Skip to content

Dataset · xwm

Cleaned UI annotations and screenshots from video tutorials

xwm/WildGUI

Large corpus of screenshots and reworked interface annotations from web videos for GUI agents.

What's inside

This is a reworked and cleaned release of annotations for the WildGUI dataset, originally gathered from online instructional videos. It contains images (screenshots) and English text annotations ranging from 10 to 100 million records. The data are regenerated and reformatted for easy reproduction and analysis.

What you can do with it

  • Pre-training multimodal GUI agents to recognize interface elements and actions
  • Training models to map text action descriptions to screen regions
Languages
English
Size
10M–100M rows
Downloads this month
44.9K
Likes
8

How to load it

from datasets import load_dataset

ds = load_dataset("xwm/WildGUI")