The paper introduces Fashion-MNIST, a new dataset designed to serve as a direct replacement for the MNIST dataset for benchmarking machine learning algorithms. Fashion-MNIST consists of 28 × 28 grayscale images of 70,000 fashion products from 10 categories, with 7,000 images per category. The dataset is freely available and shares the same image size, data format, and training-testing split structure as MNIST. It aims to provide a more challenging classification task compared to MNIST, which has been trained to high accuracy (over 99.7%). The dataset is based on images from Zalando's website, converted through a series of steps including trimming, resizing, sharpening, extending, and negating. The class labels are derived from silhouette codes manually labeled by fashion experts. The dataset is divided into a training set of 60,000 images and a test set of 10,000 images. The paper includes benchmark results for various algorithms on both Fashion-MNIST and MNIST, demonstrating the dataset's utility for evaluating machine learning models.The paper introduces Fashion-MNIST, a new dataset designed to serve as a direct replacement for the MNIST dataset for benchmarking machine learning algorithms. Fashion-MNIST consists of 28 × 28 grayscale images of 70,000 fashion products from 10 categories, with 7,000 images per category. The dataset is freely available and shares the same image size, data format, and training-testing split structure as MNIST. It aims to provide a more challenging classification task compared to MNIST, which has been trained to high accuracy (over 99.7%). The dataset is based on images from Zalando's website, converted through a series of steps including trimming, resizing, sharpening, extending, and negating. The class labels are derived from silhouette codes manually labeled by fashion experts. The dataset is divided into a training set of 60,000 images and a test set of 10,000 images. The paper includes benchmark results for various algorithms on both Fashion-MNIST and MNIST, demonstrating the dataset's utility for evaluating machine learning models.