Digits recognition

Python application to recognize handwritten digits in real time!

Project's motivations

I've always been interested in Artificial Intelligence and came across 3B1B's video about neural networks

That's why I started with one of the most basic NN project, digits recognition.

Installation

• Run python -m pip install -r .\requirements.txt
• Run python app.py
  • If it gives an error, run python model.py first

Controls

• Left click to draw
• Right click to erase
• Middle click to reset

Preview

Pygame window

Recognition preview

Model Description

Layers

1. Conv2D: A 2D convolutional layer with 16 filters, a kernel size of (3, 3), and a ReLU activation function. This layer expects an input shape of (28, 28, 1).
2. Dropout: A dropout layer with a rate of 0.5, which helps prevent overfitting by randomly setting input units to 0 during training.
3. MaxPooling2D: A 2D max pooling layer with a pool size of (2, 2), which reduces the spatial dimensions of the input.
4. Conv2D: Another 2D convolutional layer with 16 filters, a kernel size of (3, 3), and a ReLU activation function.
5. MaxPooling2D: Another 2D max pooling layer with a pool size of (2, 2).
6. Flatten: A flatten layer that reshapes the input into a 1D array, which is necessary before passing it to the dense layers.
7. Dropout: Another dropout layer with a rate of 0.5.
8. Dense: A fully connected layer with 132 units and a ReLU activation function.
9. Dense: The output layer, which is a fully connected layer with 10 units and a softmax activation function. This layer outputs the probability distribution over the 10 digits.

Hyper parameters

• Optimizer: Adam
• Loss function: Categorical sparse cross-entropy
• Dropout rate: 0.5

Regularization techniques

• Dropout
• Data Augmentation
  • Width shift range=0.3
  • Height_shift_range=0.3