Skip to content

Latest commit

 

History

History
89 lines (68 loc) · 4.33 KB

README.md

File metadata and controls

89 lines (68 loc) · 4.33 KB
Raw

Official PyTorch Implementation of "WordStylist: Styled Verbatim Handwritten Text Generation with Latent Diffusion Models" - ICDAR 2023

ArXiv Paper

ICDAR Paper

Abstract: Text-to-Image synthesis is the task of generating an image according to a specific text description. Generative Adversarial Networks have been considered the standard method for image synthesis virtually since their introduction. Denoising Diffusion Probabilistic Models are recently setting a new baseline, with remarkable results in Text-to-Image synthesis, among other fields. Aside its usefulness per se, it can also be particularly relevant as a tool for data augmentation to aid training models for other document image processing tasks. In this work, we present a latent diffusion-based method for styled text-to-text-content-image generation on word-level. Our proposed method is able to generate realistic word image samples from different writer styles, by using class index styles and text content prompts without the need of adversarial training, writer recognition, or text recognition. We gauge system performance with the Fréchet Inception Distance, writer recognition accuracy, and writer retrieval. We show that the proposed model produces samples that are aesthetically pleasing, help boosting text recognition performance, and get similar writer retrieval score as real data.

Dataset & Pre-processing

Download the data/words.tgz of IAM Handwriting Database: https://fki.tic.heia-fr.ch/databases/iam-handwriting-database.

Then, pre-process the word images by running:

python prepare_images.py

Before running the prepare_images.py code make sure you have changed the iam_path and save_dir to the corresponding data/words.tgz and the path to save the processed images.

Training from scratch

To train the diffusion model run:

python train.py --iam_path path/to/processed/images --save_path path/to/save/models/and/results

Trained Model

We provide the weights of a trained model, which you can download from: trained_model (Download models folder with the 3 pt. files).

Sampling - Regenerating IAM

If you want to regenerate the full IAM training set you can run:

python full_sampling.py --save_path path/to/save/generated/images --models_path /path/to/trained/models

Sampling - Single image

If you want to generate a single word with a random style you can run:

python sampling.py --save_path path/to/save/generated/images --models_path /path/to/trained/models --words ['hello']

Citation

If you find the code useful for your research, please cite our paper:

@inproceedings{nikolaidou2023wordstylist,
  title={Wordstylist: styled verbatim handwritten text generation with latent diffusion models},
  author={Nikolaidou, Konstantina and Retsinas, George and Christlein, Vincent and Seuret, Mathias and Sfikas, Giorgos and Smith, Elisa Barney and Mokayed, Hamam and Liwicki, Marcus},
  booktitle={International Conference on Document Analysis and Recognition},
  pages={384--401},
  year={2023},
  organization={Springer}
}

@article{nikolaidou2023wordstylist,
  title={{WordStylist: Styled Verbatim Handwritten Text Generation with Latent Diffusion Models}},
  author={Nikolaidou, Konstantina and Retsinas, George and Christlein, Vincent and Seuret, Mathias and Sfikas, Giorgos and Smith, Elisa Barney and Mokayed, Hamam and Liwicki, Marcus},
  journal={arXiv preprint arXiv:2303.16576},
  year={2023}
}

Acknowledgements

We would like to thank the researchers of Stable Diffusion, GANwriting, SmartPatch, and HTR best practices for releasing their code.