The official GitHub repository for the paper on DocSynth: A Layout Guided Approach for Controllable Document Image Synthesis

The work will feature in 16th ICDAR 2021 (Lausanne, Switzerland).

Getting Started

Step 1: Clone this repository and change directory to repository root

git clone https://github.com/biswassanket/synth_doc_generation.git
cd synth_doc_generation

Step 2: Make sure you have conda installed. If you do not have conda, here's the magic command to install miniconda.

curl -o ./miniconda.sh -O https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh
chmod +x ./miniconda.sh
./miniconda.sh -b -u -p .

Step 3: Create an environment to run the project and install required dependencies.

• To create conda environment: conda env create -f environment.yml

Step 4: Activate the conda environment

conda activate layout2im

Step 5: Downloading Dataset

• To download PubLayNet dataset: curl -o <YOUR_TARGET_DIR>/publaynet.tar.gz https://dax-cdn.cdn.appdomain.cloud/dax-publaynet/1.0.0/publaynet.tar.gz

Step 5: Downloading the trained models

• Download the trained models to checkpoints/pretrained/.

1. trained model on PubLayNet

Step 6: Testing

Testing on PubLayNet dataset:

$ python layout2im/test.py --dataset publaynet --coco_dir datasets/publaynet \
                           --saved_model checkpoints/pretrained/publaynet_netG.pkl \
                           --results_dir checkpoints/pretrained_results_publaynet

Step 7: Training

$ python layout2im/train.py

Results

1. T-SNE Visualization of Synthetic Document Images

2. Diverse results generated from the same document layout

3. Examples of interactive generation of documents by the user

Citation

If you find this code useful in your research then please cite

@inproceedings{biswas2021docsynth,
  title={DocSynth: A Layout Guided Approach for Controllable Document Image Synthesis},
  author={Biswas, Sanket and Riba, Pau and Llad{\'o}s, Josep and Pal, Umapada},
  booktitle={International Conference on Document Analysis and Recognition (ICDAR)},
  year={2021}
}

Acknowledgement

Our project has adapted and borrowed the code structure from layout2im. We thank the authors. This research has been partially supported by the Spanish projects RTI2018-095645-B-C21, and FCT-19-15244, and the Catalan projects 2017-SGR-1783, the CERCA Program / Generalitat de Catalunya and PhD Scholarship from AGAUR (2021FIB-10010).

Authors

• Sanket Biswas
• Pau Riba

Conclusion

Thank you and sorry for the bugs!