djkgfhkfdj::q!

Merge branch 'master' of github.com:Unity-Technologies/Keras-GAN

infogan/infogan.py

@@ -179,7 +179,7 @@ class INFOGAN():
                 # Select a random half batch of images
                 idx = np.random.randint(0, X_train.shape[0], batch_size)
                 imgs = X_train[idx]
-
+                
                 # Sample noise and categorical labels
                 sampled_noise, sampled_labels = self.sample_generator_input(batch_size)
                 gen_input = np.concatenate((sampled_noise, sampled_labels), axis=1)
@@ -220,6 +220,7 @@ class INFOGAN():
             gen_input = np.concatenate((sampled_noise, label), axis=1)
             gen_imgs = self.generator.predict(gen_input)
             self.save_samples("generated_" + str(i), gen_imgs)
+
             #gen_imgs = 0.5 * gen_imgs + 0.5
             #for j in range(r):
             #    formatted = (gen_imgs[j,:,:] * 255 / np.max(gen_imgs[j,:,:])).astype('uint8')

unity/unity.py

@@ -26,17 +26,28 @@ def encode_word_presence(words, all_words):
 		array[i] = 1.0
 	return array.tolist()
 
+def augment(x_train, y_train, resize):
+	images_to_augments = len(x_train)
+	for i in range(images_to_augments):
+		im = Image.fromarray(x_train[i].astype('uint8').reshape((resize,resize,3)))
+		f1 = im.transpose(Image.FLIP_LEFT_RIGHT)
+		f2 = im.transpose(Image.FLIP_TOP_BOTTOM)
+		x_train.append(np.asarray(f1, dtype="int32"))
+		x_train.append(np.asarray(f2, dtype="int32"))
+		y_train.append(y_train[i])
+		y_train.append(y_train[i])
+
 def semantic_maps(resize=None):
 	seen_labels = list()
-	x_train = np.array([load_image(f, resize) for f in glob.glob("../data/semantic_maps/train/img/*.png")])
-	x_test = np.array([load_image(f, resize) for f in glob.glob("../data/semantic_maps/test/img/*.png")])
-	y_train = np.array([load_labels(f, seen_labels) for f in glob.glob("../data/semantic_maps/train/txt/*.txt")])
-	y_test = np.array([load_labels(f, seen_labels) for f in glob.glob("../data/semantic_maps/test/txt/*.txt")])
-	
+	x_train = ([load_image(f, resize) for f in glob.glob("../data/semantic_maps/train/img/*.png")])
+	x_test = ([load_image(f, resize) for f in glob.glob("../data/semantic_maps/test/img/*.png")])
+	y_train = ([load_labels(f, seen_labels) for f in glob.glob("../data/semantic_maps/train/txt/*.txt")])
+	y_test = ([load_labels(f, seen_labels) for f in glob.glob("../data/semantic_maps/test/txt/*.txt")])
+	augment	(x_train, y_train, resize)
 	seen_labels = sorted(list(set(seen_labels)))
-	y_train = np.array([encode_word_presence(a, seen_labels) for a in y_train])
-	y_test = np.array([encode_word_presence(a, seen_labels) for a in y_test])
-	return (x_train, y_train), (x_test, y_test)
+	y_train = ([encode_word_presence(a, seen_labels) for a in y_train])
+	y_test = ([encode_word_presence(a, seen_labels) for a in y_test])
+	return (np.array(x_train), np.array(y_train)), (np.array(x_test), np.array(y_test))
 
 def get_number_of_unique_classes():
 	seen_labels = list()
@@ -45,4 +56,4 @@ def get_number_of_unique_classes():
 	return unique_labels
 
 def semantic_maps_shape():
-	return 256, 3, get_number_of_unique_classes()
+	return 64, 3, get_number_of_unique_classes()