Research map / Vision-language and multimodal / CLIP and zero-shot
Towards Calibrating Prompt Tuning of Vision- Language Models
Ashshak Sharifdeen, Fahad Shamshad, Muhammad Akhtar Munir, Abhishek Basu, Mohamed Ismithdeen, Jeyapriyan Jeyamohan, Chathurika Silva, Karthik Nandakumar, Muhammad Haris Khan
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 mean-variance margin regularization to cross-entropy (ablation) | MaPLe · novel classes, avg ECE | ECE (%): 5.76 -> 5.09 | Improved |
| add moment-matching loss on top of margin regularization (ablation) | MaPLe · novel classes, avg ECE | ECE (%): 5.09 -> 4.48 | Improved |
| add direct L1 embedding alignment to margin loss (instead of moment matching) (ablation) | MaPLe · novel classes, avg ECE | ECE (%): 5.09 -> 5.17; accuracy 75.74 -> 74.65 | No clear effect |
| add margin and moment-matching regularizers | CoOp · base classes, avg ECE | ECE (%): 6.35 -> 2.93 | Improved |
| add margin and moment-matching regularizers | MaPLe · base classes | ECE (%) and accuracy (%): ECE 3.19 -> 2.78; accuracy 82.41 -> 82.75 | Improved |
| add margin and moment-matching regularizers | MaPLe · novel classes, avg ECE | ECE (%): 5.76 -> 4.23 | Improved |
| add margin and moment-matching regularizers | CoOp · novel classes, avg ECE | ECE (%): 12.45 -> 4.79 | Improved |
| add margin and moment-matching regularizers | KgCoOp · base classes, avg ECE | ECE (%): 6.02 -> 5.47 | Improved |
| add margin and moment-matching regularizers | KgCoOp · novel classes, avg ECE | ECE (%): 3.71 -> 3.33 | Improved |
| add margin and moment-matching regularizers | CoOp · Aircraft base classes | ECE (%): 25.70 -> 4.96 | Improved |
| apply temperature scaling post hoc instead of train-time regularization | CoOp · base classes, avg ECE | ECE (%): 6.35 -> 2.96 | Improved |
| apply MBLS label smoothing baseline | CoOp · base classes, avg ECE | ECE (%): 6.35 -> 9.04 | Hurt |
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.
- MaPLe — MaPLe: Multi-Modal Prompt Learning (CVPR 2023) · better on 3 of 4 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
- C-TPT: Calibrated Test-Time Prompt Tuning for Vision-Language Models via Text Feature Dispersion (ICLR 2024 · similarity 0.93)
- Understanding and Mitigating Miscalibration in Prompt Tuning for Vision-Language Models (ICML 2025 · similarity 0.92)
- Amend to Alignment: Decoupled Prompt Tuning for Mitigating Spurious Correlation in Vision-Language Models (ICML 2024 · similarity 0.92)
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- ZIP: An Efficient Zeroth-order Prompt Tuning for Black-box Vision-Language Models (ICLR 2025 · similarity 0.90)
- Open-Vocabulary Calibration for Fine-tuned CLIP (ICML 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.
