deepghs/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

# 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

# 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:

# 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

@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}
}