A handwriting recognition project with CNN structure (LeNet).
By default, it uses MNIST dataset to train. The trained model is saved in the ./saved_model directory by default.
To use the model, use the use_model.py file, it by default recognize the images saved in the ./imgs directory, and generate output to the file ./output.txt with the format "img_path -> number \n"
As an example:
gives output:
./imgs/1.png -> 1
./imgs/2.png -> 2
./imgs/3.png -> 3
Note that the images should have black background and their number written in white color, and resized to the dimension 28x28 before using the model to recognize them.