fashion-mnist

Why we included this project

Fashion-MNIST is a common benchmark for image classifiers that want a step up from handwritten digits. It packages 70,000 28x28 grayscale images of Zalando clothing into the same train/test split as the original MNIST, so any pipeline that already loads MNIST can switch over with minimal code changes. The ten clothing categories are far more varied than digits, which means your accuracy numbers won't hit a misleading 99.7% ceiling and real model weaknesses tend to surface. The repository includes loader scripts and published benchmark results, and the dataset is available through PyTorch, Keras, TensorFlow, and other libraries, so getting started takes only a few lines. It is a reference set for comparing architectures, not a deployable service.