Dimensions

.travis.yml

@@ -1,7 +1,7 @@
 before_script:
   - sudo apt-add-repository ppa:octave/stable --yes
   - sudo apt-get update -y
-  - sudo apt-get install octave -y
+  - sudo apt-get install octave --force-yes
   - sudo apt-get install liboctave-dev -y
 script:
   - sh -c "octave tests/runalltests.m"

CNN/cnnbp.m

@@ -23,7 +23,9 @@
     for l = (n - 1) : -1 : 1
         if strcmp(net.layers{l}.type, 'c')
             for j = 1 : numel(net.layers{l}.a)
-                net.layers{l}.d{j} = net.layers{l}.a{j} .* (1 - net.layers{l}.a{j}) .* (expand(net.layers{l + 1}.d{j}, [net.layers{l + 1}.scale net.layers{l + 1}.scale 1]) / net.layers{l + 1}.scale ^ 2);
+                xscale = net.layers{l + 1}.xscale;
+                yscale = net.layers{l + 1}.yscale;
+                net.layers{l}.d{j} = net.layers{l}.a{j} .* (1 - net.layers{l}.a{j}) .* (expand(net.layers{l + 1}.d{j}, [xscale yscale 1]) / (xscale*yscale));
             end
         elseif strcmp(net.layers{l}.type, 's')
             for i = 1 : numel(net.layers{l}.a)

CNN/cnnff.m

@@ -21,8 +21,8 @@
         elseif strcmp(net.layers{l}.type, 's')
             %  downsample
             for j = 1 : inputmaps
-                z = convn(net.layers{l - 1}.a{j}, ones(net.layers{l}.scale) / (net.layers{l}.scale ^ 2), 'valid');   %  !! replace with variable
-                net.layers{l}.a{j} = z(1 : net.layers{l}.scale : end, 1 : net.layers{l}.scale : end, :);
+                z = convn(net.layers{l - 1}.a{j}, net.layers{l}.meanFilter, 'valid');
+                net.layers{l}.a{j} = z(1 : net.layers{l}.xscale : end, 1 : net.layers{l}.yscale : end, :);
             end
         end
     end

CNN/cnnsetup.m

@@ -1,14 +1,17 @@
 function net = cnnsetup(net, x, y)
     inputmaps = 1;
     mapsize = size(squeeze(x(:, :, 1)));
-
-    for l = 1 : numel(net.layers)   %  layer
+	
+	for l = 1 : numel(net.layers)   %  layer
         if strcmp(net.layers{l}.type, 's')
-            mapsize = mapsize / net.layers{l}.scale;
+            mapsize = [mapsize(1)/net.layers{l}.xscale, mapsize(2)/net.layers{l}.yscale];
             assert(all(floor(mapsize)==mapsize), ['Layer ' num2str(l) ' size must be integer. Actual: ' num2str(mapsize)]);
             for j = 1 : inputmaps
                 net.layers{l}.b{j} = 0;
             end
+			%these are the subsampling filters for each layer
+			net.layers{l}.meanFilter = ones(net.layers{l}.xscale, net.layers{l}.yscale) / (net.layers{l}.xscale*net.layers{l}.yscale);
+
         end
         if strcmp(net.layers{l}.type, 'c')
             mapsize = mapsize - net.layers{l}.kernelsize + 1;

tests/test_cnn_gradients_are_numerically_correct.m

@@ -1,12 +1,29 @@
 function test_cnn_gradients_are_numerically_correct
+%squared average kernels
 batch_x = rand(28,28,5);
 batch_y = rand(10,5);
 cnn.layers = {
     struct('type', 'i') %input layer
     struct('type', 'c', 'outputmaps', 2, 'kernelsize', 5) %convolution layer
-    struct('type', 's', 'scale', 2) %sub sampling layer
+    struct('type', 's', 'xscale', 2, 'yscale', 2) %sub sampling layer
     struct('type', 'c', 'outputmaps', 2, 'kernelsize', 5) %convolution layer
-    struct('type', 's', 'scale', 2) %subsampling layer
+    struct('type', 's', 'xscale', 2, 'yscale', 2) %subsampling layer
+};
+cnn = cnnsetup(cnn, batch_x, batch_y);
+
+cnn = cnnff(cnn, batch_x);
+cnn = cnnbp(cnn, batch_y);
+cnnnumgradcheck(cnn, batch_x, batch_y);
+
+%non squared average kernels
+batch_x = rand(28,600,5);
+batch_y = rand(10,5);
+cnn.layers = {
+    struct('type', 'i') %input layer
+    struct('type', 'c', 'outputmaps', 2, 'kernelsize', 5) %convolution layer
+    struct('type', 's', 'xscale', 2, 'yscale', 2) %sub sampling layer
+    struct('type', 'c', 'outputmaps', 2, 'kernelsize', 5) %convolution layer
+    struct('type', 's', 'xscale', 2, 'yscale', 2) %subsampling layer
 };
 cnn = cnnsetup(cnn, batch_x, batch_y);
 

tests/test_example_CNN.m

@@ -15,12 +15,11 @@
 cnn.layers = {
     struct('type', 'i') %input layer
     struct('type', 'c', 'outputmaps', 6, 'kernelsize', 5) %convolution layer
-    struct('type', 's', 'scale', 2) %sub sampling layer
+    struct('type', 's', 'xscale', 2, 'yscale', 2) %sub sampling layer
     struct('type', 'c', 'outputmaps', 12, 'kernelsize', 5) %convolution layer
-    struct('type', 's', 'scale', 2) %subsampling layer
+    struct('type', 's', 'xscale', 2, 'yscale', 2) %subsampling layer
 };
 
-
 opts.alpha = 1;
 opts.batchsize = 200;
 opts.numepochs = 7;