VLM Evaluation

Large-scale construction of medical vision-language model (VLM) benchmarks is increasingly feasible with richly annotated imaging datasets and large language models (LLMs), yet existing automation largely focuses on generating evaluation items within predefined benchmark specifications. We study the broader problem of automatically deriving the specification itself: what to evaluate, which annotations support each task, and how to translate this evidence into reliable evaluation items. We formulate benchmark construction as constrained compilation, in which the benchmark specification is progressively derived from evaluation requirements, heterogeneous annotations, and medical knowledge. Based on this formulation, we introduce MedBenchAgent, a multi-agent framework with a Benchmark Intermediate Representation (BIR) that encodes task definitions, evidence mappings, evaluation protocols, and item specifications across construction stages. MedBenchAgent separates planning, which derives and verifies the specification, from instantiation, which constructs and audits items under the locked specification. MedBenchAgent achieves a Task-Space F1 of 90.9%, outperforming direct task induction (79.2-80.0%) and prior-guided induction (85.1%); 994 of 1,000 sampled items from correctly identified tasks pass human audit. We further demonstrate portability to a specialized medical domain and evaluate twelve VLMs, revealing task- and setting-specific variation obscured by aggregate scores. These results establish constrained compilation as a scalable and auditable framework for medical VLM benchmark construction beyond question generation.

Token-Pruning accelerates Vision-Language Models by removing redundant visual tokens, yet its safety implications remain underexplored. In this work, we present the first comprehensive safety evaluation of Token-Pruning mechanisms and find that: most pruning strategies significantly degrade safety as pruning ratios increase, whereas Query-based Compression shows the opposite, with extreme pruning (up to 99.8%), unexpectedly improves model safety. This sharp contrast prompts a key question: How do different Token-Pruning strategies reshape model safety behavior, and is it possible to enhance safety without sacrificing acceleration? To answer this, we identify an unrecognized mechanism, termed Pruning-Induced Malicious Amplification, where removal of background tokens triggers a side effect: forcing the model's attention to collapse onto a few retained malicious anchors within the foreground, inadvertently amplifying their toxic semantics under jailbreak. To address that, we propose an inference-time and plug-and-play Safety-Aware Pruning (SAP) mechanism that counteracts such dominance via three steps: (1) identifying malicious anchors, (2) restoring pruned benign tokens, and (3) reallocating excessive attention from malicious anchors to benign tokens. Extensive experiments across three safety and four utility benchmarks demonstrate that SAP mitigates pruning-induced vulnerabilities, i.e., reducing ASR by up to 62%, without compromising efficiency or utility.

Vision-language models increasingly reason through crops, regions, and tool-produced observations. Yet an observation can influence the answer without benefiting the candidate it supports. We study candidate-bound visual contribution: valid evidence should help, invalidating its supporting relation should remove its additional effect, and valid rebinding should redirect that effect to the newly supported candidate. We introduce CROSS-Bench, a benchmark of 28,000 decision problems, with matched invalidation and rebinding tests on a dedicated evaluation subset. Our RIVET interface preserves evidence identity and uncertainty, composes a candidate-conditioned response, and separately controls its strength. Shared-evidence experiments show that task accuracy and evidence ownership can diverge. Under matched capacity and training, RIVET increases normalized effect transfer from 0.512 to 0.651 where clean evidence has a positive effect. The advantage persists on common evaluation examples and across repeated decision-layer fits. With evidence predicted from raw inputs, RIVET improves CROSS-Bench accuracy by an average of 5.70 pp across four frozen backbones, relative to the same models without auxiliary evidence. These results separate the utility of visual evidence from the candidate-specific destination of its effect.

Frozen vision-language models increasingly provide safety signals for reinforcement learning. Their use assumes that similarity to language describing danger indicates the hazard itself. Yet policy return and collision rate cannot reveal whether a score detects hazards or responds to correlated features of the scene. VLM-based methods have reported gains in driving and safe-RL benchmarks by converting image-text similarity into rewards, costs, or confidence weights. Such signals promise to reduce reliance on manually designed feedback. They may also reflect prompt structure, embedding geometry, or camera viewpoint, leaving their safety meaning unverified. To address this gap, we present a controlled evaluation of a frozen CLIP prompt-margin safety score. We apply the score to trajectories generated by policies that never receive it, match pre-contact observations to contact-free observations with comparable hazard geometry, and vary the captions, encoder, and camera view. Across three policies, 180 episodes, and 130 isolated contact onsets, the score decreases for about twenty steps before contact. Mechanism controls indicate that the score mainly tracks resemblance to the scene shared by its captions and changes with caption separation and camera view. A constant-confidence control retains the lower catastrophe-rate point estimate, so policy gains do not establish hazard perception.

Scribble-based interaction provides a lightweight and intuitive way for users to specify image editing intents in interactive editing tools. However, current image editing models based on VLMs or LLMs struggle to understand and execute edits based solely on scribble inputs. To systematically study this problem, we construct a new benchmark, ScribbleEdit, that evaluates the ability of image editing models to perform image editing conditioned on scribbles. This task requires both a deep understanding of the intention of the scribble and an accurate interpretation of its spatial information. In ScribbleEdit, we design an automated data construction pipeline and introduce a dedicated evaluation protocol that explicitly measures intention alignment. Our analysis reveals that existing VLM/LLM-based editing models fail to accurately capture scribble intentions. To guide future progress on scribble-only image editing, we propose a simple yet effective soft-token baseline, which enhances the model's understanding of scribble semantics and outperforms standard image editing models on our benchmark. Our evaluation and baseline together provide a concrete foundation for assessing and improving the scribble-driven image editing.

We adapt Stevens's power law to measure the implicit ability of AI models to read visualizations, which can reveal the built-in perceptual mechanisms of algorithmic models. In our pilot study, AIs see no legend. In the color conditions, no colormap name is provided either. GPT-5.5 first views a reference visual representation and estimates its magnitude, then estimates the magnitude of each subsequent image of the same representation relative to that reference. Our evaluation of twelve visual variables makes how algorithmic models read visual encodings measurable, comparable with human perception, and more transparent to humans.

Agentic VLM pipelines increasingly pass the same static visual evidence through multiple specialist agents and tools. This design creates an orchestration-level redundancy mode: semantically unchanged images are repeatedly reconstructed as image-conditioned requests at the VLM API boundary. We call this phenomenon visual orchestration tax and develop a measurement-to-certification framework for visual evidence reuse in agentic VLM pipelines. The audit side defines $\mathrm{M1}_{\mathrm{trace}}$ to count raw visual-evidence touches and M2 to measure structural touch redundancy, with query-level distributions, bootstrap confidence intervals, and paired quality tests. Across SeeingEye and MAMMQA on chart, document, general-VQA, and multi-modal-QA tasks, audits reveal 66.8-75.6% visual-evidence touch redundancy, and every audited query exceeds the predefined gate. The certification side introduces SharedVisCache, a contract-aware evidence reuse hook keyed by image content, preprocessing fingerprint, and encoder assumptions. On SeeingEye, contract validation certifies 75.0-75.5% repeated touches as reusable while preserving 350/350 output strings and $Δ\mathrm{M5}{=}0$. At the physical layer, certified hits reduce $F_{\mathrm{vision}}$ from 800 to 200 in ChartQA-200 trace replay and from 200 to 50 inside live SeeingEye translator-stage physical integration, preserving 800/800 replay strings and 200/200 integrated call outputs. The results position visual reuse as a measurable, behavior-preserving property of agent orchestration and define an agent-layer contract that makes backend prefix or token reuse semantically interpretable.

Fine-grained emotion recognition supports therapy tools and social robots, but it needs facial data, which raises privacy and data-protection concerns. EmoNet-Face-HQ answers that with generated portraits, expert-rated over a $40$-category taxonomy far finer than the usual six to eight basic emotions. Under the protocol it ships with, vision-language models (VLMs) score poorly on that taxonomy, and the benchmark concludes that a dedicated fine-tuned model is necessary: Empathic-Insight-Face (EIF; Small/Large). We show that off-the-shelf VLMs match or beat that fine-tuned model when the answer is not generated but read from the logits, as one binary query per category. We keep the benchmark's images, taxonomy and ratings, and change only how the answer is read. Experts agree at $κ_w = 0.468$ on the five categories they measure most reliably. Generatively, no interval among eleven open-weight VLMs lies entirely above that anchor ($κ_w=0.268$-$0.486$). Under verification all eleven clear it, each of them significantly better at $κ_w=0.507$-$0.586$. Three also significantly beat EIF sitting at $κ_w = 0.551$ (Small; $0.534$ Large). The gain comes from the graded probability and not from asking a yes/no question: as a control, thresholding those same probabilities to yes/no costs 142% of the average gains and drops binarization below generative elicitation to $κ_w=0.254$-$0.423$. A replication on real photographs (FACES) is weaker and mixed: of the ten models that pass a validity gate, six gain, three are neutral to positive and one is negative, so the effect is not confined to synthetic data.

Brain MRI interpretation is a longitudinal clinical reasoning problem: radiologists compare serial studies, integrate information across MRI sequences, localize findings within volumetric anatomy, and translate this evidence into report-grounded assessments. Existing medical VQA and 3D imaging benchmarks capture important parts of this workflow, but often evaluate brain MRI through isolated images, static volumes, or ungrounded report-style answers, thereby obscuring failures in the evidence chain that support clinical validity. We introduce BrainTRACE, a report-grounded benchmark for evaluating whether vision-language models can trace the evidence structure required for longitudinal brain MRI interpretation. BrainTRACE contains 7,273 scored VQA instances derived from 1,778 longitudinal patients, 7,299 MRI studies, and approximately 29k co-registered 3D MRI sequence volumes. The benchmark is organized by five levels of clinical reasoning, from acquisition recognition to case-level synthesis, and by evidence demands covering longitudinal comparison, report-grounded references, multi-sequence integration, and volumetric spatial evidence. BrainTRACE supports rendered inputs compatible with standard VLM interfaces, a 3D-evidence condition, and a decomposed case-reasoning track that audits six steps in a longitudinal evidence chain. Evaluation of 20 VLM configurations shows that current systems can identify isolated visual cues but rarely compose them into grounded longitudinal interpretations. We release the benchmark specification, evaluation lists, scoring implementation, scoring rubrics, and audit-record format to support reproducible progress in brain MRI VLM evaluation.

Thinking-enabled vision-language models (VLMs) report ever-higher accuracy on spatial benchmarks, yet final-answer scores cannot reveal whether a correct prediction reflects faithful spatial reasoning or a linguistic shortcut. We introduce SpatialChain, a dataset of 28,350 training and 899 test examples pairing spatially-oriented GQA questions with scene-graph-grounded reasoning chains, retained only when the generated answer matches the symbolic ground truth, and a two-axis evaluation combining objective chain-overlap metrics with a scene-graph-aware LLM judge that scores faithfulness and completeness independently of the final answer. Applied to nine thinking-enabled VLMs, the protocol surfaces three findings invisible to standard accuracy: (i) four of nine models achieve $\geq$79% VQA accuracy while exhibiting shortcut rates above 39%, i.e., correct answers whose reasoning the judge marks as unfaithful; (ii) chain quality significantly predicts answer correctness for seven of nine models, but the two exceptions (Claude Sonnet 4.6, InternVL3.5-8B) reveal qualitatively distinct failure modes, terse output vs. verbose-decorative reasoning, that benchmark accuracy alone conflates; (iii) SFT on SpatialChain improves Qwen3-VL-8B by +6.2 pp in-domain and reduces its shortcut rate to 22%, while a stylistic specialization effect on external benchmarks motivates replay-augmented training as mitigation. The faithfulness judge is validated against 198 human-annotated items, where judge-human agreement matches human-human agreement, and against a second judge from a different provider, which preserves the model ranking ($ρ$ = 0.88). Data, generation scripts, and evaluation code are released at https://github.com/spatialchain/SpatialChainBenchmark.

CLIP-like vision-language models remain a cornerstone of multimodal systems, yet their scores stay near chance on directed spatial relations, such as whether one object is left of another. We call this failure readout blindness and analyze, theoretically and empirically, why deployed scores miss the direction: when scoring rules treat the subject and object symmetrically, direction cancels regardless of encoder training. Guided by this analysis, we introduce Antisymmetric Displacement Readout (ADR), which aligns caption words with image patches in the frozen features and scores each relation by the signed displacement between matched object centroids. Notably, ADR succeeds without additional training or learned parameters, thereby demonstrating that directional information remains in the frozen encoder. However, text and world priors can inflate accuracy, so we further introduce prior deflation, which measures the benefit of the image-text pairing as the grounded gain over a null that pairs each item with an unrelated image. Extensive experiments across encoder families show that ADR substantially improves over deployed scores, which remain near chance on most direction-balanced sets even for fine-tuned encoders. Compared with more complex readouts, ADR outperforms the evaluated MLLM likelihood readouts and is competitive with their chat inference at a small fraction of the computation. These results support our claim that directional information can be recovered from frozen features by an appropriate readout. Our implementation and evaluation kit will be publicly available.

Vision-language models (VLMs) can detect that an object has rotated across views, but cannot reliably tell by how much. We introduce OR-Bench, a fine-grained benchmark for object-rotation reasoning with eight tasks covering rotation detection, rotation magnitude estimation, and multi-view rotation reasoning. Across 12 VLMs, the gap is stark: the strongest models approach 100% accuracy on detection, yet even coarse magnitude estimation is near chance. When asked for exact angles, models place 91.8--100% of their predictions on just $0^\circ$, $90^\circ$, and $180^\circ$, a failure we term canonical-angle collapse. This collapse persists even without visual input. Representation probing shows that missing information is only part of the explanation. Although rotation information becomes less recoverable at finer granularity, substantial coarse-grained information remains, and a simple linear probe outperforms the models' generated answers. This suggests that VLMs underuse rotation information they already encode. We therefore propose RotationCue, a lightweight decoder that recovers coarse rotation information from the VLM's own frozen representations and feeds it back to the model as intermediate textual context. Across three VLMs, RotationCue improves every model--task combination on OR-Bench, raising macro-average accuracy by 7.9--12.6 points while preserving general capabilities.

Vision-language models (VLMs) are increasingly trained to generate structured outputs like points and bounding boxes that downstream interfaces, agents, and robots can act on, yet safety alignment of this output channel has not been systematically analyzed. We study visual grounding safety by repurposing three safety benchmarks spanning direct harm (VLSU), social bias (BBQ-V), and situational safety (Asimov-2.0) into 15,401 matched pairs of harmful requests that differ only in the requested output: a free-text answer (VQA) or a grounding (point or bounding box). Across five VLMs, models that refuse a harmful request posed as a question often comply when the same request asks for a grounding: averaged over models, grounding refusal trails VQA refusal by 31-59 percentage points, depending on the domain, and safety system prompts do not close this gap. We propose a fine-tuning approach that combines grounding-form refusals with capability grounding data and self-distilled benign data to counter over-refusal. For Qwen3-VL-8B and VisionReasoner-7B, it improves grounding refusal by 77-95 percentage points on VLSU and BBQ-V and by 64-85 points on the held-out Asimov-2.0 domain, while also improving VQA refusal, preserving grounding capability, and keeping over-refusal limited. Representation analysis shows that fine-tuning moves harmful requests toward each model's refusal direction, most strongly for grounding, while leaving benign requests near the harmless reference.

Although the advancement of vision-language models (VLMs) has endowed robots with enhanced environmental understanding and task reasoning, a comprehensive evaluation methodology is important to advance the integration of VLMs in robotic navigation and manipulation. However, current benchmarks lack a comprehensive method to evaluate diverse robotic tasks, and evaluation metrics remain relatively constrained, making it difficult to assess the embodied capabilities of VLMs in a thorough and fine-grained manner. To address this issue, we propose RMMBench, an evaluation benchmark that requires robots to understand language instructions and perform long-horizon tasks in continuous spaces. RMMBench seamlessly integrates high- and low-level embodied tasks into a unified framework, constructing a "navigation-manipulation" task suite comprising 70 canonical task scenarios that range from localized manipulation to long-horizon composite navigation. The results reveal that leading VLMs still face major challenges in spatial localization when performing mobile manipulation tasks, and also highlight the necessity of enhancing the spatial perception capability of robots during long-horizon interactions. RMMBench can be accessed at https://mxxq-stack.github.io/rmmbench-project/

Evaluating vision encoders requires metrics that reliably predict their downstream performance in multimodal large language models (MLLMs). Although recent studies have shown that cross-modal metrics can better capture such performance, unimodal metrics remain the dominant choice in practice. In this work, we revisit cross-modal evaluation of vision encoders through large-scale experiments. We identify important limitations in both the experimental design and methodological formulation of prior approaches. After addressing these limitations and introducing simple improvements, we propose RAVEL, a training-free method based on cross-modal nearest-neighbor retrieval. Despite its simplicity, RAVEL achieves state-of-the-art performance across our experiments, outperforming prior methods by a substantial margin. Our results demonstrate that simple cross-modal metrics, when evaluated under a careful and comprehensive setup, can provide a strong basis for evaluating vision encoders for MLLMs.

Reliable visual reasoning requires composing multiple visual observations and returning consistent answers to logically equivalent questions. We introduce Hob-VL, a benchmark for visually grounded Boolean reasoning. Hob-VL comprises two tasks: (1) evaluating whether a Boolean rule holds in an image, and (2) identifying the (unique) object satisfying a Boolean description. Hob-VL contains 6,000 human-verified balanced Yes/No questions, each defined by a Boolean combination of ten visual statements, across 1,000 generated scenes and 46 diverse labeled photographs, along with 1,000 object-identification questions over the same photographs. Our question families are deliberately constructed to challenge reasoning through misleading local cues and nested logical operations, and include symbolic and structured natural-language presentations. Across eight model configurations with thinking disabled or minimized, Boolean accuracy ranges from 48.52% to 50.57%, while the identification accuracy reaches at most 43.0%. A thinking-enabled GLM configuration achieves uneven gains while retaining substantial errors and inconsistencies. Hob-VL exposes these failures through executable reference answers and matched evaluations.

Vision-language models (VLMs) increasingly act as judges that pick the best of several generated images, so their choices decide what users see. Such judges are usually validated by score agreement with human ratings, not by the images they return. We audit VLM judges as decision-makers: on 300 culturally situated prompts, we compare the returned image with human ratings the judge never sees and with random choice from the same candidates, and repeat every decision with the candidates reordered. A 4B-parameter judge barely beats random and falls short of a CLIP similarity baseline. It picks the first image shown in 49% of calls (chance: 28%), and reordering changes its choice on 60% of prompts. For this judge, agreement across orders is informative: decisions that survive reordering are much better than random, whereas agreement with a weaker second judge keeps the wrong ones. An 8B judge shows almost no position bias and outperforms CLIP, yet for it the same filter mostly discards good decisions. Agreement helps only when it targets the judge's failure mode, so filters must be re-audited whenever the judge changes. The 4B judge's slight rise in stereotype ratings is no longer detectable after aggregating across orders or with the larger judge.

Existing synthetic image evaluators typically provide only a scalar quality score and do not identify the image regions that support it. We introduce VIEScore2, a unified evaluator for image generation and editing tasks with optional conditioning images. VIEScore2 represents an image as an N x N grid and jointly predicts quality scores and defect locations in a single model pass. Its text-native grid representation provides a common interface for heterogeneous spatial supervision and enables directly verifiable post-training objectives. We train on 38K examples spanning score-only, localization-only, and joint supervision across generation and editing tasks. Starting from supervised fine-tuning, we further apply GRPO to improve defect localization using rewards that combine cell-level Dice overlap, score accuracy, and output-format validity. A parameter-free parser converts the structured predictions into readable explanations. On the primary suite, VIEScore2 achieves an overall-score SRCC of 0.601, compared with 0.491 for Gemini-3-Flash, the strongest zero-shot general-purpose VLM baseline under matched inputs. For defect localization, VIEScore2 outperforms both general-purpose VLMs and specialized spatial evaluators on three of six benchmarks in per-image grid IoU and ranks among the top three on five, including datasets beyond its training sources.

An answer candidate in a masked diffusion MLLM can stabilize while its rationale is still unfolding. We distinguish retrospective stabilization of the logged candidate from token commitment, and examine these two clocks relative to rationale generation. Analyzing our results across three visual question-answering benchmarks, we find that 89.4-98.1% of the rationale-side canvas remains unwritten at stabilization in single-block, EOS-suppressed LaViDa runs. On VBench, reducing block length from 128 to 8 changes this fraction from 89.4% to 1.7%, together with answer coverage and the eligible observation window. Under EOS-enabled prompting, direct instructions improve Nemotron's overall accuracy by 15.0 and 19.5 percentage points on M3CoT and ScienceQA, but reduce LaViDa/VBench accuracy by 11.0 points. A symmetric decomposition associates the larger absolute component of each change with coverage rather than conditional accuracy. Matched-canvas image ablations measure visual sensitivity alongside answer stabilization, separating the two temporal readouts. Together, these measurements distinguish answer stabilization, rationale unfolding, and visual sensitivity, and identify coverage as the larger component of the prompting differences.

Recaptioned image-text corpora are now standard for text-to-image (T2I) training, with vision--language model (VLM) captioners replacing sparse alt-text by dense descriptions. A recaptioned corpus is a supervision distribution induced by a documented captioning policy ($π$), captioner ($V_c$), and source corpus ($C$). Length-correlated proxies miss caption-register artifacts and downstream T2I benchmarks entangle the corpus with training choices, so this distribution is hard to audit at corpus scale. We introduce a reusable matched-budget audit framework for recaptioned supervision distributions $D_{π,V_c,C}$: at a fixed text budget of $B = 64$ it reports a five-axis profile spanning prompt-side coverage, image-conditioned faithfulness, and caption-surface health, with claimed controllable basic units (CBU) as the common claim unit. We instantiate the framework on seven paired comparisons over five public source corpora. Across the four cross-corpus pairs, the released surface raises supported CBU per caption by $+3.39$ to $+6.36$ under both Qwen and Gemma Judges, and on CC12M the same framework exposes a long-vs-dense frontier that is consistent across both judges and four budgets. We release the audited multi-source recap corpus ($\approx$ 490M) together with the audit-artifact bundle.

Frontier vision-language models (VLMs) increasingly estimate scenes, ground interactions, and generate executable actions. How far these native capabilities support embodied generalism across diverse tasks remains unclear. We introduce Embodied Agent Arena to assess seven VLM agents across Geometry, Spatial Reasoning, Affordance, Task Planning, and Manipulation. The arena contains 1,000 cases drawn from 32 established sources and GeoProbe, our new benchmark for geometric estimation on Blender renders and real-scene images. A minimal harness preserves source observations and operations while leaving perception, reasoning, and action selection to the model. Separate measures of metric precision, functional grounding, and native goal completion connect local competence to complete task outcomes. Astra's strengths in precise estimation and usable-contact localization coexist with endpoint errors in tracing and low household-task completion. Supplementary comparisons of richer observations and multi-round review show model- and task-dependent effects. Current VLM agents thus fall short of embodied generalism: they often make partial progress without satisfying all task goals within allotted time and interaction budgets.

Vision-Language Models (VLMs) enable video annotation at scale, but costs accumulate quickly: processing a typical 60-second short-form video at one frame per second requires millions of tokens. To reduce costs, researchers rely on heuristics such as sampling a subset of frames, compressing videos into image grids, or using only a single modality. However, it remains unclear which heuristics save cost, and whether they preserve the downstream conclusions these annotations enable. To address this gap, we conduct a systematic evaluation of these heuristics using short-form videos, on two computational social science (CSS) tasks: sentiment and topic classification. We evaluate each configuration along three axes the literature typically treats separately: classification accuracy, validity of downstream inference, and per-video token cost. First, we find that accuracy and validity diverge: the highest-accuracy configuration can produce wrong conclusions. Second, modality value is not guaranteed: text alone can yield strong performance, indicating that adding modalities can add cost without adding signal. Finally, we find that cost can be decoupled from video length when annotating short-form videos: a single $2\times8$ image grid built via simple shot-transition detection approaches full-video understanding ($κ$ within~.05), at $\sim 15\%$ of the token cost. Based on these findings, we derive guidelines that can enable cost-aware VLM annotation in CSS.

Advancements in augmented reality (AR) continue to foster innovative solutions, facilitating novel methodologies within educational systems, healthcare delivery, and risk-mitigation protocols. However, optimizing for end-user immersion and comfort remains challenging, as AR head-mounted displays contend with constrained scene geometry, spatial jitter, and temporal instability. User studies are the standard AR evaluation method for visual quality, but their cost, diminishing scalability, and inflexibility pose bottlenecks during iterative application design. To address this problem, we present an automated framework for AR content evaluation and refinement, built on vision-language models (VLMs), to evaluate and predict the visual fidelity of AR scenes as perceived by users. First, we introduce RateAR, a benchmark of AR images and videos collected across diverse scenes and environmental conditions, with good-to-excellent reliability (ICC(2,5) >= .90) across perceptual factors, including object placement, scale, and shadow consistency. Subsequently, we evaluate eleven commercial VLMs on the crafted benchmark. Results support that VLM-based quality predictions strongly correlate with human subjective judgments, achieving Spearman's rank-order correlations of up to 0.8695. An ablation study further suggests that, compared to other prompting strategies, our contextual prompting yields better alignment with human ratings while balancing introduced complexity cues. Building on these findings, we construct an automated AR content adjustment system and conduct a 21-participant user study. More than 90% of participants found that the system improved placement and size coherence of virtual content.

Qualitative comparison figures are central evidence in computer vision papers, and vision-language models (VLMs) are increasingly used to judge them. Yet existing benchmarks score only scalar quality or overall preference, so a judge can be rewarded for picking the preferred image for the wrong visual reason. We introduce VisionQ, the first benchmark built from peer-reviewed CV comparison figures that grounds every judgment in a named visual criterion: each question states the criterion, and a judge is credited only when it selects the output the authors identify as best on that criterion. We call this task criterion-conditioned visual discrimination. VisionQ comprises (1) a corpus of 1,409 CVPR and ICCV papers with 1,800+ validated comparison figures and 3,911 hand-annotated data points linking method crops to author-stated visual claims; (2) a six-axis, 51-leaf taxonomy of the visual criteria behind qualitative judgment; (3) a criterion-conditioned evaluation protocol that hides method names, captions, and paper identity and reports accuracy per criterion; and (4) VisionQ-Judge, a DPO-tuned Gemma-4-E4B judge trained on symmetric evidence pairs, which reduces last-option predictions by 7.0pp and improves accuracy by 2.5pp on a held-out test set. Evaluating 20 open- and closed-source VLM judges, we find that the strongest reach only 63.1% accuracy (chance 32.2%) and that reliability varies sharply across criteria. Code: https://github.com/ReML-AI/visionq. Data: https://huggingface.co/datasets/visionq-anon-2026/VisionQ-1k.

Vision-Language Models (VLMs) have shown strong multimodal reasoning capabilities, yet whether they truly capture the physical consistency underlying real-world dynamics remains unclear. Existing benchmark paradigms often suffer from fragmented evaluation, focusing on isolated cognitive stages while overlooking the inherent synergy between perception, reasoning, and physical judgment. The lack of a holistic perspective limits the ability to diagnose whether VLMs can reliably evaluate the physical authenticity of emerging generative models. To address these issues, we introduce PhysVista, a benchmark designed to evaluate physical intelligence in VLMs through a closed cognitive loop framework inspired by the human seeing-reasoning-assessment process. PhysVista restores this loop by jointly evaluating physical state perception, physical dynamics reasoning, and physical plausibility assessment. It further distinguishes event-level reasoning and scale-level reasoning to enable fine-grained analysis of physical understanding. In addition, PhysVista incorporates both real-world and AI-generated videos, allowing evaluation across diverse domains and emerging generative scenarios. Extensive experiments across a diverse set of VLMs reveal substantial limitations in physical reasoning and plausibility assessment, highlighting a persistent gap between visual recognition and genuine physical understanding, and pointing toward more principled designs for physically grounded multimodal intelligence.

Can a text-to-3D leaderboard change when every generated scene stays fixed? We audit this question for rendered-image evaluation, where camera settings and caption wording become part of the measurement protocol. Across 300 frozen scenes from six generators, we vary eight render and caption factors for 19 alignment evaluators plus one perceptual-quality control, then test four targeted scene degradations. Peak configuration variance exceeds between-generator variance for 17/19 alignment evaluators, with prompt-bootstrap lower bounds above 1 for 11/19. Rankings are more stable than scores, yet 18/19 evaluators change their point-estimate winner under some configuration. Pairwise protocol margin envelopes show which comparisons keep their direction across the tested settings. Selected pairs have opposite pointwise intervals, but no reversal survives simultaneous inference over the full search. Thus the observed winner changes are descriptive, not confirmed changes in generator superiority. Sensitivity remains separate: no evaluator, even the prompt-free control, exceeds 67% tie-adjusted directional discrimination on layout scrambling, which is diagnostic rather than human-validated ground truth. The audit separates score stability, decision uncertainty, and targeted sensitivity, and recommends reporting (generator, score, card ID) with protocol-dependent comparisons and selection-aware uncertainty.

Robot navigation in dynamic, human-centered environments requires socially-compliant decisions grounded in robust scene understanding. Recent Vision-Language Models (VLMs) exhibit promising capabilities such as object recognition, common-sense reasoning, and contextual understanding, capabilities that align with the nuanced requirements of social robot navigation. However, it remains unclear whether VLMs can accurately understand complex social navigation scenes (e.g., inferring the spatial-temporal relations among agents and human intentions), which is essential for safe and socially compliant robot navigation. While some recent works have explored the use of VLMs in social robot navigation, no existing work systematically evaluates their ability to meet these necessary conditions. In this paper, we introduce the Social Navigation Scene Understanding Benchmark (SocialNav-SUB), a Visual Question Answering (VQA) dataset and benchmark designed to evaluate VLMs for scene understanding in real-world social robot navigation scenarios. SocialNav-SUB provides a unified framework for evaluating VLMs against human and rule-based baselines across VQA tasks requiring spatial, spatiotemporal, and social reasoning in social robot navigation. Through experiments with state-of-the-art VLMs, we find that while the best-performing VLM achieves an encouraging probability of agreeing with human answers, it still underperforms simpler rule-based approach and human consensus baselines, indicating critical gaps in social scene understanding of current VLMs. Our benchmark sets the stage for further research on foundation models for social robot navigation, offering a framework to explore how VLMs can be tailored to meet real-world social robot navigation needs. An overview of this paper along with the code and data can be found at https://larg.github.io/stars.

Vision-language models (VLMs) are increasingly used for AI-generated image (AIGI) detection, providing natural-language explanations for authenticity judgments. However, their ability to interpret text within images may also expose these judgments to misleading semantic cues. We systematically evaluate typographic attack strategies across detection-oriented, open-weight, and commercial VLMs, considering both real-to-fake and fake-to-real attacks. Our results show that reasoning modes generally exhibit greater vulnerability than direct modes and that attack effectiveness exhibits pronounced directional asymmetry. Moreover, larger models tend to exhibit higher clean detection accuracy but also higher attack success rates. We further examine attack robustness under image and text transformations and investigate whether overlays indicating the correct class can aid error correction. Together, these analyses characterize how typographic attacks influence authenticity judgments and expose limitations of current VLM-based AIGI detection systems.

Matching the same vehicle across front and rear cameras is difficult because the cameras do not share a view and the vehicle's appearance changes substantially. We introduce Front2Back-ReID, a benchmark of 500 manually verified vehicle handovers from 20 recording sequences in South Africa. Each example asks a model to match a vehicle highlighted in a front-camera image to the same vehicle among at least three candidates in a later rear-camera image. We evaluate seven zero-shot vision-language models, four image-retrieval baselines, and 25 human participants. Models are tested using full front RGB images, cropped target vehicles, and binary silhouettes. The strongest VLM achieved 76.6 percent Rank-1 accuracy on target crops, compared with 74.0 percent for the frozen SigLIP2 baseline; this difference was not statistically clear. Human participants achieved 94.0 percent accuracy with full images and 92.2 percent with target crops. Under our evaluation setup, enabling reasoning improved accuracy across all three input conditions for every model evaluated in both modes. We also found that VLMs generally performed worse on full scenes than on target crops. These results show that general-purpose VLMs do not yet consistently outperform strong visual retrieval for front-to-rear vehicle matching, while humans remain substantially more reliable.

This paper presents an explainable approach to pedestrian routing, in which perceived safety is estimated from street-level imagery through an explicit natural-language intermediate representation. A vision--language model caption is generated and stored before any scoring is undertaken, and the perceived-risk class is derived entirely from structured features of that stored text, so that every segment score remains inspectable by the user. Nine captioning conditions across five model families are benchmarked against a direct Contrastive Language--Image Pre-training (CLIP) image-embedding baseline under an identical downstream pipeline, and the caption-mediated representation is found to reach parity with the image embedding rather than to trail it. The approach was deployed over 654,115 images covering 36 electoral wards in two locations in Northern England (Manchester and Huddersfield). Independent field validation against 3,669 locally collected ratings of 494 images across 70 participant sessions established agreement that is statistically significant but modest, at $r=0.262$, against a measured noise ceiling of 0.737 imposed by disagreement between raters. A single-use confirmatory test then found that a pipeline 44% stronger on the supervised benchmark did not produce measurable improvement in the field ($r=0.250$, $p=0.84$), so the benchmark gains did not predict the deployment gains in this case. Routing behaviour varies systematically with journey length. There is negligible change below 1,km, reaching a median increase of 12.78% in low-risk route length for a median detour of 2.73% on journeys of 3 to 6 km.