Description

This is fun weekend project intended to test the idea of engine help without hacking into website communication protocol.

High level idea:

• take screenshot of the browser with Lichess board
• detect board, pieces, board orientation and who's turn to move using computer vision and neural network techniques
• send current position to Stockfish and get best move estimate

Disclaimer

This is not intended for serious cheating and I don't encourage you to use it for playing internet chess. Moreover, I never use and stand against any kind of cheating during chess games.

Usage

Open Lichess web page for approximate 2/3 of screen width.

Choose black background settings and usual pieces. The board size should be large enough to take almost all height. The clocks should be visible (it is used to detect who's turn).

Open console window to the right of browser window and run python console.py

Console takes screenshot, detects position and runs Stockfish for 1 second. The result and

Requirements

• Python 3.6
• Stockfish engine

Installation

Install virtualenv with python3.6.

pip install -r requirements.txt

Generate training data

To detect each piece we need to generate boards full of pieces for each board background and piece style.

Inside nn/data folder run simple server

php -S localhost:9000

and navigate Google Chrome to http://localhost:9000. The page will render board for one piece set and all board styles. Images are saved to pngs/ folder by save.php script.

Changing let i = 0 in index.html and refreshing page will yield board images for other piece set (doing it in a loop causes web browser hanging).

Generated pngs are stored here.

These pngs are used to train simple convolution neural network which detects individual piece.