Diverse Data Augmentation with Diffusions for Effective Test-time Prompt Tuning
This work addresses the challenge of adapting pre-trained models to unseen domains during inference, offering an incremental improvement for tasks like zero-shot classification under distribution shifts.
The paper tackles the problem of limited data diversity and prediction fidelity in test-time prompt tuning for vision-language models by proposing DiffTPT, which uses diffusion models for diverse data augmentation and a cosine similarity-based filtration technique, resulting in an average 5.13% improvement in zero-shot accuracy over state-of-the-art methods.
Benefiting from prompt tuning, recent years have witnessed the promising performance of pre-trained vision-language models, e.g., CLIP, on versatile downstream tasks. In this paper, we focus on a particular setting of learning adaptive prompts on the fly for each test sample from an unseen new domain, which is known as test-time prompt tuning (TPT). Existing TPT methods typically rely on data augmentation and confidence selection. However, conventional data augmentation techniques, e.g., random resized crops, suffers from the lack of data diversity, while entropy-based confidence selection alone is not sufficient to guarantee prediction fidelity. To address these issues, we propose a novel TPT method, named DiffTPT, which leverages pre-trained diffusion models to generate diverse and informative new data. Specifically, we incorporate augmented data by both conventional method and pre-trained stable diffusion to exploit their respective merits, improving the models ability to adapt to unknown new test data. Moreover, to ensure the prediction fidelity of generated data, we introduce a cosine similarity-based filtration technique to select the generated data with higher similarity to the single test sample. Our experiments on test datasets with distribution shifts and unseen categories demonstrate that DiffTPT improves the zero-shot accuracy by an average of 5.13\% compared to the state-of-the-art TPT method. Our code and models will be publicly released.