kanjivg-ML

Note

These tools are based on the KanjiVG project.

This is an unofficial implementation to generate datasets for Machine learning experiments from the Kanjivg project.

Overview

• Source Data: KanjiVG (Kanji Vector Graphics)
• Status: Unofficial, for research/personal use.

Setup

• python 3.10>=
• ubuntu

1. Clone the KanjiVG-ML repository:

git clone https://github.com/rishiyama/kanjivg-ML
cd kanjivg-ML

2. Clone the KanjiVG repository and initialize it:

git clone https://github.com/KanjiVG/kanjivg.git
# fix kanjivg/__init__.py to import kanjivg
bash scripts/init.sh 

3. Install any required dependencies:

cairo:

pip install CairoSVG 
apt install libcairo2

Optional:

if you can get the output like this, then you are ready to use the kanjivg and kanjivg-ML package.

$ python example.py 
Is 0x4E00 a kanji? True

Usage

Generate a dataset:

python run.py

and also, you can customize the parameters of png-images, such as width, height, and save directory by using the following command:

# same as default
python run.py --path ./kanjivg/kanji --width 256 --height 256 --save_dir ./output

result

output
|-- kanji
|   |-- png
|   |-- png_white
|   `-- svg
`-- other
    |-- png
    |-- png_white
    `-- svg

• kanji: contains the kanji images by filtering.
• other: contains the non-kanji images by filtering.
• png: contains the kanji images in PNG format with a transparent background.
• png_white: contains the kanji images in PNG format with a white background.
• svg: contains the kanji images in SVG format, using simplified SVG paths.

Acknowledgments

This project is heavily reliant on the fantastic work done by the KanjiVG project.