AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

From Hallucination to Grounding: Diagnosing Visual Spatial Intelligence via CRISP

Zhixing Li, Yinan Yu

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add visual input to text-only model (ablation)Gemini-3-Flash · CRISP QAQA score: 7.94 -> +45.47 gain (drop -45.47 without vision)Improved
add visual input to text-only model (ablation)Qwen2.5-VL-7B · CRISP QAQA score: 28.47 text-only, gain 16.31Improved
add visual input to text-only model (ablation)Qwen2.5-VL-7B · CRISP SGCSGC score: 39.46 text-only, gain 3.24Improved
add visual input to text-only model (ablation)Qwen3-VL-8B · CRISP SGCSGC score: 34.75 text-only, gain 27.60Improved
add visual input to text-only model (ablation)LLaVA-OV-1.5-8B · CRISP SGCSGC score: 43.20 text-only, gain 7.13Improved
add visual input to text-only model (ablation)Gemini-2.5-Pro · CRISP SGCSGC score: 37.05 text-only, gain 27.53Improved
inject ground-truth 3D scene graph alongside imageGemini 2.5 Flash · CRISP QAaccuracy: 30.3% surgeImproved
feed model's own predicted scene graph alongside imageVLMs (most) · CRISP QAaccuracyHurt
feed predicted scene graph alongside imageGemini 2.5 Flash · CRISP QAaccuracyImproved
inject predicted scene graph on compositional-logic questionsVLMs except LLaVA-OneVision-1.5 · CRISP compositional logicaccuracyImproved
inject ground-truth scene graph on compositional logicproprietary models · CRISP compositional logicaccuracy: >50% gainImproved
inject ground-truth scene graph on compositional logicopen-source models · CRISP compositional logicaccuracyNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy