Skip to content

Dataset · guangyil

Filtered LAION-COCO Aesthetic Subset

guangyil/laion-coco-aesthetic

8.56M filtered English text-image pairs from LAION-COCO with aesthetic and watermark scores.

What's inside

A 10% English sample of LAION-COCO containing 8,563,753 filtered records. Links and special tokens were removed from texts, and images were filtered by resolution (above 384x384), aesthetic score (above 4.75), and watermark probability (below 0.5), with metrics preserved in the table.

What you can do with it

  • Training and fine-tuning text-to-image models with an aesthetic focus
  • Training image-to-text captioning models
  • Training classifiers to evaluate visual appeal and detect watermarks
Tasks
Image → text, Text → image
Languages
English
Size
1M–10M rows
Downloads this month
306
Likes
43

How to load it

from datasets import load_dataset

ds = load_dataset("guangyil/laion-coco-aesthetic")