Counterfactual Evaluation of VLMs

VLM: Vision-Language Model

Ask about this topic

Overview

Counterfactual evaluation studies whether vision-language models use visual evidence by changing an image, video, or context in a controlled way and checking whether the answer responds as expected. Common approaches pair factual and edited examples, use image ablations or misleading text, and measure answer flips, consistency, grounding, and confidence; some also trace internal activations or train models on evidence-dependent changes. These tests are used to diagnose hallucination, spatial and causal reasoning, chart interpretation, medical grounding, and safety decisions. A central challenge is separating genuine evidence use from correct guesses, language shortcuts, and unperceived edits, so robust evaluation needs multiple perturbations and controls for perceptibility and blind baselines.

Latest papers

CardsList

61 papers20 papers in the last four weeks, up 54% on the four weeks before. 0.1% of all new papers.

Sep 6, 2026

Separating Capability from Confidence: Grounded Dual-State Calibration for GRPO-Trained Medical Vision-Language Models

Medical vision-language models (VLMs) require confidence that reflects both answer correctness and patient-specific visual evidence. Recent GRPO-based methods optimize verbalized confidence together with answer generation. However, this joint optimization may interfere with answer learning and drive confidence toward near-binary values. Verbalized confidence also provides no explicit assessment of visual support. We therefore separate capability learning from confidence estimation and propose DualRead. DualRead builds on the insight that reliability can be read from the actor's internal states at critical moments in the answering process. It freezes the GRPO-trained actor and combines pre-answer solvability with a post-answer assessment of the generated answer and its visual support. To further assess whether confidence reflects visual grounding, we introduce Counterfactual Confidence Grounding AUC (CCG-AUC). It measures whether confidence decreases when real-image substitution changes the actor from correct to incorrect. Across two VLM backbones and both in- and out-of-distribution medical VQA benchmarks, DualRead improves correctness discrimination and calibration over verbalized confidence while preserving answer accuracy. CCG-AUC reveals whether confidence responds to answer-relevant visual evidence rather than primarily to non-visual cues.

Yangyang Xie, Ke Hao, Jiaqi Liu +2

Shanghai Jiao Tong University, Shanghai, China · Medical Image Insights Co. Ltd., Shanghai, China

Confidence Estimation for VLMsVLM CalibrationMedical VLMsMedical VQACounterfactual Evaluation of VLMs
AskPDF
Sep 5, 2026

One Perturbation Is Not Enough: Identifiability and Blind Baselines for Behavioral AI Evaluation

Behavioral evaluations perturb an input and read the induced change in the output in order to certify that a system uses that input. We show that the number of perturbations such a certificate requires is fixed, and that reporting a single perturbation cannot supply it. Where a response ratio is a property of the policy rather than of the test items, the behavioral record is a linear measurement of an exponent vector recording how much the output depends on each input, so perturbations identify input use exactly when their logarithms span the input space. At least nn are needed for nn inputs, an incomplete design confuses precisely the policies differing along the kernel of its design matrix, and sharpening a perturbation never substitutes for adding an independent one. We also derive in closed form the score such a test awards a policy that reads nothing, which is far from zero and which none of the probes we survey reports. Instantiating this where the correct response is fixed by dimensional analysis, we run a complete identifying set of three perturbations on three vision--language models reporting a physical quantity from video. All three score far below their own blind bound rather than above it, because each defaults to one of a small set of round calibration values that never matches what the scale asserts; none moves its relabeling response by a single exponent, and none is separable from the same model instructed to ignore the video.

Rasul Khanbayov, Mariam Sohail, Ahmed Abdala +1

College of Science and Engineering, Hamad Bin Khalifa University, Doha, Qatar

Counterfactual Evaluation of VLMs
AskPDF
Sep 3, 2026

When Do Frozen VLMs Respond to Image-Free Object-Token Edits? An Answer-Key-Free Protocol and What It Reveals

Answering what-if queries about a scene with a VLM usually means injecting the assumption as text or repainting the scene with a generative model. We instead move the edit to the representation level, before the model input. The image is abstracted into a set of object-level tokens, and the original image never enters the VLM. This design rests on an open question: when do frozen VLMs actually respond to such token edits? We introduce an answer-key-free protocol: no post-edit answer is annotated. It scores edits whose answers are logically determined, and audits itself by reversing each scoreable choice. The protocol reveals three structures. The response is not free: explicit edit teaching, not ordinary VQA training, produces it in dense scenes and multiplies it in sparse ones, on all three operations. Once on, it is governed by token cleanliness and density, with deployable detector+segmenter tokens competitive with the oracle and outperforming it on VRSBench. And reading is a separable axis: the image-free token route preserves 92-96% of a matched patch-token baseline's free-text VQA, and the answers measurably depend on the tokens. The response, cleanliness, and reading structures are sign-preserved across two remote-sensing datasets (iSAID, VRSBench) and three frozen LM backbones. We release the probe generator, records, judge logs, and code.

Wonbin Son, Gyumun Choi, Junil Seo +3

Changwon National University · Chung-Ang University

VLM EvaluationCounterfactual Evaluation of VLMsVLM Reasoning
AskPDF
Sep 2, 2026

Beyond the Verdict: Evidence-Aligned Evaluation of Visual Prompt-Injection Guardrails

Verdict-only evaluation does not reveal whether a vision-language model (VLM) used the visual evidence that should support its decision. We study this problem in web-agent guardrails, where a VLM judges whether on-screen text conflicts with a user instruction. We introduce Mind2Web-Injection, a benchmark of 9,954 instruction-screenshot pairs with instruction-relative labels, pixel-exact evidence boxes, and matched image-side counterfactuals. Across six VLMs, two models with nearly identical average precision differ ninefold in Evidence-Aligned Detection (EAD), the fraction of attacks both detected and correctly localized. To test whether a verdict depends on the command cited as evidence, we replace the instruction with one that endorses that command. Qwen3-VL-32B, the strongest open-weight localizer, returns aligned in only 58.7% of cases, whereas GPT-5.6-luna does so in 99.9%. To diagnose these failures, we propose two training-free interventions. ReadGate improves grounding without changing verdicts, while CmdCompare tests whether explicit instruction-command comparison resolves instruction-side inconsistency. These results motivate reporting verdict correctness, evidence localization, and counterfactual responsiveness separately.

Suyoung Lee, Myungsub Choi

Tynapse, Seoul, Republic of Korea

Cultural Safety Evaluation of VLMsVLM EvaluationPrompt Injection DetectionCounterfactual Evaluation of VLMsWeb Agents
AskPDF
Sep 1, 2026

Don't Just Look, Intervene: Perturbation Based Region Labeling for VQA Images

Vision Language Models (VLMs) should rely on visual evidence that directly determines the correct answer, but supervision for grounding visual reasoning is often expensive to obtain manually or tied to dataset-specific annotation primitives. We instead introduce model-causal visual evidence as an annotation target, defined as the set of image regions whose counterfactual intervention changes a model's answer distribution for a given image-question pair. Based on this principle, we introduce Counterfactual Search for Grounding Regions (CSGR). CSGR is a scalable pipeline that proposes candidate regions, perturbs them, measures their effect on answer sensitivity, and aggregates this evidence across multiple judges to approximate answer-critical regions in VQA data. To assess whether CSGR annotations contain a useful supervision signal, we plug them into three existing grounding-aware training routines: attention steering, Visual CoTfinetuning, and latent visual reasoning. These experiments test whether the proposed annotation scheme can provide a useful supervision signal across multiple ways of consuming region labels, rather than introducing a new way of using them. Across competing automatic region-labeling mechanisms, CSGR annotations provide the most consistent gains over Cross Entropy-only finetuning in both in-domain and out-of-domain evaluations, indicating that the proposed labeling scheme captures useful region-level information.

Marko Jojic, Zhaonan Li, Ben Zhou

Arizona State University

Vision-Language GroundingVisual Question AnsweringCounterfactual Evaluation of VLMs
AskPDF
Aug 26, 2026

GGSS: Geodesic-Gated Spherical Steering for Inference-Time Debiasing of Generative Vision-Language Models

Generative vision-language models (VLMs) are increasingly used in human-centered settings, yet they can produce demographically biased outputs even when images differ only in controlled attributes such as perceived race or gender. However, existing inference-time debiasers were largely designed for static embeddings or CLIP-like models rather than generative VLMs. We propose GGSS---Geodesic-Gated Spherical Steering---a norm-preserving intervention that discovers a counterfactual bias subspace on the unit hypersphere, steers visual tokens along geodesic arcs, and uses an adaptive gate to focus correction on tokens that carry stronger demographic signal. We evaluate four generative VLMs against ten adapted inference-time debiasing baselines and prompt-based mitigation under a single operating-point protocol across categorical, pairwise, and occupation-gender bias tests, while also measuring general visual-language capability. GGSS achieves the lowest average bias on all four models, significant on three of four backbones under paired permutation tests, while preserving MMStar accuracy within +/- 0.6 p.p. of the unsteered baseline. Code is available at https://github.com/dukesun99/GGSS.

Yiqun Sun, Junyu Chen, Pengfei Wei +1

Magellan Technology Research Institute (MTRI) · National University of Singapore

Activation SteeringCounterfactual Evaluation of VLMs
AskPDF
Aug 24, 2026

EviSafe: Evidence-Grounded Safety Evaluation for Vision-Language Models

Vision-language model safety benchmarks typically evaluate only final responses: whether a model refuses, warns, or complies. This outcome-level view cannot tell whether a model is safe for the right multimodal reason. Safelooking behavior may reflect keyword-triggered refusal, missed visual hazards, or over-refusal of benign-sensitive inputs. We introduce EviSafe, an evidence-grounded framework for VLM safety that jointly evaluates natural user-facing behavior, explicit grounding in textual and visual evidence, and behavioral sensitivity to counterfactual changes in safety-critical evidence. EviSafeBench instantiates the framework as a controlled benchmark with 1,181 gold image-text scenarios and 2,452 targeted counterfactual variants across eight safety domains and eight risk-source types. Each scenario includes a gold safety decision, evidence annotations, a safe-response policy, and counterfactual interventions. The three-probe protocol queries models with natural-response, evidencereporting, and counterfactual-response prompts, then scores them using an evidence-aware judge. Across eleven evaluated VLMs, natural severity accuracy ranges from 27.6% to 52.8%, relaxed diagnostic consistency from 6.1% to 29.3%, and unsafe-to-safe counterfactual transition success from 30.4% to 58.4%. These gaps show that the evaluated VLMs are not reliably safe for the right multimodal reason and motivate evaluation beyond refusal counts.

Xuetong Li, Gaofeng Liu

School of Mathematical Sciences, MOE Key Lab of Scientific and Engineering Computing, Shanghai Center for Applied Mathematics, Shanghai Jiao Tong University · Department of Automation, Shanghai Jiao Tong University

Cultural Safety Evaluation of VLMsVLM SafetyVLM EvaluationVLM Safety AlignmentCounterfactual Evaluation of VLMs
AskPDF
Aug 17, 2026

Fidelity Preference, Not Demographic Preference: A Pixel-Level Attribute-Sensitivity Audit of Image Aesthetic/Preference Scorers

Text-to-image systems use learned aesthetic scorers to filter training data and guide generation, but whether these scores encode demographic attributes as objective quality is unclear. We audit four scorers (LAION-Aesthetics, PickScore, ImageReward, HPSv2) using pixel-level interventions on skin tone and body type in synthetic and real images. Our key finding is that along skin-lightness, the dominant effect is fidelity preference: unaltered images score highest, and perturbations in either direction are penalized (inverted-U). Placebo arms show this penalty is not an artifact of the skin operator, as applying the same CIELAB L* shift to non-skin regions yields similar penalty magnitudes. However, the penalty is operator-dependent and holds for all operators only for LAION-Aes. Critically, audits on synthetic images alone are misleading: LAION-Aes shows strong preference for darker skin on synthetic faces, but on 1470 real faces the preference reverses and becomes much smaller, and amplification becomes non-significant. Across scorers, synthetic results do not transfer -- reversing for LAION-Aes and HPSv2, attenuating for PickScore. We contribute a reproducible benchmark with artifact control and synthetic/real cross-validation, and an auditability criterion for pixel-level causal isolation (valid for skin tone, not for body type due to deformation). Population-stratified analysis shows fidelity-penalty asymmetry is not robust across groups after FDR correction except for HPSv2. Our findings show naive synthetic audits misjudge bias direction and magnitude, and only within-image causal isolation on real data can distinguish true demographic bias from fidelity preference.

Mingyang Xu

Peking University Beijing, China

Algorithmic BiasAlgorithmic FairnessImage Quality AssessmentCounterfactual Evaluation of VLMs
AskPDF
Aug 12, 2026

Same Chart, Different Story: Bias in Vision-Language Chart Interpretation

Vision-language models (VLMs) are increasingly used to interpret charts and generate natural-language explanations for socially consequential data. However, they may produce different narratives for the same chart when only the referenced social group changes, reinforcing stereotypes and misleading decisions. Despite these risks, no benchmark exists for systematically evaluating bias in chart interpretation across social dimensions. We introduce ChartBias, the first benchmark for auditing bias in VLM-based chart interpretation. ChartBias contains 820 manually curated real-world charts spanning six attributes: race, income, age, religion, immigration status, and gender, yielding 4,319 valid chart, attribute instances and 8,638 paired generations where the chart is fixed and only the group term is swapped. Across 12 proprietary and open-source VLMs, totaling 155,484 model responses, we find three widespread failure modes: narrative shift (same chart, different narratives), group hallucination (assigning a chart to a group without evidence), and preference polarity (favourable trends often linked to one group). We further propose a multi-agent mitigation framework that serves as a strong baseline by separating chart-grounded evidence extraction from group-conditioned generation and using a counterfactual judge to verify that group-driven differences are supported by the chart. The framework substantially reduces narrative shift while preserving chart-grounded reasoning. Our findings show that evaluating chart understanding requires measuring not only accuracy, but also fairness and consistency across social groups. We release ChartBias at https://github.com/vis-nlp/ChartBiasBench.

Mizanur Rahman, Huan Wu, Arash Asgari +2

York University · Vector Institute · Connected Minds

Cultural Bias in VLMsVLM EvaluationChart UnderstandingCounterfactual Evaluation of VLMs
AskPDF
Aug 5, 2026

CARGO-VL: Counterfactual Arbitration with Risk-Constrained Group Optimization for Vision-Language Models

Vision-language systems combine images with retrieved text, but these sources can disagree or jointly fail to support an answer. Reliable models must identify the trustworthy source and abstain when neither is adequate. Existing post-training objectives score instances independently and therefore do not enforce coherent behavior under counterfactual evidence changes. We introduce CARGO-VL, a group-relative framework that optimizes matched variants covering aligned, image-correct, text-correct, and both-wrong (A/V/T/N) evidence states as one bundle. Its objective couples condition-wise correctness with transition rewards for answer invariance, source equivariance, and answer-to-abstention switching, while a primal-dual controller balances unsafe answers against excessive deferral. We also contribute XMC (eXtended Modal Conflict), a four-condition conflict training resource, and evaluate transfer on CMC-Bench and Modality-Bias. Across multiple seeds, CARGO-VL improves conflict handling, unsupported-answer avoidance, and modality balance over pointwise baselines. Ablations identify complementary benefits from relational transition signals and adaptive risk control, supporting counterfactual consistency as a practical objective for reliable multimodal evidence arbitration.

De Jiang, Zhengyang Zhang, Kehong Yuan +1

VLM RobustnessCounterfactual Evaluation of VLMsConstrained Optimization
AskPDF
Jul 1, 2026

MindEdit-Bench: Benchmarking Object-Level Counterfactual Spatial Reasoning in VLMs from In-the-Wild Photos

Benchmarks for vision-language models (VLMs) mostly test observational spatial reasoning: models describe relations already visible in the input. Existing what-if tasks typically vary the observer while keeping the scene fixed. Can VLMs instead predict the consequences of hypothetically moving or rotating an object? We introduce MindEdit-Bench, a benchmark of six spatial reasoning tasks built from three-photo smartphone triplets of newly captured indoor scenes via an automatic in-the-wild 3D scene-graph extraction pipeline. Four tasks probe perception and perspective transformation over observed structure; two new tasks, L4 (spatial editing) and L5 (cross-view visibility editing), probe object-level counterfactual reasoning, where correct answers are absent from all input images. Each question provides 8-24 structured answer choices, enabling answer-letter-level diagnosis of spatial and fallback errors. The benchmark covers 120 private indoor scenes not drawn from public datasets, reducing public-data pretraining-overlap risk. Across 15 VLMs on 1,003 human-verified questions, task-wise mean VLM accuracy is only 8%-31%, versus 81%-97% human majority-vote accuracy. The pooled human--best-VLM gap is 53 pp, with at least 39 pp on every task. The structured answer space further reveals non-uniform failures, including weaker camera-depth-axis inference and fallback behavior on difficult visibility-editing cases.

Leyuan Yu, Xiao Tang, Minghao Liu +6

ZODA · Tongji University · Zhejiang University +1

VLM EvaluationSpatial Reasoning BenchmarksVisual Spatial ReasoningCounterfactual Evaluation of VLMs3D Spatial ReasoningVisually Grounded Reasoning+1
AskPDF
Jun 23, 2026

Ill-Posed by Design: Probing Evidence Use in VLMs

Counterfactual analysis is widely used to study evidence use in vision-language models, but its diagnostic value is limited on well-posed tasks: when several cues independently support the same answer, removing one may not change the prediction. We propose monocular metric object-size estimation as an ill-posed diagnostic setting for evidence selection: because physical size cannot be determined from a single uncalibrated image, models must rely on imperfect cues category priors, target appearance, local context, apparent image size, and scene geometry. We assemble Metric VQA (10,81310{,}813 dimension queries from Objectron and 331331 tape-measured in-the-wild scenes) and evaluate 1212 open-weight VLMs (33--397397,B parameters) with counterfactual analysis decomposing six visual and language evidence channels. Even the largest VLMs tested (Qwen3-VL-235B, Qwen3.5-397B, InternVL3.5-241B) trail a text-only frontier LLM on the in-the-wild split. The diagnostic analysis shows: target identity is the most load-bearing cue, target pixels and local context help only some models, apparent size shifts predictions without a directional readout, and global scene geometry is largely unused. We analyze LoRA fine-tuning as an actionable intervention specific to metric estimation: while the task is learnable, the models do not learn to leverage scene geometry.

Boaz Meivar, Shaked Perek, Shani Shvartzman +2

1Tel Aviv University · 2IBM Research

VLM EvaluationCounterfactual Evaluation of VLMsVLM InterpretabilityVisually Grounded Reasoning
AskPDF
Jun 7, 2026

How Many Counterfactuals Does It Take? Probing VLM Hallucinations Through Circuits and Causal Effects

Visual Language Models (VLMs) are known to produce hallucinated predictions that are not grounded in visual evidence, yet existing approaches lack a principled understanding of how robust such predictions are under counterfactual perturbations. In this work, we study the sample complexity of counterfactual robustness for hallucinated outputs in VLMs. We define a causal influence metric based on log-probability differences between factual, counterfactual, and activation-patched runs, and use it to characterize the stability of hallucinated predictions. By leveraging circuit discovery techniques (CD-T), we identify model components responsible for these predictions and track their activation differences across counterfactual samples. We then derive empirical bounds on the minimum number of counterfactual samples m required to reliably detect instability in hallucinated outputs, using concentration inequalities and variance estimates of the causal influence distribution.

Abhivansh Gupta, Simardeep Singh, Advika Sinha +2

Data Science Group, IIT Roor- kee, Roorkee, Uttarakhand, India.

Hallucination Detection in VLMsVLM RobustnessCounterfactual Evaluation of VLMsVLM HallucinationVLM Interpretability
AskPDF
Jun 5, 2026

VisualFLIP: Do Predictions Depend on Task-Critical Visual Evidence in Multimodal Reasoning?

When a multimodal large language model answers a visual reasoning question correctly, is the prediction actually supported by the task-critical visual evidence? Correct answers can coexist with flawed reasoning, making accuracy alone an incomplete test of grounding. We introduce VisualFLIP, a paired benchmark with 1,374 images arranged as same-question perturbation pairs across cardinality, attribute, spatial, and logic tasks. Each pair keeps the question fixed but minimally changes the evidence so the gold answer deterministically flips. We evaluate 24 MLLMs with pair accuracy, which requires solving both sides of a pair, and Collapse Rate (CR), which measures how often a model that solves at least one side repeats the same non-empty answer for both images. Together, these metrics show that paired correctness and evidence dependence are related but distinct: capable models can still fail to update after task-critical visual changes, and collapse becomes more severe for some models when the edited image follows an earlier answer in a sequential setting. Further details are available on our project page: https://didizhu-judy.github.io/VisualFLIP/

Didi Zhu, Changrui Chen, Stefanos Zafeiriou +1

Imperial College London

VLM EvaluationMultimodal GroundingVLM RobustnessCounterfactual Evaluation of VLMsMultimodal ReasoningVisually Grounded Reasoning+3
AskPDF
May 26, 2026

Chartographer: Counterfactual Chart Generation for Evaluating Vision-Language Models

Chart question-answering (QA) benchmarks aim to pose questions that require visual reasoning to correctly answer, but models can often reach solutions through shortcuts or prior familiarity with a chart based on their own background knowledge. To strictly evaluate visual reasoning, we propose counterfactual charts where the chart-question task remains fixed, but underlying chart and the corresponding answer are varied. We introduce Chartographer, a framework to reverse engineer charts into executable code, validate reconstruction fidelity, generate seed-controlled counterfactual variants, and derive new answers from executable QA logic. We apply this framework to existing chart QA datasets and evaluate proprietary and open-source vision-language models (VLMs), measuring variation sensitivity and generalizability. Counterfactual charts reveal failures hidden by single-chart performance: VLMs often fail to generalize after answering the original chart correctly. We find failures are most prevalent when updated charts require novel visual reasoning pathways.

Yifan Jiang, Dae Yon Hwang, Jesse C. Cresswell +1

University of Waterloo · Vector Institute · Layer 6 AI

VLM EvaluationData VisualizationCounterfactual Evaluation of VLMsChart QAVLM ReasoningVisually Grounded Reasoning
AskPDF
May 18, 2026

CounterCount: Diagnosing Knowledge Bias in How Vision-Language Models Count Attributes

Do Vision-Language Models (VLMs) count what they see, or what they know about the world? We introduce CounterCount, a diagnostic dataset of paired factual and counterfactual (CF) images, in which the CF image contains an atypical count of a subject's attribute (e.g., a rabbit with four ears), together with annotations localizing the counted attributes. Humans answer both types of images nearly perfectly (98.9%98.9\% and 97.5%97.5\%), but across 16 VLMs (14 open-weight and 2 closed-source), average accuracy drops from 91.6%91.6\% on factual to 49.5%49.5\% on CF images, with most errors stemming from responses that match the canonical count, therefore revealing a bias towards knowledge priors rather than reliance on visual information. Based on these findings, we perform an analysis that reveals two interesting insights behind this gap. First, models exhibit a ``System-1''-like response behavior, falling back on canonical counts after a shallow recognition of the subject. Second, model activations partially contain the visual evidence necessary to answer correctly, yet models underutilize it, attending less to the counted attributes when they provide biased answers. Input manipulations that break visual context, introduce cueing or reasoning, or direct amplification of the model's internal attention to relevant regions can reduce such biases. CounterCount goes beyond showing that VLMs fall back on knowledge priors, revealing that once a familiar context is recognized, models favor prior knowledge over available visual evidence rather than failing to encode that evidence.

Reem Alzahrani, Bushra Bin Hemid, Hassan Alshanqiti +4

King Abdullah University of Science and Technology, Thuwal, Saudi Arabia. · University of Edinburgh

VLM EvaluationVision-Language ModelsCounterfactual Evaluation of VLMsObject CountingVisually Grounded ReasoningAttention Mechanisms
AskPDF
May 11, 2026

C-CoT: Counterfactual Chain-of-Thought with Vision-Language Models for Safe Autonomous Driving

Safety-critical planning in complex environments, particularly at urban intersections, remains a fundamental challenge for autonomous driving. Existing methods, whether rule-based or data-driven, frequently struggle to capture complex scene semantics, infer potential risks, and make reliable decisions in rare, high-risk situations. While vision-language models (VLMs) offer promising approaches for safe decision-making in these environments, most current approaches lack reflective and causal reasoning, thereby limiting their overall robustness. To address this, we propose a counterfactual chain-of-thought (C-CoT) framework that leverages VLMs to decompose driving decisions into five sequential stages: scene description, critical object identification, risk prediction, counterfactual risk reasoning, and final action planning. Within the counterfactual reasoning stage, we introduce a structured meta-action evaluation tree to explicitly assess the potential consequences of alternative action combinations. This self-reflective reasoning establishes causal links between action choices and safety outcomes, improving robustness in long-tail and out-of-distribution scenarios. To validate our approach, we construct the DeepAccident-CCoT dataset based on the DeepAccident benchmark and fine-tune a Qwen2.5-VL (7B) model using low-rank adaptation. Our model achieves a risk prediction recall of 81.9%, reduces the collision rate to 3.52%, and lowers L2 error to 1.98 m. Ablation studies further confirm the critical role of counterfactual reasoning and the meta-action evaluation tree in enhancing safety and interpretability.

Kefei Tian, Yuansheng Lian, Kai Yang +2

College of Transportation, Tongji University, Shanghai, China · Department of Civil Engineering, Tsinghua University, Beijing 100084, China · School of Vehicle and Mobility, Tsinghua University, Beijing, China +1

VLM RobustnessCounterfactual Evaluation of VLMsAutonomous Driving Safety EvaluationAutonomous DrivingVLM ReasoningAutonomous Driving Planning
AskPDF
Apr 10, 2026

Cross-Cultural Value Attribution in Large Vision-Language Models

The rapid adoption of large vision-language models (LVLMs) in recent years has been accompanied by growing fairness concerns due to their propensity to reinforce harmful societal stereotypes. While significant attention has been paid to such fairness concerns in the context of social biases, relatively little prior work has examined the presence of stereotypes in LVLMs related to cultural contexts such as religion, nationality, and socioeconomic status. In this work, we aim to narrow this gap by investigating how LVLM judgments about a person's moral, ethical, and political values vary across cultural contexts presented in images. We conduct a multi-dimensional analysis of such value judgments in popular LVLMs using counterfactual image sets, which depict the same person across different cultural contexts. Our evaluation framework pairs descriptive analyses (Moral Foundations Theory categorization, lexical analyses, and value sensitivity) with a novel grounding analysis that compares LVLM cross-context variation against two large-scale human surveys (MFQ-2 and WVS Wave 7). Across 4.8 million LVLM generations, we identify three survey-grounding bias patterns that replicate across multiple architecturally diverse models. Additional ablations show that nationality grounding is text-dominant while religion and socioeconomic grounding depend strongly on the image, and that image conditioning can amplify survey-grounding bias patterns.

Phillip Howard, Xin Su, Kathleen C. Fraser

Thoughtworks · University of Ottawa

VLM EvaluationCounterfactual Evaluation of VLMsCultural Bias in VLMsCultural Bias in Language Models
AskPDF
Mar 31, 2026

Seeing the Evidence, Missing the Answer: Tool-Guided Vision-Language Models on Visual Illusions

Vision-language models (VLMs) exhibit a systematic bias when confronted with classic optical illusions: they overwhelmingly predict the illusion as "real" regardless of whether the image has been counterfactually modified. We present a tool-guided inference framework for the DataCV 2026 Challenge (Tasks I and II) that addresses this failure mode without any model training. An off-the-shelf vision-language model is given access to a small set of generic image manipulation tools: line drawing, region cropping, side-by-side comparison, and channel isolation, together with an illusion-type-routing system prompt that prescribes which tools to invoke for each perceptual question category. Critically, every tool call produces a new, immutable image resource appended to a persistent registry, so the model can reference and compose any prior annotated view throughout its reasoning chain. Rather than hard-coding illusion-specific modules, this generic-tool-plus-routing design yields strong cross-structural generalization: performance remained consistent from the validation set to a test set containing structurally unfamiliar illusion variants (e.g., Mach Bands rotated from vertical to horizontal stacking). We further report three empirical observations that we believe warrant additional investigation: (i) a strong positive-detection bias likely rooted in imbalanced illusion training data, (ii) a striking dissociation between pixel-accurate spatial reasoning and logical inference over self-generated annotations, and (iii) pronounced sensitivity to image compression artifacts that compounds false positives.

Xuesong Wang, Harry Wang

Wayne State University · University of Michigan

VLM RobustnessReasoning for VLMsTool Use in VLMsCounterfactual Evaluation of VLMs
AskPDF
Mar 30, 2026

Prompts Without Evidence: How Neuroimaging Mentions Shift Clinical Vision-Language Model Predictions

Trustworthy clinical AI must use real evidence and avoid relying on surface-level artifacts. We evaluate 12 open-weight vision-language models (VLMs) on two clinical neuroimaging cohorts for binary classification of affective disorders and cognitive decline. Both cohorts include structural magnetic resonance imaging (MRI) acquired under their original research protocols. Prior work does not establish the included neuroimaging inputs as reliable stand-alone diagnostic evidence for the present tasks. Nevertheless, when neuroimaging context is introduced, smaller VLMs gain up to 0.66 F1 under the evaluated augmented conditions, becoming competitive with models an order of magnitude larger. Confidence estimation shows that most of the calibration improvement for the analyzed smaller models occurs after the MRI reference is added to the prompt, before any image is supplied. Our preliminary expert case study finds that faithfulness remains low in every condition examined, with the reviewed model introducing unverified clinical details. Finally, in our single-model intervention, preference alignment suppresses MRI-referencing behavior but reduces the augmented-condition advantage, leaving the underlying issue unresolved. These results caution against reading surface metric gains as evidence of true multimodal integration, with direct implications for clinical VLM deployment.

Doan Nam Long Vu, Simone Balloccu

Natural Language Processing for Expert Domains (ExpNLP), Technical University of Darmstadt

Medical VLMsPrompt SensitivityFaithfulness in VLMsVLM CalibrationCounterfactual Evaluation of VLMs
AskPDF
Mar 30, 2026

CDH-Bench: A Commonsense-Driven Hallucination Benchmark for Evaluating Visual Fidelity in Vision-Language Models

Vision-language models (VLMs) achieve strong performance on many benchmarks, yet a basic reliability question remains underexplored: when visual evidence conflicts with commonsense, do models follow what is shown or what commonsense suggests? A characteristic failure in this setting is that the model overrides visual evidence and outputs the commonsense alternative. We term this phenomenon commonsense-driven hallucination (CDH). To evaluate it, we introduce CDH-Bench, a benchmark designed to create explicit visual evidence--commonsense conflicts. CDH-Bench covers three dimensions: counting anomalies, relational anomalies, and attribute anomalies. We evaluate frontier VLMs under binary Question Answering (QA) and multiple-choice QA, and report metrics including Counterfactual Accuracy (CF-Acc), Commonsense Accuracy (CS-Acc), Counterfactual Accuracy Drop (CFAD), Commonsense Collapse Rate (CCR), and Relative Prior Dependency (RPD). Results show that even strong models remain vulnerable to prior-driven normalization under visual evidence--commonsense conflict. CDH-Bench provides a controlled diagnostic of visual fidelity under visual evidence--commonsense conflict.

Kesheng Chen, Yamin Hu, Qi Zhou +2

Guangdong Provincial Key Laboratory of Novel Security Intelligence Technologies, Institute of Cyberspace Security, School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen, China

Emotional Hallucination in VLMsVLM HallucinationVLM EvaluationFaithfulness in VLMsVisual Evidence Sensitivity in VLMsCounterfactual Evaluation of VLMs
AskPDF
Mar 22, 2026

CVT-Bench: Probing Spatial-State Integrity through Counterfactual Viewpoint Transformations

Multimodal large language models (MLLMs) perform strongly on isolated spatial tasks, but whether their predictions remain persistent and mutually coherent across viewpoints and competing scenes is unclear. We formalize this behavioral property as spatial-state integrity and introduce CVT-Bench, a factorial diagnostic suite spanning two domains (CVT-Synthetic, CVT-Real), four context regimes (CVT-Isolated, irrelevant filler, attribute filler, CVT-Competing), three representations (Image, Text/BBox, Scene Graph), and ten viewpoint conditions (nine azimuths from 0∘0^\circ to 360∘360^\circ at 45∘45^\circ increments, plus Top) with target views withheld. Across 200 scenes and 11,833 relational queries, we measure counterfactual accuracy, cycle consistency, the normalized survival coefficient, spatial realizability, and context-specific interference. Five state-of-the-art MLLMs exhibit rapid persistence loss and frequently produce jointly unrealizable states despite high local accuracy, with failures amplified by competing contexts and natural-scene complexity. Matched irrelevant and attribute filler controls show that prompt length, position, or loss of scene access alone are insufficient explanations. On average, Text/BBox improves accuracy, persistence, and realizability, while Scene Graph provides complementary gains; neither eliminates the instability. Thus, isolated spatial accuracy substantially overestimates robustness, establishing spatial-state integrity as a distinct evaluation target. The complete benchmark and codebase will be publicly released.

Shanmukha Vellamcheti, Uday Kiran Kothapalli, Disharee Bhowmick +1

Department of Computer Science and Software Engineering, Auburn University Auburn, AL 36849, USA

Spatial Reasoning BenchmarksMulti-View ConsistencyVLM RobustnessCounterfactual Evaluation of VLMs
AskPDF
Mar 20, 2026

FREAK: A Fine-grained Hallucination Evaluation Benchmark for Advanced MLLMs

Multimodal Large Language Models (MLLMs) suffer from hallucinations. Existing hallucination evaluation benchmarks are often limited by over-simplified tasks leading to saturated metrics, or insufficient diversity that fails to adequately assess the hallucination extent in state-of-the-art multimodal models. To address this gap, we propose FREAK, a comprehensive multimodal benchmark designed for fine-grained hallucination assessment in MLLMs. Through high-quality photorealistic images featuring fine-grained counter-commonsense edits, FREAK innovatively evaluates hallucination phenomena in detailed visual perception of MLLMs. Extensive experiments on FREAK show severe hallucination issues in SOTA models regarding detailed visual perception. To enable deeper investigation, we curate a controlled subset to indirectly evaluate the model's ability to perceive target detailed information. Through systematic evaluation of prevailing Chain-of-Thought (CoT) prompting techniques within this task, we reveal critical insights regarding hallucination patterns and model reasoning processes.

Zhihan Yin, Jianxin Liang, Yueqian Wang +3

Wangxuan Institute of Computer Technology, Peking University · State Key Laboratory of General Artificial Intelligence

VLM HallucinationVLM EvaluationCounterfactual Evaluation of VLMs
AskPDF
Mar 19, 2026

To See or To Please: Uncovering Visual Sycophancy and Split Beliefs in VLMs

When VLMs answer correctly, do they genuinely rely on visual information? We introduce a Tri-Layer Diagnostic Framework with three per-sample metrics: Latent Anomaly Detection, Visual Necessity Score, and Competition Score, which disentangle perception, dependency, and alignment failures. Across 9 VLMs and 9,000 model-sample pairs under counterfactual blind, noise, and conflict interventions, 72.9% of samples exhibit Visual Sycophancy, a Split Beliefs pattern in which internal evidence is preserved yet a hallucinated answer is decoded, while zero samples show Robust Refusal, indicating that current alignment training has eliminated refusal as a decoding outcome. Scaling within the Qwen-VL family, both within- and across-generation, monotonically reduces Language Shortcuts but amplifies Visual Sycophancy, showing that scale and newer post-training alone cannot resolve the grounding problem. Diagnostic scores further enable a training-free selective-prediction strategy yielding up to +9.5 percentage points accuracy at 50% coverage.

Rui Hong, Shuxue Quan

George Mason University, Fairfax, VA, USA · Independent Researcher

Selective Prediction for VLMsSycophancy in VLMsVision-Language ModelsVision-Language GroundingVisual Evidence Sensitivity in VLMsHallucination Detection in VLMs+1
AskPDF
Mar 8, 2026

Scaling Test-Time Robustness of Vision-Language Models via Self-Critical Inference Framework

The emergence of Large Language Models (LLMs) has driven rapid progress in multi-modal learning, particularly in the development of Large Vision-Language Models (LVLMs). However, existing LVLM training paradigms place excessive reliance on the LLM component, giving rise to two critical robustness challenges: language bias and language sensitivity. To address both issues simultaneously, we propose a novel Self-Critical Inference (SCI) framework that extends Visual Contrastive Decoding by conducting multi-round counterfactual reasoning through both textual and visual perturbations. This process further introduces a new strategy for improving robustness by scaling the number of counterfactual rounds. Moreover, we also observe that failure cases of LVLMs differ significantly across models, indicating that fixed robustness benchmarks may not be able to capture the true reliability of LVLMs. To this end, we propose the Dynamic Robustness Benchmark (DRBench), a model-specific evaluation framework targeting both language bias and sensitivity issues. Extensive experiments show that SCI consistently outperforms baseline methods on DRBench, and that increasing the number of inference rounds further boosts robustness beyond existing single-step counterfactual reasoning methods.

Kaihua Tang, Jiaxin Qi, Jinli Ou +2

Tongji University, China · Computer Network Information Center, CAS, China · University of Chinese Academy of Sciences, China +1

VLM RobustnessTest-Time Scaling for VLMsVLM ReasoningCounterfactual Evaluation of VLMsVLM EvaluationContrastive Decoding
AskPDF
Mar 3, 2026

VB: Visibility Benchmark for Visibility and Perspective Reasoning in Images

We present VB, a benchmark that tests whether vision-language models can determine what is and is not visible in a photograph, and abstain when a human viewer cannot reliably answer. Each item pairs a single photo with a short yes/no visibility claim; the model must output VISIBLY_TRUE, VISIBLY_FALSE, or ABSTAIN, together with a confidence score. Items are organized into 100 families using a 2x2 design that crosses a minimal image edit with a minimal text edit, yielding 300 headline evaluation cells. Unlike prior unanswerable-VQA benchmarks, VB tests not only whether a question is unanswerable but why (via reason codes tied to specific visibility factors), and uses controlled minimal edits to verify that model judgments change when and only when the underlying evidence changes. We score models on confidence-aware accuracy with abstention (CAA), minimal-edit flip rate (MEFR), confidence-ranked selective prediction (SelRank), and second-order perspective reasoning (ToMAcc); all headline numbers are computed on the strict XOR subset (three cells per family, 300 scored items per model). We evaluate nine models spanning flagship and prior-generation closed-source systems, and open-source models from 8B to 12B parameters. GPT-4o and Gemini 3.1 Pro effectively tie for the best composite score (0.728 and 0.727), followed by Gemini 2.5 Pro (0.678). The best open-source model, Gemma 3 12B (0.505), surpasses one prior-generation closed-source system. Text-flip robustness exceeds image-flip robustness for six of nine models, and confidence calibration varies substantially: GPT-4o and Gemini 2.5 Pro achieve similar accuracy yet differ sharply in selective prediction quality.

Neil Tripathi

New York University

VLM EvaluationVLM ReasoningCounterfactual Evaluation of VLMs
AskPDF
Mar 3, 2026

Beyond Accuracy: Evaluating Visual Grounding In Multimodal Medical Reasoning

Recent work shows that text-only reinforcement learning with verifiable rewards (RLVR) can match or outperform image-text RLVR on multimodal medical VQA benchmarks, suggesting current evaluation protocols may fail to measure causal visual dependence. We introduce a counterfactual evaluation framework using real, blank, and shuffled images across four medical VQA benchmarks: PathVQA, PMC-VQA, SLAKE, and VQA-RAD. Beyond accuracy, we measure Visual Reliance Score (VRS), Image Sensitivity (IS), and introduce Hallucinated Visual Reasoning Rate (HVRR) to detect cases where models generate visual claims despite producing image-invariant answers. Our findings reveal that RLVR improves accuracy while degrading visual grounding: text-only RLVR achieves negative VRS on PathVQA (-0.09), performing better with mismatched images, while image-text RLVR reduces image sensitivity to 39.8% overall despite improving accuracy. On VQA-RAD, both variants achieve 63% accuracy through different mechanisms: text-only RLVR retains 81% performance with blank images, while image-text RLVR shows only 29% image sensitivity. Models generate visual claims in 68-74% of responses, yet 38-43% are ungrounded (HVRR). These findings demonstrate that accuracy-only rewards enable shortcut exploitation, and progress requires grounding-aware evaluation protocols and training objectives that explicitly enforce visual dependence.

Anas Zafar, Leema Krishna Murali, Ashish Vashist

The University of Texas MD Anderson Cancer Center · Cohere Labs Community · Eisai Inc. +1

Visual GroundingMedical VQACounterfactual Evaluation of VLMsRL for VLMsHallucination Detection in VLMs
AskPDF
Mar 2, 2026

Cultural Counterfactuals: Evaluating Cultural Biases in Large Vision-Language Models with Counterfactual Examples

Large Vision-Language Models (LVLMs) have grown increasingly powerful in recent years, but can also exhibit harmful biases. Prior studies investigating such biases have primarily focused on demographic traits related to the visual characteristics of a person depicted in an image, such as their race or gender. This has left biases related to cultural differences (e.g., religion, socioeconomic status), which cannot be readily discerned from an individual's appearance alone, relatively understudied. A key challenge in measuring cultural biases is that determining which group an individual belongs to often depends upon cultural context cues in images, and datasets annotated with cultural context cues are lacking. To address this gap, we introduce Cultural Counterfactuals: a high-quality synthetic dataset containing nearly 60k counterfactual images for measuring cultural biases related to religion, nationality, and socioeconomic status. To ensure that cultural contexts are accurately depicted, we generate our dataset using an image-editing model to place people of different demographics into real cultural context images. This enables the construction of counterfactual image sets which depict the same person in multiple different contexts, allowing for precise measurement of the impact that cultural context differences have on LVLM outputs. We demonstrate the utility of Cultural Counterfactuals for quantifying cultural biases in popular LVLMs.

Phillip Howard, Xin Su, Kathleen C. Fraser

Thoughtworks · University of Ottawa

VLM EvaluationCounterfactual Evaluation of VLMsLarge Vision-Language ModelsCultural Bias in VLMs
AskPDF
Feb 27, 2026

Half-Truths Break Similarity-Based Retrieval

When a text description is extended with an additional detail, image-text similarity should drop if that detail is wrong. We show that CLIP-style dual encoders often violate this intuition: appending a plausible but incorrect object or relation to an otherwise correct description can increase the similarity score. We call such cases half-truths. On COCO, CLIP prefers the correct shorter description only 34.0% of the time, and performance drops to 27.8% when the added detail is a relation. We trace this vulnerability to weak supervision on caption parts: contrastive training aligns full sentences but does not explicitly enforce that individual entities and relations are grounded. We propose CS-CLIP (Component-Supervised CLIP), which decomposes captions into entity and relation units, constructs a minimally edited foil for each unit, and fine-tunes the model to score the correct unit above its foil while preserving standard dual-encoder inference. CS-CLIP raises half-truth accuracy to 76.4% and improves average performance on established compositional benchmarks by 5.8 points, suggesting that reducing half-truth errors aligns with broader gains in compositional understanding.

Bora Kargi, Arnas Uselis, Seong Joon Oh

University of Tübingen, Tübingen AI Center · Konrad Zuse School of Excellence in Learning and Intelligent Systems (ELIZA)

Contrastive Learning for VLMsVision-Language ModelsVision-Language RetrievalVision-Language AlignmentCompositional Representation LearningCounterfactual Evaluation of VLMs
AskPDF
Feb 19, 2026

When Vision Overrides Language: Evaluating and Mitigating Counterfactual Failures in VLAs

Vision-Language-Action models (VLAs) promise to ground language instructions in robot control, yet in practice often fail to faithfully follow language. When presented with instructions that lack strong scene-specific supervision, VLAs suffer from counterfactual failures: they act based on vision shortcuts induced by dataset biases, repeatedly executing well-learned behaviors and selecting objects frequently seen during training regardless of language intent. To systematically study it, we introduce LIBERO-CF, the first counterfactual benchmark for VLAs that evaluates language following capability by assigning alternative instructions under visually plausible LIBERO layouts. Our evaluation reveals that counterfactual failures are prevalent yet underexplored across state-of-the-art VLAs. We propose Counterfactual Action Guidance (CAG), a simple yet effective dual-branch inference scheme that explicitly regularizes language conditioning in VLAs. CAG combines a standard VLA policy with a language-unconditioned Vision-Action (VA) module, enabling counterfactual comparison during action selection. This design reduces reliance on visual shortcuts, improves robustness on under-observed tasks, and requires neither additional demonstrations nor modifications to existing architectures or pretrained models. Extensive experiments demonstrate its plug-and-play integration across diverse VLAs and consistent improvements. For example, on LIBERO-CF, CAG improves π0.5π_{0.5} by 9.7% in language following accuracy and 3.6% in task success on under-observed tasks using a training-free strategy, with further gains of 15.5% and 8.5%, respectively, when paired with a VA model. In real-world evaluations, CAG reduces counterfactual failures of 9.4% and improves task success by 17.2% on average.

Yu Fang, Yuchun Feng, Dong Jing +5

University of North Carolina at Chapel Hill

VLM RobustnessCounterfactual Evaluation of VLMsRobotic ManipulationVision-Language-Action ModelsLanguage-Conditioned Robot ControlRobustness of VLA Models
AskPDF