deepghs/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

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

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

@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}
}