This project was being conducted as part of the Advanced Python (FS2023) course at the University of Bern. It aims at building a multi-class classifier for clothing images from the Fashion-MNIST dataset using a convolutional neural network built in Pytorch.
Fashion-MNIST is a dataset of Zalando article photos, with 60,000 examples in the training set and 10,000 examples in the test set. Each sample is a 28x28 grayscale image with a label from one of ten classes.
The classes are the following:
- 0: T-shirt/top
- 1: Trouser
- 2: Pullover
- 3: Dress
- 4: Coat
- 5: Sandal
- 6: Shirt
- 7: Sneaker
- 8: Bag
- 9: Ankle boot
Although all necessary packages can be installed manually, we recommend using Anaconda.
The environment file lists all packages and their versions. If you already have anaconda installed, simply create a new environment using:
conda env create -f environment_lx.yml
conda env create -f environment_macOS.yml
Check that the environment has been correctly installed with
conda env list
It should be listed under the name pytorch.
To run our base version of the model, first specify your desired batch size, # of cores and # epochs in the main.py file.
Please note that if you change the base batch size, you also need to change it in scripts/training.py. We intend on fixing this issue later.
Once the parameters are set, simply run main.py.
The model is saved in models, plots of input images, loss and accuracy are saved in data.
If you wish to modify the loss function, optimizer parameters or model architecture, head to scripts/training.py
Some key resources:
Detailed breakdown of the CNN architecture
Authors: Léa Brönnimann & Mansour Faye