Dataset · Ejafa
Cleaned selection of image links from LAION-POP
Ejafa/ye-pop
A filtered and human-verified set of 550 thousand image links with descriptions.
What's inside
The dataset is a thoroughly cleaned version of LAION-POP with 100k to 1M English records. It is split into 11 parts of 50,000 image URLs each, passed NSFW filtering and manual review, with the first part additionally processed using Gemini-Pro.
What you can do with it
- Training and fine-tuning text-to-image generative models on aesthetic data
- Training multimodal models and visual content quality filters
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 481
- Likes
- 41
Models trained on it
- thwri/CogFlorence-2.2-LargeImage → text 3.9K
- thwri/CogFlorence-2.1-LargeImage → text 3.6K
- thwri/CogFlorence-2-Large-FreezeImage → text 439
- etri-vilab/koala-lightning-700mText → image 37
- etri-vilab/koala-lightning-1.7bText → image 33
- etri-vilab/koala-lightning-1bText → image 26
- dngerle/CogFlorence-2.2-LargeImage → text 8
- dddv587/CogFlorence-2.2-LargeImage → text 8
How to load it
from datasets import load_dataset
ds = load_dataset("Ejafa/ye-pop")