Synthetic_Data_Engine_For_Text_Recognition

This repository contains code for Max Jaderberg's Text Renderer (https://arxiv.org/pdf/1406.2227.pdf) for generating synthetic cropped text images which can be used to train text recognition models like CRNN (https://arxiv.org/pdf/1507.05717.pdf) (https://github.com/bgshih/crnn)

Dependencies(Successfully tested on Ubuntu 14.04 LTS)

• Anaconda Python 2.7
  https://www.continuum.io/downloads

• Opencv 3.1
  conda install -c menpo opencv3=3.1.0

• Pygame
  conda install -c cogsci pygame

• LMDB
  conda install -c conda-forge python-lmdb

To Run the Code :

• Change the paths in these files
  • Synthetic_Data_Engine_For_Text_Recognition/SVT/font_path_list.txt
  • Synthetic_Data_Engine_For_Text_Recognition/SVT/icdar_2003_train.txt
  • Synthetic_Data_Engine_For_Text_Recognition/SVT/svt.txt
  • Synthetic_Data_Engine_For_Text_Recognition/text_renderer/generate_word_training_data.py
• cd Synthetic_Data_Engine_For_Text_Recognition/text_renderer/
• python generate_word_training_data.py (This code would generate the images and the lmdb files for training text recognition models like CRNN)

Examples