ResNet50V2模型复现

经典网络/ResNetV2/ResNet50.py

+import tensorflow as tf
+import tensorflow.keras.layers as layers
+from tensorflow.keras.models import Model
+from plot_model import plot_model
+
+"""A residual block.
+  Arguments:
+      x: input tensor.
+      filters: integer, filters of the bottleneck layer.
+      kernel_size: default 3, kernel size of the bottleneck layer.
+      stride: default 1, stride of the first layer.
+      conv_shortcut: default False, use convolution shortcut if True,
+        otherwise identity shortcut.
+      name: string, block label.
+  Returns:
+    Output tensor for the residual block.
+  """
+def block2(x, filters, kernel_size=3, stride=1, conv_shortcut=False, name=None):
+    preact = layers.BatchNormalization(name=name + '_preact_bn')(x)
+    preact = layers.Activation('relu', name=name + '_preact_relu')(preact)
+
+    if conv_shortcut:
+        shortcut = layers.Conv2D(4 * filters, 1, strides=stride, name=name + '_0_conv')(preact)
+    else:
+        shortcut = layers.MaxPooling2D(1, strides=stride)(x) if stride > 1 else x
+
+    x = layers.Conv2D(filters, 1, strides=1, use_bias=False, name=name + '_1_conv')(preact)
+    x = layers.BatchNormalization(name=name + '_1_bn')(x)
+    x = layers.Activation('relu', name=name + '_1_relu')(x)
+
+    x = layers.ZeroPadding2D(padding=((1, 1), (1, 1)), name=name + '_2_pad')(x)
+    x = layers.Conv2D(filters,
+                      kernel_size,
+                      strides=stride,
+                      use_bias=False,
+                      name=name + '_2_conv')(x)
+    x = layers.BatchNormalization(name=name + '_2_bn')(x)
+    x = layers.Activation('relu', name=name + '_2_relu')(x)
+
+    x = layers.Conv2D(4 * filters, 1, name=name + '_3_conv')(x)
+    x = layers.Add(name=name + '_out')([shortcut, x])
+    return x
+
+
+def stack2(x, filters, blocks, stride1=2, name=None):
+    x = block2(x, filters, conv_shortcut=True, name=name + '_block1')
+    for i in range(2, blocks):
+        x = block2(x, filters, name=name + '_block' + str(i))
+    x = block2(x, filters, stride=stride1, name=name + '_block' + str(blocks))
+    return x
+
+
+def ResNet50V2(include_top=True,  # 是否包含位于网络顶部的全连接层
+               preact=True,  # 是否使用预激活
+               use_bias=True,  # 是否对卷积层使用偏置
+               weights='imagenet',
+               input_tensor=None,  # 可选的keras张量，用作模型的图像输入
+               input_shape=None,
+               pooling=None,
+               classes=1000,  # 用于分类图像的可选类数
+               classifier_activation='softmax'):# 分类层激活函数
+    img_input = layers.Input(shape=input_shape)
+    x = layers.ZeroPadding2D(padding=((3, 3), (3, 3)), name='conv1_pad')(img_input)
+    x = layers.Conv2D(64, 7, strides=2, use_bias=use_bias, name='conv1_conv')(x)
+
+    if not preact:
+        x = layers.BatchNormalization(name='conv1_bn')(x)
+        x = layers.Activation('relu', name='conv1_relu')(x)
+
+    x = layers.ZeroPadding2D(padding=((1, 1), (1, 1)), name='pool1_pad')(x)
+    x = layers.MaxPooling2D(3, strides=2, name='pool1_pool')(x)
+
+    x = stack2(x, 64, 3, name='conv2')
+    x = stack2(x, 128, 4, name='conv3')
+    x = stack2(x, 256, 6, name='conv4')
+    x = stack2(x, 512, 3, stride1=1, name='conv5')
+
+    if preact:
+        x = layers.BatchNormalization(name='post_bn')(x)
+        x = layers.Activation('relu', name='post_relu')(x)
+    if include_top:
+        x = layers.GlobalAveragePooling2D(name='avg_pool')(x)
+        x = layers.Dense(classes, activation=classifier_activation, name='predictions')(x)
+    else:
+        if pooling == 'avg':
+            x = layers.GlobalAveragePooling2D(name='avg_pool')(x)
+        elif pooling == 'max':
+            x = layers.GlobalMaxPooling2D(name='max_pool')(x)
+
+    model = Model(img_input, x)
+    return model
+
+
+if __name__ == '__main__':
+    model = ResNet50V2(input_shape=(224, 224, 3))
+    model.summary()
+    plot_model(model,to_file='../img/resnet50v2.png')

经典网络/ResNetV2/test/resnetv2.py

+import tensorflow as tf
+import tensorflow.keras.layers as layers
+from tensorflow.keras.models import Model
+from plot_model import plot_model
+
+"""A residual block.
+  Arguments:
+      x: input tensor.
+      filters: integer, filters of the bottleneck layer.
+      kernel_size: default 3, kernel size of the bottleneck layer.
+      stride: default 1, stride of the first layer.
+      conv_shortcut: default False, use convolution shortcut if True,
+        otherwise identity shortcut.
+      name: string, block label.
+  Returns:
+    Output tensor for the residual block.
+  """
+def block2(x, filters, kernel_size=3, stride=1, conv_shortcut=False, name=None):
+    preact = layers.BatchNormalization(name=name + '_preact_bn')(x)
+    preact = layers.Activation('relu', name=name + '_preact_relu')(preact)
+
+    if conv_shortcut:
+        shortcut = layers.Conv2D(4 * filters, 1, strides=stride, name=name + '_0_conv')(preact)
+    else:
+        shortcut = layers.MaxPooling2D(1, strides=stride)(x) if stride > 1 else x
+
+    x = layers.Conv2D(filters, 1, strides=1, use_bias=False, name=name + '_1_conv')(preact)
+    x = layers.BatchNormalization(name=name + '_1_bn')(x)
+    x = layers.Activation('relu', name=name + '_1_relu')(x)
+
+    x = layers.ZeroPadding2D(padding=((1, 1), (1, 1)), name=name + '_2_pad')(x)
+    x = layers.Conv2D(filters,
+                      kernel_size,
+                      strides=stride,
+                      use_bias=False,
+                      name=name + '_2_conv')(x)
+    x = layers.BatchNormalization(name=name + '_2_bn')(x)
+    x = layers.Activation('relu', name=name + '_2_relu')(x)
+
+    x = layers.Conv2D(4 * filters, 1, name=name + '_3_conv')(x)
+    x = layers.Add(name=name + '_out')([shortcut, x])
+    return x
+
+
+def stack2(x, filters, blocks, stride1=2, name=None):
+    x = block2(x, filters, conv_shortcut=True, name=name + '_block1')
+    for i in range(2, blocks):
+        x = block2(x, filters, name=name + '_block' + str(i))
+    x = block2(x, filters, stride=stride1, name=name + '_block' + str(blocks))
+    return x
+
+
+def ResNet50V2(include_top=True,  # 是否包含位于网络顶部的全连接层
+               preact=True,  # 是否使用预激活
+               use_bias=True,  # 是否对卷积层使用偏置
+               weights='imagenet',
+               input_tensor=None,  # 可选的keras张量，用作模型的图像输入
+               input_shape=None,
+               pooling=None,
+               classes=1000,  # 用于分类图像的可选类数
+               classifier_activation='softmax'):# 分类层激活函数
+    img_input = layers.Input(shape=input_shape)
+    x = layers.ZeroPadding2D(padding=((3, 3), (3, 3)), name='conv1_pad')(img_input)
+    x = layers.Conv2D(64, 7, strides=2, use_bias=use_bias, name='conv1_conv')(x)
+
+    if not preact:
+        x = layers.BatchNormalization(name='conv1_bn')(x)
+        x = layers.Activation('relu', name='conv1_relu')(x)
+
+    x = layers.ZeroPadding2D(padding=((1, 1), (1, 1)), name='pool1_pad')(x)
+    x = layers.MaxPooling2D(3, strides=2, name='pool1_pool')(x)
+
+    x = stack2(x, 64, 3, name='conv2')
+    x = stack2(x, 128, 4, name='conv3')
+    x = stack2(x, 256, 6, name='conv4')
+    x = stack2(x, 512, 3, stride1=1, name='conv5')
+
+    if preact:
+        x = layers.BatchNormalization(name='post_bn')(x)
+        x = layers.Activation('relu', name='post_relu')(x)
+    if include_top:
+        x = layers.GlobalAveragePooling2D(name='avg_pool')(x)
+        x = layers.Dense(classes, activation=classifier_activation, name='predictions')(x)
+    else:
+        if pooling == 'avg':
+            x = layers.GlobalAveragePooling2D(name='avg_pool')(x)
+        elif pooling == 'max':
+            x = layers.GlobalMaxPooling2D(name='max_pool')(x)
+
+    model = Model(img_input, x)
+    return model
+
+
+if __name__ == '__main__':
+    model = ResNet50V2(input_shape=(224, 224, 3))
+    model.summary()
+    plot_model(model,to_file='../img/resnet50v2.png')

经典网络/SKNet注意力机制/SKNet-keras.py

+# -*- coding: utf-8 -*-
+import tensorflow as tf
+
+from tensorflow.keras import layers
+from plot_model import plot_model
+
+def SKConv(M=2, r=16, L=32, G=32, name='skconv'):
+
+  def wrapper(inputs):
+    inputs_shape = tf.shape(inputs)
+    b, h, w = inputs_shape[0], inputs_shape[1], inputs_shape[2]
+    filters = inputs.get_shape().as_list()[-1]
+    d = max(filters//r, L)
+
+    x = inputs
+
+    xs = []
+    for m in range(1, M+1):
+      if G == 1:
+        _x = layers.Conv2D(filters, 3, dilation_rate=m, padding='same',
+                          use_bias=False, name=name+'_conv%d'%m)(x)
+      else:
+        c = filters // G
+        _x = layers.DepthwiseConv2D(3, dilation_rate=m, depth_multiplier=c, padding='same',
+                                   use_bias=False, name=name+'_conv%d'%m)(x)
+
+        _x = layers.Reshape([h, w, G, c, c], name=name+'_conv%d_reshape1'%m)(_x)
+        _x = layers.Lambda(lambda x: tf.reduce_sum(x, axis=-1),
+                          output_shape=[b, h, w, G, c],
+                          name=name+'_conv%d_sum'%m)(_x)
+        _x = layers.Reshape([h, w, filters],
+                           name=name+'_conv%d_reshape2'%m)(_x)
+
+
+      _x = layers.BatchNormalization(name=name+'_conv%d_bn'%m)(_x)
+      _x = layers.Activation('relu', name=name+'_conv%d_relu'%m)(_x)
+
+      xs.append(_x)
+
+    U = layers.Add(name=name+'_add')(xs)
+    s = layers.Lambda(lambda x: tf.reduce_mean(x, axis=[1,2], keepdims=True),
+                      output_shape=[b, 1, 1, filters],
+                      name=name+'_gap')(U)
+
+    z = layers.Conv2D(d, 1, name=name+'_fc_z')(s)
+    z = layers.BatchNormalization(name=name+'_fc_z_bn')(z)
+    z = layers.Activation('relu', name=name+'_fc_z_relu')(z)
+
+    x = layers.Conv2D(filters*M, 1, name=name+'_fc_x')(z)
+    x = layers.Reshape([1, 1, filters, M],name=name+'_reshape')(x)
+    scale = layers.Softmax(name=name+'_softmax')(x)
+
+    x = layers.Lambda(lambda x: tf.stack(x, axis=-1),
+                      output_shape=[b, h, w, filters, M],
+                      name=name+'_stack')(xs) # b, h, w, c, M
+    x = Axpby(name=name+'_axpby')([scale, x])
+
+    return x
+  return wrapper
+
+class Axpby(layers.Layer):
+  """
+  Do this:
+    F = a * X + b * Y + ...
+    Shape info:
+      a:  B x 1 x 1 x C
+      X:  B x H x W x C
+      b:  B x 1 x 1 x C
+      Y:  B x H x W x C
+      ...
+      F:  B x H x W x C
+  """
+  def __init__(self, **kwargs):
+        super(Axpby, self).__init__(**kwargs)
+
+  def build(self, input_shape):
+        super(Axpby, self).build(input_shape)  # Be sure to call this at the end
+
+  def call(self, inputs):
+    """ scale: [B, 1, 1, C, M]
+        x: [B, H, W, C, M]
+    """
+    scale, x = inputs
+    f = tf.multiply(scale, x, name='product')
+    f = tf.reduce_sum(f, axis=-1, name='sum')
+    return f
+
+  def compute_output_shape(self, input_shape):
+    return input_shape[0:4]
+
+
+if __name__ == '__main__':
+  from tensorflow.keras.layers import Input
+  from tensorflow.keras.models import Model
+
+  inputs = Input([224, 224, 32])
+  x = SKConv(3, G=1)(inputs)
+
+  m = Model(inputs, x)
+  m.summary()
+  plot_model(m,to_file='img/SKNet.png')
+
+  import numpy as np
+
+  X = np.random.random([2, 224, 224, 32]).astype(np.float32)
+  y = m.predict(X)