satoyutaka/yolov8n-handwritten-japanese-ingredients

YOLOv8n Handwritten Japanese Ingredients Detection

This model is a fine-tuned version of YOLOv8n specifically trained to detect handwritten Japanese ingredient regions (text blocks) from images.

Model Description

• Task: Object Detection
• Base Model: YOLOv8n (nano)
• Target Class: text (handwritten text regions)
• Category: OCR Pre-processing / Vision

Intended Use

This model is designed to be the "vision" component of a mobile application that calculates nutritional information from handwritten ingredient lists. It identifies text regions before they are passed to an OCR engine.

Training Data & Environment

• Dataset: 30 high-quality handwritten images (20 Train / 5 Val / 5 Test).
• Hardward: MacBook Air (M4) using MPS (Metal Performance Shaders).
• Training Epochs: 100

Performance

Fine-tuned on 20 images, the model achieved a significant improvement compared to the baseline stock YOLOv8 model.

Metric	Value
mAP50	0.978
Precision	0.966
Recall	0.981

Usage

You can use this model with the ultralytics library:

from ultralytics import YOLO

# Load the model
model = YOLO("satoyutaka/yolov8n-handwritten-japanese-ingredients")

# Run inference
results = model.predict("path/to/your/image.jpg")

# Show results
results[0].show()

---

Created by satoyutaka.