New Yorker Caption Contest
New Yorker Caption Contest is a multi-modal dataset combining images and sophisticated humorous captions, used to assess how AIs understand visual and textual context.
Approximately 149,000 humorous images and captions, 7.91 GB, Parquet format
CC-BY 4.0
Description
The dataset New Yorker Caption Contest includes nearly 150,000 images from the famous New Yorker comedy legends competition, along with captions created by humans. It challenges AI models to capture complex nuances and unexpected relationships between image and text.
What is this dataset for?
- Test and train models to understand multimodal humor
- Improving the generation of relevant and creative legends
- Exploring the complex contextual understanding between images and text
Can it be enriched or improved?
This dataset can be enriched by additional annotations on types of humor or by cultural metadata to further analyze the models.
🔎 In summary
🧠 Recommended for
- Multimodality researchers
- NLP developers
- Creative AI
🔧 Compatible tools
- Hugging Face Transformers
- CLIP
- BLIP
- Multimodal frameworks
💡 Tip
Explore the most unexpected legends to improve the model's creativity.
Frequently Asked Questions
Does this dataset contain images or just captions?
It contains images accompanied by humorous captions created by humans.
Can this dataset be used for commercial use?
Yes, licensed under CC-BY 4.0, commercial use authorized with attribution.
What is the main challenge posed by this dataset to AI models?
Understand the complex relationships and subtle humor between images and texts.
