manga109_yolo
| Property | Value |
|---|---|
| Model Type | YOLO (You Only Look Once) |
| Latest Version | v2023.12.07 |
| Model Variants | Nano to Extra Large |
| Repository | Hugging Face |
What is manga109_yolo?
manga109_yolo is a specialized computer vision model designed for manga image analysis. It utilizes the YOLO architecture to detect and classify four key elements in manga pages: body, face, frame, and text. The model comes in multiple variants optimized for different computational requirements, from lightweight nano versions to high-capacity extra-large versions.
Implementation Details
The model family includes multiple variants with different parameter counts and computational requirements: Extra Large (258G FLOPS, 68.2M params), Large (165G FLOPS, 43.6M params), Medium (79.1G FLOPS, 25.9M params), Small (28.7G FLOPS, 11.1M params), and Nano (8.2G FLOPS, 3.01M params). All variants maintain impressive F1 scores ranging from 0.88 to 0.92.
- Advanced object detection architecture using YOLO framework
- Optimized for manga-specific element detection
- Multiple model sizes for different deployment scenarios
- High precision and recall across all variants
Core Capabilities
- Body Detection: Identifies character bodies in manga panels
- Face Detection: Locates and identifies character faces
- Frame Detection: Recognizes panel boundaries and layout elements
- Text Detection: Identifies text areas including speech bubbles and captions
- High Performance: Achieves mAP50 scores of up to 0.95 on benchmark datasets
Frequently Asked Questions
Q: What makes this model unique?
The model's specialization in manga content analysis, combined with its variety of deployment options from nano to extra-large variants, makes it uniquely versatile for manga processing applications. Its high F1 scores (0.88-0.92) across all variants demonstrate robust performance regardless of model size.
Q: What are the recommended use cases?
This model is ideal for manga digitization projects, content analysis, automated translation preprocessing, and manga research applications. The different variants allow users to choose between computational efficiency and maximum accuracy based on their specific needs.
