Hand-Written Digit Recognizer

Draw a digit (0-9) and let the vision model recognize it!

Make sure to use the space given by the square.

Model trained on MNIST database.

Powered by TensorFlow & Flask

Model Architecture

Convolutional Neural Network trained on MNIST dataset

Layer Visualizations

visualkeras

Visualkeras Model Architecture

Click image to enlarge

keras plot_model

Keras Plot Model Architecture

Click image to enlarge

Network Overview

This CNN model consists of:

  • Input Layer: 28×28 grayscale images
  • 3 Convolutional Blocks: Feature extraction with ReLU activation
  • 2 MaxPooling Layers: Spatial dimension reduction
  • Flatten Layer: Convert 2D features to 1D vector
  • 2 Dense Layers: Classification with dropout regularization
  • Output Layer: 10 neurons (digits 0-9) with softmax activation
Test Accuracy
~99%
Training Epochs
10
Dataset
MNIST