AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

Improving Calibration in Test-Time Prompt Tuning for Vision-Language Models via Data-Free Flatness-Aware Prompt Pretraining

Hyeonseo Jang, Jaebyeong Jeon, Joong-Won Hwang, Kibok Lee

CVPR 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Prompt learning for CLIP · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
add data-free flatness-aware prompt pretraining (FPP) before TPTTPT · Fine-grained classificationECE: 11.67 -> 4.13Improved
add FPP before TPTTPT · Fine-grained classificationaccuracy: 64.62 -> 65.37Improved
add alignment loss to flatness loss (ablation)FPP · Fine-grained classificationzero-shot accuracy: 4.65 -> 64.35Improved
add flatness loss to alignment loss (ablation)FPP · Fine-grained classificationECE: 7.05 -> 4.13Improved
add first perturbation term eps1 in flatness loss (ablation)FPP · Fine-grained classificationECE: 5.86 -> 4.13Improved
add second perturbation term eps2 in flatness loss (ablation)FPP · Fine-grained classificationECE: 4.64 -> 4.13Improved
pretrain with Gaussian noise instead of class names (ablation)FPP · Fine-grained classificationaccuracy: 65.37 -> 65.28No clear effect
pretrain with ImageNet class names instead of true class names (ablation)FPP · Fine-grained classificationaccuracy: 65.37 -> 64.97No clear effect
add SAM to TPTTPT · Fine-grained classificationaccuracy: 64.62 -> 61.59Hurt
add SAM to TPTTPT · Fine-grained classificationECE: 11.67 -> 4.15Improved
add ASAM to TPTTPT · Fine-grained classificationECE: 11.67 -> 9.78Improved
add FPP instead of O-TPT regularizationTPT · Natural distribution shiftsECE: 4.88 -> 4.63Improved

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy