NewsImages in MediaEval 2025 – Comparing Image Retrieval and Generation for News Articles

The NewsImages challenge in MediaEval 2025 explores matching news articles with fitting thumbnails. Participants receive a large set of English-language articles from international publishers. Given the text of a news article, the goal of the challenge is to (1) retrieve or (2) generate a fitting image. In two subtasks, participating teams create approaches to fully automate the retrieval/generation workflows as well as develop approaches that feature a human-in-the-loop. The overall quality of the image recommendation is evaluated in a crowd-sourced online event where image fit and relevance are assessed. This task overview paper provides details on the challenge setup, the crowdsourced online evaluation, and technical aspects of the group submissions.