Research map / Vision-language and multimodal / CLIP and zero-shot
Doubly Debiased Test-Time Prompt Tuning for Vision-Language Models
Fei Song, Yi Li, Rui Wang, Jiahuan Zhou, Changwen Zheng, Jiangmeng 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 dynamic retrieval-augmented modulation (RAM) (ablation) | D2TPT · Natural distribution shifts, OOD average | top-1 accuracy: 64.57 -> 64.82 | Improved |
| add RAM (ablation) | D2TPT · Cross-dataset generalization, average | top-1 accuracy: 67.08 -> 68.27 | Improved |
| add confidence-based weighted ensemble (assumed second marked column) (ablation) | D2TPT · Cross-dataset generalization, average | top-1 accuracy: 68.27 -> 68.43 | Improved |
| add cross-modal consistency distillation (assumed third marked column) (ablation) | D2TPT · Cross-dataset generalization, average | top-1 accuracy: 68.43 -> 68.93 | Improved |
| add all D2TPT modules to entropy-minimization baseline (ablation) | CLIP ViT-B/16 test-time prompt tuning · Natural distribution shifts, OOD average | top-1 accuracy: 64.57 -> 65.25 | Improved |
| use D2TPT instead of TPS | CLIP ViT-B/16 · Cross-dataset generalization, average | top-1 accuracy: 66.95 -> 68.93 | Improved |
| use D2TPT instead of TDA on EuroSAT | CLIP ViT-B/16 · EuroSAT | top-1 accuracy: 58.00 -> 54.92 | Hurt |
| use D2TPT instead of PromptAlign on OxfordPets | CLIP ViT-B/16 · OxfordPets | top-1 accuracy: 90.76 -> 87.55 | Hurt |
| use D2TPT instead of TPT regarding speed | CLIP ViT-B/16 · Flowers102 | inference throughput FPS: 6.37 -> 5.71 | Hurt |
| use D2TPT instead of TPS on natural distribution shifts | CLIP ViT-B/16 · Natural distribution shifts, OOD average | top-1 accuracy: 64.15 -> 65.25 | 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.
- DynaPrompt — DynaPrompt: Dynamic Test-Time Prompt Tuning (ICLR 2025) · 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
- DynaPrompt: Dynamic Test-Time Prompt Tuning (ICLR 2025 · similarity 0.93)
- Black-Box Test-Time Prompt Tuning for Vision-Language Models (AAAI 2025 · similarity 0.93)
- C-TPT: Calibrated Test-Time Prompt Tuning for Vision-Language Models via Text Feature Dispersion (ICLR 2024 · similarity 0.92)
- Robust Test-Time Adaptation for Zero-Shot Prompt Tuning (AAAI 2024 · similarity 0.91)
- On the Test-Time Zero-Shot Generalization of Vision-Language Models: Do We Really Need Prompt Learning? (CVPR 2024 · similarity 0.91)
- Learning to Learn Better Visual Prompts (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.
