AGIS-Net

Introduction

This is the official PyTorch implementation of the Artistic Glyph Image Synthesis via One-Stage Few-Shot Learning.

paper | supplementary material

Abstract

Automatic generation of artistic glyph images is a challenging task that attracts many research interests. Previous methods either are specifically designed for shape synthesis or focus on texture transfer. In this paper, we propose a novel model, AGIS-Net, to transfer both shape and texture styles in one-stage with only a few stylized samples. To achieve this goal, we first disentangle the representations for content and style by using two encoders, ensuring the multi-content and multi-style generation. Then we utilize two collaboratively working decoders to generate the glyph shape image and its texture image simultaneously. In addition, we introduce a local texture refinement loss to further improve the quality of the synthesized textures. In this manner, our one-stage model is much more efficient and effective than other multi-stage stacked methods. We also propose a large-scale dataset with Chinese glyph images in various shape and texture styles, rendered from 35 professional-designed artistic fonts with 7,326 characters and 2,460 synthetic artistic fonts with 639 characters, to validate the effectiveness and extendability of our method. Extensive experiments on both English and Chinese artistic glyph image datasets demonstrate the superiority of our model in generating high-quality stylized glyph images against other state-of-the-art methods.

Model Architecture

Skip Connection	Local Discriminator

Some Results

Prerequisites

• Linux
• CPU or NVIDIA GPU + CUDA cuDNN
• Python 3
• PyTorch 0.4.0+

Get Started

Installation

1. Install PyTorch, torchvison and dependencies from https://pytorch.org
2. Install python libraries visdom and dominate:

   pip install visdom
   pip install dominate

3. Clone this repo:

   git clone -b master --single-branch https://github.com/hologerry/AGIS-Net
   cd AGIS-Net

4. Download the offical pre-trained vgg19 model: vgg19-dcbb9e9d.pth, and put it under the models/ folder

Datasets

The datasets server is down, you can download the datasets from PKU Disk, Dropbox or MEGA. Download the datasets using the following script, four datasets and the raw average font style glyph image are available.

It may take a while, please be patient

bash ./datasets/download_dataset.sh DATASET_NAME

• base_gray_color English synthesized gradient glyph image dataset, proposed by MC-GAN.
• base_gray_texture English artistic glyph image dataset, proposed by MC-GAN.
• skeleton_gray_color Chinese synthesized gradient glyph image dataset by us.
• skeleton_gray_texture Chinese artistic glyph image dataset proposed by us.
• average_skeleton Raw Chinese avgerage font style (skeleton) glyph image dataset proposed by us.

Please refer to the data for more details about our datasets and how to prepare your own datasets.

Model Training

• To train a model, download the training images (e.g., English artistic glyph transfer)

  bash ./datasets/download_dataset.sh base_gray_color
  bash ./datasets/download_dataset.sh base_gray_texture

• Train a model:

  1. Start the Visdom Visualizer

     python -m visdom.server -port PORT

     PORT is specified in train.sh

  2. Pretrain on synthesized gradient glyph image dataset

     bash ./scripts/train.sh base_gray_color GPU_ID

     GPU_ID indicates which GPU to use.

  3. Fineture on artistic glyph image dataset

     bash ./scripts/train.sh base_gray_texture GPU_ID DATA_ID FEW_SIZE

     DATA_ID indicates which artistic font is fine-tuned.
     FEW_SIZE indicates the size of few-shot set.

     It will raise an error saying:

     FileNodeFoundError: [Error 2] No such file or directory: 'chechpoints/base_gray_texture/base_gray_texture_DATA_ID_TIME/latest_net_G.pth
     

     Copy the pretrained model to above path

     cp chechpoints/base_gray_color/base_gray_color_TIME/latest_net_* chechpoints/base_gray_texture/base_gray_texture_DATA_ID_TIME/

     And start train again. It will works well.

Model Testing

• To test a model, copy the trained model from checkpoint to pretrained_models folder (e.g., English artistic glyph transfer)

  cp chechpoints/base_gray_color/base_gray_texture_DATA_ID_TIME/latest_net_* pretrained_models/base_gray_texture_DATA_ID/

• Test a model

  bash ./scripts/test_base_gray_texture.sh GPU_ID DATA_ID

Acknowledgements

This code is inspired by the BicycleGAN.

Special thanks to the following works for sharing their code and dataset.

• pix2pix
• MSGAN
• MC-GAN
• TET-GAN
• EMD

Citation

If you find our work is helpful, please cite our paper:

@article{Gao2019Artistic,
  author = {Gao, Yue and Guo, Yuan and Lian, Zhouhui and Tang, Yingmin and Xiao, Jianguo},
  title = {Artistic Glyph Image Synthesis via One-Stage Few-Shot Learning},
  journal = {ACM Trans. Graph.},
  issue_date = {November 2019},
  volume = {38},
  number = {6},
  year = {2019},
  articleno = {185},
  numpages = {12},
  url = {http://doi.acm.org/10.1145/3355089.3356574},
  publisher = {ACM}
} 

Copyright

The code and dataset are only allowed for PERSONAL and ACADEMIC usage.