BLIP helper

A script for tagging picture with DeepDanbooru and BLIP

This script is NOT intended to be publish to PyPI. Clone this repository and run it directly, since I can't make sense of the package management of Python. Help wanted!

Usage

git clone https://github.com/crosstyan/blip_helper
cd blip_helper
git submodule update --init --recursive
pip install -r requirements.txt
pip install repo/DeepDanbooru
# using conda is recommended
# You would need pycocotoools
# https://anaconda.org/conda-forge/pycocotools
python run.py --path /path/to/image

See args_parser.py for more options.

TODO

• Add support for metal for BLIP (based on PyTorch)
• Add support for half precision for BLIP
• Add support for choosing GPU for BLIP (trivial but why?)
• A workflow for tagging
• filter out unwanted tags like orginal anime character

Misc

• Get started with tensorflow-metal for MacOS user

Choose Catagories

check repo/DeepDanbooru/deepdanbooru/commands/download_tags.py

category_definitions = [
    {
        "category_name": "General",
        "category": "general",
        "path": os.path.join(project_path, "tags-general.txt"),
    },
    {
       'category_name': 'Artist',
       'category': 'artist',
       'path': os.path.join(path, 'tags-artist.txt'),
    },
    {
       'category_name': 'Copyright',
       'category': 'copyright',
       'path': os.path.join(path, 'tags-copyright.txt'),
    },
    {
        "category_name": "Character",
        "category": "character",
        "path": os.path.join(project_path, "tags-character.txt"),
    },
]

Comment out the ones you don't want to apply. Usually I just leave general on.

MacOS

• arm64 support for M1

git clone https://github.com/tensorflow/io tfio
cd tfio
python3 setup.py -q bdist_wheel --project tensorflow_io_gcs_filesystem
pip install .