[ADD] Generator class

.gitignore

@@ -256,6 +256,6 @@ dmypy.json
 # Pyre type checker
 .pyre/
 
+# User ignores
 dataset/sources/~$Labels.docx
-ml/generator.py
-utils.py
+jupyter_config_itmo.py

facades.py

@@ -20,9 +20,10 @@
 keras.backend.set_image_data_format('channels_last')
 
 
-class Facades:  # TODO: Predict  # TODO: Comments
+class Facades:  # TODO: Predict
+    """Class for segmentation buildings' facades using Deep Learning"""
     def __init__(self, dim=(720, 1080), n_channels=3, use_datagenerator=True):
-        """Initialize main parameters of Facades class
+        """Initialize main parameters of Facades class s
         :param dim: Images resolution in format (height, width)
         :param n_channels: Number of channels in image (1 or 3)
         """
@@ -78,11 +79,12 @@ def training_init(self, dataset_dir="dataset", batch_size=8):
 
             # Initialize data generators
             seed = 909  # to transform image and masks with same augmentation parameter.
-            image_generator = data_datagen.flow_from_directory(dataset_dir + "train/data/", class_mode=None, seed=seed,
-                                                               batch_size=batch_size, target_size=self.dim)
-            masks_generator = masks_datagen.flow_from_directory(dataset_dir + "train/masks/", class_mode=None, seed=seed,
-                                                                batch_size=batch_size, target_size=self.dim,
-                                                                color_mode="grayscale")
+            image_generator = data_datagen.flow_from_directory(dataset_dir + "train/data/", class_mode=None,
+                                                               seed=seed, batch_size=batch_size,
+                                                               target_size=self.dim)
+            masks_generator = masks_datagen.flow_from_directory(dataset_dir + "train/masks/", class_mode=None,
+                                                                seed=seed, batch_size=batch_size,
+                                                                target_size=self.dim, color_mode="grayscale")
 
             self.train_generator = zip(image_generator, masks_generator)
 
@@ -117,11 +119,10 @@ def training_init(self, dataset_dir="dataset", batch_size=8):
         self.model = Model(base_model.input, output_layer)
         # keras.utils.plot_model(self.model, 'model.png', show_shapes=True)
 
-    def train(self, epochs=1, number_of_GPUs=0, remote_monitor_ip="localhost:4959"):
+    def train(self, epochs=1, number_of_GPUs=0):
         """Train model
         :param epochs: Number of epochs to train model
         :param number_of_GPUs: Number of devices to train on multiple GPUs
-        :param remote_monitor_ip: IP where to run Keras RemoteMonitor (default: localhost:4959)  # TODO
         """
         # Check training init before train
         if not self.training_init_check:
@@ -151,6 +152,11 @@ def train(self, epochs=1, number_of_GPUs=0, remote_monitor_ip="localhost:4959"):
 
 
 if __name__ == "__main__":
+    # Initialize main class
     facades = Facades((512, 512))
+
+    # Initialize training dataset
     facades.training_init("dataset/", batch_size=5)
+
+    # Run model training
     facades.train(10)

jupyter_config_itmo.py

@@ -2,4 +2,4 @@
 c.NotebookApp.port = 4958
 c.NotebookApp.password = u'sha1:1236633a1c55:9f2f5745eb1da1304369e01b65e5ad6d64f1cb13'
 
-c.NotebookApp.open_browser = False
+c.NotebookApp.open_browser = False

ml/__init__.py

@@ -1 +1,2 @@
+from .generator import Generator
 from .residual_model import residual_model

ml/generator.py

@@ -0,0 +1,52 @@
+import numpy as np
+from cv2 import imread
+import os
+from keras.utils import Sequence
+
+
+class Generator(Sequence):
+    """Generates data for Keras"""
+
+    def __init__(self, dataset_dir, batch_size=16, dim=(1920, 1080), n_channels=3, shuffle=True):
+        """Initialization"""
+        data_folder = dataset_dir + "/data"
+        self.data_paths = os.listdir(data_folder)
+
+        labels_path = dataset_dir + "/masks"
+        self.labels_paths = os.listdir(labels_path)
+
+        self.batch_size = batch_size
+        self.dim = dim
+        self.n_channels = n_channels
+
+        self.shuffle = shuffle
+
+        self.len = len(self.data_paths)
+
+    def __len__(self):
+        """Denotes the number of batches per epoch
+        :return: number of batches per epoch
+        """
+        return len(np.floor(self.len / self.batch_size))
+
+    def __getitem__(self, index):
+        """Generate one batch of data
+        :param index: index of the batch
+        :return: X and y when fitting. X only when predicting
+        """
+        # Generate indexes of the batch
+        indexes = self.indexes[index * self.batch_size:(index + 1) * self.batch_size]
+
+        # Download data by indexes
+        X, y = [], []
+        for i in indexes:
+            X.append(imread(self.data_paths[i]))
+            y.append(get_label(self.labels_paths[i]))
+
+        return X, y
+
+    def on_epoch_end(self):
+        """Updates indexes after each epoch"""
+        self.indexes = np.arange(self.len)
+        if self.shuffle:
+            np.random.shuffle(self.indexes)

utils.py

@@ -0,0 +1,77 @@
+import os
+
+import numpy as np
+from PIL import Image
+from tqdm import tqdm
+
+
+# def fix_dataset_labels(labels_path, colors=None):
+#     if colors is None:
+#         colors = {  # RGB
+#             (0, 0, 255): 0,  # Wall
+#             (0, 0, 170): 1,  # Background
+#             (255, 255, 0): 2,  # Window (closed)
+#             (0, 85, 255): 2,  # Window
+#             (0, 170, 255): 7,  # Door
+#             (170, 0, 0): 8,  # Shop
+#             (170, 255, 85): 3,  # Balcony
+#             (255, 85, 0): 6,  # Molding
+#             (255, 3, 0): 11,  # Pillar
+#             (0, 255, 255): 12,  # Cornice
+#             (85, 255, 170): 4,  # Sill
+#             (255, 170, 0): 9,  # Deco
+#             # (): 11,  # Blind
+#         }
+#
+#         try:
+#             os.mkdir(labels_path + "_FIXED/")
+#         except:
+#             pass
+#
+#         files = [f for f in os.listdir(labels_path) if f[-4:] in [".jpg", ".png"]]
+#         for id, image_name in enumerate(tqdm(files)):
+#             image_path = labels_path + "/" + image_name
+#             image = cv.imread(image_path)
+#
+#             for h in range(len(image)):
+#                 for w in range(len(image[h])):
+#                     c = image[h][w]
+#                     for key, value in colors.items():
+#                         if c[2] in range(key[0] - 3, key[0] + 3) and c[1] in range(key[1] - 2, key[1] + 3) and \
+#                                 c[0] in range(key[2] - 2, key[2] + 3):
+#                             image[h][w] = value
+#                             break
+#
+#             image = cv.cvtColor(image, cv.COLOR_BGR2GRAY)
+#             cv.imwrite(labels_path + "_FIXED/" + image_name, image)
+
+
+# def show_batch_photos(facades_class):
+#     x_batch, y_batch = next(facades_class.train_generator)
+#     for i in range(0, facades_class.batch_size):
+#         x = x_batch[i]
+#         y = y_batch[i]
+#
+#         plt.imshow(x)
+#         plt.show()  # TODO: Check needing two
+#         plt.imshow(y)
+#         plt.show()
+
+
+def load_images_from_folder(dir, shape):
+    files = [file for file in os.listdir(dir) if file.endswith((".png", ".jpg"))]
+    result = []
+
+    for image_name in tqdm(files, dir):
+        image_path = dir + "/" + image_name
+        image = Image.open(image_path)
+        image_resized = image.resize(shape[:-1], Image.ANTIALIAS)
+        image_numpy = np.array(image_resized)
+        if shape[-1] == 1:
+            image_numpy = np.expand_dims(image_numpy, axis=-1)
+
+        result.append(image_numpy)
+
+    return np.array(result)
+
+