MobileVLM

Paper Link

MobileVLM: A Vision-Language Model for Better Intra- and Inter-UI Understanding

News

• 2024.11.12 - Partial training data and random walk code for Mobile3M released!
• 2024.10.4 - Test data for Mobile3M released!
• 2024.9.26 - Our work accepted by EMNLP 2024 Findings!

1. Quick Start

Requirements

• transformers==4.32.0
• accelerate
• tiktoken
• einops
• transformers_stream_generator==0.0.4
• scipy
• torchvision
• pillow
• tensorboard
• matplotlib

2. Mobile3M Dataset

Training Data

Training data is available at the following link: data. We will gradually upload data for all apps.

Corpus Collection Script

To start collecting data, run the script main/corpus/googleCreatDataset/arm_graph_para_lock.py.

Example usage:

python googleCreatDataset/arm_graph_para_lock.py --device_name 10.53.89.79:6532 --systemPort 8112 --appid 8201 --command_executorhttp://127.0.0.1:4812/wd/hub--appPackage com.lucky.luckyclient --name_en lucky --diff_max 0.5 --diff_png 0.3 --waitadb 8 --prefix lucky0_3_1_2_ --recheck -1

Parameter Descriptions

• device_name: Name of the emulator.
• appid: Storage ID of the app being collected, e.g., 8201.
• command_executor: Appium system endpoint URL.
• --diff_max 0.5 --diff_png 0.3: Page similarity thresholds for differentiating screens.
• --prefix lucky0_3_1_2_: Distributed starting path for data collection.
• --recheck -1: Specifies whether to recheck previously collected data. Set to -1 for no recheck.

Data Generation Code for Each Task

The code for generating data for each task can be found in the following directories:

Our Test Data

Our test data is available at data.

4. License

The dataset of this project is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.

The source code of the this is licensed under the Apache 2.0 license.

Summary of Terms

• Attribution: You must give appropriate credit, provide a link to the license, and indicate if changes were made.
• NonCommercial: You may not use the material for commercial purposes.
• ShareAlike: If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original.

License Badge

5. Citation

If you'd like to use our benchmark or cite this paper, please kindly use the reference below:

@article{wu2024mobilevlm,
  title={Mobilevlm: A vision-language model for better intra-and inter-ui understanding},
  author={Wu, Qinzhuo and Xu, Weikai and Liu, Wei and Tan, Tao and Liu, Jianfeng and Li, Ang and Luan, Jian and Wang, Bin and Shang, Shuo},
  journal={arXiv preprint arXiv:2409.14818},
  year={2024}
}