You can first draw any digit in the white pane using the mouse and the application can parse the result when clicking the "Recognize" button.
There are 2 solutions offering the same UI but interacting with different backends.
- MnistDemo.FlaskServer
Set up a Python Web server which calls TensorFlow for image classification based on Checkpoint. - MnistDemo.TensorFlowSharp
Freeze the inference graph and wrap it as a .NET class library based on the 3rd-party project titled TensorFlowSharp.
Visual Studio 2015+ with:
- Python Tools
- C# and WPF
Runtime:
- .NET Framework 4.5+
- Python 3.5+