Fashion-MNIST is a new image dataset consisting of 70,000 grayscale images of 10 fashion categories, each with 7,000 images. It is designed as a direct replacement for the original MNIST dataset, offering the same image size, data format, and split structure. The dataset is freely available at https://github.com/zalandoresearch/fashion-mnist. The MNIST dataset, introduced in 1998, has become a standard benchmark in deep learning due to its simplicity and accessibility. However, it is now considered too simple for modern deep learning models, which can achieve over 99.7% accuracy on MNIST. Fashion-MNIST provides a more challenging task, as it involves classifying images of clothing items rather than handwritten digits. Fashion-MNIST is derived from Zalando's website, where each product is photographed from multiple angles. The images are processed through a series of steps to standardize them to 28x28 grayscale. These steps include converting to PNG, trimming edges, resizing, sharpening, extending, negating intensities, and converting to grayscale. The dataset includes 60,000 training images and 10,000 test images. Each image is labeled with a silhouette code, which is manually assigned by Zalando's fashion experts. The dataset is structured similarly to MNIST, with images and labels stored in the same format. Experiments show that Fashion-MNIST provides a more challenging benchmark for machine learning algorithms compared to MNIST. The dataset is intended to be a drop-in replacement for MNIST, allowing researchers to easily adapt their models to this new dataset. The paper also compares results on both datasets, showing that Fashion-MNIST offers a more realistic and challenging testbed for machine learning algorithms.Fashion-MNIST is a new image dataset consisting of 70,000 grayscale images of 10 fashion categories, each with 7,000 images. It is designed as a direct replacement for the original MNIST dataset, offering the same image size, data format, and split structure. The dataset is freely available at https://github.com/zalandoresearch/fashion-mnist. The MNIST dataset, introduced in 1998, has become a standard benchmark in deep learning due to its simplicity and accessibility. However, it is now considered too simple for modern deep learning models, which can achieve over 99.7% accuracy on MNIST. Fashion-MNIST provides a more challenging task, as it involves classifying images of clothing items rather than handwritten digits. Fashion-MNIST is derived from Zalando's website, where each product is photographed from multiple angles. The images are processed through a series of steps to standardize them to 28x28 grayscale. These steps include converting to PNG, trimming edges, resizing, sharpening, extending, negating intensities, and converting to grayscale. The dataset includes 60,000 training images and 10,000 test images. Each image is labeled with a silhouette code, which is manually assigned by Zalando's fashion experts. The dataset is structured similarly to MNIST, with images and labels stored in the same format. Experiments show that Fashion-MNIST provides a more challenging benchmark for machine learning algorithms compared to MNIST. The dataset is intended to be a drop-in replacement for MNIST, allowing researchers to easily adapt their models to this new dataset. The paper also compares results on both datasets, showing that Fashion-MNIST offers a more realistic and challenging testbed for machine learning algorithms.