Table of Contents

1. About The Project
2. Getting Started
   • Installation
3. Usage
4. License
5. Contact

About The Project

Sign language generation and recognition using synthetic data augmentation. Repository for the models and experiments used in the paper Sign Generation for Data Augmentation.

Conditional Human motion prediction model:

• CMLPe

Sign language recognition models:

• Mamba
• Transformer

(back to top)

Getting Started

Generate and classify sign language gestures with HandCraft. To do it follow these steps:

Installation

Clone the repo

git clone https://github.com/okason97/HandCraft.git

(back to top)

Usage

1. Create a config file. Examples in /src/configs
2. Use /script.sh to use the models.

Sign Language Recognition

./script.sh classification mamba original128-pad LSFB

Sign Language Generation

./script.sh cond_prediction CsiMLPe depth_big_noise_0.1 LSFB

Example of a generated sequence:

(back to top)

Publication

Paper: HandCraft: Dynamic Sign Generation for Synthetic Data Augmentation

Citation

If you use this work in your research, please cite:

@misc{rios2025handcraftdynamicsigngeneration,
      title={HandCraft: Dynamic Sign Generation for Synthetic Data Augmentation}, 
      author={Gaston Gustavo Rios and Pedro Dal Bianco and Franco Ronchetti and Facundo Quiroga and Oscar Stanchi and Santiago Ponte Ahón and Waldo Hasperué},
      year={2025},
      eprint={2508.14345},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2508.14345}, 
}

License

Distributed under the MIT License. See LICENSE.txt for more information.

(back to top)

Contact

Gaston Rios - [email protected]

(back to top)