cs-winui

Windows ML WinUI 3 Image Classification Sample

This sample demonstrates how to use Windows ML in a WinUI 3 packaged or unpackaged application for image classification. The application showcases Windows UI design with automatic demo functionality and execution provider configuration.

Features

• Automatic Demo Mode - Loads and classifies included sample image on startup
• Execution Provider Configuration - View available providers (NPU, GPU, CPU) and configure download settings
• Real-time Inference - Classify images using the SqueezeNet model with immediate results
• Results Display - Show top 5 predictions with confidence percentages
• Model Reloading - Reload model with different provider settings
• Modern UI - Clean WinUI 3 interface with contemporary design elements

Key Technologies

• WinUI 3 - Windows UI framework for packaged applications
• Windows ML - ONNX Runtime integration via Windows App SDK
• Shared Helpers - Uses WindowsML.Shared library for consistent ML operations
• Packaged Deployment - MSIX deployment model with app manifest

Building and Running

1. Open the main solution: WindowsML-Samples.sln
2. Set the WindowsMLSample project as the startup project
3. Build and run (F5)

Usage

1. Launch Application - The app automatically loads and displays execution provider information
2. Auto-Demo - Sample image is automatically loaded and classified on startup
3. Configure Providers - Use "Allow Provider Download" checkbox to enable automatic EP downloads
4. Reload Model - Click "Reload Model" to reinitialize with new provider settings
5. View Results - See the top 5 predictions displayed automatically

Expected Output

Top 5 Predictions:
-------------------------------------------
Label                            Confidence
-------------------------------------------
French bulldog                       45.07%
bull mastiff                         35.05%
boxer                                 2.85%
pug                                   1.01%
American Staffordshire terrier        0.57%
-------------------------------------------

Code Structure

The sample uses the shared WindowsML.Shared library for consistency across samples:

// Initialize execution providers
var catalog = ExecutionProviderCatalog.GetDefault();
await ModelManager.InitializeExecutionProvidersAsync(allowDownload: allowDownload);

// Load and preprocess image
using var videoFrame = await ImageProcessor.LoadImageFileAsync(imagePath);
var inputTensor = await ImageProcessor.PreprocessImageAsync(videoFrame);

// Run inference
using var results = InferenceEngine.RunInference(_session, inputTensor);
var output = InferenceEngine.ExtractResults(_session, results);

// Process results
var topResults = ResultProcessor.GetTopPredictions(output, _labels, 5);

Related Samples

• WPF Version: cs-wpf - Same functionality with WPF UI framework
• WinForms Version: cs-winforms - Windows Forms application
• Console Version: cs/CSharpConsoleDesktop - Command-line interface

Model Information

This sample uses the SqueezeNet model:

• Purpose: Image classification with 1000 categories
• Input: 224x224 RGB images
• Output: Probability distribution over ImageNet classes
• Size: Lightweight model optimized for performance

Deployment

The application can be deployed as:

• Development: Direct run from Visual Studio
• Sideloading: MSIX package for deployment to other devices