Jl-wei's picture
Update image download and verification instructions and research-use notice
0d25dbe verified
|
Raw History Blame Contribute Delete
2.04 kB

PhoneWatchUI: Google Play

Layout

google_play/
├── dataset/
│   ├── apps.jsonl
│   ├── images.jsonl
│   └── images/<app_folder>/<position>.<extension>
└── script/
    ├── app_collect.py
    ├── watch_app_filter.py
    ├── image_download.py
    ├── device_filter.py
    └── requirements.txt

Fields

apps.jsonl contains app_id, store, title, description, source_url, and store-specific metadata. images.jsonl contains gallery-position identity, dimensions, source URL, SHA-256, relative image_path, and device labels. Paths are relative to the directory containing images.jsonl; unavailable files have null paths. device is provisionally watch for square images, phone for non-squares with long/short ratio ≥ 1.5, and other otherwise or without dimensions. device_label_source records the aspect-ratio method and threshold. See all field definitions.

Usage

Use Python 3.10+ and run from google_play/. Create package_ids.txt with one package ID per line, and choose a fresh output directory:

python3 -m pip install -r script/requirements.txt
python3 script/app_collect.py --ids package_ids.txt --output fresh/candidates.jsonl
python3 script/watch_app_filter.py --input fresh/candidates.jsonl --output fresh/dataset/apps.jsonl
python3 script/image_download.py --apps fresh/dataset/apps.jsonl --output fresh/dataset
python3 script/device_filter.py --input fresh/dataset/images.jsonl --output fresh/dataset/images.jsonl --phone-ratio 1.5

The first three scripts collect app metadata, select watch-related apps, and download images. image_download.py produces intermediate records without device or device_label_source; device_filter.py adds them and keeps every record by default. To export one device subset, add --device watch (or phone/other) and use a different output path, such as fresh/dataset/watch-images.jsonl. Use --help on each script for options.