Create and process the celebA dataset in (32,32,3) dimension

snknitin · Apr 26, 2018 · 91891a2 · 91891a2
1 parent f1c1bee
commit 91891a2
Show file tree

Hide file tree

Showing 3 changed files with 21 additions and 11 deletions.
diff --git a/Code/data.py b/Code/data.py
@@ -3,13 +3,7 @@
 import pandas as pd
 import numpy as np
 from keras.datasets import mnist, cifar10
-
-
-
-
-
-
-
+import shared_utils as su
 
 
 def load_dataset(hps):
@@ -34,9 +28,8 @@ def load_dataset(hps):
     if hps.module == 'celeba':
         # load CelebA data
         width, height, channels = hps.train_celeba_dimensions
-        (X_train, y_train), (X_test, y_test) = celeba.load_data()
-        # rescale -1 to 1
-        X_train = (X_train.astype(np.float32) - 127.5) / 127.5
+        X_train = su.load_data()
+
 
 
     # defining input dims

diff --git a/Code/shared_utils.py b/Code/shared_utils.py
@@ -1,6 +1,7 @@
 import os
 import pandas as pd
 import numpy as np
+import h5py
 from keras import backend as K
 # import skimage
 # from skimage import data, color, exposure
@@ -63,3 +64,19 @@ def squash(vectors, axis=-1):
     return scale * vectors
 
 
+def load_data():
+    """
+    Loads the CelebA dataset from the hdf5 file and processes it
+    :return:
+    """
+
+    with h5py.File(os.path.join(os.path.dirname(os.getcwd()),"Data/celebA/CelebA_32_data.h5"), "r") as hf:
+        # Loading the data as floats
+        X_real_train = hf["data"][:].astype(np.float32)
+        # Transpose to make channels the last dimension
+        X_real_train = X_real_train.transpose(0, 2, 3, 1)
+        # Normalizing the pixels
+        X_real_train = (X_real_train- 127.5) / 127.5
+        np.random.shuffle(X_real_train)
+        # Split to 80%
+        return X_real_train[:162080]
diff --git a/Data/create_celeba.py b/Data/create_celeba.py
@@ -93,7 +93,7 @@ def check_HDF5(size=64):
                         help='Plot the images to make sure the data processing went OK')
     args = parser.parse_args()
 
-    data_dir = "../../data/processed"
+    data_dir = os.path.join(os.getcwd(),"celebA/")
 
     build_HDF5(args.jpeg_dir, size=args.img_size)