AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Visual Prompt Discovery via Semantic Exploration

Jaechang Kim, Yotaro Shimose, Zhao Wang, Kuang-Da Wang, Jungseul Ok, Shingo Takamatsu

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 0 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 SEVEX-discovered visual prompt instead of naive inferenceGemini-2.5-flash · BlindTest+BLINK averageaccuracy: 71.6 -> 78.9Improved
Use zero-shot SketchPad tool-use instead of naiveGemini-2.5-flash · BlindTest averageaccuracy: 65.6 -> 47.4Hurt
Use SketchPad on LineIntersectionsGemini-2.5-flash · LineIntersectionsaccuracy: 73.0 -> 33.3Hurt
SEVEX vs SketchPad+APE exploration costexploration · 9-task averageexploration cost: 738k -> 85k tokens per iterationImproved
SEVEX vs SketchPad+APE generalization gapexploration · 9-task averagedev-test accuracy gap: SketchPad+APE 67.8 -> 57.7; SEVEX 83.3 -> 78.9Improved
Run SEVEX on GPT-4oGPT-4o · BlindTest averageaccuracy: 38.8 -> 62.4Improved
Run SEVEX on Claude Sonnet 4.0Claude Sonnet 4.0 · BlindTest averageaccuracy: 66.8 -> 75.2Improved
SketchPad vs SEVEX on Claude CircledLetterClaude Sonnet 4.0 · CircledLetteraccuracy: SketchPad 90.7 vs SEVEX 80.5Mixed
Transfer visual prompt found with GPT to Claudevisual prompt · LineIntersectionsaccuracy: 66.3 -> 35.0Hurt
Transfer Claude-found prompt to GPT-4ovisual prompt · LineIntersectionsaccuracy: GPT-found 57.1 vs Claude-found 31.0 on GPT-4o; naive 8.2Mixed
Remove tree structure, semantic backpropagation, or abstract idea space (ablation)SEVEX · Jigsaw and CircledLetter explorationdev accuracy over iterationsImproved
Restrict toolbox to basic and drawing primitives (ablation)SEVEX · BlindTest/BLINKaccuracyNo 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