[ENHANCE] nnf.core.iters for Core NN Framework

Signed-off-by: Nadith Pathirage <[email protected]>

+nnf/+core/+iters/+memory/BigDataNumpyArrayIterator.m

@@ -54,7 +54,7 @@
 
             % TODO: Apply necessary transofmraiton
             %%x = self.image_data_generator.random_transform(x)
-            %x = self.image_data_generator.standardize(x)
+            x = self.image_data_generator.standardize(x);
         end
 
         %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

+nnf/+core/+iters/+memory/MemDataIterator.m

@@ -99,7 +99,8 @@
             % zca_whitening. Currently supported only for in memory datasets.
             if (self.imdata_pp_.featurewise_center || ...
                 self.imdata_pp_.featurewise_std_normalization || ...
-                self.imdata_pp_.zca_whitening)
+                self.imdata_pp_.zca_whitening || ...
+                (~isempty(self.pp_params_) && self.pp_params_.isKey('mapminmax')))
                 self.imdata_pp_.fit(db, ...
                                     self.imdata_pp_.augment, ...
                                     self.imdata_pp_.rounds, ...

+nnf/+core/+iters/ImageDataGenerator.m

@@ -189,7 +189,7 @@
             end
 
             % x is a single image, so it doesn't have image number at index 0
-            img_channel_index = self.channel_index - 1
+            img_channel_index = self.channel_index - 1;
             if self.samplewise_center
                 M = mean(x, img_channel_index);
                 x = bsxfun(@minus, x, M);

+nnf/+core/+iters/ImageDataPreProcessor.m

@@ -13,7 +13,8 @@
 
     properties (SetAccess = protected)
         fn_gen_coreiter_;
-        nrm_vgg16_;
+        pp_params_;
+        mapminmax_setting_;
     end
 
     properties (SetAccess = protected, Dependent)
@@ -42,15 +43,8 @@
 
             self = [email protected](pp_params);            
             self.fn_gen_coreiter_ = fn_gen_coreiter;
-
-            % VGG16Model specific pre-processing param
-            if (isempty(pp_params)) 
-                self.nrm_vgg16_ = false;
-            elseif (pp_params.isKey('normalize_vgg16'))                
-                self.nrm_vgg16_ = pp_params.get('normalize_vgg16');
-            else
-                self.nrm_vgg16_ = false;
-            end            
+            self.pp_params_ = pp_params;
+            self.mapminmax_setting_ = [];            
         end
 
         %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
@@ -61,20 +55,30 @@ function apply(self, settings)
             self.mean = settings.mean;
             self.std = settings.std;
             self.principal_components = settings.principal_components;
-            self.map_min_max = settings.map_min_max;
+            self.mapminmax_setting_ = settings.mapminmax_setting_;
             self.whiten = settings.whiten;
         end
 
         %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
         function x = standardize(self, x)
             % Standardize data sample.
-            if (self.nrm_vgg16_)
+
+            % New custom pre-processing param
+            % VGG16Model specific pre-processing param
+            if (~isempty(self.pp_params_) && ...
+                    self.pp_params_.isKey('normalize_vgg16') && ...
+                    self.pp_params_.get('normalize_vgg16')) 
                 x(0, :, :) = x(0, :, :) - 103.939;
                 x(1, :, :) = x(1, :, :) - 116.779;
                 x(2, :, :) = x(2, :, :) - 123.68;
             end
-
+            
             x = [email protected](self, x);
+
+            % Map min max normalization
+            if (~isempty(self.mapminmax_setting_))
+                x = mapminmax('apply', x', self.mapminmax_setting_)';
+            end
         end
 
         %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
@@ -86,7 +90,13 @@ function fit(self, X, augment, rounds, seed)
             if (nargin < 3); augment = false; end
 
             [email protected](self, X, augment, rounds, seed)
+
             % Perform whitening/mapminmax/etc
+            if (~isempty(self.pp_params_) && ...
+                    self.pp_params_.isKey('mapminmax'))
+                    minmax_range = self.pp_params_.get('mapminmax');
+                [~, self.mapminmax_setting_] = mapminmax(X', minmax_range(1), minmax_range(2)); 
+            end
         end
 
         %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
@@ -118,7 +128,6 @@ function fit(self, X, augment, rounds, seed)
             if (nargin < 4); nb_class = []; end
             if (nargin < 3); y = []; end
 
-
             if (isempty(self.fn_gen_coreiter_))
                 core_iter = NumpyArrayIterator(X, y, nb_class, self, params);
                 return;
@@ -190,7 +199,7 @@ function fit(self, X, augment, rounds, seed)
             value.mean = self.mean;
             value.std = self.std;
             value.principal_components = self.principal_components;
-            value.map_min_max = self.map_min_max;
+            value.mapminmax_setting_ = self.mapminmax_setting_;
             value.whiten = self.whiten;
         end