Research map / Vision-language and multimodal / Cross-modal learning
PromptHub: Enhancing Multi-Prompt Visual In-Context Learning with Locality-Aware Fusion, Concentration and Alignment
Tianci Luo, Jinpeng Wang, Shiyu Qin, Niu Lian, Yan Feng, Bin Chen, Chun Yuan, Shu-Tao Xia
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 utilization loss L_u (ablation) | PromptHub N=16 · Pascal-5i segmentation | mean mIoU: 46.71 -> 47.81 | Improved |
| add semantic integrity loss L_s (ablation) | PromptHub N=16 · Pascal-5i segmentation | mean mIoU: 45.84 -> 47.81 | Improved |
| add label prediction loss L_p (ablation) | PromptHub N=16 · Pascal-5i segmentation | mean mIoU: 11.44 -> 47.81 | Improved |
| use Gaussian locality prior instead of Laplacian prior (ablation) | PromptHub N=16 · Pascal-5i segmentation | mean mIoU: 47.72 -> 47.81 | No clear effect |
| use locality-aware fusion instead of global fusion (ablation) | PromptHub N=16 · Pascal-5i segmentation | mean mIoU: 44.64 -> 47.81 | Improved |
| use locality-aware fusion instead of convolution-based fusion (ablation) | PromptHub N=16 · Pascal-5i segmentation | mean mIoU: 46.95 -> 47.81 | Improved |
| add data augmentation (ablation) | PromptHub N=16 · Pascal-5i segmentation | mean mIoU: 47.07 -> 47.81 | Improved |
| add data augmentation (ablation) | PromptHub N=1 · Pascal VOC2012 detection | mIoU: 43.52 -> 44.51 | Improved |
| increase prompts from N=1 to N=16 | PromptHub · Pascal-5i segmentation | mean mIoU: 45.17 -> 47.81 | Improved |
| replace Condenser with PromptHub fusion under N=16 | multi-prompt VICL · Pascal-5i segmentation (trained on COCO-5i) | mean mIoU: 40.52 -> 42.17 | Improved |
| replace hyperparameter sigma with query-adaptive sigma (ablation) | PromptHub N=1 · Pascal-5i segmentation | mean mIoU: 45.17 -> 45.86 | No clear effect |
| replace Condenser with PromptHub under flipped-query misalignment | fusion module N=16 · Spatial misalignment | performance drop: -1.39 -> -0.66 | Improved |
Closest papers by meaning
- Embracing Collaboration Over Competition: Condensing Multiple Prompts for Visual In-Context Learning (CVPR 2025 · similarity 0.92)
- Exploring Task-Level Optimal Prompts for Visual In-Context Learning (AAAI 2025 · similarity 0.90)
- VIPAMIN: Visual Prompt Initialization via Embedding Selection and Subspace Expansion (NeurIPS 2025 · similarity 0.88)
- Towards Global Optimal Visual In-Context Learning Prompt Selection (NeurIPS 2024 · similarity 0.88)
- Test-Time Visual In-Context Tuning (CVPR 2025 · similarity 0.88)
- CoPL: Contextual Prompt Learning for Vision-Language Understanding (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.
