Research map / Vision-language and multimodal / CLIP and zero-shot
Progressive Visual Prompt Learning with Contrastive Feature Re-formation
Xu Chen, Yuhan Zhu, Haocheng Shen, Boheng Chen, Yixuan Liao, Xiaoxin Chen, Limin Wang
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 contrastive feature re-formation loss to ProVP (ablation) | ProVP · Base-to-new (11 datasets avg) | new accuracy: 69.57 -> 73.22 | Improved |
| add contrastive feature re-formation loss to ProVP (ablation) | ProVP · Base-to-new (11 datasets avg) | harmonic mean: 76.57 -> 78.76 | Improved |
| add contrastive feature re-formation loss to ProVP (ablation) | ProVP · Base-to-new (11 datasets avg) | base accuracy: 85.14 -> 85.20 | No clear effect |
| use lambda=1.0 re-formation weight instead of 0 (ablation) | ProVP-Ref · Base-to-new (DTD) | new accuracy: 52.82 -> 59.06 | Improved |
| use lambda=0.8 re-formation weight instead of 0 (ablation) | ProVP-Ref · Base-to-new (StanfordCars) | harmonic mean: 71.03 -> 73.71 | Improved |
| use lambda=1.0 re-formation weight instead of 0 (ablation) | ProVP-Ref · Base-to-new (DTD) | base accuracy: 84.76 -> 83.95 | Hurt |
| use lambda=1.0 re-formation weight instead of 0 (ablation) | ProVP-Ref · DTD 16-shot | accuracy: 73.70 -> 73.62 | No clear effect |
| use alpha=0.1 instead of 0.9 progressive decay (ablation) | ProVP-Ref · DTD 16-shot | accuracy: 61.96 -> 73.62 | Improved |
| use alpha=0.1 instead of 0.01 progressive decay (ablation) | ProVP-Ref · StanfordCars 16-shot | accuracy: 82.88 -> 84.85 | Improved |
| use alpha=0.1 instead of 0.9 progressive decay (ablation) | ProVP-Ref · StanfordCars 16-shot | accuracy: 69.04 -> 84.85 | Improved |
| use ProVP-Ref instead of CoCoOp | CLIP ViT-B/16 · Cross-dataset transfer (ImageNet source, 10 targets avg) | accuracy: 65.74 -> 66.49 | Improved |
| use ProVP-Ref instead of CoOp | CLIP ViT-B/16 · Cross-dataset transfer (EuroSAT target) | accuracy: 46.39 -> 51.95 | Improved |
Closest papers by meaning
- Make Prompts Adaptable: Bayesian Modeling for Vision-Language Prompt Learning with Data-Dependent Prior (AAAI 2024 · similarity 0.91)
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.91)
- LaViP: Language-Grounded Visual Prompting (AAAI 2024 · similarity 0.91)
- SA²VP: Spatially Aligned-and-Adapted Visual Prompt (AAAI 2024 · similarity 0.90)
- LoR-VP: Low-Rank Visual Prompting for Efficient Vision Model Adaptation (ICLR 2025 · similarity 0.90)
- Concept-Guided Prompt Learning for Generalization in Vision-Language Models (AAAI 2024 · similarity 0.90)
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.
