Improving Zero-shot Generalization of Learned Prompts via Unsupervised Knowledge Distillation

TL;DR

This paper introduces KDPL, a novel unsupervised knowledge distillation method that enhances zero-shot generalization of learned prompts in vision-language models without requiring labeled data.

Contribution

It proposes a new prompt learning approach that leverages unsupervised knowledge distillation, eliminating the need for annotated samples during adaptation.

Findings

KDPL improves zero-shot domain and cross-dataset generalization.

It enhances base-to-novel class transfer without labeled data.

The method is effective even without knowledge of class names.

Abstract

Vision-Language Models (VLMs) demonstrate remarkable zero-shot generalization to unseen tasks, but fall short of the performance of supervised methods in generalizing to downstream tasks with limited data. Prompt learning is emerging as a parameter-efficient method for adapting VLMs, but state-of-the-art approaches require annotated samples. In this paper we propose a novel approach to prompt learning based on unsupervised knowledge distillation from more powerful models. Our approach, which we call Knowledge Distillation Prompt Learning (KDPL), can be integrated into existing prompt learning techniques and eliminates the need for labeled examples during adaptation. Our experiments on more than ten standard benchmark datasets demonstrate that KDPL is very effective at improving generalization of learned prompts for zero-shot domain generalization, zero-shot cross-dataset generalization, and zero-shot base-to-novel class generalization problems. KDPL requires no ground-truth labels for adaptation, and moreover we show that even in the absence of any knowledge of training class names it can be used to effectively transfer knowledge. The code is publicly available at https://github.com/miccunifi/KDPL.