kanji-dataset / README.md
LLParallax's picture
Update README.md
1efdc8b verified
metadata
dataset_info:
  features:
    - name: image
      dtype: image
    - name: text
      dtype: string
  splits:
    - name: train
      num_bytes: 305177414.32
      num_examples: 6410
  download_size: 303927894
  dataset_size: 305177414.32
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

How to generate dataset

Download and extract definitions and matching SVG files

https://www.edrdg.org/kanjidic/kanjidic2.xml.gz
https://github.com/KanjiVG/kanjivg/releases/download/r20220427/kanjivg-20220427.xml.gz

Then run those scripts with python

python render_images.py
python create_dataset.py