zalandoresearch/fashion-mnist

Project overview

Fashion-MNIST is a dataset that replaces the original MNIST benchmark with a more challenging image classification task while preserving the same image size and train/test structure. It loads directly in mainstream machine learning libraries, making it applicable for researchers and educators evaluating algorithms without manual data downloads.

Project type

Model Development · Data Processing · Evaluation & Observability

Deployment

Refer to project documentation

License

MIT

Best for

• Researchers and educators evaluating machine learning classifiers who need a more challenging alternative to MNIST that preserves the same image size and train/test split structure.

Key capabilities

• A dataset replacing MNIST with the same image size and structure for training and testing splits.
• An automatic benchmarking system based on scikit-learn that covers 129 classifiers (but no deep learning).
• Utility to load the data easily via Python using NumPy.

Limitations and risks

• The official automatic benchmarking system covers classifiers but does not include deep learning.

Getting started

• Data can be loaded directly in popular ML libraries without manual download. Open library documentation (such as PyTorch or Keras) for the dataset and call the dataset loader function to begin training.

Evidence and sources

• README: `Fashion-MNIST` is a dataset of [Zalando](https://jobs.zalando.com/tech/)'s article images—consisting of a training set of 60,000 examples and a test set of 10,000 examples. Each…
• GitHub project description: A MNIST-like fashion product database. Benchmark :point_down:
• README: **MNIST is too easy.** Convolutional nets can achieve 99.7% on MNIST. Classic machine learning algorithms can also achieve 97% easily.
• README: **MNIST can not represent modern CV tasks**, as noted in [this April 2017 Twitter thread](https://twitter.com/fchollet/status/852594987527045120), deep learning expert/Keras autho…
• README: We intend `Fashion-MNIST` to serve as a direct **drop-in replacement** for the original [MNIST dataset](http://yann.lecun.com/exdb/mnist/) for benchmarking machine learning algori…

Reviewed at: 2026-07-18T00:36:55Z