Align Your Prompts: Test-Time Prompting with Distribution Alignment for Zero-Shot Generalization

Test-Time adaptation of foundational Vision-Language models using prompt learning.

We design a distribution alignment strategy for Vision Language models to improve test-time adaptation by explicitly handling the distribution shift in test data. We use a single test sample to adapt multi-modal prompts at test time, by minimizing the feature distribution shift to bridge the gap in the test domain

Paper accepted at NeurIPS 2023. Paper and code here.