Fashion Classifier

A DeepLearning model developed using PyTorch for classifying images from the Fashion MNIST data set. This is my first time using PyTorch and intended to teach me some basic concepts of it.

Performance 📊

I tuned the hyper-parameters for 20 trials with 5 epochs each and then trained the model for 25 epochs. This resulted in an accuracy of 93.19% for unseen test data.
The following graph shows a sample of the test data along with the models predictions. As you can see, the model is pretty confident for all predictions that it got correct, but also relatively confident for the one prediction that it got wrong, which I assume is due to the similarity of Shirts and Pullovers.

Installation 💻

Option 1 with Docker 🐳

Since I have not deployed the Docker image for this project to Docker Hub, you need to create the image yourself.

1. Clone the repo:
   git clone https://github.com/qhilipp/FashionClassifier
2. cd into the repo:
   cd FashionClassifier
3. Create the docker image:
   docker build -t fashion-classifier:latest .
4. Run the image:
   docker run --name fashion-classifier-container -v $(pwd):/app fashion-classifier:latest python3.13 main.py --load model

Option 2 without Docker 🐳❌

Make sure you have Python 3.13 installed and the 'python' command pointing to version 3.13.

1. Clone the repo:
   git clone https://github.com/qhilipp/FashionClassifier
2. cd into the repo:
   cd FashionClassifier
3. Create a virtual environment
   python -m venv env or with Conda conda create -n fashion-classifier python=3.13
4. Activate it
   source evn/bin/activate or with Conda conda activate
5. Install the packages
   pip install -r requirements.txt
6. Run
   python main.py --load model

Arguments 🚀

Argument	Shortform	Type	Description	Default Value
--load	-l	str	The name of the file from which the model should be loaded.	None
--save	-s	str	The name of the file to which the model should be saved.	None
--device	-d	str	The device on which PyTorch should perform the tensor calculations (cpu or cuda, if available).	'cpu'
--epochs	-e	int	The number of epochs used to train the model.	20
--trials	-t	int	The number of trials to find optimal hyperparameters.	5
--trial_epochs	-te	int	The number of epochs per trial when finding optimal hyperparameters.	2
--load_hyper_parameters	-lh	str	The name of a JSON file from which the model should load its hyperparameters.	None
--save_hyper_parameters	-sh	str	The name of a JSON file to which the model should save its hyperparameters.	None