Updated SRGAN with changes in vgg19 output, updated to Pillow Image reader from scipy.misc.imread, and updated the dataset link

.gitignore

@@ -7,5 +7,8 @@
 *.hdf5
 .DS_Store
 */datasets
+*.idea
+*.ipynb
+*.zip
 
 __pycache__

srgan/data_loader.py

@@ -1,6 +1,7 @@
 import scipy
 from glob import glob
 import numpy as np
+from PIL import Image
 import matplotlib.pyplot as plt
 
 class DataLoader():
@@ -18,13 +19,13 @@ def load_data(self, batch_size=1, is_testing=False):
         imgs_hr = []
         imgs_lr = []
         for img_path in batch_images:
-            img = self.imread(img_path)
+            img = Image.open(img_path)
 
             h, w = self.img_res
             low_h, low_w = int(h / 4), int(w / 4)
 
-            img_hr = scipy.misc.imresize(img, self.img_res)
-            img_lr = scipy.misc.imresize(img, (low_h, low_w))
+            img_hr = np.array(img.resize(self.img_res))
+            img_lr = np.array(img.resize((low_h, low_w)))
 
             # If training => do random flip
             if not is_testing and np.random.random() < 0.5:
@@ -38,7 +39,3 @@ def load_data(self, batch_size=1, is_testing=False):
         imgs_lr = np.array(imgs_lr) / 127.5 - 1.
 
         return imgs_hr, imgs_lr
-
-
-    def imread(self, path):
-        return scipy.misc.imread(path, mode='RGB').astype(np.float)

srgan/srgan.py

@@ -1,7 +1,11 @@
 """
 Super-resolution of CelebA using Generative Adversarial Networks.
 
-The dataset can be downloaded from: https://www.dropbox.com/sh/8oqt9vytwxb3s4r/AADIKlz8PR9zr6Y20qbkunrba/Img/img_align_celeba.zip?dl=0
+The dataset can be downloaded from: http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html
+Instructions to download dataset:
+1. Click on the above link, followed by the Google Drive link
+2. Click on the Img folder and save the img_align_celeba.zip
+3. Create a folder 'datasets/' and unizp the above file to this dataset folder.
 
 Instrustion on running the script:
 1. Download the dataset from the provided link
@@ -103,17 +107,12 @@ def build_vgg(self):
         Builds a pre-trained VGG19 model that outputs image features extracted at the
         third block of the model
         """
-        vgg = VGG19(weights="imagenet")
+        vgg = VGG19(weights="imagenet", input_shape=self.hr_shape, include_top=False)
         # Set outputs to outputs of last conv. layer in block 3
         # See architecture at: https://github.com/keras-team/keras/blob/master/keras/applications/vgg19.py
-        vgg.outputs = [vgg.layers[9].output]
+        outputs = vgg.layers[9].output
 
-        img = Input(shape=self.hr_shape)
-
-        # Extract image features
-        img_features = vgg(img)
-
-        return Model(img, img_features)
+        return Model(vgg.input, outputs)
 
     def build_generator(self):