anime_eye_detection

# Anime Eye Detection Dataset

## Summary

This repository provides an **ultralytics-compatible** dataset specifically designed for **anime character eye detection**. The dataset contains annotated bounding boxes for eyes in anime-style character images, making it ideal for training computer vision models to detect and localize eyes in anime artwork. The dataset is formatted to work seamlessly with **YOLOv8** and other Ultralytics frameworks, enabling researchers and developers to quickly implement and experiment with eye detection models for anime content.

The dataset includes multiple format versions to accommodate different training pipelines and requirements. The **v1.0** release contains cleaned and processed annotations, while the **v1.0-raw** versions provide the original unprocessed data for researchers who prefer to apply their own preprocessing pipelines. This dual-format approach ensures flexibility for various use cases, from rapid prototyping to detailed research experiments.

Each dataset version is available in both **COCO** and **YOLOv8** formats, providing compatibility with a wide range of object detection frameworks. The COCO format supports general-purpose detection pipelines, while the YOLOv8 format is optimized specifically for Ultralytics implementations. This comprehensive format coverage makes the dataset accessible to researchers using different technical stacks and preferences.

The dataset focuses on the challenging task of **anime eye detection**, which presents unique characteristics compared to real-world eye detection. Anime eyes often feature exaggerated proportions, diverse artistic styles, and non-realistic color schemes, making this dataset particularly valuable for studying domain-specific computer vision applications. Keywords: **anime**, **eye-detection**, **object-detection**, **ultralytics**, **yolov8**.

## Usage

The dataset is provided in multiple formats for different use cases:

### For YOLOv8 Users
```bash
# Download and extract the YOLOv8 formatted dataset
wget https://huggingface.co/datasets/deepghs/anime_eye_detection/resolve/main/v1.0.yolov8.zip
unzip v1.0.yolov8.zip

# Use with Ultralytics YOLOv8
from ultralytics import YOLO

# Load and train model
model = YOLO('yolov8n.pt')
model.train(data='path/to/extracted/dataset/data.yaml', epochs=100, imgsz=640)
```

### For COCO Format Users
```bash
# Download and extract the COCO formatted dataset
wget https://huggingface.co/datasets/deepghs/anime_eye_detection/resolve/main/v1.0.coco.zip
unzip v1.0.coco.zip

# Use with any COCO-compatible framework
import json

with open('path/to/annotations/instances_train.json', 'r') as f:
coco_annotations = json.load(f)
```

### Raw Data Versions
For researchers who prefer to apply custom preprocessing:
```bash
# Raw YOLOv8 format
wget https://huggingface.co/datasets/deepghs/anime_eye_detection/resolve/main/v1.0-raw.yolov8.zip

# Raw COCO format
wget https://huggingface.co/datasets/deepghs/anime_eye_detection/resolve/main/v1.0-raw.coco.zip
```

## Dataset Structure

The repository contains the following files:

- **v1.0.coco.zip**: Processed dataset in COCO format
- **v1.0.yolov8.zip**: Processed dataset in YOLOv8 format
- **v1.0-raw.coco.zip**: Raw dataset in COCO format
- **v1.0-raw.yolov8.zip**: Raw dataset in YOLOv8 format

Each zip file contains the complete dataset with images and annotations in the specified format, ready for immediate use in training object detection models.

## Original Content

ultralytics-compatible dataset for anime character eyes bboxes

# Citation

```bibtex
@misc{anime_eye_detection,
title = {Anime Eye Detection Dataset},
author = {deepghs},
howpublished = {\url{https://huggingface.co/datasets/deepghs/anime_eye_detection}},
year = {2023},
note = {Ultralytics-compatible dataset for anime character eye detection with bounding box annotations},
abstract = {This repository provides an ultralytics-compatible dataset specifically designed for anime character eye detection. The dataset contains annotated bounding boxes for eyes in anime-style character images, making it ideal for training computer vision models to detect and localize eyes in anime artwork. The dataset is formatted to work seamlessly with YOLOv8 and other Ultralytics frameworks, enabling researchers and developers to quickly implement and experiment with eye detection models for anime content. The dataset includes multiple format versions to accommodate different training pipelines and requirements, with both processed and raw data available in COCO and YOLOv8 formats.},
keywords = {anime, eye-detection, object-detection, ultralytics, yolov8}
}
```

# 动漫人物眼部检测数据集

## 概述
本仓库推出一款**兼容Ultralytics（ultralytics）**的专项数据集，专为动漫人物眼部检测任务打造。该数据集收录动漫风格人物图像的眼部标注边界框，堪称训练计算机视觉模型以检测并定位动漫作品眼部区域的理想载体。其格式可与**YOLOv8**及其他Ultralytics框架无缝适配，助力研究人员与开发者快速部署并测试面向动漫内容的眼部检测模型。

本数据集提供多种格式版本，以适配不同的训练流程与需求。**v1.0**版本包含经过清理与预处理的标注数据，而**v1.0-raw**版本则提供原始未处理的数据，供偏好自定义预处理流程的研究人员使用。这种双格式设计可兼顾快速原型开发与精细研究实验等各类场景需求，为不同使用场景提供充足灵活性。

各数据集版本均提供**COCO（COCO）**与**YOLOv8**两种格式，可兼容多种主流目标检测框架。其中COCO格式适配通用目标检测流程，而YOLOv8格式则针对Ultralytics实现做了专项优化。这种全面的格式覆盖，让不同技术栈与使用偏好的研究人员都能便捷使用该数据集。

本数据集聚焦于**动漫眼部检测**这一极具挑战性的任务，与真实场景下的眼部检测相比，该任务具备独特的视觉特性。动漫眼部往往拥有夸张的比例、多样的艺术风格与非写实的色彩搭配，这使得本数据集在研究领域专属的计算机视觉应用方面具备极高价值。关键词：**动漫（anime）**、**眼部检测（eye-detection）**、**目标检测（object-detection）**、**Ultralytics（ultralytics）**、**YOLOv8（yolov8）**

## 使用指南
本数据集提供多种格式版本，适配不同使用场景：

### 针对YOLOv8用户
bash
# 下载并解压YOLOv8格式的数据集
wget https://huggingface.co/datasets/deepghs/anime_eye_detection/resolve/main/v1.0.yolov8.zip
unzip v1.0.yolov8.zip

# 搭配Ultralytics YOLOv8使用
from ultralytics import YOLO

# 加载并训练模型
model = YOLO('yolov8n.pt')
model.train(data='path/to/extracted/dataset/data.yaml', epochs=100, imgsz=640)

### 针对COCO格式用户
bash
# 下载并解压COCO格式的数据集
wget https://huggingface.co/datasets/deepghs/anime_eye_detection/resolve/main/v1.0.coco.zip
unzip v1.0.coco.zip

# 可与任意兼容COCO的框架搭配使用
import json

with open('path/to/annotations/instances_train.json', 'r') as f:
coco_annotations = json.load(f)

### 原始数据版本
针对偏好自定义预处理流程的研究人员：
bash
# 原始YOLOv8格式
wget https://huggingface.co/datasets/deepghs/anime_eye_detection/resolve/main/v1.0-raw.yolov8.zip

# 原始COCO格式
wget https://huggingface.co/datasets/deepghs/anime_eye_detection/resolve/main/v1.0-raw.coco.zip

## 数据集结构
本仓库包含以下文件：
- **v1.0.coco.zip**：COCO格式的预处理后数据集
- **v1.0.yolov8.zip**：YOLOv8格式的预处理后数据集
- **v1.0-raw.coco.zip**：COCO格式的原始数据集
- **v1.0-raw.yolov8.zip**：YOLOv8格式的原始数据集

每个压缩包均包含对应格式下的完整数据集，包含图像与标注文件，可直接用于目标检测模型的训练。

## 原始内容说明
兼容Ultralytics的动漫人物眼部边界框数据集

## 引用
bibtex
@misc{anime_eye_detection,
title = {Anime Eye Detection Dataset},
author = {deepghs},
howpublished = {url{https://huggingface.co/datasets/deepghs/anime_eye_detection}},
year = {2023},
note = {Ultralytics-compatible dataset for anime character eye detection with bounding box annotations},
abstract = {This repository provides an ultralytics-compatible dataset specifically designed for anime character eye detection. The dataset contains annotated bounding boxes for eyes in anime-style character images, making it ideal for training computer vision models to detect and localize eyes in anime artwork. The dataset is formatted to work seamlessly with YOLOv8 and other Ultralytics frameworks, enabling researchers and developers to quickly implement and experiment with eye detection models for anime content. The dataset includes multiple format versions to accommodate different training pipelines and requirements, with both processed and raw data available in COCO and YOLOv8 formats.},
keywords = {anime, eye-detection, object-detection, ultralytics, yolov8}
}