The Visual Insensitivity Gap: Diagnosing When Vision-Language Models Fail to Use Visual Evidence

• URL: http://arxiv.org/abs/2609.00868v1
• Date: Tue, 01 Sep 2026 08:01:57 GMT
• Title: The Visual Insensitivity Gap: Diagnosing When Vision-Language Models Fail to Use Visual Evidence
• Abstract summary: We name this phenomenon the Visual Insensitivity Gap and quantify it with a per-sample Visual Sensitivity Index (VSI)<n>We show this assumption fails on 40%--97% of samples across six VLMs and three perceptual benchmarks.
• Score: 1.7844382164707184
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Vision-language models are evaluated by aggregate accuracy on multimodal benchmarks, a practice that implicitly assumes the model uses its visual input. We show this assumption fails on 40%--97% of samples across six VLMs and three perceptual benchmarks: blurring the question-relevant visual region leaves the next-token distribution nearly unchanged. We name this phenomenon the Visual Insensitivity Gap and quantify it with a per-sample Visual Sensitivity Index (VSI). The gap is a property of samples, not of models: VSI ranks correlate across models (grand-mean Spearman rho=+0.40, permutation p<10^-3), so the same samples are flagged insensitive by VLMs sharing no architectural detail beyond a contrastively pretrained vision tower. The mechanism is concrete: on the insensitive samples, a linear probe on each model's own vision tower distinguishes perturbed from clean images at 0.72--0.79 accuracy, yet the model's argmax token changes on only 2%--11% of the same samples, an encoder--LLM gap above 0.65 on every model. Mapping VSI's diagnostic utility cell by cell surfaces a strong regime (multi-choice reasoning on capable VLMs: AUROC=0.85--0.87) and a weak regime (well-calibrated factuality, where softmax confidence already leads). VSI is not a universal best abstention signal; it is a sample-intrinsic indicator of vision-ignoring failure, best used as a conditional ensemble component.

Related papers

• NumerosityVLM: A Cognitively Inspired Benchmark for Interpreting Numerosity Representations in Vision-Language Models [40.71918318095227]
  Vision-factor models (VLMs) achieve strong performance on high-level multimodal tasks, yet numerosity perception remains poorly understood.<n>We introduce a cognitively inspired diagnostic benchmark, NumerosityVLM, comprising 10,800 synthetic images across six controlled conditions.
  arXiv  Detail & Related papers  (2026-08-15T22:01:21Z)
• Zero-Shot Vision-Language Models for Classroom Engagement Recognition: A Benchmark Study of Prompt Sensitivity and Cross-Dataset Generalization [0.44787896002954924]
  We present a systematic benchmark that evaluates five widely-used Vision-Language Models (VLMs)<n>Our experiments reveal three primary failure modes of zero-shot VLMs for engagement recognition.
  arXiv  Detail & Related papers  (2026-06-20T03:53:40Z)
• Consistent Yet Wrong: Evidence Insensitivity in Spatial Vision-Language Models [29.66841995436342]
  Modern vision-language models (VLMs) remain unreliable on metric distance queries.<n>We introduce textbfViewDiag, a controlled multi-view evaluation protocol built from Hypersim, ScanNet, and KITTI360.<n>We show that stable predictions may reflect prior-driven collapse rather than evidence-sensitive reasoning.
  arXiv  Detail & Related papers  (2026-06-01T18:06:08Z)
• Unveiling the Visual Counting Bottleneck in Vision-Language Models [49.591496870141846]
  This work deconstructs visual counting into three cognitive stages: visual individuation, magnitude awareness, and symbolic mapping.<n>Using synthetic Go boards and linear probes, we demonstrate that visual backbones maintain robust, linearly separable representations of quantity well into the extrapolation regime.<n>We pinpoint the collapse to the symbolic mapping stage, where the model fails to project valid visual magnitudes onto symbolic tokens.
  arXiv  Detail & Related papers  (2026-05-28T16:20:29Z)
• Seeing without Looking: Do Vision-Language Benchmarks Really Test Vision? [10.315515647818009]
  Benchmark accuracy is often assumed to reflect grounded visual understanding in vision-language models.<n>We systematically investigate this mismatch in a set of open-source vision-language models.<n>We find that although VLMs do incorporate visual input, their predictions are less sensitive to the loss of fine-grained visual evidence.
  arXiv  Detail & Related papers  (2026-05-21T17:35:04Z)
• To See or To Please: Uncovering Visual Sycophancy and Split Beliefs in VLMs [1.9349092719498848]
  Tri-Layer Diagnostic Framework disentangles hallucination sources via three metrics: Latent Anomaly Detection, Visual Necessity Score, and Competition Score.<n>Our taxonomy reveals that 69.6% of samples exhibit Visual Sycophancy--models detect visual anomalies but hallucinate to satisfy user expectations--while zero samples show Robust Refusal.
  arXiv  Detail & Related papers  (2026-03-19T00:15:05Z)
• VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models [64.56065206447788]
  Vision-language models (VLMs) achieve strong performance on standard, high-quality datasets.<n>We present VLM-RobustBench, a benchmark spanning 49 augmentation types across noise, blur, weather, digital, and geometrics.<n>We show that low-severity spatial perturbations often degrade performance more than visually severe photometric corruptions.
  arXiv  Detail & Related papers  (2026-03-06T10:58:02Z)
• Same Answer, Different Representations: Hidden instability in VLMs [65.36933543377346]
  We introduce a representation-aware and frequency-aware evaluation framework that measures internal embedding drift, spectral sensitivity, and structural smoothness.<n>We apply this framework to modern Vision Language Models (VLMs) across the SEEDBench, MMMU, and POPE datasets.
  arXiv  Detail & Related papers  (2026-02-06T12:24:26Z)
• SSVP: Synergistic Semantic-Visual Prompting for Industrial Zero-Shot Anomaly Detection [55.54007781679915]
  We propose Synergistic Semantic-Visual Prompting (SSVP), that efficiently fuses diverse visual encodings to elevate model's fine-grained perception.<n>SSVP achieves state-of-the-art performance with 93.0% Image-AUROC and 92.2% Pixel-AUROC on MVTec-AD, significantly outperforming existing zero-shot approaches.
  arXiv  Detail & Related papers  (2026-01-14T04:42:19Z)
• Fluid: Scaling Autoregressive Text-to-image Generative Models with Continuous Tokens [53.99177152562075]
  Scaling up autoregressive models in vision has not proven as beneficial as in large language models. We focus on two critical factors: whether models use discrete or continuous tokens, and whether tokens are generated in a random or fixed order using BERT- or GPT-like transformer architectures. Our results show that while all models scale effectively in terms of validation loss, their evaluation performance -- measured by FID, GenEval score, and visual quality -- follows different trends.
  arXiv  Detail & Related papers  (2024-10-17T17:59:59Z)
• Vision Transformers are Robust Learners [65.91359312429147]
  We study the robustness of the Vision Transformer (ViT) against common corruptions and perturbations, distribution shifts, and natural adversarial examples. We present analyses that provide both quantitative and qualitative indications to explain why ViTs are indeed more robust learners.
  arXiv  Detail & Related papers  (2021-05-17T02:39:22Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.