WGAN (GP): Resolves #37. + clean up of handling input shapes of latent-to-image models

README.md

@@ -58,7 +58,7 @@ $ python3 acgan.py
 ### Adversarial Autoencoder
 Implementation of _Adversarial Autoencoder_.
 
-[Code](aae/adversarial_autoencoder.py)
+[Code](aae/aae.py)
 
 Paper: https://arxiv.org/abs/1511.05644
 

aae/aae.py

@@ -60,7 +60,7 @@ class AdversarialAutoencoder():
 
         img = Input(shape=self.img_shape)
 
-        h = Flatten(input_shape=self.img_shape)(img)
+        h = Flatten()(img)
         h = Dense(512)(h)
         h = LeakyReLU(alpha=0.2)(h)
         h = Dense(512)(h)

bgan/bgan.py

@@ -22,6 +22,7 @@ class BGAN():
         self.img_cols = 28
         self.channels = 1
         self.img_shape = (self.img_rows, self.img_cols, self.channels)
+        self.latent_dim = 100
 
         optimizer = Adam(0.0002, 0.5)
 
@@ -51,11 +52,9 @@ class BGAN():
 
     def build_generator(self):
 
-        noise_shape = (100,)
-
         model = Sequential()
 
-        model.add(Dense(256, input_shape=noise_shape))
+        model.add(Dense(256, input_dim=self.latent_dim))
         model.add(LeakyReLU(alpha=0.2))
         model.add(BatchNormalization(momentum=0.8))
         model.add(Dense(512))
@@ -69,18 +68,16 @@ class BGAN():
 
         model.summary()
 
-        noise = Input(shape=noise_shape)
+        noise = Input(shape=(self.latent_dim,))
         img = model(noise)
 
         return Model(noise, img)
 
     def build_discriminator(self):
 
-        img_shape = (self.img_rows, self.img_cols, self.channels)
-
         model = Sequential()
 
-        model.add(Flatten(input_shape=img_shape))
+        model.add(Flatten(input_shape=self.img_shape))
         model.add(Dense(512))
         model.add(LeakyReLU(alpha=0.2))
         model.add(Dense(256))
@@ -88,7 +85,7 @@ class BGAN():
         model.add(Dense(1, activation='sigmoid'))
         model.summary()
 
-        img = Input(shape=img_shape)
+        img = Input(shape=self.img_shape)
         validity = model(img)
 
         return Model(img, validity)
@@ -121,7 +118,7 @@ class BGAN():
             idx = np.random.randint(0, X_train.shape[0], half_batch)
             imgs = X_train[idx]
 
-            noise = np.random.normal(0, 1, (half_batch, 100))
+            noise = np.random.normal(0, 1, (half_batch, self.latent_dim))
 
             # Generate a half batch of new images
             gen_imgs = self.generator.predict(noise)
@@ -136,7 +133,7 @@ class BGAN():
             #  Train Generator
             # ---------------------
 
-            noise = np.random.normal(0, 1, (batch_size, 100))
+            noise = np.random.normal(0, 1, (batch_size, self.latent_dim))
 
             # The generator wants the discriminator to label the generated samples
             # as valid (ones)
@@ -154,7 +151,7 @@ class BGAN():
 
     def sample_images(self, epoch):
         r, c = 5, 5
-        noise = np.random.normal(0, 1, (r * c, 100))
+        noise = np.random.normal(0, 1, (r * c, self.latent_dim))
         gen_imgs = self.generator.predict(noise)
 
         # Rescale images 0 - 1

dcgan/dcgan.py

@@ -52,7 +52,7 @@ class DCGAN():
 
         model = Sequential()
 
-        model.add(Dense(128 * 7 * 7, activation="relu", input_shape=(self.latent_dim,)))
+        model.add(Dense(128 * 7 * 7, activation="relu", input_dim=self.latent_dim))
         model.add(Reshape((7, 7, 128)))
         model.add(UpSampling2D())
         model.add(Conv2D(128, kernel_size=3, padding="same"))

gan/gan.py

@@ -20,6 +20,7 @@ class GAN():
         self.img_cols = 28
         self.channels = 1
         self.img_shape = (self.img_rows, self.img_cols, self.channels)
+        self.latent_dim = 100
 
         optimizer = Adam(0.0002, 0.5)
 
@@ -54,7 +55,7 @@ class GAN():
 
         model = Sequential()
 
-        model.add(Dense(256, input_shape=noise_shape))
+        model.add(Dense(256, input_dim=self.latent_dim))
         model.add(LeakyReLU(alpha=0.2))
         model.add(BatchNormalization(momentum=0.8))
         model.add(Dense(512))
@@ -68,18 +69,16 @@ class GAN():
 
         model.summary()
 
-        noise = Input(shape=noise_shape)
+        noise = Input(shape=(self.latent_dim,))
         img = model(noise)
 
         return Model(noise, img)
 
     def build_discriminator(self):
         
-        img_shape = (self.img_rows, self.img_cols, self.channels)
-
         model = Sequential()
 
-        model.add(Flatten(input_shape=img_shape))
+        model.add(Flatten(input_shape=self.img_shape))
         model.add(Dense(512))
         model.add(LeakyReLU(alpha=0.2))
         model.add(Dense(256))
@@ -87,7 +86,7 @@ class GAN():
         model.add(Dense(1, activation='sigmoid'))
         model.summary()
 
-        img = Input(shape=img_shape)
+        img = Input(shape=self.img_shape)
         validity = model(img)
 
         return Model(img, validity)

lsgan/lsgan.py

@@ -20,6 +20,7 @@ class LSGAN():
         self.img_cols = 28
         self.channels = 1
         self.img_shape = (self.img_rows, self.img_cols, self.channels)
+        self.latent_dim = 100
 
         optimizer = Adam(0.0002, 0.5)
 
@@ -50,11 +51,9 @@ class LSGAN():
 
     def build_generator(self):
 
-        noise_shape = (100,)
-
         model = Sequential()
 
-        model.add(Dense(256, input_shape=noise_shape))
+        model.add(Dense(256, input_dim=self.latent_dim))
         model.add(LeakyReLU(alpha=0.2))
         model.add(BatchNormalization(momentum=0.8))
         model.add(Dense(512))
@@ -68,18 +67,16 @@ class LSGAN():
 
         model.summary()
 
-        noise = Input(shape=noise_shape)
+        noise = Input(shape=(self.latent_dim,))
         img = model(noise)
 
         return Model(noise, img)
 
     def build_discriminator(self):
 
-        img_shape = (self.img_rows, self.img_cols, self.channels)
-
         model = Sequential()
 
-        model.add(Flatten(input_shape=img_shape))
+        model.add(Flatten(input_shape=self.img_shape))
         model.add(Dense(512))
         model.add(LeakyReLU(alpha=0.2))
         model.add(Dense(256))
@@ -88,7 +85,7 @@ class LSGAN():
         model.add(Dense(1))
         model.summary()
 
-        img = Input(shape=img_shape)
+        img = Input(shape=self.img_shape)
         validity = model(img)
 
         return Model(img, validity)
@@ -114,7 +111,7 @@ class LSGAN():
             idx = np.random.randint(0, X_train.shape[0], half_batch)
             imgs = X_train[idx]
 
-            noise = np.random.normal(0, 1, (half_batch, 100))
+            noise = np.random.normal(0, 1, (half_batch, self.latent_dim))
 
             # Generate a half batch of new images
             gen_imgs = self.generator.predict(noise)
@@ -129,7 +126,7 @@ class LSGAN():
             #  Train Generator
             # ---------------------
 
-            noise = np.random.normal(0, 1, (batch_size, 100))
+            noise = np.random.normal(0, 1, (batch_size, self.latent_dim))
 
             # The generator wants the discriminator to label the generated samples
             # as valid (ones)
@@ -147,7 +144,7 @@ class LSGAN():
 
     def sample_images(self, epoch):
         r, c = 5, 5
-        noise = np.random.normal(0, 1, (r * c, 100))
+        noise = np.random.normal(0, 1, (r * c, self.latent_dim))
         gen_imgs = self.generator.predict(noise)
 
         # Rescale images 0 - 1

sgan/sgan.py

@@ -21,6 +21,7 @@ class SGAN():
         self.img_cols = 28
         self.channels = 1
         self.num_classes = 10
+        self.latent_dim = 100
 
         optimizer = Adam(0.0002, 0.5)
 
@@ -55,7 +56,7 @@ class SGAN():
 
         model = Sequential()
 
-        model.add(Dense(128 * 7 * 7, activation="relu", input_dim=100))
+        model.add(Dense(128 * 7 * 7, activation="relu", input_dim=self.latent_dim))
         model.add(Reshape((7, 7, 128)))
         model.add(BatchNormalization(momentum=0.8))
         model.add(UpSampling2D())
@@ -71,18 +72,16 @@ class SGAN():
 
         model.summary()
 
-        noise = Input(shape=(100,))
+        noise = Input(shape=(self.latent_dim,))
         img = model(noise)
 
         return Model(noise, img)
 
     def build_discriminator(self):
 
-        img_shape = (self.img_rows, self.img_cols, self.channels)
-
         model = Sequential()
 
-        model.add(Conv2D(32, kernel_size=3, strides=2, input_shape=img_shape, padding="same"))
+        model.add(Conv2D(32, kernel_size=3, strides=2, input_shape=self.img_shape, padding="same"))
         model.add(LeakyReLU(alpha=0.2))
         model.add(Dropout(0.25))
         model.add(Conv2D(64, kernel_size=3, strides=2, padding="same"))
@@ -101,7 +100,7 @@ class SGAN():
         model.add(Flatten())
         model.summary()
 
-        img = Input(shape=img_shape)
+        img = Input(shape=self.img_shape)
 
         features = model(img)
         valid = Dense(1, activation="sigmoid")(features)
@@ -142,7 +141,7 @@ class SGAN():
             imgs = X_train[idx]
 
             # Sample noise and generate a half batch of new images
-            noise = np.random.normal(0, 1, (half_batch, 100))
+            noise = np.random.normal(0, 1, (half_batch, self.latent_dim))
             gen_imgs = self.generator.predict(noise)
 
             valid = np.ones((half_batch, 1))
@@ -161,7 +160,7 @@ class SGAN():
             #  Train Generator
             # ---------------------
 
-            noise = np.random.normal(0, 1, (batch_size, 100))
+            noise = np.random.normal(0, 1, (batch_size, self.latent_dim))
             validity = np.ones((batch_size, 1))
 
             # Train the generator
@@ -176,7 +175,7 @@ class SGAN():
 
     def sample_images(self, epoch):
         r, c = 5, 5
-        noise = np.random.normal(0, 1, (r * c, 100))
+        noise = np.random.normal(0, 1, (r * c, self.latent_dim))
         gen_imgs = self.generator.predict(noise)
 
         # Rescale images 0 - 1

wgan/wgan.py

@@ -61,7 +61,7 @@ class WGAN():
 
         model = Sequential()
 
-        model.add(Dense(128 * 7 * 7, activation="relu", input_shape=(self.latent_dim,)))
+        model.add(Dense(128 * 7 * 7, activation="relu", input_dim=self.latent_dim))
         model.add(Reshape((7, 7, 128)))
         model.add(UpSampling2D())
         model.add(Conv2D(128, kernel_size=4, padding="same"))
@@ -76,7 +76,7 @@ class WGAN():
 
         model.summary()
 
-        noise = Input(shape=noise_shape)
+        noise = Input(shape=(self.latent_dim,))
         img = model(noise)
 
         return Model(noise, img)
@@ -106,7 +106,7 @@ class WGAN():
 
         model.summary()
 
-        img = Input(shape=img_shape)
+        img = Input(shape=self.img_shape)
         validity = model(img)
 
         return Model(img, validity)

wgan_gp/wgan_gp.py

@@ -128,7 +128,7 @@ class WGANGP():
 
         model = Sequential()
 
-        model.add(Dense(128 * 7 * 7, activation="relu", input_shape=(self.latent_dim,)))
+        model.add(Dense(128 * 7 * 7, activation="relu", input_dim=self.latent_dim))
         model.add(Reshape((7, 7, 128)))
         model.add(UpSampling2D())
         model.add(Conv2D(128, kernel_size=4, padding="same"))
@@ -143,7 +143,7 @@ class WGANGP():
 
         model.summary()
 
-        noise = Input(shape=noise_shape)
+        noise = Input(shape=(self.latent_dim,))
         img = model(noise)
 
         return Model(noise, img)
@@ -173,7 +173,7 @@ class WGANGP():
 
         model.summary()
 
-        img = Input(shape=img_shape)
+        img = Input(shape=self.img_shape)
         validity = model(img)
 
         return Model(img, validity)