A data-free pretraining step that places prompts in flatter loss regions improves calibration and performance when used as initialization for test-time prompt tuning of vision-language models.
Towards theoretically understand- 10 ing why sgd generalizes better than adam in deep learn- ing
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Improving Calibration in Test-Time Prompt Tuning for Vision-Language Models via Data-Free Flatness-Aware Prompt Pretraining
A data-free pretraining step that places prompts in flatter loss regions improves calibration and performance when used as initialization for test-time prompt tuning of vision-language models.