The Chinese Visual Embeddings are trained for the purpose of visual representations for the Chinese written signs. It is part of a larger project Graphein by Qianxun Chen

• Download 'NotoSansCJKsc-Regular' and 'NotoSansCJKtc-Regular' from Google Noto Fonts and save them to folder 'fonts'
• Create a new virtual environment with python 3 virtualenv --python=python3.6 venv
• Activate the environment activate source venv/bin/activate
• Install all the packages pip3 install -r requirements.txt

python preprocess.py -img

• An earlier version of the embeddings was trained with python CNN.py
• The latest version of the embeddings can be trained with python CNN_cuda_multiLabel.py

• Generate tsv files to preview embeddings in Embedding Projector python embeddings/VC/generateTSV.py
• Generate word2vec format embeddings python embeddings/VC/tsvs2txt.py

dict.csv is created based on the dictionary.txt file from the project, makemehanzi.