Research map / Vision-language and multimodal / CLIP and zero-shot
Cascade Prompt Learning for Visual-Language Model Adaptation
Ge Wu, Xin Zhang, Zheng Li, Zhaowei Chen, Jiajun Liang, Jian Yang, Xiang Li
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 CasPL boosting prompt to PromptSRC | PromptSRC · Base-to-novel, average over 11 datasets | HM: 79.97 -> 82.69 | Improved |
| add CasPL boosting prompt to CoOp | CoOp · Base-to-novel, average over 11 datasets | HM: 71.66 -> 79.30 | Improved |
| add CasPL boosting prompt to CoOp | CoOp · Base-to-novel, ImageNet | HM: 71.92 -> 72.29 | Improved |
| add CasPL to MaPLe | MaPLe · Domain generalization, ImageNet-S | accuracy: 49.15 -> 48.81 | Hurt |
| add CasPL to PromptSRC | PromptSRC · Domain generalization, target average | accuracy: 63.90 -> 64.44 | Improved |
| use in-domain instead of out-domain unlabeled data for the boosting prompt (ablation) | PromptSRC + CasPL · 10 datasets few-shot | accuracy: 84.56 -> 85.52 | Improved |
| split 16 prompt tokens into 8 frozen boosting plus 8 adapting (ablation) | PromptSRC · Base-to-novel, equal total VLP tokens | HM: 80.34 -> 83.51 | Improved |
| increase prompt depth in both phases from 1 to 12 (ablation) | CasPL · Base-to-novel, 10 datasets | HM: 79.95 -> 83.51 | Improved |
| keep boosting prompts frozen in the second phase (ablation) | CasPL · Base-to-novel | HM | Improved |
| add both boosting and adapting prompts versus adapting alone (ablation) | CLIP · Base-to-novel, average over 11 datasets | HM: 79.97 -> 82.69 | Improved |
| train boosting prompt with ViT-L/14 teacher on small student (ablation) | PromptSRC (ViT-B/16) · Base-to-novel, average over 11 datasets | HM: 82.69 vs PromptSRC ViT-L/14 83.17 | Mixed |
| increase unlabeled images per class for boosting prompt on DTD (ablation) | CasPL · DTD base-to-novel | HM: 62.68 -> 76.44 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- MaPLe — MaPLe: Multi-Modal Prompt Learning (CVPR 2023) · better on 1 of 2 shared results
- LaFTer — LaFTer: Label-Free Tuning of Zero-shot Classifier using Language and Unlabeled Image Collections (NeurIPS 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.90)
- Make Prompts Adaptable: Bayesian Modeling for Vision-Language Prompt Learning with Data-Dependent Prior (AAAI 2024 · similarity 0.90)
- Surrogate Prompt Learning: Towards Efficient and Diverse Prompt Learning for Vision-Language Models (ICML 2025 · similarity 0.89)
- CoPL: Contextual Prompt Learning for Vision-Language Understanding (AAAI 2024 · similarity 0.89)
- PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 2024 · similarity 0.89)
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.89)
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.
