Visual Evidence Sensitivity in VLMs

Spatial reasoning is fundamental to robotics, autonomy, and embodied AI, yet modern vision-language models (VLMs) remain unreliable on metric distance queries. A common assumption is that consistent predictions across viewpoints reflect geometric grounding. We test this assumption and find the opposite: leading VLMs often produce view-invariant and consistent answers even when those answers are incorrect, indicating weak coupling between predictions and viewpoint-specific visual evidence. We introduce ViewDiag, a controlled multi-view evaluation protocol built from Hypersim, ScanNet, and KITTI360, comprising 176 object-pair tracks across 80 scenes with 2--10 views per track. The protocol evaluates models along three axes: metric accuracy, distributional concentration, and internal collapse, the last of which is assessed using a latent feature probe. Across diverse models, we observe a consistent pattern of high prediction stability paired with substantial error, clustering in a regime characterized by strong consistency but low accuracy. \noindent These results challenge the common use of cross-view consistency as a proxy for geometric understanding. Instead, we show that stable predictions may reflect prior-driven collapse rather than evidence-sensitive reasoning. ViewDiag provides a controlled benchmark and diagnostic framework for evaluating whether spatial VLMs are not only accurate, but also meaningfully coupled to visual evidence.

Visual inputs are often assumed to improve language understanding in multimodal models. We examine this assumption by asking whether vision-language models (VLMs) can distinguish useful visual evidence from incidental image context in lexical judgments. We use human concreteness and imagery ratings because they span words with varying expected visual relevance, from abstract and low-imagery words to concrete and high-imagery words. We find that real-image contexts do not yield consistent gains and often hurt alignment with human ratings, most sharply when visual evidence is least relevant. Through probing and canonical correlation analysis, complemented by an attribution case study, we find that real-image contexts are associated with representational shifts and greater sensitivity to spurious visual cues, coinciding with weaker recoverability of the targeted lexical properties. We further show that instructing models to focus solely on textual content at inference time can reduce this degradation, with the clearest gains on these vulnerable subsets. Our findings suggest that current instruction-tuned VLMs need better calibration of when visual context should inform lexical judgments.

Benchmark accuracy is often implicitly assumed to reflect grounded visual understanding in vision-language models (VLMs), yet it remains unclear to what extent such scores truly reflect reliance on visual evidence. Motivated by a surprising observation that removing a substantial fraction of image tokens only degrades model performance very slightly on a widely used hallucination benchmark, we systematically investigate this mismatch in a set of open-source VLMs. Our analysis spans multiple levels of granularity, spanning global visual degradation, localized occlusion, question reformulation, answer-space expansion, and decision-level analyses beyond standard accuracy. We further complement these behavioral results with a layer-wise analysis of vision-token geometry. Throughout the experiments, we find that although VLMs do incorporate visual input, their predictions are less sensitive to the loss of fine-grained visual evidence that standard accuracy should have suggested. Even when the final prediction remains unchanged, the model's internal support for the correct answer may already be weakened. We further complement a representation-level analysis, which shows increasing similarity among visual tokens in deeper layers, providing a possible explanation for our findings. Together, these results suggest that current benchmarks are not sufficient to reliably evaluate fine-grained visual grounding in VLMs.

Vision-Language Models (VLMs) often produce self-reflective statements like "let me check the figure again" during reasoning. Do such statements trigger genuine visual re-examination, or are they merely learned textual patterns? We investigate this via VisualSwap, an image-swap probing framework: after a model reasons over an image, we replace it with a visually similar but semantically different one and test whether the model notices. We introduce VS-Bench, 800 image pairs curated from MathVista, MathVerse, MathVision, and MMMU-Pro. Experiments on Qwen3-VL, Kimi-VL, and ERNIE-VL reveal a striking failure: models overwhelmingly miss the swap, with accuracy dropping by up to 60%. Counterintuitively, thinking models are nearly 3x more vulnerable than their instructed counterparts, and scaling offers no mitigation. Multi-turn user instructions restore visual grounding, but self-generated reflective statements during continuous generation do not. Attention analysis explains why: user instructions substantially elevate attention to visual tokens, whereas self-reflection does not. Current VLMs tend to say rather than actually see when claiming to perform visual re-examination. Our code and dataset are available at the project page: https://visualswap.github.io

Vision-Language Models (VLMs) achieve strong cross-modal performance, yet recent evidence suggests they over-rely on textual descriptions while under-utilizing visual evidence -- a phenomenon termed ``text shortcut learning.'' We propose an adversarial evaluation framework that quantifies this cross-modal dependency by measuring accuracy degradation (Drop) when semantically conflicting text is paired with unchanged images. Four adversarial strategies -- shape_swap, color_swap, position_swap, and random_text -- are applied to a controlled geometric-shapes dataset ($n{=}1{,}000$). We compare three configurations: Baseline CLIP (ViT-B/32), LoRA fine-tuning, and LoRA Optimized (integrating Hard Negative Mining, Label Smoothing, layer-wise learning rates, Cosine Restarts, curriculum learning, and data augmentation). The optimized model reduces average Drop from 27.5% to 9.8% (64.4% relative improvement, $p{<}0.001$) while maintaining 97% normal accuracy. Attention visualization and embedding-space analysis confirm that the optimized model attends more to visual features and achieves tighter cross-modal alignment.

When a Vision-Language Model (VLM) sees a blue banana and answers "yellow", is the problem of perception or arbitration? We explore the question in ten VLMs with various sizes and reveal an Encoding-Grounding Dissociation: models that fail to report what they see (and thus provide a wrong answer) still encode the visual evidence as strongly as models that provide the correct answer. Using Multimodal Arbitration Crossover (MAC) analysis with layer-by-layer Logit Lens probing, we track the competition between visual and prior signals across every layer of each model. We show that visual attributes can be linearly decodable from early layers (AUC > 0.86). The accuracy remains nearly identical for both successful and failed samples. However, the gap in the final-layer logit - not the strength of encoding - better predicts grounding outcomes with a correlation of $ρ=$ 0.847. After having studied when VLMs base their answers on image clues rather than prior knowledge, we want to understand the causal relationships. We establish causality through full-sequence activation patching. The standard last-token interventions in LLM interpretability do not affect VLMs. In contrast, replacing the full token sequence at layers identified by MAC alters 60 to 84% of outputs. Partial-token decomposition shows that image tokens carry almost all of the causal impact, while text tokens have none. Scaling addresses the remaining architectural differences to achieve perfect retention. Moving from diagnosis to intervention, we show that training-free activation steering - both linear and sparse autoencoder-guided - in early layers can improve visual grounding by up to +3.8% with degrading performance in some setups. Overall, these findings lead to a clear conclusion: VLMs already see well, but the challenge is acting on what they see. Targeted interventions can help to bridge this gap.

Vision-Language Models (VLMs) are increasingly proposed for autonomous driving tasks, yet their performance on sequential driving scenes remains poorly characterized, particularly regarding how input configurations affect their capabilities. We introduce VENUSS (VLM Evaluation oN Understanding Sequential Scenes), a framework for systematic sensitivity analysis of VLM performance on sequential driving scenes, establishing baselines for future research. Building upon existing datasets, VENUSS extracts temporal sequences from driving videos, and generates structured evaluations across custom categories. By comparing 25+ existing VLMs across 2,600+ scenarios, we reveal how even top models achieve only 57% accuracy, not matching human performance under similar constraints (65%) and exposing significant capability gaps. Our analysis shows that VLMs excel with static object detection but struggle with understanding vehicle dynamics and temporal relations. VENUSS offers the first systematic sensitivity analysis of VLMs focused on how input image configurations - resolution, frame count, temporal intervals, spatial layouts, and presentation modes - affect performance on sequential driving scenes. Supplementary material available at https://TUM-AVS.github.io/VENUSS/.

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.

Vision-Language Models (VLMs) have been shown to be blind, often underutilizing their visual inputs even on tasks that require visual reasoning. In this work, we demonstrate that VLMs are selectively blind. They modulate the amount of attention applied to visual inputs based on linguistic framing even when alternative framings demand identical visual reasoning. Using visual attention as a probe, we quantify how framing alters both the amount and distribution of attention over the image. Constrained framings, such as multiple choice and yes/no, induce substantially lower attention to image context compared to open-ended, reduce focus on task-relevant regions, and shift attention towards uninformative tokens. We further demonstrate that this attention misallocation is the principal cause of degraded accuracy and cross-framing inconsistency. Building on this mechanistic insight, we introduce a lightweight prompt-tuning method using learnable tokens that encourages the robust, visually grounded attention patterns observed in open-ended settings, improving visual grounding and improving performance across framings.

When Vision-Language Models (VLMs) answer correctly, do they rely on the image? We introduce a Tri-Layer Diagnostic Framework that asks three questions of every (image, question) pair when the image is blanked: does the model register that the image is missing (Latent Anomaly Detection, LAD), does its output distribution depend on the image (Visual Necessity Score, VNS), and does it nonetheless commit to a content answer? The cascade assigns each pair of any existing benchmark to one of four modes: Perceptual Blindness, Language Shortcut, Visual Sycophancy (the blank image registers and matters, yet the model commits to a fabricated answer), or Robust Refusal. Across nine open-weight VLMs and 9,000 model-sample pairs, 28.1% are Visual Sycophancy, 18.6% Language Shortcut, 5.6% Perceptual Blindness, and 47.8% Robust Refusal. The modes are strongly model-specific: the Qwen3-VL models refuse on blank input in 93-98% of pairs, whereas Pixtral-12B commits to an answer in 72%. Within-generation scaling increases Visual Sycophancy in both Qwen2.5-VL (7B->72B: 25.5->40.1%) and Llama-3.2-Vision (11B->90B: 34.3->48.0%), while the newer Qwen3-VL generation nearly eliminates it. A blinded human study confirms that LAD and VNS track what annotators observe, and that 80% of Visual-Sycophancy responses carry no textual sign that the image was blank.

While chain-of-thought (CoT) reasoning has substantially improved multimodal large language models (MLLMs) on complex reasoning tasks, existing approaches largely rely on long textual reasoning trajectories and provide limited mechanisms for learning stable visual attention policies. Our analysis shows that current MLLMs exhibit weak visual focus: early-stage visual misalignment is rarely corrected during subsequent reasoning, leading to error propagation and failed inferences. We argue that this limitation stems from inadequate credit assignment for visual attention during training. To address this issue, we propose SAYO, a visual reasoning model trained with a reinforcement learning (RL) framework that introduces a region-level visual attention-based reward. This reward explicitly aligns optimization signals with visually grounded reasoning steps, enabling the model to learn more reliable attention behaviors. Extensive experiments across multiple multimodal benchmarks demonstrate that SAYO consistently improves performance on diverse reasoning and perception tasks. Project is open at https://github.com/Cratileo/Sayo.

Recent large vision-language models (LVLMs) can generate vision-text multimodal chain-of-thought (MCoT) traces after reinforcement fine-tuning (RFT). However, we observe that the visual information in MCoT is often inaccurate even when answers are correct, and task accuracy remains nearly unchanged despite the visual information being corrupted. This indicates a lack of faithfulness in the vision part of MCoT reasoning. We attribute this to the RL reward design in RFT, which solely incentivizes the format of interleaved vision-text cues, encouraging the model to incorporate visual information into its text reasoning steps without considering its correctness. In this paper, we first probe MCoT faithfulness by measuring how much the prediction changes when its visual and textual thoughts are intervened. Surprisingly, the model's predictions remain nearly unchanged under visual intervention but change significantly under textual intervention, indicating that the visual evidence is largely ignored. To further analyze the visual information, we introduce an automated LVLM-based evaluation metric that quantifies the faithfulness of visual cues from two perspectives: relevance and sufficiency. Our evaluation reveals that the visual information in current MCoT traces is simultaneously irrelevant and insufficient. To address this issue, we propose Sufficient-Component Cause Model (SCCM) learning. This approach encourages the MCoT to generate sufficient yet minimal visual components that are independently capable of leading to the correct answer. The proposed SCCM is annotation-free and plug-and-play, compatible with various RFT for MCoT. Empirical results demonstrate that SCCM substantially improves the visual faithfulness across a suite of fine-grained perception and reasoning benchmarks. Code is available at https://github.com/EugeneLiu01/Faithful_Thinking_with_Image.

Vision-language models (VLMs) can increasingly handle extended contexts and answer questions about long videos--but can they understand the narrative structure of movies as humans do? To investigate this question, we introduce MF$^2$ (Movie Facts and Fibs), a fully manually curated benchmark designed to evaluate whether models can identify, recall, and comprehend information central to the movie's storyline, using both the visual and textual modalities. MF$^2$ includes over 50 open-licensed movies, each paired with manually constructed sets of claim pairs: one true claim (fact) and one plausible but false (fib), totaling over 850 pairs. Our analysis across 20 open-source and 2 proprietary models reveals a substantial gap relative to human performance. While subtitles provide a strong signal for narrative understanding, other textual sources (e.g., synopses) are less informative. In contrast, when visual information is distributed across the entire movie and jointly embedded with subtitles, models struggle to identify and effectively use the relevant content. However, when the visual input is restricted to claim-relevant visual content, models improve significantly, indicating substantial room for improvement in processing large multimodal contexts and distinguishing relevant from irrelevant information. Overall, our findings highlight persistent challenges in multimodal narrative understanding for current VLMs. We release our data and code to support further research in this direction.