Object Detection pass.

src/TensorFlowNET.Core/Sessions/BaseSession.cs

@@ -54,19 +54,19 @@ namespace Tensorflow
             status.Check(true);
         }
 
-        public virtual NDArray run(object fetches, params FeedItem[] feed_dict)
+        public virtual NDArray[] run(object fetches, params FeedItem[] feed_dict)
         {
             return _run(fetches, feed_dict);
         }
 
-        public virtual NDArray run(object fetches, Hashtable feed_dict = null)
+        public virtual NDArray[] run(object fetches, Hashtable feed_dict = null)
         {
             var feed_items = feed_dict == null ? new FeedItem[0] :
                 feed_dict.Keys.OfType<object>().Select(key => new FeedItem(key, feed_dict[key])).ToArray();
             return _run(fetches, feed_items);
         }
 
-        private NDArray _run(object fetches, FeedItem[] feed_dict = null)
+        private NDArray[] _run(object fetches, FeedItem[] feed_dict = null)
         {
             var feed_dict_tensor = new Dictionary<object, object>();
             var feed_map = new Dictionary<object, object>();

src/TensorFlowNET.Core/Sessions/_ElementFetchMapper.cs

@@ -17,6 +17,7 @@
 using NumSharp;
 using System;
 using System.Collections.Generic;
+using System.Numerics;
 
 namespace Tensorflow
 {
@@ -45,9 +46,9 @@ namespace Tensorflow
         /// </summary>
         /// <param name="values"></param>
         /// <returns></returns>
-        public override NDArray build_results(List<NDArray> values)
+        public override NDArray[] build_results(List<NDArray> values)
         {
-            NDArray result = null;
+            NDArray[] result = null;
 
             if (values.Count > 0)
             {
@@ -55,26 +56,53 @@ namespace Tensorflow
                 switch (ret)
                 {
                     case NDArray value:
-                        result = value;
+                        result = new[] { value };
+                        break;
+#if _REGEN
+        %types=["sbyte", "bool", "byte", "short", "ushort", "int", "uint", "long", "ulong", "float", "double", "Complex"]
+        %foreach types%
+                    case #1 value:
+                        result = new[] { NDArray.Scalar(value) };
+                        break;
+         %
+#else
+                    case sbyte value:
+                        result = new[] { NDArray.Scalar(value) };
+                        break;
+                    case bool value:
+                        result = new[] { NDArray.Scalar(value) };
+                        break;
+                    case byte value:
+                        result = new[] { NDArray.Scalar(value) };
                         break;
                     case short value:
-                        result = value;
+                        result = new[] { NDArray.Scalar(value) };
+                        break;
+                    case ushort value:
+                        result = new[] { NDArray.Scalar(value) };
                         break;
                     case int value:
-                        result = value;
+                        result = new[] { NDArray.Scalar(value) };
+                        break;
+                    case uint value:
+                        result = new[] { NDArray.Scalar(value) };
                         break;
                     case long value:
-                        result = value;
+                        result = new[] { NDArray.Scalar(value) };
+                        break;
+                    case ulong value:
+                        result = new[] { NDArray.Scalar(value) };
                         break;
                     case float value:
-                        result = value;
+                        result = new[] { NDArray.Scalar(value) };
                         break;
                     case double value:
-                        result = value;
+                        result = new[] { NDArray.Scalar(value) };
                         break;
-                    case string value:
-                        result = value;
+                    case Complex value:
+                        result = new[] { NDArray.Scalar(value) };
                         break;
+#endif
                     default:
                         break;
                 }

src/TensorFlowNET.Core/Sessions/_FetchHandler.cs

@@ -57,7 +57,7 @@ namespace Tensorflow
             _final_fetches = _fetches;
         }
 
-        public NDArray build_results(BaseSession session, NDArray[] tensor_values)
+        public NDArray[] build_results(BaseSession session, NDArray[] tensor_values)
         {
             var full_values = new List<NDArray>();
             if (_final_fetches.Count != tensor_values.Length)
@@ -123,8 +123,7 @@ namespace Tensorflow
                     }
                     else
                     {
-                        //full_values.Add(value.GetNDArrays());
-                        full_values.Add(value);
+                        full_values.Add(value[np.arange(0, value.shape[0])]);
                     }
                 }
                 i += 1;

src/TensorFlowNET.Core/Sessions/_FetchMapper.cs

@@ -37,23 +37,9 @@ namespace Tensorflow
                 return new _ElementFetchMapper(fetches, (List<NDArray> fetched_vals) => fetched_vals[0]);
         }
 
-        public virtual NDArray build_results(List<NDArray> values)
+        public virtual NDArray[] build_results(List<NDArray> values)
         {
-            // if they're all scalar value
-            bool isAllScalars = values.Count(x => x.ndim == 0) == values.Count;
-            if (isAllScalars)
-            {
-                var type = values[0].dtype;
-                switch(Type.GetTypeCode(type))
-                {
-                    case TypeCode.Single:
-                        return np.array(values.Select(x => (float)x).ToArray());
-                    default:
-                        throw new NotImplementedException("build_results");
-                }
-            }
-
-            return np.stack(values.ToArray());
+            return values.ToArray();
         }
 
         public virtual List<ITensorOrOperation> unique_fetches()

src/TensorFlowNET.Core/TensorFlowNET.Core.csproj

@@ -37,7 +37,7 @@ Docs: https://tensorflownet.readthedocs.io</Description>
 15. Fix Tensor memory leak.
 16. Rename with to tf_with that is only used to build graph purpose.
 17. Graph inherit from DisposableObject.</PackageReleaseNotes>
-    <LangVersion>7.2</LangVersion>
+    <LangVersion>7.3</LangVersion>
     <FileVersion>0.10.11.0</FileVersion>
     <PackageLicenseFile>LICENSE</PackageLicenseFile>
     <PackageRequireLicenseAcceptance>true</PackageRequireLicenseAcceptance>

src/TensorFlowNET.Core/Tensors/Tensor.Creation.cs

@@ -531,8 +531,8 @@ namespace Tensorflow
             var handle = TF_NewTensor(dtype, dims: nd.shape.Select(i=>(long)i).ToArray(), num_dims: nd.ndim, data: ptr, len: (UIntPtr)num_bytes, deallocator: _nothingDeallocator, ref _deallocatorArgs);
             IsMemoryOwner = false;
             return handle;
-        }
-
+        }
+
         public unsafe Tensor(byte[][] buffer, long[] shape)
         {
             int size = 0;

src/TensorFlowNET.Core/Tensors/Tensor.cs

@@ -394,9 +394,8 @@ namespace Tensorflow
 
         protected override void DisposeManagedState()
         {
-            if (gcHandle.IsAllocated)
-                gcHandle.Free();
         }
+
         protected override void DisposeUnManagedState(IntPtr handle)
         {
             if(handle != IntPtr.Zero)

src/TensorFlowNET.Core/Tensors/TensorShape.cs

@@ -71,12 +71,12 @@ namespace Tensorflow
                 return new TensorShape();
             else
             {
-                var concatenate_dims = new int[NDim + other.NDim];
-                for (int i = 0; i < NDim; i++)
+                var concatenate_dims = new int[ndim + other.ndim];
+                for (int i = 0; i < ndim; i++)
                     concatenate_dims[i] = dims[i];
 
-                for (int i = 0; i < other.NDim; i++)
-                    concatenate_dims[NDim + i] = other.dims[i];
+                for (int i = 0; i < other.ndim; i++)
+                    concatenate_dims[ndim + i] = other.dims[i];
 
                 return new TensorShape(concatenate_dims);
             }

src/TensorFlowNET.Core/Tensors/tensor_util.cs

@@ -212,7 +212,7 @@ namespace Tensorflow
                             if (values.GetType().IsArray)
                                 nparray = np.array((string[])values, np_dt);
                             else
-                                nparray = Convert.ToString(values);
+                                nparray = NDArray.FromString(Convert.ToString(values));
                             break;
                         case "Boolean":
                             if (values.GetType().IsArray)
@@ -267,7 +267,10 @@ namespace Tensorflow
             if (numpy_dtype == TF_DataType.TF_STRING && !(values is NDArray))
             {
                 if (values is string str)
+                {
                     tensor_proto.StringVal.Add(Google.Protobuf.ByteString.CopyFromUtf8(str));
+                    tensor_proto.TensorShape = tensor_util.as_shape(new int[0]);
+                }
                 else if (values is string[] str_values)
                     tensor_proto.StringVal.AddRange(str_values.Select(x => Google.Protobuf.ByteString.CopyFromUtf8(x)));
                 else if(values is byte[] byte_values)

src/TensorFlowNET.Core/ops.py.cs

@@ -377,7 +377,7 @@ namespace Tensorflow
                            "`eval(session=sess)`.");
             }
 
-            return session.run(tensor, feed_dict);
+            return session.run(tensor, feed_dict)[0];
         }
 
         /// <summary>

test/TensorFlowNET.Examples/BasicModels/LinearRegression.cs

@@ -91,16 +91,16 @@ namespace TensorFlowNET.Examples
                     {
                         var c = sess.run(cost, 
                             new FeedItem(X, train_X),
-                            new FeedItem(Y, train_Y));
-                        Console.WriteLine($"Epoch: {epoch + 1} cost={c} " + $"W={sess.run(W)} b={sess.run(b)}");
+                            new FeedItem(Y, train_Y))[0];
+                        Console.WriteLine($"Epoch: {epoch + 1} cost={c} " + $"W={sess.run(W)[0]} b={sess.run(b)[0]}");
                     }
                 }
 
                 Console.WriteLine("Optimization Finished!");
                 var training_cost = sess.run(cost,
                     new FeedItem(X, train_X),
-                    new FeedItem(Y, train_Y));
-                Console.WriteLine($"Training cost={training_cost} W={sess.run(W)} b={sess.run(b)}");
+                    new FeedItem(Y, train_Y))[0];
+                Console.WriteLine($"Training cost={training_cost} W={sess.run(W)[0]} b={sess.run(b)[0]}");
 
                 // Testing example
                 var test_X = np.array(6.83f, 4.668f, 8.9f, 7.91f, 5.7f, 8.7f, 3.1f, 2.1f);
@@ -108,7 +108,7 @@ namespace TensorFlowNET.Examples
                 Console.WriteLine("Testing... (Mean square loss Comparison)");
                 var testing_cost = sess.run(tf.reduce_sum(tf.pow(pred - Y, 2.0f)) / (2.0f * test_X.shape[0]),
                     new FeedItem(X, test_X), 
-                    new FeedItem(Y, test_Y));
+                    new FeedItem(Y, test_Y))[0];
                 Console.WriteLine($"Testing cost={testing_cost}");
                 var diff = Math.Abs((float)training_cost - (float)testing_cost);
                 Console.WriteLine($"Absolute mean square loss difference: {diff}");

test/TensorFlowNET.Examples/BasicModels/LogisticRegression.cs

@@ -162,7 +162,7 @@ namespace TensorFlowNET.Examples
             var (batch_xs, batch_ys) = mnist.Train.GetNextBatch(10);
             var results = sess.run(output, new FeedItem(input, batch_xs[np.arange(1)]));
 
-            if (results.argmax() == (batch_ys[0] as NDArray).argmax())
+            if (results[0].argmax() == (batch_ys[0] as NDArray).argmax())
                 print("predicted OK!");
             else
                 throw new ValueError("predict error, should be 90% accuracy");

test/TensorFlowNET.Examples/BasicModels/NearestNeighbor.cs

@@ -64,7 +64,7 @@ namespace TensorFlowNET.Examples
                 foreach(int i in range(Xte.shape[0]))
                 {
                     // Get nearest neighbor
-                    long nn_index = sess.run(pred, new FeedItem(xtr, Xtr), new FeedItem(xte, Xte[i]));
+                    long nn_index = sess.run(pred, new FeedItem(xtr, Xtr), new FeedItem(xte, Xte[i]))[0];
                     // Get nearest neighbor class label and compare it to its true label
                     int index = (int)nn_index;
 

test/TensorFlowNET.Examples/ImageProcessing/ImageRecognitionInception.cs

@@ -51,7 +51,7 @@ namespace TensorFlowNET.Examples
                 {
                     sw.Restart();
 
-                    var results = sess.run(output_operation.outputs[0], new FeedItem(input_operation.outputs[0], nd));
+                    var results = sess.run(output_operation.outputs[0], new FeedItem(input_operation.outputs[0], nd))[0];
                     results = np.squeeze(results);
                     int idx = np.argmax(results);
 

test/TensorFlowNET.Examples/ImageProcessing/ObjectDetection.cs

@@ -78,9 +78,7 @@ namespace TensorFlowNET.Examples
 
             var results = sess.run(outTensorArr, new FeedItem(imgTensor, imgArr));
 
-            NDArray[] resultArr = results.GetNDArrays();
-
-            buildOutputImage(resultArr);
+            buildOutputImage(results);
         }
 
         public void PrepareData()
@@ -110,7 +108,7 @@ namespace TensorFlowNET.Examples
             var dims_expander = tf.expand_dims(casted, 0);
 
             using (var sess = tf.Session(graph))
-                return sess.run(dims_expander);
+                return sess.run(dims_expander)[0];
         }
 
         private void buildOutputImage(NDArray[] resultArr)
@@ -122,12 +120,13 @@ namespace TensorFlowNET.Examples
             Bitmap bitmap = new Bitmap(Path.Join(imageDir, "input.jpg"));
 
             var scores = resultArr[2].AsIterator<float>();
+            var boxes = resultArr[1].GetData<float>();
+            var id = np.squeeze(resultArr[3]).GetData<float>();
             for (int i=0; i< scores.size; i++)
             {
                 float score = scores.MoveNext();
                 if (score > MIN_SCORE)
                 {
-                    float[] boxes = resultArr[1].GetData<float>().ToArray();
                     float top = boxes[i * 4] * bitmap.Height;
                     float left = boxes[i * 4 + 1] * bitmap.Width;
                     float bottom = boxes[i * 4 + 2] * bitmap.Height;
@@ -141,8 +140,7 @@ namespace TensorFlowNET.Examples
                         Height = (int)(bottom - top)
                     };
 
-                    var id = (int)resultArr[3].GetValue(i);
-                    string name = pbTxtItems.items.Where(w => w.id == id).Select(s=>s.display_name).FirstOrDefault();
+                    string name = pbTxtItems.items.Where(w => w.id == id[i]).Select(s=>s.display_name).FirstOrDefault();
 
                     drawObjectOnBitmap(bitmap, rect, score, name);
                 }

test/TensorFlowNET.Examples/ImageProcessing/RetrainImageClassifier.cs

@@ -455,7 +455,7 @@ namespace TensorFlowNET.Examples
             // First decode the JPEG image, resize it, and rescale the pixel values.
             var resized_input_values = sess.run(decoded_image_tensor, new FeedItem(image_data_tensor, new Tensor(image_data, TF_DataType.TF_STRING)));
             // Then run it through the recognition network.
-            var bottleneck_values = sess.run(bottleneck_tensor, new FeedItem(resized_input_tensor, resized_input_values));
+            var bottleneck_values = sess.run(bottleneck_tensor, new FeedItem(resized_input_tensor, resized_input_values))[0];
             bottleneck_values = np.squeeze(bottleneck_values);
             return bottleneck_values;
         }

test/TensorFlowNET.Examples/TextProcessing/CnnTextClassification.cs

@@ -244,7 +244,7 @@ namespace TensorFlowNET.Examples
                             [is_training] = false
                         };
                         var result1 = sess.run(accuracy, valid_feed_dict);
-                        float accuracy_value = result1;
+                        float accuracy_value = result1[0];
                         sum_accuracy += accuracy_value;
                         cnt += 1;
                     }

test/TensorFlowNET.UnitTest/ConstantTest.cs

@@ -98,9 +98,9 @@ namespace TensorFlowNET.UnitTest
             {
                 var result = sess.run(tensor);
 
-                Assert.AreEqual(result.shape[0], 3);
-                Assert.AreEqual(result.shape[1], 2);
-                Assert.IsTrue(Enumerable.SequenceEqual(new int[] { 0, 0, 0, 0, 0, 0 }, result.Data<int>()));
+                Assert.AreEqual(result[0].shape[0], 3);
+                Assert.AreEqual(result[0].shape[1], 2);
+                Assert.IsTrue(Enumerable.SequenceEqual(new int[] { 0, 0, 0, 0, 0, 0 }, result[0].Data<int>()));
             }
 
             // big size
@@ -109,13 +109,13 @@ namespace TensorFlowNET.UnitTest
             {
                 var result = sess.run(tensor);
 
-                Assert.AreEqual(result.shape[0], 200);
-                Assert.AreEqual(result.shape[1], 100);
+                Assert.AreEqual(result[0].shape[0], 200);
+                Assert.AreEqual(result[0].shape[1], 100);
 
-                var data = result.Data<int>();
+                var data = result[0].Data<int>();
                 Assert.AreEqual(0, data[0]);
                 Assert.AreEqual(0, data[500]);
-                Assert.AreEqual(0, data[result.size - 1]);
+                Assert.AreEqual(0, data[result[0].size - 1]);
             }
         }
 
@@ -127,9 +127,9 @@ namespace TensorFlowNET.UnitTest
             {
                 var result = sess.run(ones);
 
-                Assert.AreEqual(result.shape[0], 3);
-                Assert.AreEqual(result.shape[1], 2);
-                Assert.IsTrue(new[] { 1, 1, 1, 1, 1, 1 }.SequenceEqual(result.Data<int>()));
+                Assert.AreEqual(result[0].shape[0], 3);
+                Assert.AreEqual(result[0].shape[1], 2);
+                Assert.IsTrue(new[] { 1, 1, 1, 1, 1, 1 }.SequenceEqual(result[0].Data<int>()));
             }
         }
 
@@ -142,9 +142,9 @@ namespace TensorFlowNET.UnitTest
             {
                 var result = sess.run(halfes);
 
-                Assert.AreEqual(result.shape[0], 3);
-                Assert.AreEqual(result.shape[1], 2);
-                Assert.IsTrue(new[] { .5, .5, .5, .5, .5, .5 }.SequenceEqual(result.Data<double>()));
+                Assert.AreEqual(result[0].shape[0], 3);
+                Assert.AreEqual(result[0].shape[1], 2);
+                Assert.IsTrue(new[] { .5, .5, .5, .5, .5, .5 }.SequenceEqual(result[0].Data<double>()));
             }
         }
 
@@ -161,10 +161,10 @@ namespace TensorFlowNET.UnitTest
             using (var sess = tf.Session())
             {
                 var result = sess.run(tensor);
-                var data = result.Data<int>();
+                var data = result[0].Data<int>();
 
-                Assert.AreEqual(result.shape[0], 2);
-                Assert.AreEqual(result.shape[1], 3);
+                Assert.AreEqual(result[0].shape[0], 2);
+                Assert.AreEqual(result[0].shape[1], 3);
                 Assert.IsTrue(Enumerable.SequenceEqual(new int[] { 3, 1, 1, 2, 1, 3 }, data));
             }
         }
@@ -177,7 +177,7 @@ namespace TensorFlowNET.UnitTest
             var c = a * b;
 
             var sess = tf.Session();
-            double result = sess.run(c);
+            double result = sess.run(c)[0];
             sess.close();
 
             Assert.AreEqual(6.0, result);

test/TensorFlowNET.UnitTest/GradientTest.cs

@@ -41,7 +41,7 @@ namespace TensorFlowNET.UnitTest
                 var grad = tf.gradients(y, x);
                 Assert.AreEqual(grad[0].name, "gradients/AddN:0");
 
-                float r = sess.run(grad[0]);
+                float r = sess.run(grad[0])[0];
                 Assert.AreEqual(r, 1.4f);
             }
         }
@@ -57,7 +57,7 @@ namespace TensorFlowNET.UnitTest
                 var grad = tf.gradients(y, x);
                 Assert.AreEqual(grad[0].name, "gradients/AddN:0");
 
-                float r = sess.run(grad[0]);
+                float r = sess.run(grad[0])[0];
                 Assert.AreEqual(r, 14.700001f);
             });
         }
@@ -94,7 +94,7 @@ namespace TensorFlowNET.UnitTest
 
             using (var sess = tf.Session(graph))
             {
-                var r = sess.run(slice);
+                var r = sess.run(slice)[0];
 
                 Assert.IsTrue(Enumerable.SequenceEqual(r.shape, new[] { 2, 1, 2 }));
                 Assert.IsTrue(Enumerable.SequenceEqual(r[0].GetData<int>(), new[] { 11, 13 }));

test/TensorFlowNET.UnitTest/PlaceholderTest.cs

@@ -17,7 +17,7 @@ namespace TensorFlowNET.UnitTest
             {
                 var result = sess.run(y, 
                     new FeedItem(x, 2));
-                Assert.AreEqual((int)result, 6);
+                Assert.AreEqual((int)result[0], 6);
             }
         }
     }

test/TensorFlowNET.UnitTest/VariableTest.cs

@@ -16,7 +16,7 @@ namespace TensorFlowNET.UnitTest
             {
                 session.run(x.initializer);
                 var result = session.run(x);
-                Assert.AreEqual(10, (int)result);
+                Assert.AreEqual(10, (int)result[0]);
             }
         }
 
@@ -81,7 +81,7 @@ namespace TensorFlowNET.UnitTest
             using (var session = tf.Session())
             {
                 session.run(model);
-                int result = session.run(y);
+                int result = session.run(y)[0];
                 Assert.AreEqual(result, 4);
             }
         }
@@ -98,11 +98,11 @@ namespace TensorFlowNET.UnitTest
             sess.run(init);
 
             var result = sess.run(variable);
-            Assert.IsTrue((int)result == 31);
+            Assert.IsTrue((int)result[0] == 31);
 
             var assign = variable.assign(12);
             result = sess.run(assign);
-            Assert.IsTrue((int)result == 12);
+            Assert.IsTrue((int)result[0] == 12);
         }
 
         [TestMethod]
@@ -139,7 +139,7 @@ namespace TensorFlowNET.UnitTest
                 for(int i = 0; i < 5; i++)
                 {
                     x = x + 1;
-                    result = session.run(x);
+                    result = session.run(x)[0];
                     print(result);
                 }
             }