mk8dx-table-reader 0.6.4

A package to extract player names and scores from MK8DX tables using YOLO and ONNX-based EasyOCR.

MK8DX Table Reader

A Python package to extract player names and scores from Mario Kart 8 Deluxe end-game tables using YOLO object detection and ONNX-based EasyOCR.

Installation

Option 1: Lightweight ONNX Version (Recommended)

# Install the package (without heavy PyTorch dependencies)
pip install mk8dx-table-reader

# Download ONNX EasyOCR models (~17 MB)
python download_easyocr_models.py

Total size: ~50 MB

Option 2: PyTorch Version (Fallback)

# Install with PyTorch-based EasyOCR
pip install mk8dx-table-reader[pytorch-ocr]

Total size: ~2 GB

Features

• Extract player names from MK8DX end-game screenshots
• Read scores and positions from tables
• Uses YOLO for object detection
• NEW: ONNX-based EasyOCR for lightweight inference (3-4x faster, 97% smaller)
• Fallback to PyTorch EasyOCR if ONNX models not available
• OCR powered by optimized ONNX models and Keras models

Requirements

Core Requirements

• Python >= 3.11.11
• OpenCV
• ONNXRuntime
• Ultralytics YOLO
• NumPy
• Pillow

Optional Requirements

• EasyOCR (PyTorch-based, only if not using ONNX models)

Usage

from mk8dx_table_reader import Fullreader
import PIL

# Initialize reader (automatically uses ONNX if models are available)
reader = Fullreader()

# Load an image
img = PIL.Image.open("screenshot.png")

# Extract player names and scores
names, scores = reader.fullOCR(img)

print("Players:", names)
print("Scores:", scores)

ONNX EasyOCR Benefits

Feature	PyTorch EasyOCR	ONNX EasyOCR
Package Size	~2 GB	~17 MB
Inference Speed	Baseline	3-4x faster
Memory Usage	~2 GB RAM	~50 MB RAM
Accuracy	100%	>95%

See EASYOCR_ONNX_GUIDE.md for more details.

License

See LICENSE.md for details.

Author

Julien ABADIE - [e-mail]

GitHub

https://github.com/PrizeWhipser/shortcat.tips