A homemade digit recognition system
Includes data generation, training, teaching, and prediction
Read about MNIST here
Why do this in the first place? Doesn't a better MNIST dataset already exist?
Yes. Much larger, much higher accuracy MNIST datasets exist. This was created for two reasons:
- I wanted experience generating a dataset myself
- I wanted to see what conditions (image resolution, dataset size, training time) would contribute to better results
Conv2D (15x15, relu) -> Flatten -> Dense (64, relu) -> Dense (10, softmax)
Using a convolution layer is very common for image based classifiers. It allows for a better understanding of spacially related information.