Merge pull request #114 from sumitdvlp/AvgPool

avg pool 2d

src/Native/LibTorchSharp/THSNN.cpp

@@ -148,6 +148,18 @@ Tensor THSNN_adaptiveAvgPool2DApply(const Tensor tensor, const int length, const
     return new torch::Tensor(torch::adaptive_avg_pool2d(*tensor, at::IntList(outputSize, length)));
 }
 
+Tensor THSNN_avgPool2DApply(const Tensor tensor,
+	const int kernelSizeLength,
+	const int64_t* kernelSize,
+	const int strideLength,
+	const int64_t* stride) 
+{
+	return new torch::Tensor(torch::avg_pool2d(
+		*tensor,
+		at::IntList(kernelSize, kernelSizeLength),
+		at::IntList(stride, strideLength)));
+}
+
 Tensor THSNN_logSoftMaxApply(const Tensor tensor, const int64_t dimension)
 {
     return new torch::Tensor(torch::log_softmax(*tensor, dimension));

src/Native/LibTorchSharp/THSNN.h

@@ -71,6 +71,14 @@ EXPORT_API(Tensor) THSNN_maxPool2DApply(
 // Applies a 2D adaptive average pooling over an input signal composed of several input planes.
 EXPORT_API(Tensor) THSNN_adaptiveAvgPool2DApply(const Tensor tensor, const int length, const int64_t* outputSize);
 
+// Applies a avgpool 2d on the input tensor. 
+EXPORT_API(Tensor) THSNN_avgPool2DApply(
+	const Tensor tensor,
+	const int kernelSizeLength,
+	const int64_t* kernelSize,
+	const int strideLength,
+	const int64_t* stride);
+
 // Applies a log soft max on the input tensor. 
 EXPORT_API(Tensor) THSNN_logSoftMaxApply(const Tensor tensor, const int64_t dimension);
 

src/TorchSharp/NN/AvgPool2D.cs

@@ -0,0 +1,26 @@
+using System;
+using System.Runtime.InteropServices;
+using TorchSharp.Tensor;
+
+namespace TorchSharp.NN
+{
+    public class AvgPool2D : FunctionalModule<AdaptiveAvgPool2D>
+    {
+        private readonly long[] _kernelSize;
+        private readonly long[] _stride;
+
+        internal AvgPool2D(long[] kernelSize, long[] stride) : base()
+        {
+            _kernelSize = kernelSize;
+            _stride = stride ?? new long[0];
+        }
+
+        [DllImport("LibTorchSharp")]
+        private static extern IntPtr THSNN_avgPool2DApply(IntPtr tensor, int kernelSizeLength, long[] kernelSize, int strideLength, long[] stride);
+
+        public override TorchTensor Forward(TorchTensor tensor)
+        {
+            return new TorchTensor(THSNN_avgPool2DApply(tensor.Handle, _kernelSize.Length, _kernelSize, _stride.Length, _stride));
+        }
+    }
+}

src/TorchSharp/NN/Module.cs

@@ -155,6 +155,14 @@ static public AdaptiveAvgPool2D AdaptiveAvgPool2D(params long[] outputSize)
             return new AdaptiveAvgPool2D(outputSize);
         }
 
+        static public TorchTensor AvgPool2D(TorchTensor x, long[] kernelSize, long[] stride = null)
+        {
+            using (var m = new AvgPool2D(kernelSize, stride))
+            {
+                return m.Forward(x);
+            }
+        }
+
         static public TorchTensor AdaptiveAvgPool2D(TorchTensor x, params long[] outputSize)
         {
             using (var a = new AdaptiveAvgPool2D(outputSize))

test/TorchSharpTest/TorchSharp/TorchSharp.cs

@@ -3,6 +3,7 @@
 using System.Linq;
 using System.Runtime.InteropServices;
 using TorchSharp.JIT;
+using TorchSharp.NN;
 using TorchSharp.Tensor;
 using Xunit;
 
@@ -1226,5 +1227,21 @@ public void TestMNISTLoaderWithEpochs()
                 Assert.Equal(size * epochs, i * 32);
             }
         }
+
+        [Fact]
+        public void AvgPool2DObjectInitialized()
+        {
+            TorchTensor ones = FloatTensor.Ones(new long[] { 2, 2, 2 });
+            var obj = NN.Module.AvgPool2D(ones, new long[] { 2 }, new long[] { 2 });
+            Assert.Equal(typeof(TorchTensor), obj.GetType());
+        }
+
+        [Fact]
+        public void MaxPool2DObjectInitialized()
+        {
+            TorchTensor ones = FloatTensor.Ones(new long[] { 2, 2, 2 });
+            var obj = NN.Module.MaxPool2D(ones, new long[] { 2 }, new long[] { 2 });
+            Assert.Equal(typeof(TorchTensor), obj.GetType());
+        }
     }
-}
+}

test/TorchSharpTest/TorchSharpTest.csproj

@@ -47,7 +47,7 @@
   </ItemGroup>
 
   <ItemGroup Condition="'$(NativeTargetArchitecture)' == 'x64' and $([MSBuild]::IsOSPlatform('linux'))">
-    <NativeAssemblyReference Include="torch" ExtraExtension=".1"/>
+    <NativeAssemblyReference Include="torch" ExtraExtension=".1" />
     <NativeAssemblyReference Include="gomp-8bba0e50" ExtraExtension=".1" />
   </ItemGroup>