Handwriting-Generation-Using-Recurrent-Neural-Networks

This is an implementation of handwriting generation with use of recurrent neural networks (LSTM & GRU) using torch and python. The work is based on Alex Graves paper published in 2013.

Installation

Clone the GitHub repository and install the dependencies.

• Install

  • Anaconda (for creating and activating a separate environment)
  • pytest==3.2.1
  • numpy=1.13.3
  • matplotlib
  • tqdm==4.17.1
  • colorama==0.3.9
  • scikit-learn==0.19.1
  • pytorch==0.2.0 -c soumith

• Clone the repo and go to the directory

$ git clone https://github.com/AizazSharif/Handwriting-Generation-Using-Recurrent-Neural-Networks.git
$ cd Handwriting-Generation-Using-Recurrent-Neural-Networks

Training

The pretrained model is saved in models/ directory. Once you get the concept of Conditional and Unconditional Handwriting Generation you can train your own model by changing the setting in configurations.py.

For training the model use :

python main.py --train_conditional
python main.py --train_unconditional

Validation

Validation can be simply done by running:

python main.py --validate_conditional --conditional_model_path /Handwriting-Generation-Project/models/conditional.pt
python main.py --validate_unconditional --unconditional_model_path /Handwriting-Generation-Project/models/unconditional.pt

Credit

This project is a part of Deep Learning related task and all credit goes to Lyrebird for introducing me to this incredible work.