Fashion-MNIST: Grayscale Zalando Product Images in 10 Categories

Fashion-MNIST contains grayscale images of Zalando products across 10 label categories including apparel and footwear. Created by Zalando Research, it serves as a benchmark for machine learning algorithms and is designed as a drop-in replacement for the original MNIST.

Use Cases

• Benchmarking classification accuracy using the 10 fashion category labels
• Training generative adversarial networks to produce synthetic 28x28 fashion images
• Testing unsupervised clustering on the 784-pixel grayscale feature vectors

Strengths

• Direct compatibility with MNIST data loaders and scripts
• 10 distinct fashion categories providing higher complexity than handwritten digits
• MIT license allows for unrestricted research and commercial application

Limitations

• Low 28x28 resolution limits applicability to modern high-definition vision tasks
• Grayscale format excludes color-based feature analysis
• Restricted to Zalando's specific product catalog and photography style

Provenance

Source

Zalando Research via GitHub

Collection Method

Derived from Zalando's product database

Freshness

Last updated 2022-06-13; the dataset is a static benchmark.