anime_blurry_preparation

# Anime Blurry Preparation Dataset

## Summary

The **Anime Blurry Preparation Dataset** is a specialized **binary classification** dataset designed for distinguishing between clear and blurry 3D anime-style images. This dataset provides a comprehensive collection of image pairs where each clear image has a corresponding blurry counterpart, making it ideal for training and evaluating image quality assessment models. The dataset employs **automatic annotation** techniques using BLIP (Bootstrapping Language-Image Pre-training) to generate reliable labels for the classification task.

This dataset specifically focuses on **3D anime-style** content, addressing the unique challenges of image quality assessment in computer-generated anime imagery. The clear images represent high-quality, sharp renderings while the blurry counterparts simulate various types of image degradation that commonly occur during rendering, compression, or transmission processes. The **binary classification** framework makes this dataset particularly useful for developing models that can automatically detect and filter out low-quality images in anime content pipelines.

The dataset organization follows a structured approach with separate directories for clear and blurry images, each containing thousands of carefully curated samples. With a total size of approximately 5GB, this dataset provides substantial training data for developing robust image quality assessment models. The **automatic annotation** process ensures consistent labeling while maintaining the dataset's scalability and reproducibility for research and development purposes.

## Dataset Structure

The dataset is organized in a tar archive with the following structure:

- `clear/` - Directory containing clear, high-quality 3D anime images
- `blurry/` - Directory containing blurry versions of the same images

Each image file follows a consistent naming convention with hash-based identifiers to ensure uniqueness and prevent duplicates.

## Dataset Statistics

- **Total Size**: 5,088,880,640 bytes (approximately 5GB)
- **File Format**: JPEG images
- **Organization**: Binary classification (clear vs blurry)
- **Content Type**: 3D anime-style images
- **Annotation Method**: Automatic annotation using BLIP

## Original Content

binary classification dataset (clear-blurry) for 3d images,
auto-annotated by blip, just for testing, not recommended for production model training

## Citation

```bibtex
@misc{anime_blurry_preparation,
title = {Anime Blurry Preparation Dataset},
author = {deepghs},
howpublished = {\url{https://huggingface.co/datasets/deepghs/anime_blurry_preparation}},
year = {2023},
note = {Binary classification dataset for clear vs blurry 3D anime images with automatic BLIP annotation},
abstract = {The Anime Blurry Preparation Dataset is a specialized binary classification dataset designed for distinguishing between clear and blurry 3D anime-style images. This dataset provides a comprehensive collection of image pairs where each clear image has a corresponding blurry counterpart, making it ideal for training and evaluating image quality assessment models. The dataset employs automatic annotation techniques using BLIP to generate reliable labels for the classification task. This dataset specifically focuses on 3D anime-style content, addressing the unique challenges of image quality assessment in computer-generated anime imagery. The clear images represent high-quality, sharp renderings while the blurry counterparts simulate various types of image degradation that commonly occur during rendering, compression, or transmission processes.},
keywords = {binary classification, image-classification, computer-vision, anime, 3d}
}
```

# 动漫模糊预处理数据集（Anime Blurry Preparation Dataset）

## 摘要
**动漫模糊预处理数据集**是一款专为区分清晰与模糊3D动漫风格图像设计的专用**二分类（binary classification）**数据集。本数据集提供了全面的图像对集合，每张清晰图像均配有对应的模糊版本，非常适合用于训练与评估图像质量评估模型。本数据集采用基于**BLIP（Bootstrapping Language-Image Pre-training，自举语言图像预训练）**的自动标注（automatic annotation）技术，为该分类任务生成可靠的标签。

本数据集专门聚焦于**3D动漫风格（3D anime-style）**内容，解决了计算机生成动漫图像领域中图像质量评估的独特挑战。其中清晰图像代表高质量、高锐度的渲染结果，而模糊图像则模拟了渲染、压缩或传输过程中常见的各类图像退化现象。该二分类框架非常适用于开发可在动漫内容处理流程中自动检测并过滤低质量图像的模型。

本数据集采用结构化组织方式，分别为清晰图像与模糊图像设置独立目录，每个目录均包含数千份精心筛选的样本。本数据集总容量约5GB，可为开发鲁棒的图像质量评估模型提供充足的训练数据。其自动标注流程确保了标注的一致性，同时为研究与开发场景提供了良好的可扩展性与可复现性。

## 数据集结构
本数据集以tar归档格式组织，结构如下：
- `clear/`：存储清晰、高质量3D动漫图像的目录
- `blurry/`：存储对应图像模糊版本的目录

所有图像文件均遵循统一的命名规范，采用基于哈希的标识符以确保唯一性并避免重复。

## 数据集统计信息
- **总容量**：5088880640字节（约5GB）
- **文件格式**：JPEG图像
- **组织形式**：二分类（清晰vs模糊）
- **内容类型**：3D动漫风格图像
- **标注方法**：采用BLIP进行自动标注

## 原始说明
本数据集为面向3D图像的（清晰-模糊）二分类数据集，由BLIP自动标注，仅用于测试，不建议用于生产模型训练。

## 引用
bibtex
@misc{anime_blurry_preparation,
title = {Anime Blurry Preparation Dataset},
author = {deepghs},
howpublished = {url{https://huggingface.co/datasets/deepghs/anime_blurry_preparation}},
year = {2023},
note = {Binary classification dataset for clear vs blurry 3D anime images with automatic BLIP annotation},
abstract = {The Anime Blurry Preparation Dataset is a specialized binary classification dataset designed for distinguishing between clear and blurry 3D anime-style images. This dataset provides a comprehensive collection of image pairs where each clear image has a corresponding blurry counterpart, making it ideal for training and evaluating image quality assessment models. The dataset employs automatic annotation techniques using BLIP to generate reliable labels for the classification task. This dataset specifically focuses on 3D anime-style content, addressing the unique challenges of image quality assessment in computer-generated anime imagery. The clear images represent high-quality, sharp renderings while the blurry counterparts simulate various types of image degradation that commonly occur during rendering, compression, or transmission processes.},
keywords = {binary classification, image-classification, computer-vision, anime, 3d}
}