This sample uses the MNIST model from the Model Zoo: https://github.com/onnx/models/tree/master/vision/classification/mnist
(TODO: add a CI build for this sample)
- MacOS Catalina
- Xcode 11
- Compiled libonnxruntime.dll / lib
Command line:
$ xcodebuild -project SwiftMnist.xcodeproj
From Xcode, open SwiftMnist.xcodeproj and run with Command-R.
Just draw a number on the surface, when you lift your finger from the mouse or the trackpad, the guess will be displayed.
Note that when drawing numbers requiring multiple drawing strokes, the model will be run at the end of each stroke with probably wrong predictions (but it's amusing to see and avoids needing to press a 'run model' button).
(Add once it is added)
MNIST's input is a {1,1,28,28} shaped float tensor, which is basically a 28x28 floating point grayscale image (0.0 = background, 1.0 = foreground).
MNIST's output is a simple {1,10} float tensor that holds the likelihood weights per number. The number with the highest value is the model's best guess.
The MNIST structure uses std::max_element to do this and stores it in result_:
https://github.com/microsoft/onnxruntime-inference-examples/blob/master/c_cxx/MNIST/MNIST.cpp#L31