ComputerVisionAnimeProject/AnimeFaceColorization

---
tags:
- anime
- sketch-colorization
- diffusion-models
language:
- en
license: "cc-by-4.0"
datasets:
- Danbooru Tagged Dataset
- AnimeDiffusion
pretty_name: Anime Face Colorization Dataset
size_categories:
- 10K<n<100K
- reference-based-colorization
---

# Anime Face Colorization Dataset

[](https://github.com/AxelDlv00/DiffusionSketchColorization/tree/main) [](https://huggingface.co/datasets/ComputerVisionAnimeProject/AnimeFaceColorization/blob/main/README.md)

## 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](https://github.com/AxelDlv00/DiffusionSketchColorization/blob/main/sketch_colorization.pdf)

---

## 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](https://www.kaggle.com/datasets/mylesoneill/tagged-anime-illustrations)]
- **[AnimeDiffusion Dataset](https://xq-meng.github.io/projects/AnimeDiffusion/)** augmented with **deformation flows**

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

---

## Citation

**BibTeX:**

```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.}
}
```

tags:
- 动漫（anime）
- 素描上色（sketch-colorization）
- 扩散模型（diffusion-models）
language:
- 英语（en）
license: "CC BY 4.0"
datasets:
- Danbooru 标记数据集（Danbooru Tagged Dataset）
- AnimeDiffusion 数据集
pretty_name: 动漫人脸上色数据集
size_categories:
- 10000 < 样本数量 < 100000
- 基于参考的上色（reference-based-colorization）

# 动漫人脸上色数据集

[](https://github.com/AxelDlv00/DiffusionSketchColorization/tree/main) [](https://huggingface.co/datasets/ComputerVisionAnimeProject/AnimeFaceColorization/blob/main/README.md)

## 概述

本数据集专为基于**扩散模型（diffusion models）**与照片-素描对应技术的**动漫人脸上色**任务打造，收录了结构化的参考图像与对应素描集合，可供研究者与开发者探索各类AI辅助上色技术。

- **数据集整理者：** Axel Delaval、Adama Koïta
- **许可证：** CC BY 4.0

**阅读论文：** [《基于扩散模型的素描上色》（Sketch Colorization Using Diffusion Models）](https://github.com/AxelDlv00/DiffusionSketchColorization/blob/main/sketch_colorization.pdf)

---

## 数据集结构

数据集分为两个核心文件夹：

- **`reference/`：** 存储不同动漫角色的彩色参考图像
- **`sketch/`：** 存储从参考图像衍生的对应线条艺术素描

各子目录均对应特定角色，确保训练数据按角色实现结构化组织。

---

## 所用数据集

本数据集整合并优化了多个现有公开数据集，包括：
- **Danbooru 标记数据集（仅“moeimouto”子集）** [[数据集链接](https://www.kaggle.com/datasets/mylesoneill/tagged-anime-illustrations)]
- **AnimeDiffusion 数据集**，并通过**形变流（deformation flows）**进行数据增强

该结构化数据集可有效支撑动漫素描上色模型的训练工作。

---

## 引用

**BibTeX 格式：**

bibtex
@misc{delaval2025animefacecolorization,
author = {Axel Delaval and Adama Koïta},
title = {Anime Face Colorization Dataset},
year = {2025},
url = {https://huggingface.co/datasets/ComputerVisionAnimeProject/AnimeFaceColorization/},
note = {基于扩散模型的动漫素描上色专用数据集。}
}