A short demo that shows how to create, train, test, save and reload a CNN model.
This CNN ranks under the top 50 in the MNIST benchmark. For that, my basic NN is used, extended and turned into a CNN. The result was quite good with 99,20% accuracy in the MNIST test. This CNN model here is quite similar and based on the previous demo. But with an additional technique, Infinity Dropout, and a few more epochs of training.
With 99.42% accuracy in the MNIST test, it was even much better, nice.
To run the demo program, you must have VisualStudio2022 installed on your machine. Then just start a console application with .NET 6, copy the code and change from debug to release mode and run the demo. MNIST data and network are then managed by the code.