Skip to content

It includes two datasets that are used in the downstream tasks for evaluating UIBert: App Similar Element Retrieval data and Visual Item Selection (VIS) data. Both datasets are written TFRecords.

Notifications You must be signed in to change notification settings

google-research-datasets/uibert

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 

Repository files navigation

UI understanding datasets for UIBert

This directory contains two datasets that are used in the downstream tasks for evaluating UIBert: Learning Generic Multimodal Representations for UI Understanding: app similar UI component retrieval data (AppSim) and referring expression component retrieval data (RefExp).

Both datasets are extended from the public Rico dataset, which contains 72k mobile app UI data. They add two different types of annotations to these UIs:

  1. In AppSim, each datapoint contains two UIs of similar category and the annotation of two semantically similar UI elements on them, such as a “Menu” buttons that appear on two UIs.
  2. In RefExp, each datapoint contains a UI and a referring expression of a UI element on it, such as “Red button on the top”.

We use unique image ids in both datasets to represent the UI and they can be used to retrieve the original UI data in Rico.

Data format

Both data are saved as TFRecords, which makes it easy to be loaded in Tensorflow.

  1. AppSim: Each TFRecord contains unique image ids of an anchor UI and a search UI. The anchor UI has one anchor UI element and the search UI has 10 UI elements, one of which is semantically similar to the anchor UI element. Detailed features are itemized as below.

    • anchor/image/anchor/bbox/{xmax|xmin|ymax|ymin}: float_list[1]. Bounding box of the anchor UIelement on the anchor UI.

    • {anchor|search}/image/id: bytes_list[1]. Image id of the anchor and search UIs.

    • search/image/candidate/bbox/{xmax|xmin|ymax|ymin}: float_list[10]. 10 candidate UI elements on the search UI. One of them is semantically similar to the anchor UI element of the anchor UI. Its index is saved in search/correct/index.

    • search/correct/index: int64_list[1]. Index of the semantically simialr UI element in the search UI candidates list.

  2. RefExp: Each TFRecord has the following features. Values under image/view_hierarchy are information extracted from UI’s view hierarchy file.

    • image/id: bytes_list[1]. Image id.

    • image/ref_exp/label: int64_list[1]. Index of the UI element that is referred to by the referring expression in the image/object/bbox/… list.

    • image/ref_exp/text: bytes_list[1]. Referring expression.

    • image/object/bbox/{xmax|xmin|ymax|ymin}: float_list of variable length. UI elements on the UI. The number of elements is denoted in image/object/num.

    • image/object/num: float_list[1]. Number of UI elements.

    • image/view_hierarchy/bbox/{xmax|xmin|ymax|ymin}: float list of variable length. Bounding boxes of leaf nodes in the view hierarchy.

    • image/view_hierarchy/class/label: int64_list. Label of the class name.

    • image/view_hierarchy/class/name: bytes_list. Processed class name string in the view hierarchy.

    • image/view_hierarchy/description bytes_list. Content description string in the view hierarchy.

    • image/view_hierarchy/id/name: bytes_list. Processed source id string in the view hierarchy.

    • image/view_hierarchy/text: byte_list. Text string in the view hierarchy.

Citation

If you use or discuss this dataset in your work, please cite our paper:

@misc{bai2021uibert,
      title={UIBert: Learning Generic Multimodal Representations for UI Understanding},
      author={Chongyang Bai and Xiaoxue Zang and Ying Xu and Srinivas Sunkara and Abhinav Rastogi and Jindong Chen and Blaise Aguera y Arcas},
      year={2021},
      eprint={2107.13731},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

License

Datasets are licensed under CC BY 4.0.

Contact us

If you have a technical question regarding the dataset or publication, please create an issue in this repository.

About

It includes two datasets that are used in the downstream tasks for evaluating UIBert: App Similar Element Retrieval data and Visual Item Selection (VIS) data. Both datasets are written TFRecords.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published