anime_dbrating

# Anime Danbooru Rating Dataset

## Summary

This dataset provides comprehensive **danbooru rating** classifications for anime-style images, organized into four distinct categories based on content safety levels. The dataset contains over 1.2 million images distributed across explicit, general, questionable, and sensitive rating classes, making it ideal for training **content moderation** systems and **image classification** models specifically tailored for anime and manga-style artwork.

The rating system follows the established **danbooru tagging** conventions, where images are categorized based on their content appropriateness. This hierarchical classification enables fine-grained control over content filtering, with clear distinctions between safe-for-work and not-safe-for-work material. The dataset's large scale and well-balanced distribution across rating categories make it particularly valuable for developing robust **automated moderation** tools in anime-focused platforms and applications.

Each image in the dataset is associated with its corresponding danbooru rating tag, providing reliable ground truth labels for supervised learning tasks. The dataset structure facilitates both multi-class classification (predicting exact rating) and binary classification (SFW vs NSFW) tasks, offering flexibility for different application requirements. The comprehensive coverage of anime art styles and diverse content types ensures models trained on this dataset can generalize effectively across various anime and manga artwork.

## Dataset Structure

The dataset is organized into four main categories based on danbooru rating tags:

- **explicit_278119.zip**: Contains 278,119 images with explicit content
- **general_341506.zip**: Contains 341,506 images with general/safe content
- **questionable_320107.zip**: Contains 320,107 images with questionable content
- **sensitive_341769.zip**: Contains 341,769 images with sensitive content

### Rating Categories

The danbooru rating system follows this classification:

- **safe + sensitive = SFW (Safe For Work)**
- **questionable + explicit = NSFW (Not Safe For Work)**

## Applications

This dataset is particularly useful for:

- Training content moderation systems for anime art platforms
- Developing SFW/NSW classification models
- Research in anime image understanding and classification
- Building automated tagging systems for anime artwork
- Content filtering applications for anime communities

## Original Content

dbrating == danbooru rating, use danbooru rating tags and danbooru subsets as training dataset
safe - sensitive - questionable - explicit, safe + sensitive == sfw, questionable + explicit == nsfw

## Citation

```bibtex
@misc{anime_dbrating,
title = {Anime Danbooru Rating Dataset},
author = {deepghs},
howpublished = {\url{https://huggingface.co/datasets/deepghs/anime_dbrating}},
year = {2023},
note = {Large-scale dataset for anime image rating classification using danbooru tagging system},
abstract = {This dataset provides comprehensive danbooru rating classifications for anime-style images, organized into four distinct categories based on content safety levels. The dataset contains over 1.2 million images distributed across explicit, general, questionable, and sensitive rating classes, making it ideal for training content moderation systems and image classification models specifically tailored for anime and manga-style artwork. The rating system follows the established danbooru tagging conventions, where images are categorized based on their content appropriateness.},
keywords = {image-classification, computer-vision, anime, danbooru, rating-classification}
}
```

# 动漫Danbooru评分数据集（Anime Danbooru Rating Dataset）

## 摘要
该数据集为动漫风格图像提供了全面的Danbooru评分（danbooru rating）分类，基于内容安全级别划分为四个独立类别。数据集包含超过120万张图像，分布在露骨（explicit）、通用（general）、可疑（questionable）、敏感（sensitive）四个评分类别中，非常适合训练专为动漫和漫画风格美术作品定制的内容审核（content moderation）系统与图像分类（image classification）模型。

该评分系统遵循已确立的Danbooru标签（danbooru tagging）规范，依据内容适宜性对图像进行分类。这种层级化分类可实现内容过滤的细粒度管控，清晰区分安全工作场所（safe-for-work, SFW）与非安全工作场所（not-safe-for-work, NSFW）的内容。该数据集规模庞大且各评分类别分布均衡，尤其适用于为专注于动漫的平台与应用开发稳健的自动化审核（automated moderation）工具。

数据集中的每张图像都配有对应的Danbooru评分标签，可为监督学习（supervised learning）任务提供可靠的真值标签（ground truth）。该数据集的结构同时支持多分类（预测精确评分）与二分类（SFW与NSFW区分）任务，可为不同应用需求提供灵活性。其对动漫艺术风格与多样内容类型的全面覆盖，确保基于该数据集训练的模型能够在各类动漫与漫画美术作品上实现有效泛化。

## 数据集结构
该数据集基于Danbooru评分标签划分为四个主要类别：
- **explicit_278119.zip**：包含278,119张露骨内容图像
- **general_341506.zip**：包含341,506张通用/安全内容图像
- **questionable_320107.zip**：包含320,107张可疑内容图像
- **sensitive_341769.zip**：包含341,769张敏感内容图像

### 评分类别
Danbooru评分系统采用如下分类规则：
- **安全（safe）+ 敏感（sensitive）= 安全工作场所（Safe For Work, SFW）**
- **可疑（questionable）+ 露骨（explicit）= 非安全工作场所（Not Safe For Work, NSFW）**

## 应用场景
该数据集尤其适用于：
- 为动漫美术平台训练内容审核系统
- 开发SFW/NSFW分类模型
- 开展动漫图像理解与分类相关研究
- 为动漫美术作品构建自动化标签系统
- 为动漫社区开发内容过滤应用

## 原始内容说明
dbrating即Danbooru评分（danbooru rating），可使用Danbooru评分标签与Danbooru子集作为训练数据集。
安全（safe）- 敏感（sensitive）- 可疑（questionable）- 露骨（explicit）；安全+敏感=SFW，可疑+露骨=NSFW。

## 引用
bibtex
@misc{anime_dbrating,
title = {Anime Danbooru Rating Dataset},
author = {deepghs},
howpublished = {url{https://huggingface.co/datasets/deepghs/anime_dbrating}},
year = {2023},
note = {采用Danbooru标签系统的动漫图像评分分类大规模数据集},
abstract = {该数据集为动漫风格图像提供了全面的Danbooru评分分类，基于内容安全级别划分为四个独立类别。数据集包含超过120万张图像，分布在露骨、通用、可疑、敏感四个评分类别中，非常适合训练专为动漫和漫画风格美术作品定制的内容审核系统与图像分类模型。该评分系统遵循已确立的Danbooru标签规范，依据内容适宜性对图像进行分类。},
keywords = {image-classification, computer-vision, anime, danbooru, rating-classification}
}