README file for the data overview and expected submission of the NewsImages Challenge at MediaEval 2026. Please refer to the official MediaEval 2026 website for the full task description and event registration details.
Note: This task summary only applies to regular paper submissions to the NewsImages challenge. If you are writing a Quest for Insight paper, then you can skip this outline and go directly to the deadlines below.
Upon successful registration, the participants are given a collection of 8,500 news articles with images (the article text is in English, from GDELT. Given a randomly selected article, the goal is to build a pipeline that combines image retrieval and image generation techniques to provide a fitting image recommendation for a given news article text. There will be a crowdsourced online event in which all participating teams rate the submitted image recommendations using a 5-point Likert scale.
The winning team is determined by the highest average image rating for the articles within the evaluation dataset. This evaluation dataset will be shared with all registered groups, together with the information on where to submit your results (see deadlines below).
The training dataset you receive includes the list of 8,500 article IDs from the 2025 iteration of the task. This allows you, for example, to create image recommendations for the same images and compare your approach with last year's results in a preliminary user study. (To that end, the test dataset includes a copy of all image IDs used in the 2025 survey.)
The test dataset for the online evaluation event contains around 800 articles. It is a mixed collection of English articles from MIND and from historical archives (scans of 19th- and 20th-century printed newspapers). We will share this together with an evaluation dataset. The difference between test and evaluation is that the test version comes without the original images.
The datasets contain a CSV with the following data on news articles:
| Attribute | Description |
|---|---|
| article_id | ID of news article. |
| article_url | Original URL of the news article. |
| article_title | Title of the news article (may include lead). |
| image_id | ID of news image (we provide a copy of the image for the train and evaluation dataset). |
Furthermore, a folder 'newsimages' containing a copy of the original thumbnails is included. The name of each JPG file corresponds to the 'image_id' associated with each news article.
For image retrieval, we recommend using Yahoo-Flickr Creative Commons 100 Million (YFCC100M) or Public Domain 12 Million (PD12M). You are free to use any other open source dataset, but you must share it with us as part of your workflow submission.
Your submission can include multiple runs/approaches. Each approach must include precisely one image recommendation for a given article ID. (If there is no valid recommendation for a given article, the recommendation will receive a default rating score of 0.)
Important: There is no restriction in terms of how many runs you can submit. However, all the runs need to be sufficiently different from one another. No two runs can have the same image recommendation for a given article ID, and all images within one run must be unique.
To submit your runs, you must provide a ZIP file [group_name].zip that is structured as follows:
[group_name] + _ + [approach_name] / [article_id] + _ + [group_name] + _ + [approach_name].png
Use the group name you registered with for the task. For each submitted approach/run, please provide a unique name (i.e., unique among your runs).
You must hand in your workflow together with the Working Notes Papers (see deadlines below). Please have a look at last year's workflows to get an idea of how to organize your codebase.
Below is the folder structure of an example ZIP file for the group 'UnstableOsmosis':
UnstableOsmosis_Submission.zip
|_ UnstableOsmosis_ZImage
| |_ 117_UnstableOsmosis_ZImage.png
| |_ …
|_ UnstableOsmosis_OpenCLIP
| |_ 117_UnstableOsmosis_OpenCLIP.png
| |_ …
|_ …
The image format must be PNG, with a target dimension of 460x260 pixels (in landscape orientation). This applies to both generated and retrieved images. If you generate images with tools like ComfyUI and then edit them afterwards (e.g., for cropping), ensure the workflow remains embedded. We advise teams not to stretch the images. Consider simply rescaling the images if cropping is not an option. (And instead of using black bars to reframe the image, you can leave the background transparent with PNGs.)
You will need to submit your runs by the deadlines indicated below. Do that by sending an email to the address that shared the dataset download link with you. It must include (1) your group name, (2) a link to download your image submissions, and (3) links to the documented code of your workflow (e.g., a link to a GitHub repository with a notebook and/or a collection of scripts). (Please note that this is something separate from the Working Notes Paper.)
We advise all teams to carefully study previous submissions. We share the code of all previous workflows together with their overview papers. Please follow the recommendations outlined in the Task Overview Paper (Release February) to ensure a competitive run submission.
Note: This section is primarily for teams that participated in the past. It summarized the main difference between the 2025 and 2026 iterations.
Teams that participated in earlier iterations of the challenge are welcome to submit a new entry. Their existing workflow, however, needs to be adapted to accommodate the changes.
- Separate training and test sets for (with the training set being the same as in 2025).
- You can use/combine automated means and human-in-the-loop approaches for your runs (i.e., merge the two objectives of the previous small and large subtasks into one main challenge task).
- Mixing of image retrieval and generation is now allowed; you no longer need to separate them.
- Image submissions for runs must be unique (no reusing images within one run) and must not use any baseline image (you can no longer use any of the original news article images in final run submissions).
- There are no original images available for the test dataset (your workflow must focus on text input).
- We encourage the submission of non-photorealistic images.
In addition, acceptance of your run is tied to participation in the online event and to taking part in the review of Working Notes Papers.
Taking part in the online evaluation event is mandatory. During the evaluation, participating teams rate the image recommendations of other teams. To do that, they are being presented with a news headline and two image recommendations. They then need to select which image is the better fit. These ratings are used to calculate an overall ranking for each article's images. The average rank of team submissions across the featured item pool then determines the overall winner of the challenge.
As part of the challenge submission, each team must write a separate Working Notes Paper that documents and outlines their approach. Please refer to the online paper template for additional information. You can look at the submissions from last year as an example.
We encourage open and reproducible science. Therefore, ask each team to share their codebase/workflows and rely exclusively on open-source/open-weighted models. Please use the examples in the designated folder to structure your code and make a pull request to contribute your workflow.
Note that each team will be asked to participate in the review process for the papers on EasyChair. Make sure you are available between the paper submission and review deadlines (see the overview below). Furthermore, we ask each group to include and refer to the following papers in their Working Notes Paper:
-
NewsImages in MediaEval 2026 – Automated Image Recommendations with Retrieval and Generation Techniques for News Articles, Heitz et al., Working Notes Proceedings of the MediaEval 2026 Workshop, 2026.
@inproceedings{heitz2026newsimages, title={NewsImages in MediaEval 2026 - Automated Image Recommendations with Retrieval and Generation Techniques for News Articles Thumbnails}, author={Heitz, Lucien and Sotic, Bruno N. and Katamjani, Ali A. and Bi, Qi and Bakker, Bram and Rossetto, Luca and Kamps, Jaap}, booktitle={Working Notes Proceedings of the MediaEval 2026 Workshop}, year={2026} } -
NewsImages in MediaEval 2025 – Comparing Image Retrieval and Generation for News Articles, Heitz et al., Working Notes Proceedings of the MediaEval 2025 Workshop, 2025.
@inproceedings{heitz2025newsimages, title={NewsImages in MediaEval 2025 – Comparing Image Retrieval and Generation for News Articles}, author={Heitz, Lucien and Rossetto, Luca and Kille, Benjamin and Lommatzsch, Andreas and Elahi, Mehdi and Dang-Nguyen, Duc-Tien}, booktitle={Working Notes Proceedings of the MediaEval 2025 Workshop}, year={2025} } -
An Empirical Exploration of Perceived Similarity between News Article Texts and Images, Heitz et al., Working Notes Proceedings of the MediaEval 2023 Workshop, 2024.
@inproceedings{heitz2024empirical, title={An Empirical Exploration of Perceived Similarity between News Article Texts and Images}, author={Heitz, Lucien and Rossetto, Luca and Bernstein, Abraham}, booktitle={Working Notes Proceedings of the MediaEval 2023 Workshop}, year={2024} }
- Registration opening and release train dataset: February 1 (registration mandatory, form available online)
- Release test and evaluation dataset: April 1
- Registration closes: April 24
- Runs due: May 1 (AoE, incl. workflows to reproduce your image recommendations)
- Online evaluation: May 7-14 (with Qualtrics, no user account required)
- Evaluation feedback: May 21
- Working Notes Paper submission: May 31* (on EasyChair, user account required)
- Review deadline: June 7**
- Camera-ready deadline: June 14 (AoE)
- MediaEval workshop: June 15-16, co-located with ACM ICMR 2026 (more information on the MediaEval website, in-person or online attendance required).
(*) We will provide you with a review/feedback on your paper within one week of submission (by June 7). Afterwards, you then have another week to prepare the camera-ready revision by June 14 (exact deadlines will be communicated by the MediaEval organizers). Please note that your paper should include a results section that is based on your performance in the online evaluation. The necessary information for this part will be forwarded to you as part of the evaluation feedback on May 21. The Working Notes Paper must describe the workflows for your submissions. It may include complementary and/or alternative approaches you tested but did not submit to the online challenge.
(**) We will notify each team once their paper has been reviewed; please make the necessary changes and upload a camera-ready version within one week.