Skip to content

Try MNIST Handwiriting recognition in your computer

Notifications You must be signed in to change notification settings

rohitashwin/mnist

Repository files navigation

MNIST Digit Recognition

A website to demonstrate handwritten digit recognition

How to run?

Requirements

  • python >= 3.10
  • node >= 1.18

Loading a demonstration

  • If you're not using an apple silicon equipped mac, change the device to cuda, cpu, or whatever suits your needs.

npm start

  • This will show the localhost port the server is running on, which you can visit using your browser.

Loading the existing network

  • If you're not using an apple silicon equipped mac, change the device to cuda, cpu, or whatever suits your needs.

python3 main.py --predict --image <path/to/28x28/json_array> --model <path/to/model>

Training a new neural network

  • If you're not using an Apple Silicon equipped mac, change the device to cuda, cpu, or whatever suits your needs.

python3 main.py --train

Network

  • Convolution
  • Maxpool
  • Convolutional
  • Maxpool
  • Fully connected
  • Fully connected
  • Softmax layer

Advantages

  • Very good translational and rotational invariance

Disadvantages

  • The model is a bit overfit and produces wrong results about 5% of the time

Screenshots

Screenshot 2023-01-01 at 5 38 04 PM Screenshot 2023-01-01 at 5 38 16 PM Screenshot 2023-01-01 at 5 38 27 PM Screenshot 2023-01-01 at 5 38 39 PM

About

Try MNIST Handwiriting recognition in your computer

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published