Research map / Vision-language and multimodal / CLIP and zero-shot
Dropout Prompt Learning: Towards Robust and Adaptive Vision-Language Models
Biao Chen, Lin Zuo, Mengmeng Jing, Kunbin He, Yuchen 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 importance weighted token dropout (IWTD) (ablation) | DroPLe baseline · Base-to-novel (11 datasets avg) | HM: 79.44 -> 81.17 | Improved |
| add vanilla dropout p=0.5 (ablation) | DroPLe baseline · Base-to-novel (11 datasets avg) | HM: 79.44 -> 77.81 | Hurt |
| add vanilla dropout p=0.3 (ablation) | DroPLe baseline · Base-to-novel (11 datasets avg) | HM: 79.44 -> 78.73 | Hurt |
| add dropblock (ablation) | DroPLe baseline · Base-to-novel (11 datasets avg) | HM: 79.44 -> 77.94 | Hurt |
| replace D_KL+L1 consistency with residual entropy regularization (ablation) | DroPLe with IWTD · Base-to-novel (11 datasets avg) | HM: 81.46 -> 82.10 | Improved |
| replace cosine consistency with residual entropy regularization (ablation) | DroPLe with IWTD · Base-to-novel (11 datasets avg) | HM: 81.37 -> 82.10 | Improved |
| add class attention score S_cls to S_self (ablation) | DroPLe · Base-to-novel (11 datasets avg) | HM: 80.66 -> 81.14 | Improved |
| add cross-modal attention score S_cross (ablation) | DroPLe · Base-to-novel (11 datasets avg) | HM: 81.14 -> 82.10 | Improved |
| use linear instead of non-linear residual formulation (ablation) | DroPLe · Base-to-novel (11 datasets avg) | HM: 80.89 -> 82.10 | Improved |
| use dropout range 10-50% instead of 20-70% (ablation) | DroPLe · Base-to-novel (11 datasets avg) | HM: 81.81 -> 82.10 | Improved |
| increase bridge tokens from 16 to 64 (ablation) | DroPLe · Base-to-novel (11 datasets avg) | HM: 81.90 -> 82.10 | Improved |
| vary lambda_0 from 0.05 to 0.10 (ablation) | DroPLe · Base-to-novel (11 datasets avg) | HM: 82.08 -> 82.10 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Tip-Adapter — Tip-Adapter: Training-Free Adaption of CLIP for Few-Shot Classification (ECCV 2022) · 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
- Enhancing Vision-Language Model Reliability with Uncertainty-Guided Dropout Decoding (NeurIPS 2025 · similarity 0.90)
- NLPrompt: Noise-Label Prompt Learning for Vision-Language Models (CVPR 2025 · similarity 0.88)
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Prompt Learning via Meta-Regularization (CVPR 2024 · similarity 0.87)
- JoAPR: Cleaning the Lens of Prompt Learning for Vision-Language Models (CVPR 2024 · similarity 0.87)
- A Similarity Paradigm Through Textual Regularization Without Forgetting (AAAI 2025 · 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.
