Research map / Vision-language and multimodal / Multimodal LLMs
VAGUE: Visual Contexts Clarify Ambiguous Expressions
Heejeong Nam, Jinwoo Ahn, Keummin Ka, Jiwan Chung, Youngjae 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 raw image input (VLM) instead of text only (LM) (ablation) | Qwen2.5-VL-Instruct (7B) · VAGUE-VCR | accuracy (%): 11.1 -> 46.8 | Improved |
| add image captions (SM) instead of text only (LM) (ablation) | GPT-4o · VAGUE-VCR | accuracy (%): 46.4 -> 69.5 | Improved |
| give raw images (VLM) instead of captions (SM) to proprietary models (ablation) | GPT-4o · VAGUE-VCR | accuracy (%): 69.5 -> 65.1 | Hurt |
| give raw images (VLM) instead of captions (SM) to open-source models (ablation) | Qwen2.5-VL-Instruct (72B) · VAGUE-VCR | accuracy (%): 55.6 -> 74.2 | Improved |
| add image or caption input for Ovis2 (16B) (ablation) | Ovis2 (16B) · VAGUE-VCR | accuracy (%): 21.9 -> 24.5 | No clear effect |
| add zero-shot chain-of-thought prompting to raw image input (ablation) | GPT-4o · VAGUE (CoT subset) | accuracy (%): 64.6 -> 66.4 | Improved |
| add zero-shot chain-of-thought prompting to raw image input (ablation) | Gemini-1.5-Pro · VAGUE (CoT subset) | accuracy (%): 60.6 -> 64.4 | Improved |
| add zero-shot chain-of-thought prompting to caption input (ablation) | GPT-4o and Gemini-1.5-Pro · VAGUE (CoT subset) | accuracy (%): GPT-4o 68.9 -> 69.5; Gemini 61.8 -> 61.0 | No clear effect |
| compare best model with human on 400-question subset | VLMs · VAGUE 400-question subset | accuracy (%): Human 94.0 vs Qwen2.5-VL-72B 72.3 | Hurt |
| model errors by type (superficial understanding vs scene hallucination) | VLMs · VAGUE 400-question subset | incorrect choice counts: SU 183 vs FS 43 vs NE 10 (LLaVA-Onevision 7B) | Mixed |
Closest papers by meaning
- ConTextual: Evaluating Context-Sensitive Text-Rich Visual Reasoning in Large Multimodal Models (ICML 2024 · similarity 0.89)
- Machine Mental Imagery: Empower Multimodal Reasoning with Latent Visual Tokens (CVPR 2026 · similarity 0.89)
- VisuRiddles: Fine-grained Perception is a Primary Bottleneck for Multimodal Large Language Models in Abstract Visual Reasoning (ICLR 2026 · similarity 0.88)
- Ref-Adv: Exploring MLLM Visual Reasoning in Referring Expression Tasks (ICLR 2026 · similarity 0.88)
- Look on Demand: A Cognitive Scheduling Framework for Visual Evidence Acquisition in Multimodal Reasoning (ICML 2026 · similarity 0.88)
- ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning (ICLR 2026 · similarity 0.88)
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.
