Handwritten-Character-Recognition

This software was made to show the beauty of neural networks, visualizing the fully connected neural network diagram, and how it's being trained. Built from the ground up without any machine learning libraries.

• What we can do

  1. Create a new network
  2. Create datasets
  3. Train the model
  4. Save the network and reopen it
  5. View and delete our own datasets

• Customization

  • Change the number of hidden layers
  • Change the drawing pad resolution

• The main visual interface

  

• Dataset viewer

  

• Training the network

  

Demo video.

Installation

1. Download the latest version
2. Save it under your favorite folder (e.g. /foo/Handwritten-Character-Recognition)
3. Navigate to /foo/Handwritten-Character-Recognition
4. Install dependencies by running pip install -r requirements.txt
5. Run python __main__.py
6. There you go!

FAQ

• To exit the application: press the "esc" key

• To reset the network:

  Search for NN_NEW in __main__.py and set it to True. After you trained the network, press the button "save" to save it.

• To reset the dataset:

  Example: to reset the dataset of zero "0", replace the contents of the file dataset/0.json with an empty list [].

Limitations

• Always in fullscreen mode
• The UI layout is customized for the 1366x768 display size

Troubleshoot

• To report bugs/issues or ask questions, you can reach me here or open an issue/pull request.

Changelog

• 1.1.0 (June 11, 2023):
  • removed carbon_plug, using mykit instead
• v1.0.1 (May 10, 2023):
  • BUG FIXED: Renamed carbon to carbon_plug to prevent conflicts with the original carbon module (if installed).

License

This project is licensed under the MIT license.