Test-time Training for Data-efficient UCDR
This work addresses the challenge of generalizing retrieval models to unknown domains with limited training data, which is an incremental improvement in the field of image retrieval.
The paper tackles the problem of universal cross-domain retrieval in a data-efficient manner by adapting pre-trained models on test data using self-supervised learning, achieving effective results as demonstrated through extensive experiments.
Image retrieval under generalized test scenarios has gained significant momentum in literature, and the recently proposed protocol of Universal Cross-domain Retrieval is a pioneer in this direction. A common practice in any such generalized classification or retrieval algorithm is to exploit samples from many domains during training to learn a domain-invariant representation of data. Such criterion is often restrictive, and thus in this work, for the first time, we explore the generalized retrieval problem in a data-efficient manner. Specifically, we aim to generalize any pre-trained cross-domain retrieval network towards any unknown query domain/category, by means of adapting the model on the test data leveraging self-supervised learning techniques. Toward that goal, we explored different self-supervised loss functions~(for example, RotNet, JigSaw, Barlow Twins, etc.) and analyze their effectiveness for the same. Extensive experiments demonstrate the proposed approach is simple, easy to implement, and effective in handling data-efficient UCDR.