Skip to content

Latest commit

 

History

History
60 lines (40 loc) · 1.38 KB

README.md

File metadata and controls

60 lines (40 loc) · 1.38 KB

WoW Screenshot OCR

wow-ocr is an OCR model to extract text from WoW screenshots.

It reads into :

  • Chat
  • Combat log
  • Nameplates
  • UI frames
  • Map

Installation

pip install wow-ocr

Usage

wow-ocr is packaged with trained weights. It's very easy to use : Try it on Colab

import wow_ocr

# 1 - Init pipeline, detector and recognizer models with trained weights
pipeline = wow_ocr.pipeline.Pipeline()


# 2 - Provide screenshots urls
images = [
    wow_ocr.tools.read(url)
    for url in [
        "https://image_url.com/1.jpg",
        "https://image_url.com/2.jpg",
    ]
]

# 3 - Get predictions
prediction_groups = pipeline.recognize(images)
# # Each list of predictions in prediction_groups is a list of
# # (word, box) tuples.

Training

The recognizer model was fine tuned to be able to work with WoW Fonts. Here is the recognizer fine tuning process : Fine Tuning Recognizer

Parsing

wow-ocr has been used to extract text from over 20,000 screenshots. You can see the parsing process here: Parsing Big Dataset