Joint-Implicit-Representation

This is the official implementation of: Joint Implicit Neural Representation for High-fidelity and Compact Vector Fonts.

Installation

Requirement

• Python 3.8 or higher
• PyTorch 1.9.0
• Torchvision 0.10.0

1. Clone the repository.

git clone https://github.com/Acc-plus/Joint-Implicit-Representation.git
git --init --recursive
cd Joint-Implicit-Representation
pip install -r requirements.txt

2. Install the extension.

cd utils/extension
python setup.py install

3. Install torchmeta.

cd torch meta
python setup.py install

4. (Recommend) Install specified version of pytorch. If you install a higher version of pytorch, the torchmeta module will not be import correctly. You might resolve the problem by commenting out line 39 in torchmeta/dataset/utils.py. This operation has been verified on torch2.0.0+cu118.

conda install pytorch==1.9.0 torchvision==0.10.0 torchaudio==0.9.0 cudatoolkit=10.2 -c pytorch

Training a model from scratch

Data preparation

1. Download VecFont dataset, following the instruction in DeepVecFont and place the test_all.pkl in ./dataset/ folder.

2. Generate the training data.

python gen_fonts.py

Training

1. Train the SDF Net.

python train.py --config utils/sdf.yaml

2. Train the CF Net.

python train.py --config utils/cf.yaml

Inference

You can download pretrained model from Google Drive.

1. Generate binary image from SDF Net.

python inf_sdf.py --config utils/sdf.yaml

2. Generate corners from CF Net.

python inf_cf.py --config utils/cf.yaml

3. Generate SVGs.

python vectorization.py

Citation

Please cite the following paper if this work helps your research:

@inproceedings{chen2023joint,
	title={Joint Implicit Neural Representation for High-fidelity and Compact Vector Fonts},
	author={Chia-Hao Chen and Ying-Tian Liu and Zhifei Zhang and Yuan-Chen Guo and Song-Hai Zhang},
    booktitle={IEEE International Conference on Computer Vision},
    year={2023}
}

Contact

If you have any questions, please contact accplusjh@gmail.com

License

Licensed under the MIT license.

Acknowledgement

• This implementation takes DIF-Net as a reference. We thank the authors for their excellent work.
• DeepVecFont