Debugged the batch normalization, inference produces actual results now

MIPT-Oulu · Aug 31, 2018 · 96cf1da · 96cf1da
1 parent 5eb93de
commit 96cf1da
Show file tree

Hide file tree

Showing 2 changed files with 30 additions and 15 deletions.
diff --git a/CNTKUNet/CNTKUNet/Models/UNet.cs b/CNTKUNet/CNTKUNet/Models/UNet.cs
@@ -231,8 +231,8 @@ private static void center(Variable features, int[] ks, int in_channels, int out
             }
 
             //Convolution blocks
-            var block1 = ConvBlock(features, ks, in_channels, out_channels, null, null, wpath, name1, true, true, false);
-            var block2 = ConvBlock(block1.Output, ks, out_channels, out_channels, null, null, wpath, name2, true, true, false);
+            var block1 = ConvBlock(features, ks, in_channels, out_channels, null, null, wpath, name1, true, true, true);
+            var block2 = ConvBlock(block1.Output, ks, out_channels, out_channels, null, null, wpath, name2, true, true, true);
 
             center = block2.Output;
         }
@@ -342,19 +342,19 @@ private static void make_bn_pars(int out_channels, NDShape shape, out Parameter
                 string vn = "bn" + S[0] + "_" + S[1] + "_" + "running_var";
                 //Initialize new weight
                 w = new Parameter(new int[] { out_channels}, DataType.Float, CNTKLib.GlorotUniformInitializer(), DeviceDescriptor.CPUDevice);
-                float[] W = make_copies(weight_fromDisk(wpath, wn),shape);
+                float[] W = make_copies(weight_fromDisk(wpath, wn),shape,true);
                 w = weight_fromFloat(w, W, new int[] { out_channels });
                 //Initialize new bias
                 b = new Parameter(new int[] { out_channels }, DataType.Float, CNTKLib.GlorotUniformInitializer(), DeviceDescriptor.CPUDevice);
-                W = make_copies(weight_fromDisk(wpath, bn), shape);
+                W = make_copies(weight_fromDisk(wpath, bn), shape, true);
                 b = weight_fromFloat(b, W, new int[] { out_channels });
                 //Initialize new running mean
                 m = new Parameter(new int[] { out_channels }, DataType.Float, CNTKLib.GlorotUniformInitializer(), DeviceDescriptor.CPUDevice);
-                W = make_copies(weight_fromDisk(wpath, mn), shape);
+                W = make_copies(weight_fromDisk(wpath, mn), shape, true);
                 m = weight_fromFloat(m, W, new int[] { out_channels });
                 //Initialize new variance
                 v = new Parameter(new int[] { out_channels }, DataType.Float, CNTKLib.GlorotUniformInitializer(), DeviceDescriptor.CPUDevice);
-                W = make_copies(weight_fromDisk(wpath, vn), shape);
+                W = make_copies(weight_fromDisk(wpath, vn), shape, true);
                 v = weight_fromFloat(v, W, new int[] { out_channels });
             }
             else
@@ -400,20 +400,34 @@ private static float[] from_kernel(float[] kernel, int dim)
         }
 
         //Copy bias to match the dimensions of the input
-        private static float[] make_copies(float[] kernel, NDShape dims)
+        private static float[] make_copies(float[] kernel, NDShape dims, bool swap_axes=false)
         {
             int K = dims[0] * dims[1];
-            float[] outarray = new float[dims[0]* dims[1]* dims[2]];
-            //Loop over number of maps
-            for(int k=0; k<dims[2]; k++)
+            float[] outarray = new float[dims[0] * dims[1] * dims[2]];
+            if (swap_axes == false)
             {
-                //Loop over spatial dimensions
-                for(int kk = 0; kk<K; kk++)
+                //Loop over number of maps
+                for (int k = 0; k < dims[2]; k++)
                 {
-                    outarray[k*K + kk] = kernel[k];
+                    //Loop over spatial dimensions
+                    for (int kk = 0; kk < K; kk++)
+                    {
+                        outarray[k * K + kk] = kernel[k];
+                    }
+                }
+            }
+            else
+            {
+                //Loop over number of maps
+                for (int k = 0; k < K; k++)
+                {
+                    //Loop over spatial dimensions
+                    for (int kk = 0; kk < dims[2]; kk++)
+                    {
+                        outarray[k*dims[2] + kk] = kernel[kk];
+                    }
                 }
             }
-
             return outarray;
         }
 

diff --git a/CNTKUNet/CNTKUNet/Program.cs b/CNTKUNet/CNTKUNet/Program.cs
@@ -14,6 +14,7 @@
 
 using CNTKUNet.Models;
 
+
 namespace CNTKUNet
 {
     class Program
@@ -22,7 +23,7 @@ class Program
         static void Main()
         {
             //Path to weights
-            string wpath = "c:\\users\\jfrondel\\Desktop\\GITS\\UNetE3bn.h5";
+            string wpath = "c:\\users\\jfrondel\\Desktop\\GITS\\UNetE3bnf.h5";
 
             //Path to test image
             string impath = "c:\\users\\jfrondel\\desktop\\GITS\\sample.png";