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
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 SEVEX-discovered visual prompt instead of naive inference | Gemini-2.5-flash · BlindTest+BLINK average | accuracy: 71.6 -> 78.9 | Improved |
| Use zero-shot SketchPad tool-use instead of naive | Gemini-2.5-flash · BlindTest average | accuracy: 65.6 -> 47.4 | Hurt |
| Use SketchPad on LineIntersections | Gemini-2.5-flash · LineIntersections | accuracy: 73.0 -> 33.3 | Hurt |
| SEVEX vs SketchPad+APE exploration cost | exploration · 9-task average | exploration cost: 738k -> 85k tokens per iteration | Improved |
| SEVEX vs SketchPad+APE generalization gap | exploration · 9-task average | dev-test accuracy gap: SketchPad+APE 67.8 -> 57.7; SEVEX 83.3 -> 78.9 | Improved |
| Run SEVEX on GPT-4o | GPT-4o · BlindTest average | accuracy: 38.8 -> 62.4 | Improved |
| Run SEVEX on Claude Sonnet 4.0 | Claude Sonnet 4.0 · BlindTest average | accuracy: 66.8 -> 75.2 | Improved |
| SketchPad vs SEVEX on Claude CircledLetter | Claude Sonnet 4.0 · CircledLetter | accuracy: SketchPad 90.7 vs SEVEX 80.5 | Mixed |
| Transfer visual prompt found with GPT to Claude | visual prompt · LineIntersections | accuracy: 66.3 -> 35.0 | Hurt |
| Transfer Claude-found prompt to GPT-4o | visual prompt · LineIntersections | accuracy: GPT-found 57.1 vs Claude-found 31.0 on GPT-4o; naive 8.2 | Mixed |
| Remove tree structure, semantic backpropagation, or abstract idea space (ablation) | SEVEX · Jigsaw and CircledLetter exploration | dev accuracy over iterations | Improved |
| Restrict toolbox to basic and drawing primitives (ablation) | SEVEX · BlindTest/BLINK | accuracy | No clear effect |
Closest papers by meaning
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- VP-Bench: A Comprehensive Benchmark for Visual Prompting in Multimodal Large Language Models (AAAI 2026 · similarity 0.89)
- Draw-and-Understand: Leveraging Visual Prompts to Enable MLLMs to Comprehend What You Want (ICLR 2025 · similarity 0.89)
- SenseSearch: Empowering Vision-Language Models with High-Resolution Agentic Search-Reasoning via Reinforcement Learning (CVPR 2026 · similarity 0.89)
- Exploring Task-Level Optimal Prompts for Visual In-Context Learning (AAAI 2025 · similarity 0.88)
- ViP-LLaVA: Making Large Multimodal Models Understand Arbitrary Visual Prompts (CVPR 2024 · 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.
