Just another real-time Sudoku solver coded in Python using OpenCV and PyTorch. Capable of tackling Sudoku puzzles in images and videos, and projecting the solutions back onto the original visuals.
After cloning the repository install the dependencies by running
pip install -r requirements.txt
If you just want to execute main.py you're good to go, as it comes with a pre-trained neural network for digit classification provided in model_file.pt.
Run the following command to verify that the setup is completed. The solved Sudoku should be displayed in a new window
python main.py -i ./images/sudoku_photo.jpg
If you want to retrain the neural network for digit classification you need to download the dataset from Kaggle. Unzip the downloaded folder, rename it to archive and paste it into the project directory. Alternatively, you can store the unzipped folder in a directory of your choice and update the path in the ASSETS_PATH variable within classifier/dataset.py. Now you should be ready to retrain the neural network using the classifier.ipynb notebook.
For solving a Sudoku puzzle in an image use
python main.py -i <PATH_TO_IMAGE>
The puzzle with the projected solution should be displayed in an new window.
For solving a Sudoku puzzle from a videocamera use
python main.py -v
The video stream should be displayed in an new window. Optionally you can pass a device index (integer) to the -v argument used by OpenCV for camera selection.
The dataset used in this project was published by Kshitij Dhama on Kaggle.