OCR Handwriting Number

Genarator Data and Traning

• gen_data_train.ipynb

Number sequence generator

Prerequisites

Install necessary libraries by executing pip install -r requirements.txt

How to use standalone

To generate a new sequence of numbers, simply run python main.py with your preferred configuration values. The generated image sequence will be automatically be saved under the working directory.

You can see the options available like so python main.py -h.

Use example: python main.py 6687 0 50 100

How to use as a function

To be able to use it on another Python script simply add like this:

from num_seq_generator import generate_numbers_sequence

... #some code

img_seq = generate_numbers_sequence(sequence, (min_spacing, max_spacing),
                                    image_width, train_imgs, train_labels)

To understand the parameters available and return value please look at the docstring.

How to execute tests

Implementation details

Normally for larger projects all the code would be arranged in a certain structure, but given the size of this codebase I decided to leave the files on the root directory. I opted to separate the code in 4 main scripts:

• get_mnist.py - Download and return MNIST dataset. Can be used to add more datasets using the current available functionality as a base, maybe even create a parent class for serving data.
• num_seq_generator.py - Contains functionality to generate a sequence of numbers. Can be used as a base to add more functions that perform other types of data generation/augmentation. Particularly I added data and labels to the function signature to make it more efficient, this way, the dataset can be loaded from any other module, giving more power to the programmer, avoiding having to hardcode a specific dataset in the function and preventing unnecessary loading of the same dataset for every sequence.
• main.py - Entry point for quick testing using a command line tool.