/PyTorchImageClassifier

A pytorch image classifier for the recognising letters from the notMNIST dataset

Primary LanguagePython

PyTorch image classifier for the notMNIST dataset

The notMNIST dataset contains 28x28px images of letters A to J in different fonts. This repository aims to serve as a sample for image classification in pytorch. It includes

  1. Custom Dataset and dataloaders
  2. Training and testing the model
  3. Creating a multiclass image classification model

Dataset

A B C D E F G H I J

The dataset contains over 18500 images for training and 469 images for testing. All images are grayscale 28x28px images.

Model used and accuracy

A convolutional neural network achieves an accuracy of 92.37% on this dataset.