Fashion-MNIST (1) is an image dataset compiled by Zalando Research to replace the MNIST dataset. It contains 70,000 28x28 grayscale images of clothing items from 10 labelled classes:
- T-shirt/top - 0
- Trouser - 1
- Pullover - 2
- Dress - 3
- Coat - 4
- Sandal - 5
- Shirt - 6
- Sneaker - 7
- Bag - 8
- Ankle boot - 9
The dataset is divided into a 60,000-image train set and 10,000-image test set. More information can be found on the GitHub page.
The data is provided in 2 .csv files contained in data.tar.gz file. The provided reader can be used to import the data as Numpy arrays in Python.
from reader import get_images
(x_train, y_train), (x_test, y_test) = get_images() This is not an exhaustive list of tasks, the points are provided in order to guide you:
Try to visualize the differences between classes, e.g. using a dimensionality reduction technique.
Test the performance of different model architectures. Tune your model to improve its performance.
Report your results using appropriate metrics. See if your model performs equally among classes. Suggest possible imporvements.
- Xiao H, Rasul K, Vollgraf R. Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms. CoRR [Internet]. 2017;abs/1708.07747. Available from: http://arxiv.org/abs/1708.07747