Research map / Vision-language and multimodal / Human and embodied modeling
Images Speak in Images: A Generalist Painter for In-Context Visual Learning
Xinlong Wang, Wen Wang, Yue Cao, Chunhua Shen, Tiejun Huang
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 |
|---|---|---|---|
| Use Painter rather than BinsFormer | NYUv2 depth estimation · NYUv2 | RMSE: 0.330 -> 0.288 | Improved |
| Use Painter rather than Pix2Seq v2 | COCO keypoint detection · COCO | AP: 64.8 -> 72.1 | Improved |
| Jointly train tasks rather than train each task separately (ablation) | Painter · ADE-20K | mIoU: 47.2 -> 49.9 | Improved |
| Jointly train tasks rather than train each task separately (ablation) | Painter · COCO keypoint detection | AP: 72.5 -> 72.1 | Hurt |
| Use searched or learned prompts instead of random prompts (ablation) | Painter · ADE-20K | mIoU: 49.3 -> 49.9 | Improved |
| Add early feature merging between input and output images (ablation) | Painter · ADE-20K | mIoU: 39.7 -> 41.2 | Improved |
| Use ViT-L instead of ViT-B (ablation) | Painter · ADE-20K | mIoU: 31.4 -> 41.2 | Improved |
| Use a light three-layer head instead of a linear head (ablation) | Painter · ADE-20K | mIoU: 38.6 -> 41.2 | Improved |
| Use smooth-L1 rather than L1 loss (ablation) | Painter · ADE-20K | mIoU: 40.6 -> 41.2 | Improved |
| Use smooth-L1 rather than L2 loss (ablation) | Painter · ADE-20K | mIoU: 26.3 -> 41.2 | Improved |
Closest papers by meaning
- InstructCV: Instruction-Tuned Text-to-Image Diffusion Models as Vision Generalists (ICLR 2024 · similarity 0.89)
- CoPL: Contextual Prompt Learning for Vision-Language Understanding (AAAI 2024 · similarity 0.87)
- Explanatory Instructions: Towards Unified Vision Tasks Understanding and Zero-shot Generalization (ICML 2025 · similarity 0.87)
- LaViP: Language-Grounded Visual Prompting (AAAI 2024 · similarity 0.87)
- In-Context Learning Can Help Vision Language Models Overcome Training Prior (NeurIPS 2026 · similarity 0.87)
- Unifying Image Processing as Visual Prompting Question Answering (ICML 2024 · similarity 0.87)
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.
