Classifying Clothes with Convolutional Neural Networks

Classifying Clothes with Convolutional Neural Networks2018-01-27T23:56:37+00:00

Project Description

The Fashion MNIST Dataset contains 70,000 grayscale samples of different clothing articles. With an External GPU Unit, I wanted to create a neural network & a convolutional neural network in Python to see if I can classify the different clothing sets.

Using Neural Layers with a Sigmoid Activation Function, I got a 0.323 Loss & 0.884 Accuracy. When I introduced three convolutional layers, I got a 0.217 Loss & 0.922 Accuracy.  Complete details about the Neural Network I created are in the attached Jupyter Notebook.

Ultimately, my ability to proceed was hampered by how my GPU was several generations behind, but I made some good progress with the resources I had. With a better GPU (which I’m going to acquire in Q1/Q2 2018), I’d like to fully test out and experiment with new neural networks ‘inspired by’ VGG and ResNets. While I tested those two in this project, I was unable to test them for as much as I should have.

Explore the Jupyter Notebook