Zero-shot composed image retrieval (ZS-CIR) is the task of finding a target image given a reference image and a text describing the desired modification. According to the paper's abstract, recent approaches to ZS-CIR typically rely on large vision-language models to generate target image captions as an intermediate step.
The paper, titled "Less is More: Concise Retrieval and Critic-based Re-ranking for Composed Image Retrieval," signals a focus on a different strategy. The title suggests that a more concise retrieval process, combined with a critic-based re-ranking stage, may outperform the common caption-generation approach. However, the abstract excerpt provided does not include details on the proposed method or experimental outcomes.
As a single-source digest, this entry can only reflect the problem statement and the direction implied by the title. Readers interested in the full methodology and results should consult the arXiv listing.