Skip to content

Latest commit

 

History

History
24 lines (15 loc) · 716 Bytes

readme.md

File metadata and controls

24 lines (15 loc) · 716 Bytes

Handwriting generation Task

Implementation of "Generating Sequences with Recurrent Neural Networks" https://arxiv.org/abs/1308.0850

Installation

conda env create -f environment.yml

Unconditional Handwriting Generation

Implemented 2 models

  1. LSTM with hidden dim of 900 and 1 layer
  2. LSTM with hidden dim of 400 and 3 layers (skip connections)

Training:

python train.py --name {name} --model {model_name} [--use_gpu] [--ckpt {/path/to/ckpt}]

model_name: unconditional_900 for 1 and unconditional_400 for 2

Conditional Handwriting Generation

Extended model 2 to generate handwriting conditioned on text input model_name: conditional_400