Neural Network Project Highlights

Explore how neural networks tackle complex tasks, from classifying images to transforming photographs.

CIFAR10 Image Classification

Trained a neural network on CIFAR10, classifying 32x32 color images across 10 categories such as airplanes and birds. I achieved a 75.2% accuracy using different layer combinations.

CIFAR10 Training Progress CIFAR10 Classification Results

Each image was evaluated for classification accuracy and probability distribution, showing the model's confidence in its predictions.

CIFAR10 Classification Probabilities

MNIST Handwritten Digit Recognition

Created a custom neural network to recognize handwritten digits with the MNIST dataset, achieving an accuracy of 99.1% through a well-thought combination of layers.

MNIST Training Progress MNIST Recognition Results

Here are some examples of the images the model struggled with detecting.

MNIST Misclassified Examples

Image Colorizer

A neural network designed to colourise black-and-white images, transforming greyscale photos into colourised images.

Original Black-and-White Image

The image to the left is the result generated by the neural network and the image to the right is the original image.

Colosized Image

Pix2Pix: Photo-to-Street-Map

Leveraged the Pix2Pix neural network to translate aerial photographs into street maps.

Pix2Pix Result