FSharp Flower Image Classification Sample

samples/fsharp/getting-started/DeepLearning_ImageClassification_Training/.gitignore

@@ -0,0 +1,2 @@
+flower_photos_small_set
+*.zip

samples/fsharp/getting-started/DeepLearning_ImageClassification_Training/ImageClassification.Train.sln

@@ -0,0 +1,34 @@
+
+Microsoft Visual Studio Solution File, Format Version 12.00
+# Visual Studio 15
+VisualStudioVersion = 15.0.26124.0
+MinimumVisualStudioVersion = 15.0.26124.0
+Project("{F2A71F9B-5D33-465A-A702-920D77279786}") = "ImageClassification.Train", "ImageClassification.Train\ImageClassification.Train.fsproj", "{85F30786-C364-4498-8CB9-208575B9A1AC}"
+EndProject
+Global
+	GlobalSection(SolutionConfigurationPlatforms) = preSolution
+		Debug|Any CPU = Debug|Any CPU
+		Debug|x64 = Debug|x64
+		Debug|x86 = Debug|x86
+		Release|Any CPU = Release|Any CPU
+		Release|x64 = Release|x64
+		Release|x86 = Release|x86
+	EndGlobalSection
+	GlobalSection(SolutionProperties) = preSolution
+		HideSolutionNode = FALSE
+	EndGlobalSection
+	GlobalSection(ProjectConfigurationPlatforms) = postSolution
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Debug|Any CPU.Build.0 = Debug|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Debug|x64.ActiveCfg = Debug|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Debug|x64.Build.0 = Debug|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Debug|x86.ActiveCfg = Debug|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Debug|x86.Build.0 = Debug|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Release|Any CPU.ActiveCfg = Release|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Release|Any CPU.Build.0 = Release|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Release|x64.ActiveCfg = Release|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Release|x64.Build.0 = Release|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Release|x86.ActiveCfg = Release|Any CPU
+		{85F30786-C364-4498-8CB9-208575B9A1AC}.Release|x86.Build.0 = Release|Any CPU
+	EndGlobalSection
+EndGlobal

samples/fsharp/getting-started/DeepLearning_ImageClassification_Training/ImageClassification.Train/ImageClassification.Train.fsproj

@@ -0,0 +1,18 @@
+<Project Sdk="Microsoft.NET.Sdk">
+
+  <PropertyGroup>
+    <OutputType>Exe</OutputType>
+    <TargetFramework>netcoreapp3.1</TargetFramework>
+  </PropertyGroup>
+
+  <ItemGroup>
+    <Compile Include="Program.fs" />
+  </ItemGroup>
+
+  <ItemGroup>
+    <PackageReference Include="Microsoft.ML" Version="1.4.0" />
+    <PackageReference Include="Microsoft.ML.ImageAnalytics" Version="1.4.0" />
+    <PackageReference Include="Microsoft.ML.Vision" Version="1.4.0" />
+    <PackageReference Include="SciSharp.TensorFlow.Redist" Version="1.15.0" />
+  </ItemGroup>
+</Project>

samples/fsharp/getting-started/DeepLearning_ImageClassification_Training/ImageClassification.Train/Program.fs

@@ -0,0 +1,143 @@
+// Learn more about F# at http://fsharp.org
+
+open System
+open System.IO
+open System.IO.Compression
+open System.Net
+open Microsoft.ML
+open Microsoft.ML.Data
+open Microsoft.ML.Transforms
+open Microsoft.ML.Vision
+
+// Define input and output schema
+[<CLIMutable>]
+type ImageData = {
+    ImagePath:string
+    Label:string
+}
+
+[<CLIMutable>]
+type ImagePrediction = {
+    ImagePath:string
+    Label: string
+    PredictedLabel:string
+}
+
+// Helper functions
+let downloadZippedImageSetAsync (fileName:string) (downloadUrl:string) (imageDownloadFolder:string) = 
+    async {
+        let zippedPath = Path.Join(imageDownloadFolder, fileName)
+        let unzippedPath = Path.Join(imageDownloadFolder,Path.GetFileNameWithoutExtension(zippedPath));
+
+        if not (File.Exists zippedPath) then
+            let client = new WebClient()
+            client.DownloadFile(Uri(downloadUrl), zippedPath)
+
+        if not (Directory.Exists unzippedPath) then
+            ZipFile.ExtractToDirectory(zippedPath, unzippedPath)
+
+        return Path.Join(unzippedPath,Path.GetFileNameWithoutExtension(fileName))
+    }
+
+let loadImagesFromDirectory (path:string) (useDirectoryAsLabel:bool) = 
+
+    let files = Directory.GetFiles(path, "*",searchOption=SearchOption.AllDirectories)
+
+    files
+    |> Array.filter(fun file -> 
+        (Path.GetExtension(file) = ".jpg") ||
+        (Path.GetExtension(file) = ".png"))
+    |> Array.map(fun file -> 
+        let mutable label = Path.GetFileName(file)
+        if useDirectoryAsLabel then
+            label <-  Directory.GetParent(file).Name
+        else
+            let mutable brk = false
+            for index in 0..label.Length do
+                while not brk do
+                    if not (label.[index] |> Char.IsLetter) then
+                        label <- label.Substring(0,index)
+                        brk <- true
+
+        {ImagePath=file; Label=label}
+    )
+
+
+[<EntryPoint>]
+let main argv =
+
+    let fileName = "flower_photos_small_set.zip"
+    let downloadUrl = "https://mlnetfilestorage.file.core.windows.net/imagesets/flower_images/flower_photos_small_set.zip?st=2019-08-07T21%3A27%3A44Z&se=2030-08-08T21%3A27%3A00Z&sp=rl&sv=2018-03-28&sr=f&sig=SZ0UBX47pXD0F1rmrOM%2BfcwbPVob8hlgFtIlN89micM%3D"
+
+    // Download Data
+    let datasetPath = 
+        __SOURCE_DIRECTORY__ 
+        |> downloadZippedImageSetAsync fileName downloadUrl
+        |> Async.RunSynchronously 
+
+    // Initialize MLContext
+    let mlContext = MLContext()
+
+    // Get List of Images        
+    let images = loadImagesFromDirectory datasetPath true
+
+    // Load Data into IDataView and Shuffle
+    let data = 
+        images 
+        |> mlContext.Data.LoadFromEnumerable
+        |> mlContext.Data.ShuffleRows
+
+    // Define preprocessing pipeline
+    let preprocessingPipeline = 
+        EstimatorChain()
+            .Append(mlContext.Transforms.Conversion.MapValueToKey("LabelAsKey","Label"))
+            .Append(mlContext.Transforms.LoadRawImageBytes("Image", null, "ImagePath"))    
+
+    // Preprocess the data
+    let preprocessedData = 
+        let processingTransformer = data |> preprocessingPipeline.Fit 
+        data |> processingTransformer.Transform
+
+    // Split data into train,validation, test sets
+    let train, validation, test = 
+        preprocessedData
+        |> ( fun originalData -> 
+                let trainValSplit = mlContext.Data.TrainTestSplit(originalData, testFraction=0.7)
+                let testValSplit = mlContext.Data.TrainTestSplit(trainValSplit.TestSet)
+                (trainValSplit.TrainSet, testValSplit.TrainSet, testValSplit.TestSet))
+
+    // Define ImageClassificationTrainer Options
+    let classifierOptions = ImageClassificationTrainer.Options()
+    classifierOptions.FeatureColumnName <- "Image" 
+    classifierOptions.LabelColumnName <- "LabelAsKey"
+    classifierOptions.ValidationSet <- validation
+    classifierOptions.Arch <- ImageClassificationTrainer.Architecture.ResnetV2101
+    classifierOptions.MetricsCallback <- Action<ImageClassificationTrainer.ImageClassificationMetrics>(fun x -> printfn "%s" (x.ToString()))
+
+    // Define model training pipeline
+    let trainingPipeline = 
+        EstimatorChain()
+            .Append(mlContext.MulticlassClassification.Trainers.ImageClassification(classifierOptions))
+            .Append(mlContext.Transforms.Conversion.MapKeyToValue("PredictedLabel","LabelAsKey"))
+
+    // Train the model
+    let trainedModel = 
+        train 
+        |> trainingPipeline.Fit
+
+    // Get Prediction IDataView
+    let predictions = test |> trainedModel.Transform
+
+    // Evaluate the model
+    let metrics = mlContext.MulticlassClassification.Evaluate(predictions,labelColumnName="LabelAsKey")
+    printfn "MacroAccurracy: %f | LogLoss: %f" metrics.MacroAccuracy metrics.LogLoss
+
+    // Save Model
+    mlContext.Model.Save(trainedModel, preprocessedData.Schema, Path.Join(__SOURCE_DIRECTORY__,"model.zip"))
+
+    // Display 5 predictions
+    mlContext.Data.CreateEnumerable<ImagePrediction>(predictions, reuseRowObject=true)
+    |> Seq.take 5
+    |> Seq.iter(fun prediction -> printfn "Original: %s | Predicted: %s" prediction.Label prediction.PredictedLabel) 
+
+    0 // return an integer exit code