Skip to content

Dataset · showlab

Desktop screenshots for training GUI agents

showlab/ShowUI-desktop

A set of PC screenshots with interface element markup and GPT-4o descriptions for UI-grounding tasks.

What's inside

The dataset includes between 1 and 10 thousand examples (about 8 thousand in the dataset name), combining images, text, and time series. Screenshots and basic markup are taken from OmniAct, then supplemented with GPT-4o descriptions of appearance, spatial relations, and interface element purpose.

What you can do with it

  • Training multimodal agents to find and click UI elements on PC
  • VLM fine-tuning for understanding desktop application structure
Size
1K–10K rows
Downloads this month
2.6K
Likes
37

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("showlab/ShowUI-desktop")