1,121,706 news images and captions collected from four major news outlets including BBC, USA Today, The Guardian, and The Washington Post. The collection spans multiple years and covers diverse topics such as politics, sports, and entertainment with real-world journalistic metadata.
Use Cases
- Train entity-aware captioning models that utilize the caption and headline fields to name specific individuals and locations
- Perform cross-source domain adaptation studies by comparing model performance across the four distinct news publishers
- Develop multi-modal retrieval systems using the image and caption pairs to match visual content with journalistic text
Strengths
- 1,121,706 images paired with natural language captions
- Sourced from 4 major news organizations: BBC, USA Today, The Guardian, and The Washington Post
- Includes metadata such as article headlines, image captions, and source attribution