snapshot...

modules/deeplearning/espcn.py

@@ -370,7 +370,7 @@ class ESPCN:
         activation = tf.nn.leaky_relu
         momentum = 0.99
 
-        num_filters = 64
+        num_filters = 32
 
         input_2d = self.inputs[0]
         print('input: ', input_2d.shape)
@@ -383,7 +383,7 @@ class ESPCN:
         conv = conv_b = tf.keras.layers.Conv2D(num_filters, kernel_size=3, padding=padding)(input_2d)
 
         if NOISE_TRAINING:
-            conv = tf.keras.layers.GaussianNoise(stddev=NOISE_STDDEV)(conv)
+            conv = conv_b = tf.keras.layers.GaussianNoise(stddev=NOISE_STDDEV)(conv)
 
         conv_b = build_conv2d_block(conv_b, num_filters, 'Residual_Block_1')
 
@@ -391,13 +391,15 @@ class ESPCN:
 
         conv_b = build_conv2d_block(conv_b, num_filters, 'Residual_Block_3')
 
-        conv_b = tf.keras.layers.Conv2D(num_filters // 2, kernel_size=3, strides=1, padding=padding)(conv_b)
+        conv_b = build_conv2d_block(conv_b, num_filters, 'Residual_Block_4')
+
+        conv_b = tf.keras.layers.Conv2D(num_filters, kernel_size=3, strides=1, padding=padding)(conv_b)
 
         conv = conv + conv_b
         print(conv.shape)
 
-        # conv = tf.keras.layers.Conv2D(num_filters * (factor ** 2), 3, padding='same')(conv)
-        conv = tf.keras.layers.Conv2D((factor ** 2), 3, padding='same')(conv)
+        conv = tf.keras.layers.Conv2D(num_filters * (factor ** 2), 3, padding='same')(conv)
+        # conv = tf.keras.layers.Conv2D((factor ** 2), 3, padding='same')(conv)
         print(conv.shape)
 
         conv = tf.nn.depth_to_space(conv, factor)