board_to_fen

Python package that converts digital chessboard image into Forsyth-Edwards notation (FEN) notation

board_to_fen is available on PyPI:

pip3 install board_to_fen

from board_to_fen.predict import get_fen_from_image_path

print(get_fen_from_image_path(PATH_TO_CHESSBOARD_IMAGE))

or, if you want you can load image object by yourself:

from PIL import Image
from board_to_fen.predict import get_fen_from_image

img = Image.open(PATH_TO_CHESSBOARD_IMAGE)

print(get_fen_from_image(img))

Note: The package uses tensorflow+keras API. They are pretty heavy.

get_fen_from_image_path takes has 3 arguments:

image_path [required]
end_of_row '/' by default
black_view False by default -> set True if chessboard is provided from black player perspective

For training You would probably want to download the source code by cloning the repository:

$ git clone https://github.com/mcdominik/board_to_fen.git

Download training data from:
I will supply url for data in the future

In the main repository dir, run

$ python3 ./board_to_fen/train_model.py

january 2023
- versions 0.0.17-25
- added simple board validation
- bug fixes
february 2023 version 0.1.0-0.1.1
- migratation from cv2 to PIL
- new function for direct image object load
- add simple tests
- bug fixes

Warnings

Image has to be provided in neutral angle (white or black player's perspective).
Image has to be square (~3% tolerance depending on image resolution).
Image can't contain paddings, board borders etc. other than 64 squares (with pieces) itself.