anime_halfbody_detection

# Anime Halfbody Detection Dataset

## Summary
This repository provides a comprehensive **anime halfbody detection** dataset specifically designed for training object detection models to identify and localize anime character halfbodies in images. The dataset contains carefully annotated bounding boxes around **anime character upper bodies** and is formatted for compatibility with popular object detection frameworks including **Ultralytics YOLOv8** and **COCO** formats.

The dataset includes both original raw data and augmented versions to support various training scenarios. The **raw dataset** contains the original annotations without any data augmentation applied, preserving the authentic distribution and characteristics of the source material. This makes it ideal for research purposes and baseline model development where data purity is essential. The augmented versions provide enhanced training data with various transformations to improve model robustness and generalization capabilities.

The dataset supports **multiple annotation formats** to accommodate different training pipelines and research requirements. The YOLOv8 format is optimized for fast training with Ultralytics frameworks, while the COCO format provides standardized JSON annotations compatible with a wide range of detection frameworks. This dual-format approach ensures maximum usability across different development environments and research workflows.

Performance highlights include precise **bounding box annotations** that accurately capture the halfbody regions of anime characters across diverse artistic styles and character designs. The dataset is particularly valuable for applications in **anime image processing**, character recognition systems, and automated content analysis tools that require reliable detection of character upper bodies in anime artwork.

## Usage

### Using with Ultralytics YOLOv8

```python
from ultralytics import YOLO

# Load the dataset
model = YOLO('yolov8n.pt')

# Train using the YOLOv8 formatted dataset
model.train(
data='path/to/v1.0.yolov8.zip',
epochs=100,
imgsz=640,
batch=16
)
```

### Using with COCO Format

```python
from datasets import load_dataset
import json

# Load COCO annotations
with open('path/to/v1.0.coco.zip/annotations.json', 'r') as f:
annotations = json.load(f)

# Process the dataset for training
# The dataset follows standard COCO object detection format
```

## Dataset Structure

The repository contains four main dataset versions:

- **v1.0-raw.coco.zip**: Original dataset in COCO format without augmentation
- **v1.0-raw.yolov8.zip**: Original dataset in YOLOv8 format without augmentation
- **v1.0.coco.zip**: Enhanced dataset in COCO format with data augmentation
- **v1.0.yolov8.zip**: Enhanced dataset in YOLOv8 format with data augmentation

## Original Content

ultralytics-compatible dataset for anime character halfbody bboxes
raw means original dataset without data augmentation

## Citation

```bibtex
@misc{anime_halfbody_detection,
title = {Anime Halfbody Detection Dataset},
author = {deepghs},
howpublished = {\url{https://huggingface.co/datasets/deepghs/anime_halfbody_detection}},
year = {2023},
note = {Ultralytics-compatible dataset for anime character halfbody bounding box detection with both raw and augmented versions},
abstract = {This repository provides a comprehensive anime halfbody detection dataset specifically designed for training object detection models to identify and localize anime character halfbodies in images. The dataset contains carefully annotated bounding boxes around anime character upper bodies and is formatted for compatibility with popular object detection frameworks including Ultralytics YOLOv8 and COCO formats. The dataset includes both original raw data and augmented versions to support various training scenarios, with the raw dataset containing original annotations without data augmentation for research purposes and baseline model development.},
keywords = {anime, halfbody, object-detection, ultralytics, yolov8}
}
```

# 动漫半身目标检测数据集

## 概述
本仓库提供了一套全面的**动漫半身目标检测**数据集，专为训练目标检测模型以识别并定位图像中的动漫角色半身区域而设计。该数据集围绕动漫角色上半身精心标注了边界框，且兼容包括**Ultralytics YOLOv8**与**COCO**格式在内的主流目标检测框架。

本数据集包含原始未经过数据增强的原始数据与增强版本，以适配多种训练场景。**原始数据集**保留了源数据的真实分布与特征，未施加任何数据增强操作的原始标注，因此非常适合对数据纯净度有严格要求的研究工作与基线模型开发。增强版本则通过各类变换生成了更丰富的训练数据，有助于提升模型的鲁棒性与泛化能力。

本数据集支持**多种标注格式**，以适配不同的训练流程与研究需求。YOLOv8格式针对Ultralytics框架的快速训练进行了优化，而COCO格式则提供了标准化的JSON标注，可兼容绝大多数目标检测框架。这种双格式设计确保了数据集在不同开发环境与研究工作流中都能发挥最大效用。

该数据集的性能亮点在于其精准的**边界框标注**，可在多样的艺术风格与角色设计下准确捕获动漫角色的半身区域。其在动漫图像处理、角色识别系统以及需要可靠检测动漫作品中角色上半身的自动化内容分析工具等场景中具有极高的应用价值。

## 使用方法

### 适配Ultralytics YOLOv8的使用方式
python
from ultralytics import YOLO

# 加载模型与数据集
model = YOLO('yolov8n.pt')

# 使用YOLOv8格式数据集开展训练
model.train(
data='path/to/v1.0.yolov8.zip',
epochs=100,
imgsz=640,
batch=16
)

### 适配COCO格式的使用方式
python
from datasets import load_dataset
import json

# 加载COCO标注文件
with open('path/to/v1.0.coco.zip/annotations.json', 'r') as f:
annotations = json.load(f)

# 针对训练需求处理数据集
# 该数据集遵循标准COCO目标检测格式规范

## 数据集结构
本仓库包含四个主要数据集版本：
- **v1.0-raw.coco.zip**：无数据增强的COCO格式原始数据集
- **v1.0-raw.yolov8.zip**：无数据增强的YOLOv8格式原始数据集
- **v1.0.coco.zip**：经过数据增强的COCO格式增强数据集
- **v1.0.yolov8.zip**：经过数据增强的YOLOv8格式增强数据集

## 原始内容说明
适用于Ultralytics框架的动漫角色半身边界框标注数据集，其中“raw”代表未经过数据增强的原始数据集。

## 引用格式
bibtex
@misc{anime_halfbody_detection,
title = {动漫半身目标检测数据集},
author = {deepghs},
howpublished = {url{https://huggingface.co/datasets/deepghs/anime_halfbody_detection}},
year = {2023},
note = {适用于Ultralytics框架的动漫角色半身边界框检测数据集，包含原始版本与增强版本},
abstract = {本仓库提供了一套全面的动漫半身目标检测数据集，专为训练目标检测模型以识别并定位图像中的动漫角色半身区域而设计。该数据集围绕动漫角色上半身精心标注了边界框，且兼容包括Ultralytics YOLOv8与COCO格式在内的主流目标检测框架。数据集包含原始未经过数据增强的数据与增强版本，以适配多种训练场景；其中原始数据集保留了源数据的原始标注，未施加任何数据增强操作，适用于对数据纯净度有严格要求的研究工作与基线模型开发。},
keywords = {动漫, 半身目标, 目标检测, Ultralytics, YOLOv8}
}