Research map / Vision-language and multimodal / CLIP and zero-shot
Noise-Aware Few-Shot Learning through Bi-directional Multi-View Prompt Alignment
Lu Niu, Cheng Xue
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 clean and noise-aware prompt pair (bi-directional) to CoOp (ablation) | CoOp · DTD (avg) | accuracy (%): 48.08 -> 49.63 | Improved |
| use explicit negative labels instead of implicit negatives (ablation) | bi-directional prompts · DTD (avg) | accuracy (%): 49.63 -> 48.82 | Hurt |
| add multiple prompt views (N=4) (ablation) | bi-directional prompts · DTD (avg) | accuracy (%): 49.63 -> 51.83 | Improved |
| use UOT for patch-to-prompt alignment (ablation) | multi-view prompts · DTD (avg) | accuracy (%): 51.83 -> 54.18 | Improved |
| replace UOT with classical OT (ablation) | multi-view prompts · DTD (avg) | accuracy (%): 54.18 -> 53.37 | Hurt |
| replace UOT with KL divergence (ablation) | multi-view prompts · DTD (avg) | accuracy (%): 54.18 -> 52.60 | Hurt |
| add full OT label refinement without adaptive threshold (ablation) | UOT multi-view · DTD (25/50/75% noise) | accuracy (%): 62.50 -> 59.60 (25%); 42.33 -> 45.77 (75%) | Mixed |
| add adaptive threshold phi for selective refinement (ablation) | OT refinement · DTD (avg) | accuracy (%): 53.38 -> 56.75 | Improved |
| increase prompts from 1 to 4, then to 8 (ablation) | NA-MVP · DTD and OxfordPets | accuracy (%): gain 1 to 4, diminishing at 8 | Mixed |
| use NA-MVP instead of NLPrompt at 0.75 symmetric noise | NA-MVP · OxfordPets | accuracy (%): 70.77 -> 86.23 | Improved |
| use NA-MVP instead of NLPrompt at 0.75 symmetric noise | NA-MVP · Flowers102 | accuracy (%): 76.80 -> 76.47 | No clear effect |
| use NA-MVP instead of NLPrompt on 4-shot real-world noise | NA-MVP · Food101N | accuracy (%): 70.57 -> 76.10 | Improved |
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.
- NLPrompt — NLPrompt: Noise-Label Prompt Learning for Vision-Language Models (CVPR 2025) · better on 11 of 12 shared results
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
- NLPrompt: Noise-Label Prompt Learning for Vision-Language Models (CVPR 2025 · similarity 0.89)
- JoAPR: Cleaning the Lens of Prompt Learning for Vision-Language Models (CVPR 2024 · similarity 0.88)
- Amend to Alignment: Decoupled Prompt Tuning for Mitigating Spurious Correlation in Vision-Language Models (ICML 2024 · similarity 0.88)
- VaMP: Variational Multi-Modal Prompt Learning for Vision-Language Models (NeurIPS 2025 · similarity 0.88)
- On the Test-Time Zero-Shot Generalization of Vision-Language Models: Do We Really Need Prompt Learning? (CVPR 2024 · similarity 0.88)
- Text as Partial Constraint: Core–Residual Alignment for Robust Vision–Language Learning (NeurIPS 2026 · 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.
