Dataset · zhongshsh
Oogiri humor and meme multimodal dataset
zhongshsh/CLoT-Oogiri-GO
Over 130,000 examples of Oogiri humor games with images and user ratings in three languages.
What's inside
The large multimodal corpus contains over 130,000 examples of three types of Oogiri humor tasks in English, Chinese, and Japanese within jsonl files. Images and text joke responses to them are used as data. For 77.95% of the examples, human preferences are preserved as the number of likes reflecting the popularity of the joke.
What you can do with it
- Training VLMs to understand humor and generate funny image captions
- Model alignment (RLHF / DPO) via human preferences based on likes
- Testing multilingual visual question answering (VQA) in non-standard tasks
- Tasks
- Visual question answering, Question answering
- Languages
- English, Chinese, Japanese
- Size
- 100K–1M rows
- Downloads this month
- 372
- Likes
- 32
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("zhongshsh/CLoT-Oogiri-GO")