This post will show how to train a custom image classification model in Azure to categorize flowers using ML.NET Model Builder. Then, you can leverage your existing .NET skills to consume the trained model inside a C# .NET Core console application. Best of all, little to no prior machine learning knowledge is required. Let’s get started!<\/p>\n
ML.NET Model Builder<\/h2>\n
ML.NET<\/a> is an open-source and cross-platform machine learning framework for .NET developers. Model Builder is a tool in Visual Studio that provides a graphical user interface that uses Automated Machine Learning (Auto ML) to train and consume custom ML.NET models for your .NET applications.<\/p>\n To follow along, you’ll need the following prerequisites:<\/p>\n To use Model Builder, make sure to enable the preview feature. In the Visual Studio Menu Bar, select Tools > Options > Environment > Preview Features<\/strong>. Then, check the Enable ML.NET Model Builder<\/strong> box.<\/p>\n The process of training models usually consists of the steps outlined below:<\/p>\n The dataset used in this guide is based on the TensorFlow flower photos dataset<\/a>. All images in this archive are licensed under the Creative Commons By-Attribution License, available at: https:\/\/creativecommons.org\/licenses\/by\/4.0\/<\/a><\/p>\n The full license information is provided in the LICENSE.txt file which is included as part of the same image set.<\/p>\n The dataset is divided into separate subfolders, one for each flower category:<\/p>\n Download the data anywhere on your PC and unzip it.<\/p>\n Model Builder requires data to be in a certain format. This dataset is already in the correct format. For more information, see Microsoft Docs to learn more about loading data with Model Builder<\/a>.<\/p>\n The application that consumes the model is a C# .NET Core console application.<\/p>\n Open Visual Studio and create a new C# .NET Core console application. In this case, I’ve called my application MBImageClassificationApp<\/em>.<\/p>\n In Solution Explorer, right-click your newly created console application project and select Add > Machine Learning<\/strong> to launch Model Builder.<\/p>\n Once you’re presented with the scenario screen, select Image Classification<\/strong> and proceed to the next step.<\/p>\n For image classification scenarios, you can choose between training locally or in the cloud. For this sample we’ll train the model in Azure. For scenarios that train on thousands of images and require a large amount of resources, it is recommended to use Azure, which provides GPU optimized virtual machines for training.<\/p>\n Choose Azure<\/strong> form the list of training environments. Then, select the Set up workspace<\/strong> button to configure your environment.<\/p>\n An Azure Machine Learning experiment is a resource that needs to be created before running Model Builder training on Azure. The experiment encapsulates the configuration and results for one or more machine learning training runs.<\/p>\n To create an Azure Machine Learning experiment, you first need to configure your environment on Azure. An experiment needs an Azure subscription, workspace, and compute to run on.<\/p>\n In the Create New Experiment dialog, choose your Azure subscription. Then, select or create a new Azure Machine Learning workspace. A workspace is an Azure Machine Learning resource that provides a central place for all Azure Machine Learning resources and artifacts created as part of a training run.<\/p>\n When you create a new workspace, the following resources are provisioned:<\/p>\n As a result, this process may take a few minutes.<\/p>\n When you choose a region, it’s recommended that you select a location close to where you or your customers are.<\/p>\n Once you’ve chosen or created your workspace, choose or create a new Azure Machine Learning compute. An Azure Machine Learning compute is a cloud-based Linux VM used for training. Learn more about compute types supported by Model Builder<\/a>. This process may take a few minutes.<\/p>\n In the Create New Experiment dialog, leave the default experiment name and select Create<\/strong>.<\/p>\n The first experiment is created and its name is registered in the workspace. Any subsequent runs – if the same experiment name is used – are logged as part of the same experiment. Otherwise, a new experiment is created.<\/p>\n If you’re satisfied with your configuration, select the Data<\/strong> button to move to the next step.<\/p>\n In the Add data screen, load your data by selecting the button next to the Select a folder text box and use the file explorer to find the unzipped directory containing the subdirectories with images.<\/p>\n Once the data is loaded, select the Train<\/strong> button to train your model.<\/p>\n In the Model Builder train screen, select the Start training<\/strong> button to start training your model.<\/p>\n At this point, your data is uploaded to Azure Storage and the training process begins. The algorithm used to train these models is a Deep Neural Network<\/a> based on the ResNet50 architecture<\/a>. The training process takes some time and the amount of time may vary depending on the size of compute selected as well as the amount of data.<\/p>\n For this sample of 3670 images, training took about 30 minutes. The first time a model is trained in Azure, you can expect a slightly longer training time because resources have to be provisioned. You can track the progress of your runs in the Azure Machine Learning portal by selecting the “Monitor current run in Azure portal” link in Visual Studio.<\/p>\n Once the model is trained, select the Evaluate<\/strong> button to move to the next step.<\/p>\n In the evaluate screen, you are able to get an overview of the results from the training process such as how long the model took to train as well as the accuracy. Additionally, you can use the “Try your model” experience to quickly check whether your model is performing as expected. All you need to do is provide an image, preferably one that the model did not use as part of training. The model classifies the image and provides the list of categories along with their probabilities with the predicted value at the top of the list.<\/p>\n If you’re satisfied with your model, select the Code<\/strong> button to add the code to make predictions.<\/p>\n In the code screen, select the Add Projects<\/strong> button to add the auto-generated projects to your solution.<\/p>\n Two projects are added to your solution with the following suffixes:<\/p>\n Open the Program.cs<\/em> file from your console application. In this case, my console application is MBImageClassificationApp<\/em>.<\/p>\n Then, replace the code with the following:<\/p>\n The first thing I’ve done is added a using<\/em> statement to reference the MBImageClassificationAppML.Model <\/em>project. To test my application, I’ve downloaded an image of tulips called Pink_Tulips<\/a> from the internet and included it in my MBImageClassificationApp<\/em> project. The model has not seen this image before. A new instance of ModelInput\u00a0<\/em>is created and the path of the image is used as the value for the ImageSource<\/em> property.<\/p>\n Then, ConsumeModel.Predict<\/em>, a helper method that loads the MLModel.zip\u00a0<\/em>file and uses PredictionEngine<\/em>, a convenience API to make predictions on a single instance of data, is called with the sample image as input. Finally, both the actual class the image belongs to, as well as the predicted flower category, are printed out to the console.<\/p>\n When you run the application, you should see output similar to the following in the console:<\/p>\n Azure Machine Learning Enterprises are currently in preview. Therefore, there is no additional surcharge at the moment. Training this model cost less than $0.50. Depending on your scenario, your costs may vary depending on the amount of data and size of your compute. For more information on pricing, see the Linux Virtual Machine Scale Sets pricing page<\/a>.<\/p>\n If you no longer plan to use the Azure resources you’ve created, delete them. This prevents you from being charged for unutilized resources that are still running.<\/p>\n Since Model Builder is still in Preview, your feedback is super important in driving the direction of this tool and ML.NET in general. We would love to hear your feedback<\/a>!<\/p>\n If you run into any issues, please let us know by creating an issue in our GitHub repo<\/a>.<\/p>\n Join the community May 29-30 for a free online conference<\/a>.<\/p>\n The conference will have a full-day hands-on workshop, followed by another day of sessions.<\/p>\n Sign up to attend<\/a> or submit a talk<\/a>.<\/p>\n See you there!<\/p>\n Watch this video to see how quick it is to train and consume an image classification model to recognize clothing garments and accessories in a Blazor web application<\/a>.<\/p>\n Check out this end-to-end sample that uses satellite images to determine land use<\/a>.<\/p>\n Learn more about Model Builder from Microsoft Docs<\/a>.<\/p>\n![\"Model](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/choose-scenario-1024x563.png\")
<\/p>\nPrerequisites<\/h2>\n
\n
Model training workflow<\/h2>\n
\n
Add the data<\/h2>\n
\n
Create a .NET Core console application<\/h2>\n
![\"Create](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/create-dotnet-console-app-1024x705.png\")
<\/p>\nChoose a scenario<\/h2>\n
![\"Add](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/launch-model-builder-1024x563.png\")
<\/p>\n<\/p>\nConfigure your training environment<\/h2>\n
Create Azure Machine Learning experiment<\/h3>\n
![\"Choose](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/create-experiment-1024x563.png\")
<\/p>\nSet up Azure environment<\/h3>\n
![\"Choose](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/choose-subscription.png\")
<\/p>\n\n
![\"Create](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/create-new-workspace.png\")
<\/p>\n![\"Create](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/create-compute.png\")
<\/p>\n![\"Create](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/create-experiment-final.png\")
<\/p>\n![\"Submit](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/submit-experiment-1024x563.png\")
<\/p>\nLoad the data<\/h2>\n
![\"Load](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/load-data-1024x563.png\")
<\/p>\nTrain the model<\/h2>\n
![\"Track](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/track-training-progress-1024x563.png\")
<\/p>\n![\"Training](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/training-complete-1024x620.png\")
<\/p>\nEvaluate your model<\/h2>\n
![\"Evaluate](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/evaluate-model-1024x620.png\")
<\/p>\nAdd the code to make predictions<\/h2>\n
![\"Add](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/add-projects-1024x620.png\")
<\/p>\n\n
\n
Use the machine learning model<\/h2>\n
using System;\r\nusing MBImageClassificationAppML.Model;\r\n\r\nnamespace MBImageClassificationApp\r\n{\r\n class Program\r\n {\r\n static void Main(string[] args)\r\n {\r\n ModelInput image = new ModelInput\r\n {\r\n ImageSource = \"Pink_Tulips.jpg\"\r\n };\r\n\r\n var prediction = ConsumeModel.Predict(image);\r\n\r\n Console.WriteLine($\"Actual: tulips | Predicted: {prediction.Prediction}\");\r\n }\r\n }\r\n}<\/pre>\n![\"Console](\"https:\/\/devblogs.microsoft.com\/dotnet\/wp-content\/uploads\/sites\/10\/2020\/04\/console-output.png\")
<\/p>\nClean up resources<\/h2>\n
\n
Share your feedback<\/h2>\n
Virtual ML.NET Conference<\/h2>\n
Additional resources<\/h2>\n