wow-ocr is an OCR model to extract text from WoW screenshots.
- Chat
- Combat log
- Nameplates
- UI frames
- Map
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.
The recognizer model was fine tuned to be able to work with WoW Fonts. Here is the recognizer fine tuning process : Fine Tuning Recognizer
wow-ocr has been used to extract text from over 20,000 screenshots. You can see the parsing process here: Parsing Big Dataset