Feed-Forward Source-Free Domain Adaptation via Class Prototypes

TL;DR

This paper introduces a fast, feed-forward method for source-free domain adaptation that computes class prototypes to improve accuracy without back-propagation, significantly reducing adaptation time.

Contribution

It proposes a novel, simple prototype-based approach for domain adaptation that eliminates the need for back-propagation, enhancing efficiency and practicality.

Findings

Achieves significant accuracy improvements over pre-trained models.

Requires only a small fraction of the time compared to existing methods.

Effectively handles domain shift using class prototypes.

Abstract

Source-free domain adaptation has become popular because of its practical usefulness and no need to access source data. However, the adaptation process still takes a considerable amount of time and is predominantly based on optimization that relies on back-propagation. In this work we present a simple feed-forward approach that challenges the need for back-propagation based adaptation. Our approach is based on computing prototypes of classes under the domain shift using a pre-trained model. It achieves strong improvements in accuracy compared to the pre-trained model and requires only a small fraction of time of existing domain adaptation methods.