision-language models (LVLMs) have enabled long-video understanding and analysis. However, processing the large number of frames in a video incurs substantial computational overhead. Existing methods reduce LVLM inference costs by scoring frame-query relevance before inference and selecting keyframes accordingly. Nevertheless, the distribution of relevant frames varies across queries, and these methods often need to score hundreds or thousands of frames. To address this limitation, we propose CSES, a training-free semantic keyframe selector that adaptively determines the numbers of frames to score and keyframes to select. CSES estimates the prominence of the frame-query relevance profile to guide active acquisition and adapt the temporal coverage of each input. It then formulates keyframe selection as a coverage problem that jointly accounts for semantic relevance, temporal redundancy, and visual redundancy. Active acquisition and keyframe selection terminate based on coverage saturation. The selection objective is monotone and submodular, enabling greedy optimization with a standard approximation guarantee. Experiments with four LVLMs on two benchmarks show that our method preserves accuracy while scoring $4$-$13\times$ fewer frames and selecting $18.4\%$-$20.5\%$ fewer input keyframes than existing baselines. CSES further achieves a $3.1$-$5.4\times$ speedup in frame selection over baselines.

Affordance grounding aims to localize the functional region for interaction, such as the handle to grasp or the button to press, rather than the whole object. This makes it more challenging than generic visual grounding because the target region is smaller, more ambiguous, and more dependent on task context, especially for compact vision-language models (VLMs) used in embodied settings. Recent sequence-level supervision and reinforcement learning improve coordinate prediction quality, yet compact autoregressive VLMs still lack reliable affordance-aware visual focus before coordinate generation: the model can produce better coordinate tokens while its cross-modal attention remains diffuse and weakly anchored to the true affordance evidence. To address it, we propose SpatialAfford, a two-stage framework that first aligns attention to the ground-truth affordance region through Spatial Attention Alignment (SAA), then refines coordinate prediction with Spatial-Aware GRPO. By explicitly teaching the model where to look before optimizing where to ground, SpatialAfford turns affordance grounding from a purely output-constrained objective into attention-grounded spatial reasoning. Across ShareRobot-Bench, ReasonAff, and PartAfford, SpatialAfford consistently improves affordance grounding, with a compact 4B model outperforming stronger 7B+ baselines.

Do frontier vision-language models present a coherent Theory-of-Mind (ToM) profile across tasks, matching the same human reference group, or does that profile fragment from one paradigm to the next? We evaluate a shared panel of nine frontier VLMs on two psychology-derived benchmarks: the Keysar Director Task (visual perspective-taking under egocentric interference) and the Frith-Happé animated triangles scored with the Castelli rubric (intention attribution from pure motion). On the Director Task, without chain-of-thought, the panel makes the egocentric error on 78% of trials like children rather than adults; variation is substantial across models, and reasoning rescues several models. On the triangles, the panel under-attributes intention: its ToM profile sits more than three times closer to the high-functioning-autistic-adult (HF-ASD) mean than to the typical-development-adult (TD) mean, while Goal-Directed and Random stay near TD. No model is nearest TD on both tasks; the model that looks adult-like on the Director Task falls on the HF-ASD side on the triangles, and the most TD-like model on the triangles is child-like on the Director Task. We report group-level descriptions, not diagnostic labels for any model.

Reinforcement learning (RL) post-training of Vision-Language-Action (VLA) models has shown strong promise for robotic manipulation. Among RL methods, critic-based approaches rely on a value estimator that predominantly operates on single-frame observations or single-frame VLM backbone latents, which is a fundamental mismatch with the partially observable nature of robot control. A naive approach to incorporate observation history into the critic incurs exponential complexity with high-dimensional visual space, and still fails because pure scalar-return regression provides insufficient supervision for learning cross-temporal dynamics. We identify the root cause as a state approximation problem: without an explicit world modeling objective, the critic's representation cannot capture the temporal structure needed for accurate value estimation. To address this, we propose the World Critic Model (WCM), built on a lightweight LeJEPA architecture; WCM jointly predicts future latent state and estimates values, such that the critic's representation is explicitly trained to capture temporal dynamics rather than merely regress scalar returns. WCM integrates seamlessly into both on-policy and off-policy training pipelines and is compatible with state-of-the-art VLA backbones including Pi0, Pi0.5, and OpenVLA-OFT. Extensive experiments on 149 tasks across four benchmarks demonstrate that WCM consistently achieves state-of-the-art performance in both in-distribution and out-of-distribution settings, with particularly strong generalization gains. We further validate WCM on seven real-world manipulation tasks using OpenVLA-OFT and Pi0.5 with off-policy RL, confirming stable deployment across diverse settings.

Vision-language-action models (VLAs), which leverage the cognition of multimodal information to infer physical-world actions, provide a generalized solution for embodied AI applications. Conventional VLAs usually concentrate on current digital cognition. While some efforts are made to enhance VLAs' reasoning capabilities by capturing temporal information, encoding the long-context history causes an efficiency-decreasing issue. To reconcile the conflict between capturing temporal information and maintaining inference efficiency in VLAs, this paper introduces FibVLA, an efficient framework featuring temporal perception of long-context history. Specifically, we leverage logarithmic hindsight sampling to both proprioceptive states and visual frames to capture long-term temporal dependencies with minimal redundancy. For the action expert, we introduce the flow matching to produce action distributions, and the Fibonacci recurrent inference strategy to generate long-range planning steps based on real-time closed-loop feedback. Experiments demonstrate that FibVLA significantly improves action smoothness and success rates without retraining large-scale visual encoders. Efficiency analysis demonstrates superior real-time responsiveness compared to video-based baselines in real-world evaluations.

To maintain common ground in cooperative conversation, humans iteratively update their beliefs as conversation participants share new information; participants who are epistemically vigilant detect when new information conflicts with prior beliefs and take steps to repair these conflicts. In order for AI systems to serve as reliable partners in complex cooperative tasks, they must similarly weigh incoming information against their own private evidence and shared context and appropriately surface inconsistencies when they arise. To measure the epistemic vigilance of vision-language models in cooperative settings, we present an information-asymmetric, dialog-based "spot-the-difference" task. Two models are privately shown one image each, and must determine through conversation whether the images are identical or, if not, identify the difference. Models routinely fail at this: they frequently overlook key evidence in their private image in favor of agreeing with their conversational partner, even when their agreement is unwarranted. We relate these violations of epistemic vigilance to the broader behavior of sycophancy, which manifests itself in cooperative goal-oriented dialog as over-accommodation and weak evidential grounding. Our results show that model steering to reduce sycophancy with a vector learned from task-agnostic sycophancy examples can reduce epistemic vigilance-related errors, making models more faithful reporters of their evidence, and in turn, more reliable partners in information-asymmetric cooperative tasks.

Multimodal video generation aims to generate and edit videos conditioned on arbitrary combinations of text, images, and videos within a single model, allowing diverse tasks to share complementary data and generative priors. Unifying these tasks requires multimodal understanding of diverse conditions, which is typically provided by a pretrained vision-language model (VLM). A key challenge is how to connect the VLM's hierarchical multimodal representations with a pretrained video diffusion transformer (DiT). Existing methods either inject features from only the final or a few manually selected VLM layers, or jointly train architecture-matched understanding and generation streams, making it difficult to reuse heterogeneous pretrained backbones. We introduce MoRoute, a unified multimodal video generation framework that formulates a frozen VLM and a pretrained video DiT with different architectures as heterogeneous experts connected through dynamic layer routing. For each input, a lightweight block-wise router enables every DiT block to select the VLM layer most relevant to its generation stage, thereby learning an adaptive correspondence between multimodal understanding and video synthesis. MoRoute further incorporates reference images and source videos directly into the DiT token sequence through unified in-context conditioning, preserving fine-grained visual details across diverse generation and editing tasks. Experiments on IntelligentVBench, OpenVE-Bench, and RefVIE-Bench show that MoRoute consistently surpasses the best competing method on each benchmark, improving the average score by 0.15, 0.18, and 0.34 on a 1-5 scale, respectively.

Infrared vision-language models (IR-VLMs) extend thermal perception to open-vocabulary classification, image captioning, and visual question answering. However, their robustness to structured thermal perturbations and the stability of cross-modal semantic alignment remain insufficiently studied. We propose QR-Structured Thermal Triggers (QR-STT), a stealthy, training-free, black-box framework for targeted semantic steering of IR-VLMs. QR-STT preserves the functional regions of a QR pattern while optimizing its internal modules, each of which is assigned a cold, neutral, or hot thermal state. The framework jointly searches module topology and rendering parameters, including position, scale, rotation, intensity, blur, and roundness. A three-stage gradient-free procedure with greedy module-flip refinement efficiently handles the mixed discrete and continuous search space. The objective promotes alignment with an attacker-selected target, suppresses source-class evidence, and regularizes QR structure and visual similarity. Experiments on multiple CLIP-style encoders show that QR-STT consistently redirects image-text alignment toward chosen concepts while maintaining visual stealth. Perturbations optimized for classification also transfer to image captioning and VQA, causing target-consistent semantic drift in generated outputs. These results identify QR-structured thermal patterns as an interpretable attack surface for language-driven infrared perception and highlight the need for robustness evaluation against structured cross-task semantic attacks.

Geometry-based grasp planners ensure physically valid grasps but ignore functional semantics, often generating grasps that are antipodal and collision-free yet practically inappropriate, for example, gripping a mug by its rim, a knife by the blade, or a bottle near its cap. These inconsistencies cause the downstream task to fail even when traditional grasp metrics are met. Existing vision-language model (VLM) approaches either depend on fine-grained, category-specific part segmentation or attempt to directly infer grasp poses, with the latter prone to spatial hallucinations. As a result, no practical, training-free framework has yet been proposed that robustly links high-level semantic reasoning to geometric grasp planning. We introduce Semantic Affordance-Guided Grasp Planning (SAGP), a training-free pipeline built on a coarse-zone abstraction layer. The method first partitions the object point cloud into spatial regions (top, middle, bottom, lateral sides, and protrusions) by applying PCA-based alignment followed by distance-driven DBSCAN clustering, entirely bypassing learned segmentation. A pre-trained VLM then assesses the grasp quality of each region through a structured zero-shot query, and the resulting zone-wise scores are fused with geometric, reachability, and task-alignment signals to re-rank antipodal grasp candidates. Experiments on YCB objects in PyBullet with a Franka Panda robot show that SAGP preserves the high success rate of geometry-only planning while substantially improving the functional appropriateness of selected grasps, particularly on asymmetric, handle-bearing objects where geometry alone is uninformative. The introduced coarse-zone abstraction offers an effective, training-free bridge between VLM-based reasoning and geometric grasp planning, without the need for fine-grained part segmentation.