Randomly transform with 50% chance

DubiousCactus · May 8, 2019 · 413c70f · 413c70f
1 parent a0191ed
commit 413c70f
Show file tree

Hide file tree

Showing 4 changed files with 10 additions and 8 deletions.
diff --git a/cnn.py b/cnn.py
@@ -144,7 +144,7 @@ def _main():
     # Generate training data with no real-time augmentation
     # train_datagen = utils.fit_flow_from_directory(rescale=1./255)
     train_datagen = utils.DroneDataGenerator(rescale=1./255,
-                                             channel_shift_range=0.15)
+                                             channel_shift_range=0.1)
 
     config = {
         'featurewise_center': True,

diff --git a/cnn_models.py b/cnn_models.py
@@ -94,9 +94,9 @@ def resnet8(img_width, img_height, img_channels, output_dim,
 
     # h_noise = GaussianNoise(0.01)(x)
     h1 = Dense(500, activation='relu', kernel_regularizer=regularizers.l2(0.001))(x)
-    h1 = Dropout(0.3)(h1)
-    h2 = Dense(500, activation='relu', kernel_regularizer=regularizers.l2(0.001))(h1)
-    h2 = Dropout(0.2)(h2)
+    h1 = Dropout(0.4)(h1)
+    h2 = Dense(100, activation='relu', kernel_regularizer=regularizers.l2(0.001))(h1)
+    h2 = Dropout(0.3)(h2)
     # Gate localization
     localization = Dense(output_dim, activation='softmax')(h2) # Logits + Softmax
 

diff --git a/img_utils.py b/img_utils.py
@@ -23,7 +23,7 @@ def load_img(path, grayscale=False):
     else:
         img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
 
-    return np.asarray(img, dtype=np.uint8)
+    return np.asarray(img, dtype=np.float64)
 
 
 def random_channel_shift(img, shiftFactor, channel_axis=2):

diff --git a/utils.py b/utils.py
@@ -276,14 +276,16 @@ def _get_batches_of_transformed_samples(self, index_array) :
             x = img_utils.load_img(os.path.join(self.directory, fname),
                                    grayscale=grayscale)
 
-            if self.image_data_generator.channelShiftFactor > 0:
+            # 50% chances of transforming the image
+            shifting = np.random.rand() <= 0.5
+            if shifting and self.image_data_generator.channelShiftFactor > 0:
                 transformed_x = img_utils.random_channel_shift(np.copy(x),
                                                                self.image_data_generator.channelShiftFactor)
             else:
                 transformed_x = x
 
-            if self.image_data_generator.channelShiftFactor > 0 and self.saved_transforms < 50:
-                Image.fromarray(x, "RGB").save(os.path.join(self.directory,
+            if shifting and self.image_data_generator.channelShiftFactor > 0 and self.saved_transforms < 50:
+                Image.fromarray(x.astype(np.uint8), "RGB").save(os.path.join(self.directory,
                                                      "img_transforms",
                                                      "original_{}.jpg".format(self.saved_transforms)))
                 Image.fromarray(transformed_x.astype(np.uint8), "RGB").save(os.path.join(self.directory,