Kuzushiji-49 Japanese Cursive Script Dataset

The Kuzushiji-49 dataset is an extension of the Kuzushiji-MNIST dataset, containing 270,912 images (28×28 grayscale) in 49 categories. The data has 48 Hiragana characters and 1 Hiragana iterative annotation class. The training set has 232,365 images and the test set has 38,547 images. This dataset can be used to study Japanese classical literature.