Danbooru Artists 10k

Dataset Summary

This dataset contains images from 10,000 artists collected from Danbooru. The images are primarily sourced from a mirror here. However, there are six images that are missing/broken in this mirror, which have been directly sourced from Danbooru. Each artist is represented by 30 images. This dataset is intended for use in metric learning or fine-grained classification tasks related to art styles. The artists are split into train, validation, and test sets at a ratio of 80%, 10%, and 10%, respectively. Similarly, each artist's 30 images are also split into train, validation, and test sets at the same ratio.

Size

300,000 images (10,000 artists * 30 images/artist)

Format

The images are stored as WebP files packaged in a10k.zip. The dataset splits are provided in split.txt. Each line of split.txt follows this format: artist_split image_split artist_name file_path.