Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models

TL;DR

This paper introduces test-time prompt tuning (TPT), a method that adaptively optimizes prompts on individual test samples to enhance zero-shot generalization of vision-language models like CLIP across diverse domains.

Contribution

The work proposes a novel test-time prompt tuning approach that improves zero-shot classification by learning adaptive prompts on the fly without additional training data.

Findings

TPT increases zero-shot top-1 accuracy by 3.6% on average across natural distribution shifts.

TPT outperforms previous prompt tuning methods that require task-specific training data.

TPT maintains competitive performance in cross-dataset generalization with unseen categories.

Abstract

Pre-trained vision-language models (e.g., CLIP) have shown promising zero-shot generalization in many downstream tasks with properly designed text prompts. Instead of relying on hand-engineered prompts, recent works learn prompts using the training data from downstream tasks. While effective, training on domain-specific data reduces a model's generalization capability to unseen new domains. In this work, we propose test-time prompt tuning (TPT), a method that can learn adaptive prompts on the fly with a single test sample. For image classification, TPT optimizes the prompt by minimizing the entropy with confidence selection so that the model has consistent predictions across different augmented views of each test sample. In evaluating generalization to natural distribution shifts, TPT improves the zero-shot top-1 accuracy of CLIP by 3.6% on average, surpassing previous prompt tuning approaches that require additional task-specific training data. In evaluating cross-dataset generalization with unseen categories, TPT performs on par with the state-of-the-art approaches that use additional training data. Project page: https://azshue.github.io/TPT.