sudosolv

Sudosolv will take a picture of a sudoku and will crop the sudoku from the image and decode this into plaintext.

To use this, you will need to train a number recognition classifier first. There are two datasets possible, either the mnist dataset, or a generated dataset based on available fonts.

Prepare the fonts dataset:

cd dataset/fonts
sh download.sh
sh create.sh

Prepare the mnist dataset:

cd dataset/mnist
sh download.sh

Train the network:

make train

Test with some sudoku image:

go run main.go decode -w trained.bin -f my_sudoku_image.jpg
go run main.go decode -w trained.bin -f my_sudoku_image.jpg --display

Or run the webserver and play around on http://localhost:8080:

go run main.go server -w trained.bin