This repository contains an implementation of the NewsImages task of MediaEval 2023, using CLIP (Contrastive Language-Image Pretraining). It aims to match news articles with the most appropriate header image.

The Mean Reciprocal Rank (MRR) is used to evaluate the rank of of each predicted image against the ground truth.

The paper describing the proposed method can be found here:

Optimizing Visual Pairings: A CLIP Framework for Precision News Image Rematching

CLIP learns a shared embedding space for images and their corresponding textual descriptions, fostering a unified understanding of multimodal relationships. The contrastive learning framework enhances the model's robustness by maximizing the similarity of correct image-text pairs and minimizing the similarity of incorrect pairs. CLIP's capability for zero-shot learning helps in scenarios with diverse datasets, such as news articles with varying topics.

The model is composed of three main sections-the image encoder, text encoder, and a module for the projection of the embeddings.

The results across datasets are summarized in the table below:

On the GDELT-P1 dataset, encompassing standard articles and their corresponding images, the model demonstrated a Mean Reciprocal Rank of 0.07839. In practical terms, this signifies that, on average, the model identifies the first correct match around the 13th position in the list of predictions for each query. Notably, its performance excelled on the GDELT-P2 dataset, primarily comprising images generated by machine-learning models. Here, the model identified the first correct match at around the 10th position.

The code returns the top 100 images for a given news heading, in order of decreasing relevance.

For example, the statement:

"Oxford researcher to watch Barbie as 'dessert' to Oppenheimer amid dual premiere"

yields the following results:

• Pooja Premnath
• Y.V. Ojus

If you find our work useful in your research, please include the following citation:

@inproceedings{premnath2023optimizing,
  title={Optimizing Visual Pairings: A CLIP Framework for Precision News Image Rematching.},
  author={Premnath, Pooja and Yenumulapalli, Venkatasai Ojus and Sivanaiah, Rajalakshmi and Suseelan, Angel Deborah},
  booktitle={MediaEval},
  year={2023}
}

1. A. Lommatzsch, B. Kille, Ö. Özgöbek, M. Elahi, D.-T. Dang-Nguyen, News images in mediaeval 2023, in: Proceedings of the MediaEval Benchmarking Initiative 2023, CEU Workshop Proceedings,2024
2. M. M. Shariatnia, Simple CLIP, 2021. doi:10.5281/zenodo.6845731.