DraftGPT

Magic: The Gathering Online OCR from arbitrary screen selection in Python

Usage

python cardfinder.py

• This will pop up a window on your main screen to select a bounding box to search for multiple cards.
• Drag from top left to bottom right, type 'w' to confirm selection
• A second window will pop up to show cards and their names as found by cv2 and identified by tesseract
• While the second window is open, you can scroll slowly through any number of cards, type 'q' to exit this window
• On close, the total unqiue list of cards seen is compiled, and card details are retrieved from Scryfall fuzzy search API
• Example usage: Youtube Example

Notes

• It's recommended to make the individual card size as large as possible in mtgo.
• Code has only been run on windows, Mac OS and Linux use at your own risk.
• Google how to set up tesseract
• If your python cannot find the executable from PATH, add the line pytesseract.pytesseract.tesseract_cmd = r'C:\Program Files\Tesseract-OCR\tesseract.exe' to the beginning of cardfinder.py
• To speed up processing, a forked pytesseract is in the repo, the change implements tesseract read from stdin rather than writing image to disk, based on this comment
• An attempt at multithreading was made in cardfinder_multithreaded.py but is much slower, even after limiting tesseract to single-threaded. Slightly more accurate, but twice as slow.

Dependencies

• pytesseract fork in repo
• PIL
• cv2
• pynput