metadata
dataset_info:
features:
- name: image_url
dtype: string
- name: image
dtype: image
- name: instruction
dtype: string
- name: bbox
sequence: float32
- name: point
sequence: float32
- name: type
dtype: string
splits:
- name: train
num_bytes: 16591347652.088
num_examples: 7496
download_size: 327573839
dataset_size: 16591347652.088
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Github | arXiv | HF Paper | Spaces | Datasets | Quick Start
ShowUI-desktop-8K is a UI-grounding dataset focused on PC-based grounding, with screenshots and annotations originally sourced from OmniAct.
We utilize GPT-4o to augment the original annotations, enriching them with diverse attributes such as appearance, spatial relationships, and intended functionality.
You can use our rewrite strategy code to augment your own data.
If you find our work helpful, please consider citing our paper.
@misc{lin2024showui,
title={ShowUI: One Vision-Language-Action Model for GUI Visual Agent},
author={Kevin Qinghong Lin and Linjie Li and Difei Gao and Zhengyuan Yang and Shiwei Wu and Zechen Bai and Weixian Lei and Lijuan Wang and Mike Zheng Shou},
year={2024},
eprint={2411.17465},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2411.17465},
}