Feed-Forward Source-Free Domain Adaptation via Class Prototypes
This work addresses the need for faster adaptation in source-free domain adaptation, which is incremental but offers practical benefits for real-world applications.
The paper tackles the problem of slow adaptation in source-free domain adaptation by proposing a feed-forward method using class prototypes, which achieves strong accuracy improvements and requires only a small fraction of the time compared to existing methods.
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.