This project was meant to be an introduction to NNs as part of the AI course that I took at Oxford University. Prior to this project, I made a KNN classifier from scratch to achieve the same goal.
Due to the simplicity and instructional nature of the task, little feature extraction is used and thus performance is limited. Many runs only achieve a ~50% success rate in classification, which, though better than the 10% random success rate, is no impressive feat.
Improvement can be found by extracting more features, using a CNN, or using edge detection.
Because Github does not allow uploads of files greater than 25MB in size, I am not able to include the dataset. It can be found here: CIFAR-10