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

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.