MobileNetV1/V2花朵识别

经典网络/MobileNet/MobileNetV2.py

@@ -11,8 +11,10 @@ from tensorflow.keras.models import Model
 from tensorflow.keras.layers import BatchNormalization
 from tensorflow.keras.layers import Conv2D, Add, ZeroPadding2D, GlobalAveragePooling2D, Dropout, Dense
 from tensorflow.keras.layers import MaxPooling2D, Activation, DepthwiseConv2D, Input, GlobalMaxPooling2D
+from tensorflow.keras.layers import ReLU
 from tensorflow.keras.applications import imagenet_utils
 from tensorflow.keras.applications.imagenet_utils import decode_predictions
+
 # from tensorflow.keras.utils.data_utils import get_file
 
 # TODO Change path to v1.1
@@ -49,23 +51,95 @@ def correct_pad(inputs, kernel_size):
 def _make_divisible(v, divisor, min_value=None):
     if min_value is None:
         min_value = divisor
+    # 与value最接近的8的倍数。若value=30，那么处理后的卷积核=((30+4)//8)*8=32
     new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    # 如果计算出的结果比原值要小，那就再加上一个倍数
     if new_v < 0.9 * v:
         new_v += divisor
+    # 返回新的卷积核个数
     return new_v
 
 
-def MobileNetV2(input_shape=(224, 224, 3),
-                alpha=1.0,
-                include_top=True,
-                weights='imagenet',
-                classes=1000):
+def _inverted_res_block(inputs, expansion, stride, alpha, filters, block_id):
+    # 取输入shape的最后一维(通道数)
+    in_channels = backend.int_shape(inputs)[-1]
+    # 超参数控制逐点卷积的卷积核个数，使用int()函数取整，防止alpha是小数
+    pointwise_conv_filters = int(filters * alpha)
+    # 由于网络结构图中特征图的个数为32、16、24、64...都是8的倍数，
+    # 保证超参数处理过后的卷积核个数能满足网络特征图个数的要求，因此需要保证逐点卷积的卷积核个数为8的倍数
+    pointwise_filters = _make_divisible(pointwise_conv_filters, 8)
+    # 先用x表示输入，如果不执行part1的话，那么x没机会接收inputs
+    x = inputs
+    prefix = 'block_{}_'.format(block_id)
+
+    # part1 数据扩张:在通道维度上升维
+    if block_id:
+        # Expand 利用1*1卷积升维(提升通道数)
+        x = Conv2D(expansion * in_channels,  # 卷积核个数，通道维度提升expansion倍数
+                   kernel_size=1,  # 卷积核大小1*1
+                   strides=1,  # 步长1
+                   padding='same',  # 卷积过程中特征图size不变
+                   use_bias=False,  # 有BN层就不需要计算偏置
+                   activation=None,
+                   name=prefix + 'expand')(x)
+        # 批标准化
+        x = BatchNormalization(epsilon=1e-3,
+                               momentum=0.999,
+                               name=prefix + 'expand_BN')(x)
+        # relu6激活函数
+        # x = Activation(relu6, name=prefix + 'expand_relu')(x)
+        x = ReLU(6.0, name=prefix + 'expand_relu')(x)
+    else:
+        prefix = 'expanded_conv_'
+
+    if stride == 2:
+        x = ZeroPadding2D(padding=correct_pad(x, 3),
+                          name=prefix + 'pad')(x)
+
+    # part2 深度可分离卷积
+    x = DepthwiseConv2D(kernel_size=3,
+                        strides=stride,
+                        activation=None,
+                        use_bias=False,
+                        padding='same' if stride == 1 else 'valid',
+                        name=prefix + 'depthwise')(x)
+    x = BatchNormalization(epsilon=1e-3,
+                           momentum=0.999,
+                           name=prefix + 'depthwise_BN')(x)
+
+    x = Activation(relu6, name=prefix + 'depthwise_relu')(x)
+
+    # part3：1*1卷积降维   压缩特征，而且不使用relu函数，保证特征不被破坏
+    x = Conv2D(pointwise_filters,  # 卷积核个数(即通道数)下降到起始状态
+               kernel_size=1,
+               padding='same',
+               use_bias=False,
+               activation=None,
+               name=prefix + 'project')(x)
+
+    x = BatchNormalization(epsilon=1e-3,
+                           momentum=0.999,
+                           name=prefix + 'project_BN')(x)
+    # 残差连接，输入和输出短接
+    # 当输入通道数=输出通道数且步长为1，进行残差边的连接
+    if in_channels == pointwise_filters and stride == 1:
+        # 将输入和输出的tensor的对应元素值相加，要求shape完全一致
+        return Add(name=prefix + 'add')([inputs, x])
+    return x
+
+# MobileNetV2网络结构
+def MobileNetV2(input_shape=(224, 224, 3),alpha=1.0,include_top=True, weights='imagenet',classes=1000):
+    # 规定超参数的取值范围
+    if alpha not in [0.5, 0.75, 1.0, 1.3]:
+        raise ValueError('alpha should use 0.5, 0.75, 1.0, 1.3')
+
     rows = input_shape[0]
 
     img_input = Input(shape=input_shape)
 
     # stem部分
     # 224,224,3 -> 112,112,32
+    # 超参数控制卷积核个数，先处理第一次卷积的卷积核个数（32*alpha）使它能被8整除
     first_block_filters = _make_divisible(32 * alpha, 8)
     x = ZeroPadding2D(padding=correct_pad(img_input, 3),
                       name='Conv1_pad')(img_input)
@@ -82,34 +156,35 @@ def MobileNetV2(input_shape=(224, 224, 3),
     x = Activation(relu6, name='Conv1_relu')(x)
 
     # 112,112,32 -> 112,112,16
-    x = _inverted_res_block(x, filters=16, alpha=alpha, stride=1,expansion=1, block_id=0)
+    # 第一个残差卷积由于expansion=1，不用上升维度，因此不用执行第part1步的标准卷积操作，所以block_id=0
+    x = _inverted_res_block(x, filters=16, alpha=alpha, stride=1, expansion=1, block_id=0)
 
     # 112,112,16 -> 56,56,24
-    x = _inverted_res_block(x, filters=24, alpha=alpha, stride=2,expansion=6, block_id=1)
-    x = _inverted_res_block(x, filters=24, alpha=alpha, stride=1,expansion=6, block_id=2)
+    x = _inverted_res_block(x, filters=24, alpha=alpha, stride=2, expansion=6, block_id=1)
+    x = _inverted_res_block(x, filters=24, alpha=alpha, stride=1, expansion=6, block_id=2)
 
     # 56,56,24 -> 28,28,32
-    x = _inverted_res_block(x, filters=32, alpha=alpha, stride=2,expansion=6, block_id=3)
-    x = _inverted_res_block(x, filters=32, alpha=alpha, stride=1,expansion=6, block_id=4)
-    x = _inverted_res_block(x, filters=32, alpha=alpha, stride=1,expansion=6, block_id=5)
+    x = _inverted_res_block(x, filters=32, alpha=alpha, stride=2, expansion=6, block_id=3)
+    x = _inverted_res_block(x, filters=32, alpha=alpha, stride=1, expansion=6, block_id=4)
+    x = _inverted_res_block(x, filters=32, alpha=alpha, stride=1, expansion=6, block_id=5)
 
     # 28,28,32 -> 14,14,64
-    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=2,expansion=6, block_id=6)
-    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=1,expansion=6, block_id=7)
-    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=1,expansion=6, block_id=8)
-    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=1,expansion=6, block_id=9)
+    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=2, expansion=6, block_id=6)
+    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=1, expansion=6, block_id=7)
+    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=1, expansion=6, block_id=8)
+    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=1, expansion=6, block_id=9)
 
     # 14,14,64 -> 14,14,96
-    x = _inverted_res_block(x, filters=96, alpha=alpha, stride=1,expansion=6, block_id=10)
-    x = _inverted_res_block(x, filters=96, alpha=alpha, stride=1,expansion=6, block_id=11)
-    x = _inverted_res_block(x, filters=96, alpha=alpha, stride=1,expansion=6, block_id=12)
+    x = _inverted_res_block(x, filters=96, alpha=alpha, stride=1, expansion=6, block_id=10)
+    x = _inverted_res_block(x, filters=96, alpha=alpha, stride=1, expansion=6, block_id=11)
+    x = _inverted_res_block(x, filters=96, alpha=alpha, stride=1, expansion=6, block_id=12)
     # 14,14,96 -> 7,7,160
-    x = _inverted_res_block(x, filters=160, alpha=alpha, stride=2,expansion=6, block_id=13)
-    x = _inverted_res_block(x, filters=160, alpha=alpha, stride=1,expansion=6, block_id=14)
-    x = _inverted_res_block(x, filters=160, alpha=alpha, stride=1,expansion=6, block_id=15)
+    x = _inverted_res_block(x, filters=160, alpha=alpha, stride=2, expansion=6, block_id=13)
+    x = _inverted_res_block(x, filters=160, alpha=alpha, stride=1, expansion=6, block_id=14)
+    x = _inverted_res_block(x, filters=160, alpha=alpha, stride=1, expansion=6, block_id=15)
 
     # 7,7,160 -> 7,7,320
-    x = _inverted_res_block(x, filters=320, alpha=alpha, stride=1,expansion=6, block_id=16)
+    x = _inverted_res_block(x, filters=320, alpha=alpha, stride=1, expansion=6, block_id=16)
 
     if alpha > 1.0:
         last_block_filters = _make_divisible(1280 * alpha, 8)
@@ -125,7 +200,7 @@ def MobileNetV2(input_shape=(224, 224, 3),
                            momentum=0.999,
                            name='Conv_1_bn')(x)
     x = Activation(relu6, name='out_relu')(x)
-
+    # [7,7,1280]->[None,1280]
     x = GlobalAveragePooling2D()(x)
     x = Dense(classes, activation='softmax',
               use_bias=True, name='Logits')(x)
@@ -155,68 +230,13 @@ def MobileNetV2(input_shape=(224, 224, 3),
     return model
 
 
-def _inverted_res_block(inputs, expansion, stride, alpha, filters, block_id):
-    # 取输入shape的最后一维(通道数)
-    in_channels = backend.int_shape(inputs)[-1]
-    pointwise_conv_filters = int(filters * alpha)
-    pointwise_filters = _make_divisible(pointwise_conv_filters, 8)
-    x = inputs
-    prefix = 'block_{}_'.format(block_id)
-
-    # part1 数据扩张
-    if block_id:
-        # Expand 利用1*1卷积升维
-        x = Conv2D(expansion * in_channels,
-                   kernel_size=1,
-                   padding='same',
-                   use_bias=False,
-                   activation=None,
-                   name=prefix + 'expand')(x)
-        x = BatchNormalization(epsilon=1e-3,
-                               momentum=0.999,
-                               name=prefix + 'expand_BN')(x)
-        x = Activation(relu6, name=prefix + 'expand_relu')(x)
-    else:
-        prefix = 'expanded_conv_'
-
-    if stride == 2:
-        x = ZeroPadding2D(padding=correct_pad(x, 3),
-                          name=prefix + 'pad')(x)
-
-    # part2 深度可分离卷积
-    x = DepthwiseConv2D(kernel_size=3,
-                        strides=stride,
-                        activation=None,
-                        use_bias=False,
-                        padding='same' if stride == 1 else 'valid',
-                        name=prefix + 'depthwise')(x)
-    x = BatchNormalization(epsilon=1e-3,
-                           momentum=0.999,
-                           name=prefix + 'depthwise_BN')(x)
-
-    x = Activation(relu6, name=prefix + 'depthwise_relu')(x)
-
-    # part3：1*1卷积降维   压缩特征，而且不使用relu函数，保证特征不被破坏
-    x = Conv2D(pointwise_filters,
-               kernel_size=1,
-               padding='same',
-               use_bias=False,
-               activation=None,
-               name=prefix + 'project')(x)
-
-    x = BatchNormalization(epsilon=1e-3,
-                           momentum=0.999,
-                           name=prefix + 'project_BN')(x)
-    # 当输入通道数=输出通道数且步长为1，进行残差边的连接
-    if in_channels == pointwise_filters and stride == 1:
-        return Add(name=prefix + 'add')([inputs, x])
-    return x
 def preprocess_input(x):
     x /= 255.
     x -= 0.5
     x *= 2.
     return x
 
+
 if __name__ == '__main__':
     model = MobileNetV2(input_shape=(224, 224, 3))
     model.summary()