Chess PhotoBoard → FEN Generator

This project converts a chessboard photo into a FEN (Forsyth–Edwards Notation) string using a CNN tile classifier.

Pipeline

Photo → Board Crop → Tile Extraction → CNN Classifier → FEN Generation

Project Structure

• inference/ → Run prediction on a board image
• training/ → Train CNN tile classifier
• pipeline/ → Full automation pipeline
• scripts/ → Image preprocessing utilities
• models/ → Trained model + class indices

Run Inference

python inference/run_full_board.py

Model

The project uses a CNN classifier trained on chessboard tile images.

Classes predicted:

• wP, wR, wN, wB, wQ, wK
• bP, bR, bN, bB, bQ, bK
• empty

Output

The system reconstructs the board and produces a FEN string describing the full chess position.

Example output:

rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR

Future Improvements

• Improve board detection robustness
• Support different board orientations
• Integrate with chess engines for move suggestions