VisualScratchpad: Inference-time Visual Concepts Analysis in Vision Language Models

• URL: http://arxiv.org/abs/2603.07335v1
• Date: Sat, 07 Mar 2026 20:52:09 GMT
• Title: VisualScratchpad: Inference-time Visual Concepts Analysis in Vision Language Models
• Abstract summary: We introduce VisualScratchpad, an interactive interface for visual concept analysis during inference.<n>We apply sparse autoencoders to the vision encoder and link the resulting visual concepts to text tokens via text-to-image attention.<n>VisualScratchpad also provides a token-latent heatmap view that suggests a sufficient set of latents for effective concept ablation in causal analysis.
• Score: 81.7389759509619
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: High-performing vision language models still produce incorrect answers, yet their failure modes are often difficult to explain. To make model internals more accessible and enable systematic debugging, we introduce VisualScratchpad, an interactive interface for visual concept analysis during inference. We apply sparse autoencoders to the vision encoder and link the resulting visual concepts to text tokens via text-to-image attention, allowing us to examine which visual concepts are both captured by the vision encoder and utilized by the language model. VisualScratchpad also provides a token-latent heatmap view that suggests a sufficient set of latents for effective concept ablation in causal analysis. Through case studies, we reveal three underexplored failure modes: limited cross-modal alignment, misleading visual concepts, and unused hidden cues. Project page: https://hyesulim.github.io/visual_scratchpad_projectpage/

Related papers

• Not Another Text Benchmark: Putting the "Visual" Back in Visual Question Answering for Large Video Models [97.0559653915626]
  We introduce three new vision-centric evaluation benchmarks in temporal frame retrieval, video future prediction, and causal memory distortion.<n>We show that current frontier models exhibit significant weakness when attempting to reason through visual queries, rather than text.
  arXiv  Detail & Related papers  (2026-09-15T12:41:08Z)
• Where To Look? : Causal Tracing of Vision Encoders in VLM [0.6980863185266349]
  Vision-language models can describe an image with remarkable accuracy, yet what visual information actually drives their answers?<n>We investigate this question through causal tracing, and we observe that highly causal vision tokens often lie outside the target region.<n>We further investigate: can these models preserve visual structure when appearance cues are removed? and find that visual cues are exploited to understand visual structures.
  arXiv  Detail & Related papers  (2026-08-11T10:17:34Z)
• How and What to Imagine? Visual Thinking in Unified Multimodal Models for Cross-View Spatial Reasoning [18.15809161363394]
  Cross-view spatial reasoning remains a weak spot for vision-language models.<n>We ask how to make visual thinking matter, and what kind of visual thinking works best.
  arXiv  Detail & Related papers  (2026-05-26T17:20:05Z)
• Chatting with Images for Introspective Visual Thinking [50.7747647794877]
  ''Chatting with images'' is a new framework that reframes visual manipulation as language-guided feature modulation.<n>Under the guidance of expressive language prompts, the model dynamically performs joint re-encoding over multiple image regions.<n>ViLaVT achieves strong and consistent improvements on complex multi-image and video-based spatial reasoning tasks.
  arXiv  Detail & Related papers  (2026-02-11T17:42:37Z)
• Thinking with Deltas: Incentivizing Reinforcement Learning via Differential Visual Reasoning Policy [75.66913260900726]
  Reinforcement Learning with Verifiable Rewards has significantly advanced reasoning capabilities in Large Language Models.<n>Existing paradigms, driven by text-centric outcome rewards, encourage models to bypass visual perception.<n>We propose textbfThinking with Deltas, a framework driven by a textbfDifferential Visual Reasoning Policy.
  arXiv  Detail & Related papers  (2026-01-11T08:25:34Z)
• Sketch-in-Latents: Eliciting Unified Reasoning in MLLMs [53.57402214935238]
  Sketch-in-Latents is a novel paradigm for unified multi-modal reasoning.<n>It generates continuous visual embeddings, termed latent sketch tokens, as visual thoughts.<n>It achieves superior performance on vision-centric tasks while exhibiting strong generalization to diverse general multi-modal benchmarks.
  arXiv  Detail & Related papers  (2025-12-18T14:29:41Z)
• ViConEx-Med: Visual Concept Explainability via Multi-Concept Token Transformer for Medical Image Analysis [5.887969742827489]
  ViConEx-Med is a transformer-based framework for visual concept explainability.<n>Our method produces concept-level localization maps while maintaining high predictive accuracy.<n> Experiments on both synthetic and real-world medical datasets demonstrate that ViConEx-Med outperforms prior concept-based models.
  arXiv  Detail & Related papers  (2025-10-11T11:24:47Z)
• Visual Graph Arena: Evaluating Visual Conceptualization of Vision and Multimodal Large Language Models [51.900488744931785]
  We introduce the Visual Graph Arena (VGA) to evaluate and improve AI systems' capacity for visual abstraction.<n>Humans achieve near-perfect accuracy across tasks, while models totally failed on isomorphism detection and showed limited success in path/cycle tasks.<n>By isolating the challenge of representation-invariant reasoning, the VGA provides a framework to drive progress toward human-like conceptualization in AI visual models.
  arXiv  Detail & Related papers  (2025-06-06T17:06:25Z)
• Sparse Autoencoders for Scientifically Rigorous Interpretation of Vision Models [27.806966289284528]
  We present a unified framework using sparse autoencoders (SAEs) to discover human-interpretable visual features.<n>We show that SAEs can reliably identify and manipulate interpretable visual features without model re-training.
  arXiv  Detail & Related papers  (2025-02-10T18:32:41Z)
• Multimodal Analysis Of Google Bard And GPT-Vision: Experiments In Visual Reasoning [0.0]
  We subject Google Bard and GPT-Vision to 64 visual tasks spanning categories like "Visual Situational Reasoning" and "Next Scene Prediction" Our findings spotlight both vision-language model's limitations.
  arXiv  Detail & Related papers  (2023-08-17T03:14:00Z)
• Localization vs. Semantics: Visual Representations in Unimodal and Multimodal Models [57.08925810659545]
  We conduct a comparative analysis of the visual representations in existing vision-and-language models and vision-only models. Our empirical observations suggest that vision-and-language models are better at label prediction tasks. We hope our study sheds light on the role of language in visual learning, and serves as an empirical guide for various pretrained models.
  arXiv  Detail & Related papers  (2022-12-01T05:00:18Z)
• Learnable Visual Words for Interpretable Image Recognition [70.85686267987744]
  We propose the Learnable Visual Words (LVW) to interpret the model prediction behaviors with two novel modules. The semantic visual words learning relaxes the category-specific constraint, enabling the general visual words shared across different categories. Our experiments on six visual benchmarks demonstrate the superior effectiveness of our proposed LVW in both accuracy and model interpretation.
  arXiv  Detail & Related papers  (2022-05-22T03:24:45Z)

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.