Add Initialize() method to TensorFlow and ImageAnalytics

src/Microsoft.ML.ImageAnalytics/EntryPoints/ImageAnalytics.cs

@@ -11,6 +11,16 @@ namespace Microsoft.ML.Runtime.ImageAnalytics.EntryPoints
 {
     public static class ImageAnalytics
     {
+        // This method is needed for the Pipeline API, since ModuleCatalog does not load entry points that are located
+        // in assemblies that aren't directly used in the code. Users who want to use ImageAnalytics components will have to call
+        // ImageAnalytics.Initialize() before creating the pipeline.
+        /// <summary>
+        /// Initialize the Image Analytics environment. Call this method before adding Image components to a learning pipeline.
+        /// </summary>
+        public static void Initialize()
+        {
+        }
+
         [TlcModule.EntryPoint(Name = "Transforms.ImageLoader", Desc = ImageLoaderTransform.Summary,
             UserName = ImageLoaderTransform.UserName, ShortName = ImageLoaderTransform.LoaderSignature)]
         public static CommonOutputs.TransformOutput ImageLoader(IHostEnvironment env, ImageLoaderTransform.Arguments input)

src/Microsoft.ML.ImageAnalytics/ImageResizerTransform.cs

@@ -17,10 +17,10 @@
 using Microsoft.ML.Runtime.Internal.Utilities;
 using Microsoft.ML.Runtime.Model;
 
-[assembly: LoadableClass(ImageResizerTransform.Summary, typeof(ImageResizerTransform), typeof(ImageResizerTransform.Arguments),
+[assembly: LoadableClass(ImageResizerTransform.Summary, typeof(IDataTransform), typeof(ImageResizerTransform), typeof(ImageResizerTransform.Arguments),
     typeof(SignatureDataTransform), ImageResizerTransform.UserName, "ImageResizerTransform", "ImageResizer")]
 
-[assembly: LoadableClass(ImageResizerTransform.Summary, typeof(ImageResizerTransform), null, typeof(SignatureLoadDataTransform),
+[assembly: LoadableClass(ImageResizerTransform.Summary, typeof(IDataTransform), typeof(ImageResizerTransform), null, typeof(SignatureLoadDataTransform),
     ImageResizerTransform.UserName, ImageResizerTransform.LoaderSignature)]
 
 [assembly: LoadableClass(typeof(ImageResizerTransform), null, typeof(SignatureLoadModel),

src/Microsoft.ML.TensorFlow/Microsoft.ML.TensorFlow.csproj

@@ -10,6 +10,7 @@
   <ItemGroup>
     <ProjectReference Include="..\Microsoft.ML.Core\Microsoft.ML.Core.csproj" />
     <ProjectReference Include="..\Microsoft.ML.Data\Microsoft.ML.Data.csproj" />
+    <ProjectReference Include="..\Microsoft.ML.ImageAnalytics\Microsoft.ML.ImageAnalytics.csproj" />
   </ItemGroup>
 
   <ItemGroup>

src/Microsoft.ML.TensorFlow/TensorFlow/TensorflowUtils.cs

@@ -5,11 +5,23 @@
 using System;
 using System.Runtime.InteropServices;
 using Microsoft.ML.Runtime.Data;
+using Microsoft.ML.Runtime.ImageAnalytics.EntryPoints;
 
 namespace Microsoft.ML.Transforms.TensorFlow
 {
-    internal partial class TensorFlowUtils
+    public static class TensorFlowUtils
     {
+        // This method is needed for the Pipeline API, since ModuleCatalog does not load entry points that are located
+        // in assemblies that aren't directly used in the code. Users who want to use TensorFlow components will have to call
+        // TensorFlowUtils.Initialize() before creating the pipeline.
+        /// <summary>
+        /// Initialize the TensorFlow environment. Call this method before adding TensorFlow components to a learning pipeline.
+        /// </summary>
+        public static void Initialize()
+        {
+            ImageAnalytics.Initialize();
+        }
+
         internal static PrimitiveType Tf2MlNetType(TFDataType type)
         {
             switch (type)
@@ -27,7 +39,7 @@ internal static PrimitiveType Tf2MlNetType(TFDataType type)
             }
         }
 
-        public static unsafe void FetchData<T>(IntPtr data, T[] result)
+        internal static unsafe void FetchData<T>(IntPtr data, T[] result)
         {
             var size = result.Length;
 

test/Microsoft.ML.Tests/Scenarios/TensorflowTests.cs

@@ -9,6 +9,7 @@
 using Microsoft.ML.Runtime.LightGBM;
 using Microsoft.ML.Trainers;
 using Microsoft.ML.Transforms;
+using Microsoft.ML.Transforms.TensorFlow;
 using System.Collections.Generic;
 using System.IO;
 using Xunit;
@@ -57,6 +58,7 @@ public void TensorFlowTransformCifarLearningPipelineTest()
             pipeline.Add(new TextToKeyConverter("Label"));
             pipeline.Add(new StochasticDualCoordinateAscentClassifier());
 
+            TensorFlowUtils.Initialize();
             var model = pipeline.Train<CifarData, CifarPrediction>();
             string[] scoreLabels;
             model.TryGetScoreLabelNames(out scoreLabels);