Anime Face Colorization Dataset

Overview

This dataset is designed for anime face colorization tasks using diffusion models and photo-sketch correspondence techniques. It provides a structured collection of reference images and corresponding sketches, enabling researchers and developers to explore various AI-assisted colorization techniques.

Curated by: Axel Delaval, Adama Koïta
License: CC BY 4.0

Read the paper: Sketch Colorization Using Diffusion Models

Dataset Structure

The dataset is organized into two main folders:

reference/: Contains colored reference images for different anime characters.
sketch/: Contains corresponding line-art sketches derived from reference images.

Each subdirectory within these folders corresponds to a specific character, ensuring structured and character-specific training data.

Datasets Used

This dataset combines and enhances multiple existing datasets, including:

Danbooru Tagged Dataset ("moeimouto" subset only) [Dataset]
AnimeDiffusion Dataset augmented with deformation flows

This structured dataset enables effective training for anime sketch colorization models.

Citation

BibTeX:

@misc{delaval2025animefacecolorization,
  author       = {Axel Delaval and Adama Koïta},
  title        = {Anime Face Colorization Dataset},
  year         = {2025},
  url          = {https://huggingface.co/datasets/ComputerVisionAnimeProject/AnimeFaceColorization/},
  note         = {Dataset for anime sketch colorization using diffusion models.}
}