Research map / Vision-language and multimodal / Human and embodied modeling
What Makes Good Examples for Visual In-Context Learning?
Yuanhan Zhang, Kaiyang Zhou, Ziwei Liu
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 |
|---|---|---|---|
| Replace random prompt selection with unsupervised nearest-example retrieval | Visual in-context learning · Pascal-5ᶦ segmentation Avg | mIoU: 27.56 -> 33.56 | Improved |
| Replace random prompt selection with supervised prompt retrieval | Visual in-context learning · Pascal-5ᶦ segmentation Avg | mIoU: 27.56 -> 35.56 | Improved |
| Replace random prompt selection with unsupervised nearest-example retrieval | Visual in-context learning · Pascal VOC single object detection | mIoU: 25.45 -> 26.84 | Improved |
| Replace random prompt selection with supervised prompt retrieval | Visual in-context learning · Pascal VOC single object detection | mIoU: 25.45 -> 28.22 | Improved |
| Replace random prompt selection with prompt retrieval | Visual in-context learning · ImageNet-2012 colorization | mse: 0.67 -> 0.63 | Improved |
| Use supervised rather than unsupervised prompt retrieval | Visual in-context learning · Pascal-5ᶦ segmentation Avg | mIoU: 33.56 -> 35.56 | Improved |
| Use supervised rather than unsupervised prompt retrieval | Visual in-context learning · MSCOCO-5ᶦ segmentation Avg (Pascal to MSCOCO shift) | mIoU: 18.02 -> 19.95 | Improved |
| Use a larger retrieval set (ablation) | Prompt retrieval methods · Pascal-5ᶦ segmentation | mIoU | Improved |
| Increase the number of in-context examples (ablation) | Random, UnsupPR, and SupPR · Pascal-5ᶦ segmentation | mIoU | Improved |
| Change the order of the in-context examples (ablation) | Visual in-context learning · Pascal-5ᶦ segmentation | mIoU | No clear effect |
| Replace cosine distance with Euclidean or Manhattan distance (ablation) | Prompt retrieval score function · Pascal-5ᶦ segmentation | mIoU | No clear effect |
| Change the retrieval backbone among CLIP, EVA, and ViT (ablation) | UnsupPR and SupPR · Pascal-5ᶦ segmentation Avg | mIoU: 33.56; 33.64; 33.83; 35.56; 35.71; 36.12 | No clear effect |
Closest papers by meaning
- Exploring Task-Level Optimal Prompts for Visual In-Context Learning (AAAI 2025 · similarity 0.89)
- Visual In-Context Prompting (CVPR 2024 · similarity 0.89)
- In-Context Learning Can Help Vision Language Models Overcome Training Prior (NeurIPS 2026 · similarity 0.89)
- Towards Global Optimal Visual In-Context Learning Prompt Selection (NeurIPS 2024 · similarity 0.89)
- Embracing Collaboration Over Competition: Condensing Multiple Prompts for Visual In-Context Learning (CVPR 2025 · similarity 0.89)
- Learning to Learn Better Visual Prompts (AAAI 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.
