NasNet架构复现--CVPR2018

经典网络/NasNet--CVPR2018/NasNet.py → 经典网络/NasNet--CVPR2018/NasNet--手动搭建.py

@@ -2,7 +2,7 @@ import tensorflow as tf
 from tensorflow.keras import layers
 from tensorflow.keras.models import Model
 from tensorflow.keras import backend
-
+from plot_model import plot_model
 
 def correct_pad(inputs, kernel_size):
     """返回一个元组，用于带有下采样的2D卷积的零填充.
@@ -33,7 +33,9 @@ def _adjust_block(p,  # 需要修改的输入张量
                   block_id=None):
     channel_dim = 1 if backend.image_data_format() == 'channels_first' else -1
     img_dim = 2 if backend.image_data_format() == 'channels_first' else -2
+
     ip_shape = backend.int_shape(ip)
+
     if p is not None:
         p_shape = backend.int_shape(p)
 
@@ -88,6 +90,7 @@ def _separable_conv_block(ip,  # 输入张量
                           kernel_size=(3, 3),  # 可分离卷积的内核大小
                           strides=(1, 1),  # 用于下采样的跨步卷积
                           block_id=None):
+    # channel_dim = 1 if backend.image_data_format() == 'channels_first' else -1
     with backend.name_scope('separable_conv_block_%s' % block_id):
         x = layers.Activation('relu')(ip)
         if strides == (2, 2):
@@ -119,8 +122,8 @@ def _separable_conv_block(ip,  # 输入张量
 
 
 # Adds a Reduction cell for NASNet-A(原论文图4)
-def _reduction_a_cell(ip,  # 输入张量
-                      p,  # 输入张量
+def _reduction_a_cell(ip,  # 输入张量x
+                      p,  # 输入张量p
                       filters,  # 输出filters的数量
                       block_id=None):
     channel_dim = 1 if backend.image_data_format() == 'channels_first' else -1
@@ -204,6 +207,7 @@ def _normal_a_cell(ip,  # 输入张量x
         h = layers.Activation('relu')(ip)
         h = layers.Conv2D(filters=filters,
                           kernel_size=(1, 1),
+                          strides=(1,1),
                           padding='same',
                           use_bias=False,
                           kernel_initializer='he_normal',
@@ -232,12 +236,13 @@ def _normal_a_cell(ip,  # 输入张量x
             x4_1 = layers.AveragePooling2D(pool_size=(3, 3),
                                            strides=(1, 1),
                                            padding='same',
-                                           name='normal-left4_%s' % block_id)(p)
+                                           name='normal_left4_%s' % block_id)(p)
             x4_2 = layers.AveragePooling2D(pool_size=(3, 3),
                                            strides=(1, 1),
                                            padding='same',
                                            name='normal_right4_%s' % block_id)(p)
             x4 = layers.add([x4_1, x4_2], name='normal_add_4_%s' % block_id)
+
         with backend.name_scope('block_5'):
             x5 = _separable_conv_block(h, filters, block_id='normal_left5_%s' % block_id)
             x5 = layers.add([x5, h], name='normal_add_5_%s' % block_id)
@@ -262,7 +267,7 @@ def NASNet(input_shape=None,  # 输入shape
     if default_size is None:
         default_size = 331
     img_input = layers.Input(shape=input_shape)
-    inputs=img_input
+    inputs = img_input
 
     filters = penultimate_filters // 24
 
@@ -293,27 +298,28 @@ def NASNet(input_shape=None,  # 输入shape
                               p,
                               filters * filter_multiplier ** 2,
                               block_id='reduce_%d' % (2 * num_blocks))
-    p=p0 if not skip_reduction else p
+    p = p0 if not skip_reduction else p
 
     for i in range(num_blocks):
-        x,p=_normal_a_cell(x,
-                           p,
-                           filters*filter_multiplier**2,
-                           block_id='%d'%(2*num_blocks+i+1))
-    x=layers.Activation('relu')(x)
+        x, p = _normal_a_cell(x,
+                              p,
+                              filters * filter_multiplier ** 2,
+                              block_id='%d' % (2 * num_blocks + i + 1))
+    x = layers.Activation('relu')(x)
 
     if include_top:
-        x=layers.GlobalAveragePooling2D()(x)
-        x=layers.Dense(classes,activation=classifier_activation,
-                       name='predictions')(x)
+        x = layers.GlobalAveragePooling2D()(x)
+        x = layers.Dense(classes, activation=classifier_activation,
+                         name='predictions')(x)
     else:
-        if pooling=='avg':
-            x=layers.GlobalAveragePooling2D()(x)
-        elif pooling=='max':
-            x=layers.GlobalMaxPooling2D()(x)
-    model=Model(inputs,x,name='NASNet')
+        if pooling == 'avg':
+            x = layers.GlobalAveragePooling2D()(x)
+        elif pooling == 'max':
+            x = layers.GlobalMaxPooling2D()(x)
+    model = Model(inputs, x, name='NASNet')
     return model
 
+
 def NASNetMobile(input_shape=None,
                  include_top=True,
                  weights='imagenet',
@@ -332,6 +338,29 @@ def NASNetMobile(input_shape=None,
                   pooling=pooling,
                   classes=classes,
                   default_size=224)
+
+def NASNetLarge(input_shape=None,
+                include_top=True,
+                weights='imagenet',
+                input_tensor=None,
+                pooling=None,
+                classes=1000,
+                classifier_activation='softmax'):
+    return NASNet(input_shape,
+                  penultimate_filters=4032,
+                  num_blocks=6,
+                  stem_block_filters=96,
+                  skip_reduction=True,
+                  filter_multiplier=2,
+                  include_top=include_top,
+                  weights=weights,
+                  input_tensor=input_tensor,
+                  pooling=pooling,
+                  classes=classes,
+                  default_size=331,
+                  classifier_activation=classifier_activation)
 if __name__ == '__main__':
-    model=NASNetMobile(input_shape=(224,224,3),classes=1000)
-    model.summary()
\ No newline at end of file
+    model = NASNetMobile(input_shape=(224, 224, 3), classes=1000)
+    # model=NASNetLarge(input_shape=(224,224,3),classes=1000)
+    model.summary()
+    # plot_model(model,to_file='img/NASNetMobile.pdf')

经典网络/NasNet--CVPR2018/NasNet--迁移学习.py

+from tensorflow.keras.applications import NASNetMobile
+
+model=NASNetMobile(include_top=True,
+                   input_shape=(224,224,3),
+                   classes=1000)
+model.summary()
\ No newline at end of file