ML-Danbooru: Anime image tags detector

[Modified] Usage:

package usage:

1. install the package: pip install -e .

2. get labels by importing function from the package:

from ml_danbooru_tagger import infer_batch_with_defaults
bench_tag_dict = infer_batch_with_defaults("bench_images_small")

package usage:

to build the package:

python setup.py bdist_wheel

CLI usage:

1. clone repo & install deps:

git clone https://github.com/troph-team/ml-danbooru-tagger && cd ml-danbooru-tagger
pip install -r requirements.txt

2. start tagging:

   • automatically downloads models

   • optimal settings preconfigured

   • tags will be saved as json file next to the tagged folder

   • for more args available see the python file

python demo_ca.py --data {IMAGE_DIR_PATH}

modifications:

• better error handling
• can be used as a pacakge on downsteraam projects
• downloads model automatically when not given
• keeps probability of tags along with string repr
• adds batched inference by default
• modified json save path

[Original Readme]

see: https://github.com/IrisRainbowNeko/ML-Danbooru