Handwriting-Imitation-and-Word-Detection

Introduction

I have basically merged two github repositories(Spatio Temporal Handwriting Imitation and EAST text detector) and turned them into an end-to-end handwriting cloning and word detection model. Instructions given below.

Step-1(Installation):

Unzip. Check the requirements.txt file in main folder and EAST-master folder, and install if any thing is missing. Any version of Tensorflow > 1.0 should be good to run the EAST text detector.

Step-2(Downloads):

• Handwriting Imitation: Before running the pipeline the trained model checkpoints have to be copied into the folder checkpoints from https://drive.google.com/open?id=11fc8b7QTSqL8oIjs7ddGutlRKEL8NBqh.

• EAST text detector: Models trained on ICDAR 2013 (training set) + ICDAR 2015 (training set): BaiduYun link GoogleDrive Resnet V1 50 provided by tensorflow slim: slim resnet v1 50

Put the east_icdar2015_resnet_v1_50_rbox file in the EAST-master folder of the main folder.

Step-3(Run):

• Provide your handwritten image of 3-4 words(for best results) in the main folder with the name "input.png". A sample image is already provided. Open the demo.sh file and write the input text and output text.
• Run the final.ipynb file.

Image with bounding boxes drawn on it and .txt files corresponding to each image(with coordinates) will be written to the output folder in the EAST-master folder.