Fashion-MNIST: 70,000 Grayscale Images of Zalando Fashion Articles

Fashion-MNIST consists of 70,000 grayscale images of Zalando fashion articles across 10 categories, released by Zalando Research in 2017. The collection includes a 60,000-sample training set and a 10,000-sample test set, each provided as 28x28 pixel arrays.

Use Cases

• Benchmarking multi-class classification models using the 10 article labels
• Training Generative Adversarial Networks (GANs) to synthesize 28x28 fashion images
• Testing clustering algorithms on the 784-dimensional pixel space

Strengths

• 70,000 total records
• 10 expert-annotated fashion categories
• Standardized 28x28 grayscale format for benchmarking

Limitations

• Low 28x28 resolution lacks fine garment details
• Grayscale format excludes color information
• Limited to 10 broad categories

Provenance

Source

Zalando Research (Arxiv 1708.07747)

Collection Method

Derived from Zalando's product image database