README.md

metadata

license: apache-2.0

DeskVision: Large Scale Desktop Region Captioning for Advanced GUI Agents

[💻Code][📝Paper] [🤗Models][🤗Data]

We have open-sourced the code for generating DeskVision data, which includes two tools, Detector and Captioner. For details and how to use them, see [📝Paper] and [💻Code]. We have also open-sourced (part of) the DeskVision data generated based on these tools. Due to data legitimacy reasons, self-built image data is presented in URL format. More data will be added in the future. We also generate Region Captions for the open source Desktop complete image data of OS-Atlas, and also open source the related annotations.

Data format description

1. DeskVision_url

{"url": xx, "region captions": xx}

2. OS-Atlas_desktop_DeskVision

{"image path": xx, "region captions": xx}

The region captions give each interactive element ID, bbox (x1,y1,x2,y2) and caption.

Citation

If you use DeskVision datasets for your research, please cite our [📝Paper]：

@misc{xu2025deskvisionlargescaledesktop,
      title={DeskVision: Large Scale Desktop Region Captioning for Advanced GUI Agents}, 
      author={Yibin Xu and Liang Yang and Hao Chen and Hua Wang and Zhi Chen and Yaohua Tang},
      year={2025},
      eprint={2503.11170},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2503.11170}, 
}