Research map / Vision-language and multimodal / CLIP and zero-shot
A-TPT: Angular Diversity Calibration Properties for Test-Time Prompt Tuning of Vision-Language Models
Shihab Aaqil Ahamed, Udaya Sampath K. Perera Miriya Thanthrige, Ranga Rodrigo, 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 angular diversity regularizer to TPT | TPT, CLIP ViT-B/16 · Fine-grained datasets (avg) | ECE: 11.60 -> 2.61 | Improved |
| add angular diversity regularizer to TPT | TPT, CLIP RN50 · Fine-grained datasets (avg) | ECE: 11.70 -> 2.92 | Improved |
| replace orthogonality loss (O-TPT) with angular diversity | TPT, CLIP ViT-B/16 · Natural distribution shifts | ECE: 4.88 -> 3.92 | Improved |
| replace orthogonality loss with angular diversity | TPT, CLIP RN50 · Natural distribution shifts | ECE: 9.69 -> 7.82 | Improved |
| A-TPT versus zero-shot CLIP on distribution shifts | CLIP RN50 · Natural distribution shifts | ECE: 7.46 -> 7.82 | No clear effect |
| add angular diversity to TPT | TPT, CLIP ViT-B/16 · Fine-grained datasets (avg) | accuracy: 65.00 -> 64.92 | No clear effect |
| add angular diversity in N>|D| classes regime | TPT, CLIP ViT-B/16 · N>|D| group | ECE: 4.27 -> 2.92 | Improved |
| combine A-TPT with supervised CoOp prompts | TPT + CoOp · Fine-grained avg | ECE: 18.36 -> 3.63 | Improved |
| combine A-TPT with CoCoOp prompts | TPT + CoCoOp · Fine-grained avg | ECE: 4.35 -> 3.22 | Improved |
| add A-TPT to medical prompt tuning (BAPLe) | MedCLIP · Covid CA | ECE: 15.64 -> 2.34 | Improved |
| add A-TPT to FPT on ISIC 2018 | FPT · ISIC 2018 | ECE: 0.2328 -> 0.0794 | Improved |
| use angular diversity in place of cosine orthogonality: compute cost | TPT · ImageNet-V2 | time per batch: 1.064 -> 1.058 | 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.
- O-TPT — O-TPT: Orthogonality Constraints for Calibrating Test-time Prompt Tuning in Vision-Language Models (CVPR 2025) · better on 6 of 6 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
- O-TPT: Orthogonality Constraints for Calibrating Test-time Prompt Tuning in Vision-Language Models (CVPR 2025 · similarity 0.93)
- C-TPT: Calibrated Test-Time Prompt Tuning for Vision-Language Models via Text Feature Dispersion (ICLR 2024 · similarity 0.91)
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.89)
- Amend to Alignment: Decoupled Prompt Tuning for Mitigating Spurious Correlation in Vision-Language Models (ICML 2024 · similarity 0.89)
- Black-Box Test-Time Prompt Tuning for Vision-Language Models (AAAI 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)
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.
