tonysy / FewShot_ChineseCharacter_GAN

Google ML Winter Camp(2020 Shanghai) Chinese Character Style Transfer with Few-shot Images based on Conditional GAN

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Chinese Character Generation with Few-shot Images

This project is design to generate personalized chinese character with few-shot images.

Installation

# Install some packages
pip install dominate

# Install the project
python setup.py build develop  

Algorithms

Data Preparation

We collect 56 fonts from the Internet.

We first extract the images from the ttf with

Model Description

Traning

Step-1: Train Font Classification Network

cd tools
bash tools/cat_embedding_train.sh

Step-2: Extract Category Embedding Features

bash tools/cat_embedding_train_feat_extraction.sh

Step-3: Train Category-aware Pix2Pix

bash tools/gan_model_version1.sh

Inference

Visualization

Server

Flask

Front-end

HTML

PyQt

Acknoledgement

About

Google ML Winter Camp(2020 Shanghai) Chinese Character Style Transfer with Few-shot Images based on Conditional GAN


Languages

Language:JavaScript 52.4%Language:CSS 37.2%Language:Python 9.7%Language:HTML 0.5%Language:Shell 0.2%