# RoSA Researchers Report Better Image Search With Long Descriptions

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/41066
Published: 2026-10-26T13:49:09.000Z
Updated: 2026-10-26T13:49:09.000Z
Section: Technology

> The framework trains models to use contextual details when prominent words in a search query are reduced.

Qiuyu Kong and three co-authors have proposed RoSA, a training framework aimed at improving image search with long descriptions by helping models use context when a query contains fewer prominent details.

Models that match text to images can rely heavily on words describing the most noticeable features. That reliance can hurt search results when those words are reduced, even though the query still contains useful contextual details. The researchers call this failure “saliency collapse.”

When salient content was reduced, retrieval performance fell even though useful context remained.

RoSA addresses the problem during model training. It estimates which objects in an image are visually prominent and uses a large language model to separate the paired description into prominent and contextual content. The framework then trains on both views, using additional objectives to align the image and text information.

In benchmark tests involving long descriptions, the researchers found that RoSA improved image-text retrieval. They also designed its additional training objectives to work with existing fine-tuning methods.