---

GANJI

This kanji does not exist.

Installation

This project uses Poetry.

Linux (using local TensorFlow)

git clone https://github.com/taskie/ganji.git
cd ganji
# change the global setting if you want to create .venv under the project directory
# poetry config virtualenvs.in-project true
poetry install
poetry install -E tensorflow

Linux (using system TensorFlow)

git clone https://github.com/taskie/ganji.git
cd ganji
poetry config virtualenvs.in-project true  # change the global setting
python3 -m venv --system-site-packages .venv
poetry install

Windows

chcp 65001
git clone https://github.com/taskie/ganji.git
cd ganji
REM (change the global setting if you want to create .venv under the project directory)
REM poetry config virtualenvs.in-project true
poetry install
REM (if you want to install TensorFlow in .venv)
poetry install -E tensorflow
REM (freetype.dll may be needed under %PATH%. see https://pypi.org/project/freetype-py/)

Usage

Initialize

poetry run ganji new -F /usr/share/fonts/OTF/ipaexm.ttf -d 0.025 -D 0.975 ipaexm
# or
cd ipaexm
poetry run ganji init -F /usr/share/fonts/OTF/ipaexm.ttf -d 0.025 -D 0.975

Train

cd ipaexm
poetry run ganji train
# or
poetry run ganji -C ipaexm train

Generate

poetry run ganji generate

Show logs

poetry run ganji log

Examples

• examples/dcgan/ipaexm

Usage (detail)

Show density of glyphs

poetry run python -m ganji.datasets \
    -F /usr/share/fonts/OTF/ipaexm.ttf -I 0 -S 40 \
    -c joyo-kanji -d 0.995 --show-density

嚇 (U+5687) 77.52
翻 (U+7FFB) 78.29
鋼 (U+92FC) 78.97625
繊 (U+7E4A) 79.169375
酬 (U+916C) 79.875625
綱 (U+7DB1) 79.9275
醜 (U+919C) 80.4075
鬱 (U+9B31) 80.62625
醸 (U+91B8) 81.4575
欄 (U+6B04) 81.908125
臓 (U+81D3) 83.02

Render glyphs

poetry run python -m ganji.datasets \
    -F /usr/share/fonts/OTF/ipaexm.ttf -I 0 -S 16 \
    -c hiragana -d 1

                        ++**++
  --                  --++--++
  ++++    ----        --++--++
  --**--  ++********++--++**++
  --**            **
  ++++            **
  ++--  --++------**++****--
  ++--    --********++----
  ++              **
--++--++          **--
--++++--      ----**--
  **++    ++**********++++
  ++++  --**------**--++**++
  ++++    ++******++    --

Tested Environments

• CPU: AMD Ryzen 7 1700 Eight-Core Processor
• GPU: NVIDIA GeForce GTX 1060 6GB
• RAM: SanMax Technologies DDR4-2400 16GB (2x8GB)
• Arch Linux (linux 5.7.4.arch1-1)
  • Python 3.8.3
  • CUDA 10.2.89
  • cuDNN 7.6.5
  • python-tensorflow-opt-cuda 2.2.0-1
• Windows 10 Home
  • Python 3.7.7
  • CUDA 10.1.243
  • cuDNN 7.6.5
• Python Libraries: see pyproject.toml or poetry.lock

References

• eriklindernoren/Keras-GAN: Keras implementations of Generative Adversarial Networks.
• [1511.06434] Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• jacobgil/keras-dcgan: Keras implementation of Deep Convolutional Generative Adversarial Networks
• はじめてのGAN
• Keras 2 で”はじめてのGAN” - Qiita