Research map / Vision-language and multimodal / Multimodal LLMs
From Hallucination to Grounding: Diagnosing Visual Spatial Intelligence via CRISP
Zhixing Li, Yinan Yu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add visual input to text-only model (ablation) | Gemini-3-Flash · CRISP QA | QA 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 QA | QA score: 28.47 text-only, gain 16.31 | Improved |
| add visual input to text-only model (ablation) | Qwen2.5-VL-7B · CRISP SGC | SGC score: 39.46 text-only, gain 3.24 | Improved |
| add visual input to text-only model (ablation) | Qwen3-VL-8B · CRISP SGC | SGC score: 34.75 text-only, gain 27.60 | Improved |
| add visual input to text-only model (ablation) | LLaVA-OV-1.5-8B · CRISP SGC | SGC score: 43.20 text-only, gain 7.13 | Improved |
| add visual input to text-only model (ablation) | Gemini-2.5-Pro · CRISP SGC | SGC score: 37.05 text-only, gain 27.53 | Improved |
| inject ground-truth 3D scene graph alongside image | Gemini 2.5 Flash · CRISP QA | accuracy: 30.3% surge | Improved |
| feed model's own predicted scene graph alongside image | VLMs (most) · CRISP QA | accuracy | Hurt |
| feed predicted scene graph alongside image | Gemini 2.5 Flash · CRISP QA | accuracy | Improved |
| inject predicted scene graph on compositional-logic questions | VLMs except LLaVA-OneVision-1.5 · CRISP compositional logic | accuracy | Improved |
| inject ground-truth scene graph on compositional logic | proprietary models · CRISP compositional logic | accuracy: >50% gain | Improved |
| inject ground-truth scene graph on compositional logic | open-source models · CRISP compositional logic | accuracy | No clear effect |
Closest papers by meaning
- Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces (ICML 2026 · similarity 0.91)
- Spatial-IQ: Deconstructing Spatial Intelligence via Hierarchical Capability Tests (NeurIPS 2026 · similarity 0.91)
- 3ViewSense: Spatial and Mental Perspective Reasoning from Orthographic Views in Vision-Language Models (ICML 2026 · similarity 0.90)
- MMSI-Bench: A Benchmark for Multi-Image Spatial Intelligence (ICLR 2026 · similarity 0.90)
- Pursuing Minimal Sufficiency in Spatial Reasoning (ICLR 2026 · similarity 0.90)
- SpinBench: Perspective and Rotation as a Lens on Spatial Reasoning in VLMs (ICLR 2026 · similarity 0.89)
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.
