Paper ID: 2403.15152

A Multimodal Approach for Cross-Domain Image Retrieval

Lucas Iijima, Nikolaos Giakoumoglou, Tania Stathaki

Cross-Domain Image Retrieval (CDIR) is a challenging task in computer vision, aiming to match images across different visual domains such as sketches, paintings, and photographs. Traditional approaches focus on visual image features and rely heavily on supervised learning with labeled data and cross-domain correspondences, which leads to an often struggle with the significant domain gap. This paper introduces a novel unsupervised approach to CDIR that incorporates textual context by leveraging pre-trained vision-language models. Our method, dubbed as Caption-Matching (CM), uses generated image captions as a domain-agnostic intermediate representation, enabling effective cross-domain similarity computation without the need for labeled data or fine-tuning. We evaluate our method on standard CDIR benchmark datasets, demonstrating state-of-the-art performance in unsupervised settings with improvements of 24.0% on Office-Home and 132.2% on DomainNet over previous methods. We also demonstrate our method's effectiveness on a dataset of AI-generated images from Midjourney, showcasing its ability to handle complex, multi-domain queries.

Submitted: Mar 22, 2024