This is the Tensorflow implementation of the SKFont: Skeleton-Driven Korean Font Synthesis with Conditional Deep Adversarial Networks.
In our research, we study the problem of font synthesis using an end-to-end conditional deep adversarial network with a small sample of Korean characters (Hangul). Hangul comprises of 11,172 characters and is composed by writing in multiple placement patterns. Traditionally, font design has required heavy-loaded human labor, easily taking one year to finish one style set. Even with the help of programmable approaches, it still takes a long time and cannot escape the limitations around the freedom to change parameters. Many trials have been attempted in deep neural network areas to generate characters without any human intervention. Our research focuses on an end-to-end deep learning model, the Skeleton-Driven Font generator (SKFont): when given 114 samples, the system automatically generates the rest of the characters in the same given font style. SKFont involves three steps: First, it generates complete target font characters by observing 114 target characters. Then, it extracts the skeletons (structures) of the synthesized characters obtained from the first step. This process drives the system to sustain the main structure of the characters throughout the whole generation processes. Finally, it transfers the style of the target font onto these learned structures. Our study resolves long overdue shortfalls such as blurriness, breaking, and a lack of delivery of delicate shapes and styles by using the ‘skeleton-driven’ conditional deep adversarial network. Qualitative and quantitative comparisons with the state-of-the-art methods demonstrate the superiority of the proposed SKFont method.
- Windows
- CPU or NVIDIA GPU + CUDA cuDNN
- python 3.6.8
- tensorflow-gpu 1.13.1
- pillow 6.1.0
-
conda create --name tutorial-TF python=3.6.8 -
conda activate tutorial-TF or activate tutorial-TF -
conda install -c anaconda tensorflow-gpu=1.13.1 -
conda env update --file tools.yml
Our model consists of three sub models namely F2F-F2S-S2F. For each model we have to prepare a paired dataset. i.e. a source to target font paired dataset, a target font to corresponing skeleton dataset, and a target skeleton to corresponding font dataset. To do this place any korean font in src_font directory and N number of target fonts in the trg_font directory. Then run the below commands for data preprocessing.
-
Generate Source font images
python ./tools/src-font-image-generator.py -
Generate Target font images
python ./tools/trg-font-image-generator.py -
Generate Target font skeleton images
python ./tools/trg-skeleton-image-generator.py -
Combine source, target, and target skeletons
python ./tools/combine_images.py --input_dir src-image-data/images --b_dir trg-image-data/images --c_dir skel-image-data/images --operation combine -
Convert images to TFRecords
python ./tools/images-to-tfrecords.py
python main.py --mode train --output_dir trained_model --max_epochs 25
To learn an unseen font style you can fine tune an already pre-trained model with the below command. If you want to generate the already learnt font styles just skip the below command.
python main.py --mode train --output_dir finetuned_model --max_epochs 500 --checkpoint trained_model/
Generate images just like before but this time use a different module for creating testing TFRecords with the below mentioned command.
- Convert images to TFRecords
python ./tools/test-images-to-tfrecords.py
python main.py --mode test --output_dir testing_results --checkpoint finetuned_model
This code is inspired by the pix2pix tensorflow project.
Special thanks to the following works for sharing their code and dataset.
Please cite our work if you like it.
Ko, D.H., Hassan, A.U., Suk, J. et al. SKFont: skeleton-driven Korean font generator with conditional deep adversarial networks. IJDAR (2021). https://doi.org/10.1007/s10032-021-00374-4
The code and other helping modules are only allowed for PERSONAL and ACADEMIC usage.