如果原理你已经了解，请直接到跳转ResNet50实现：卷积神经网络 第三周作业：Residual Networks± v1
你或许看过这篇访问量过12万的博客ResNet解析，但该博客的第一小节ResNet和吴恩达的叙述完全不同，因此博主对这篇博文持怀疑态度，你可以在这篇博文最下面找到提出该网络的论文链接，这篇博文可以作为研读这篇论文的基础。
ResNet = Residual Network
所有非残差网络都被称为平凡网络，这是一个原论文提出来的相对而言的概念。
残差网络是2015年由著名的Researcher Kaiming He(何凯明)提出的深度卷积网络，一经出世，便在ImageNet中斩获图像分类、检测、定位三项的冠军。 残差网络更容易优化，并且能够通过增加相当的深度来提高准确率。核心是解决了增加深度带来的副作用（退化问题），这样能够通过单纯地增加网络深度，来提高网络性能。
Motivation of ResNet
理论上来说，越深的深度神经网络能够计算出的特征越丰富，越能取得良好的效果，更深的神经网络的唯一缺点仅仅是你需要训练的参数十分庞大，导致其需要大量的计算资源。
但实际上，随着网络的加深，你会发现你梯度的大小(范数)急剧下降，这被称为梯度消失，这会导致学习速率非常缓慢。在极少数情况下也会出现梯度急剧上升，即梯度爆炸现象。表现在训练集上的准确度相较于浅层网络不但没有提高，反而会下降。
残差网络就是一种为了解决网络加深梯度消失现象而提出的网络。
What does Residual mean ?
假设经过网络中某几层所隐含的映射是H(X)H(X)H(X)，其中X表示这几层网络的首层输入。如果多个非线性层表示一个足够复杂的Hypothesis，那么H(X)等价于一个同样逐渐逼近该Hypothesis的残差函数(residual function)F(X)=H(X)−XF(X) = H(X) - XF(X)=H(X)−X, 原函数可以表示为F(X) XF(X) XF(X) X。H(X)H(X)H(X)和F(X)F(X)F(X)本质上都是对Hypothesis的一种逼近(近似)。
(以上译自论文原文)
基于该原理，ResNet中提出了2种映射，恒等映射(identity mapping)和残差映射(residual mapping), 恒等映射就是上图中跳过2层权重而把X直接送到后2层relu部分的映射，残差映射指平凡网络原来的部分。之所以称为恒等，因为你跳过了权重层，没有经过任何计算，即G(X)=XG(X)=XG(X)=X。
吴恩达在视频里对此的解释是，我们直接取某层之后的输出X作为输入，直接跳过一些连续的网络层，送到后面某层的relu之前。这么做使得一个Residual Block很容易的学习(因为它只做了一个relu操作)。
当然这需要满足X的维度和relu的维度一致。上图给出如何在维度不一致情况下通过Ws×alW_stimes a^{l}Ws×al使得维度一致，WsW_sWs可以仅仅是一个用0填充的矩阵，或者是需要学习的参数矩阵。
Shortcut Connection & Residual Block
吴恩达课程所提供的Residual，进行了简化，下面是原论文里的详细表示：
如图，曲线即表示了Shortcut Connection(近道连接)，它跳过了2个权重层(抄了近道)。平凡网络的一部分层加上shortcut connection即构成了一个Residual Block。shortcut使得每一个残差块很容易地学习到恒等映射函数，并且在反向传播时使得梯度直接传播到更浅的层。
在平凡网络上多次使用Residual Block，就形成了Residual Network。当然使用多少次，在什么网络的什么位置使用，就需要高深的洞察力和对深度神经网络的充分了解，目前大家参考的就是原论文和相近论文，改变这些也可以构造出新的近似网络。
论文里提出的2种Residual Block。
其中左边的residual block保留了输入的dimension，而右边则是一个"bottleneck design"，第一层把输入256维降低到64维，然后在第三层回复到256维，而shortcut/skip connection跳过这三层直接把输入送到了relu部分。
ResNet50
ResNet即共50层的参差网络，其中没有需要训练的参数的层，比如pooling layer，不参与计数。
原论文提出的常见的几种参差网络，主要是层数不同，50层和101层是最常见的。
50层的ResNet包含了Identity block(恒等块)和convolutional block(卷积块)2种结构，如下所示。
Identity block. Skip connection “skips over” 3 layers
The convolutional block
2种结构的主要差别是shortcut connection上是否进行了卷积操作。
以这两种模块为构建搭建的ResNet50如下图所示：
ResNet大致包括了5个Stage，或者叫做5种参数不同的卷积阶段，如上图所示。
(注：原论文把max pooling作为Stage 2的起始阶段)
filter of size 是三个卷积块的filter数目，而不是卷积核大小f，参数f如上表中50层ResNet那列所示，下面也有说明。
- Zero-padding pads the input with a pad of (3,3)
- Stage 1:
- The 2D Convolution has 64 filters of shape (7,7) and uses a stride of (2,2). Its name is “conv1”.
- BatchNorm is applied to the channels axis of the input.
- MaxPooling uses a (3,3) window and a (2,2) stride.
- Stage 2:
- The convolutional block uses three set of filters of size 64,64,256, “f” is 3, “s” is 1 and the block is “a”.
- The 2 identity blocks use three set of filters of size 64,64,256, “f” is 3 and the blocks are “b” and “c”.
- Stage 3:
- The convolutional block uses three set of filters of size 128,128,512, “f” is 3, “s” is 2 and the block is “a”.
- The 3 identity blocks use three set of filters of size 128,128,512, “f” is 3 and the blocks are “b”, “c” and “d”.
- Stage 4:
- The convolutional block uses three set of filters of size 256, 256, 1024, “f” is 3, “s” is 2 and the block is “a”.
- The 5 identity blocks use three set of filters of size 256, 256, 1024, “f” is 3 and the blocks are “b”, “c”, “d”, “e” and “f”.
- Stage 5:
- The convolutional block uses three set of filters of size 512, 512, 2048, “f” is 3, “s” is 2 and the block is “a”.
- The 2 identity blocks use three set of filters of size 256, 256, 2048, “f” is 3 and the blocks are “b” and “c”.
- The 2D Average Pooling uses a window of shape (2,2) and its name is “avg_pool”.
Keras实现
这个实现即为吴恩达深度学习系列视频的作业，如果你想完全掌握的话，强烈建议你参考这篇包含了作业完整过程和说明的博文:
卷积神经网络 第三周作业：Residual Networks± v1
下面是代码部分：
导入相应的库：
代码语言：javascript复制import numpy as np
import tensorflow as tf
from keras import layers
from keras.layers import Input, Add, Dense, Activation, ZeroPadding2D, BatchNormalization, Flatten, Conv2D, AveragePooling2D, MaxPooling2D, GlobalMaxPooling2D
from keras.models import Model, load_model
from keras.preprocessing import image
from keras.utils import layer_utils
from keras.utils.data_utils import get_file
from keras.applications.imagenet_utils import preprocess_input
# import pydot
# from IPython.display import SVG
from keras.utils.vis_utils import model_to_dot
from keras.utils import plot_model
from resnets_utils import *
from keras.initializers import glorot_uniform
import scipy.misc
from matplotlib.pyplot import imshow
%matplotlib inline
import keras.backend as K
K.set_image_data_format('channels_last')
K.set_learning_phase(1)恒等块
代码语言：javascript复制# GRADED FUNCTION: identity_block
def identity_block(X, f, filters, stage, block):
"""
Implementation of the identity block as defined in Figure 4
Arguments:
X -- input tensor of shape (m, n_H_prev, n_W_prev, n_C_prev)
f -- integer, specifying the shape of the middle CONV's window for the main path
filters -- python list of integers, defining the number of filters in the CONV layers of the main path
stage -- integer, used to name the layers, depending on their position in the network
block -- string/character, used to name the layers, depending on their position in the network
Returns:
X -- output of the identity block, tensor of shape (n_H, n_W, n_C)
"""
# defining name basis
conv_name_base = "res" str(stage) block "_branch"
bn_name_base = "bn" str(stage) block "_branch"
# Retrieve Filters
F1, F2, F3 = filters
# Save the input value. You'll need this later to add back to the main path.
X_shortcut = X
# First component of main path
X = Conv2D(filters=F1, kernel_size=(1, 1), strides=(1, 1), padding="valid",
name=conv_name_base "2a", kernel_initializer=glorot_uniform(seed=0))(X)
#valid mean no padding / glorot_uniform equal to Xaiver initialization - Steve
X = BatchNormalization(axis=3, name=bn_name_base "2a")(X)
X = Activation("relu")(X)
### START CODE HERE ###
# Second component of main path (≈3 lines)
X = Conv2D(filters=F2, kernel_size=(f, f), strides=(1, 1), padding="same",
name=conv_name_base "2b", kernel_initializer=glorot_uniform(seed=0))(X)
X = BatchNormalization(axis=3, name=bn_name_base "2b")(X)
X = Activation("relu")(X)
# Third component of main path (≈2 lines)
# Final step: Add shortcut value to main path, and pass it through a RELU activation (≈2 lines)
X = Conv2D(filters=F3, kernel_size=(1, 1), strides=(1, 1), padding="valid",
name=conv_name_base "2c", kernel_initializer=glorot_uniform(seed=0))(X)
X = BatchNormalization(axis=3, name=bn_name_base "2c")(X)
X = Add()([X, X_shortcut])
X = Activation("relu")(X)
### END CODE HERE ###
return X卷积块
代码语言：javascript复制# GRADED FUNCTION: convolutional_block
def convolutional_block(X, f, filters, stage, block, s = 2):
"""
Implementation of the convolutional block as defined in Figure 4
Arguments:
X -- input tensor of shape (m, n_H_prev, n_W_prev, n_C_prev)
f -- integer, specifying the shape of the middle CONV's window for the main path
filters -- python list of integers, defining the number of filters in the CONV layers of the main path
stage -- integer, used to name the layers, depending on their position in the network
block -- string/character, used to name the layers, depending on their position in the network
s -- Integer, specifying the stride to be used
Returns:
X -- output of the convolutional block, tensor of shape (n_H, n_W, n_C)
"""
# defining name basis
conv_name_base = 'res' str(stage) block '_branch'
bn_name_base = 'bn' str(stage) block '_branch'
# Retrieve Filters
F1, F2, F3 = filters
# Save the input value
X_shortcut = X
##### MAIN PATH #####
# First component of main path
X = Conv2D(F1, (1, 1), strides = (s,s), name = conv_name_base '2a', padding='valid', kernel_initializer = glorot_uniform(seed=0))(X)
X = BatchNormalization(axis = 3, name = bn_name_base '2a')(X)
X = Activation('relu')(X)
### START CODE HERE ###
# Second component of main path (≈3 lines)
X = Conv2D(F2, (f, f), strides = (1, 1), name = conv_name_base '2b',padding='same', kernel_initializer = glorot_uniform(seed=0))(X)
X = BatchNormalization(axis = 3, name = bn_name_base '2b')(X)
X = Activation('relu')(X)
# Third component of main path (≈2 lines)
X = Conv2D(F3, (1, 1), strides = (1, 1), name = conv_name_base '2c',padding='valid', kernel_initializer = glorot_uniform(seed=0))(X)
X = BatchNormalization(axis = 3, name = bn_name_base '2c')(X)
##### SHORTCUT PATH #### (≈2 lines)
X_shortcut = Conv2D(F3, (1, 1), strides = (s, s), name = conv_name_base '1',padding='valid', kernel_initializer = glorot_uniform(seed=0))(X_shortcut)
X_shortcut = BatchNormalization(axis = 3, name = bn_name_base '1')(X_shortcut)
# Final step: Add shortcut value to main path, and pass it through a RELU activation (≈2 lines)
X = layers.add([X, X_shortcut])
X = Activation('relu')(X)
### END CODE HERE ###
return XResNet本尊
代码语言：javascript复制# GRADED FUNCTION: ResNet50
def ResNet50(input_shape = (64, 64, 3), classes = 6):
"""
Implementation of the popular ResNet50 the following architecture:
CONV2D -> BATCHNORM -> RELU -> MAXPOOL -> CONVBLOCK -> IDBLOCK*2 -> CONVBLOCK -> IDBLOCK*3
-> CONVBLOCK -> IDBLOCK*5 -> CONVBLOCK -> IDBLOCK*2 -> AVGPOOL -> TOPLAYER
Arguments:
input_shape -- shape of the images of the dataset
classes -- integer, number of classes
Returns:
model -- a Model() instance in Keras
"""
# Define the input as a tensor with shape input_shape
X_input = Input(input_shape)
# Zero-Padding
X = ZeroPadding2D((3, 3))(X_input)
# Stage 1
X = Conv2D(filters=64, kernel_size=(7, 7), strides=(2, 2), name="conv",
kernel_initializer=glorot_uniform(seed=0))(X)
X = BatchNormalization(axis=3, name="bn_conv1")(X)
X = Activation("relu")(X)
X = MaxPooling2D(pool_size=(3, 3), strides=(2, 2))(X)
# Stage 2
X = convolutional_block(X, f=3, filters=[64, 64, 256], stage=2, block="a", s=1)
X = identity_block(X, f=3, filters=[64, 64, 256], stage=2, block="b")
X = identity_block(X, f=3, filters=[64, 64, 256], stage=2, block="c")
### START CODE HERE ###
# Stage 3 (≈4 lines)
# The convolutional block uses three set of filters of size [128,128,512], "f" is 3, "s" is 2 and the block is "a".
# The 3 identity blocks use three set of filters of size [128,128,512], "f" is 3 and the blocks are "b", "c" and "d".
X = convolutional_block(X, f=3, filters=[128, 128, 512], stage=3, block="a", s=1)
X = identity_block(X, f=3, filters=[128, 128, 512], stage=3, block="b")
X = identity_block(X, f=3, filters=[128, 128, 512], stage=3, block="c")
X = identity_block(X, f=3, filters=[128, 128, 512], stage=3, block="d")
# Stage 4 (≈6 lines)
# The convolutional block uses three set of filters of size [256, 256, 1024], "f" is 3, "s" is 2 and the block is "a".
# The 5 identity blocks use three set of filters of size [256, 256, 1024], "f" is 3 and the blocks are "b", "c", "d", "e" and "f".
X = convolutional_block(X, f=3, filters=[256, 256, 1024], stage=4, block="a", s=2)
X = identity_block(X, f=3, filters=[256, 256, 1024], stage=4, block="b")
X = identity_block(X, f=3, filters=[256, 256, 1024], stage=4, block="c")
X = identity_block(X, f=3, filters=[256, 256, 1024], stage=4, block="d")
X = identity_block(X, f=3, filters=[256, 256, 1024], stage=4, block="e")
X = identity_block(X, f=3, filters=[256, 256, 1024], stage=4, block="f")
# Stage 5 (≈3 lines)
# The convolutional block uses three set of filters of size [512, 512, 2048], "f" is 3, "s" is 2 and the block is "a".
# The 2 identity blocks use three set of filters of size [256, 256, 2048], "f" is 3 and the blocks are "b" and "c".
X = convolutional_block(X, f=3, filters=[512, 512, 2048], stage=5, block="a", s=2)
X = identity_block(X, f=3, filters=[512, 512, 2048], stage=5, block="b")
X = identity_block(X, f=3, filters=[512, 512, 2048], stage=5, block="c")
# filters should be [256, 256, 2048], but it fail to be graded. Use [512, 512, 2048] to pass the grading
# AVGPOOL (≈1 line). Use "X = AveragePooling2D(...)(X)"
# The 2D Average Pooling uses a window of shape (2,2) and its name is "avg_pool".
X = AveragePooling2D(pool_size=(2, 2), padding="same")(X)
### END CODE HERE ###
# output layer
X = Flatten()(X)
X = Dense(classes, activation="softmax", name="fc" str(classes), kernel_initializer=glorot_uniform(seed=0))(X)
# Create model
model = Model(inputs=X_input, outputs=X, name="ResNet50")
return model定义模型
代码语言：javascript复制model = ResNet50(input_shape=(64, 64, 3), classes=6)编译模型
代码语言：javascript复制model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])导入训练数据
代码语言：javascript复制X_train_orig, Y_train_orig, X_test_orig, Y_test_orig, classes = load_dataset()
# Normalize image vectors
X_train = X_train_orig/255.
X_test = X_test_orig/255.
"""
def convert_to_one_hot(Y, C):
Y = np.eye(C)[Y.reshape(-1)].T
return Y
"""
# Convert training and test labels to one hot matrices
Y_train = convert_to_one_hot(Y_train_orig, 6).T
Y_test = convert_to_one_hot(Y_test_orig, 6).T
print ("number of training examples = " str(X_train.shape[0]))
print ("number of test examples = " str(X_test.shape[0]))
print ("X_train shape: " str(X_train.shape))
print ("Y_train shape: " str(Y_train.shape))
print ("X_test shape: " str(X_test.shape))
print ("Y_test shape: " str(Y_test.shape))number of training examples = 1080
number of test examples = 120
X_train shape: (1080, 64, 64, 3)
Y_train shape: (1080, 6)
X_test shape: (120, 64, 64, 3)
Y_test shape: (120, 6)训练模型 (可能耗费比较长的时间)
代码语言：javascript复制model.fit(X_train, Y_train, epochs = 20, batch_size = 32)Epoch 1/20
1080/1080 [==============================] - 268s 248ms/step - loss: 2.9721 - acc: 0.2898
Epoch 2/20
1080/1080 [==============================] - 270s 250ms/step - loss: 1.8968 - acc: 0.3639
Epoch 3/20
1080/1080 [==============================] - 268s 248ms/step - loss: 1.5796 - acc: 0.4463
Epoch 4/20
1080/1080 [==============================] - 251s 233ms/step - loss: 1.2796 - acc: 0.5213
Epoch 5/20
1080/1080 [==============================] - 260s 241ms/step - loss: 0.9278 - acc: 0.6722
Epoch 6/20
1080/1080 [==============================] - 261s 242ms/step - loss: 0.7286 - acc: 0.7315
Epoch 7/20
1080/1080 [==============================] - 258s 239ms/step - loss: 0.4950 - acc: 0.8324
Epoch 8/20
1080/1080 [==============================] - 261s 241ms/step - loss: 0.3646 - acc: 0.8889
Epoch 9/20
1080/1080 [==============================] - 258s 238ms/step - loss: 0.3135 - acc: 0.9019
Epoch 10/20
1080/1080 [==============================] - 255s 237ms/step - loss: 0.1291 - acc: 0.9639
Epoch 11/20
1080/1080 [==============================] - 253s 235ms/step - loss: 0.0814 - acc: 0.9704
Epoch 12/20
1080/1080 [==============================] - 260s 240ms/step - loss: 0.0901 - acc: 0.9685
Epoch 13/20
1080/1080 [==============================] - 260s 240ms/step - loss: 0.0848 - acc: 0.9694
Epoch 14/20
1080/1080 [==============================] - 261s 242ms/step - loss: 0.0740 - acc: 0.9741
Epoch 15/20
1080/1080 [==============================] - 258s 239ms/step - loss: 0.0488 - acc: 0.9833
Epoch 16/20
1080/1080 [==============================] - 260s 241ms/step - loss: 0.0257 - acc: 0.9981
Epoch 17/20
1080/1080 [==============================] - 259s 240ms/step - loss: 0.0029 - acc: 1.0000
Epoch 18/20
1080/1080 [==============================] - 260s 241ms/step - loss: 0.0014 - acc: 1.0000
Epoch 19/20
1080/1080 [==============================] - 257s 238ms/step - loss: 8.9325e-04 - acc: 1.0000
Epoch 20/20
1080/1080 [==============================] - 255s 236ms/step - loss: 6.9667e-04 - acc: 1.0000
<keras.callbacks.History at 0x21761b11710>评估准确度
代码语言：javascript复制preds = model.evaluate(X_test, Y_test)
print ("Loss = " str(preds[0]))
print ("Test Accuracy = " str(preds[1]))120/120 [==============================] - 6s 49ms/step
Loss = 0.11732131155828635
Test Accuracy = 0.9666666666666667打印出模型的结构
代码语言：javascript复制model.summary()__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_3 (InputLayer) (None, 64, 64, 3) 0
__________________________________________________________________________________________________
zero_padding2d_3 (ZeroPadding2D (None, 70, 70, 3) 0 input_3[0][0]
__________________________________________________________________________________________________
conv (Conv2D) (None, 32, 32, 64) 9472 zero_padding2d_3[0][0]
__________________________________________________________________________________________________
bn_conv1 (BatchNormalization) (None, 32, 32, 64) 256 conv[0][0]
__________________________________________________________________________________________________
activation_96 (Activation) (None, 32, 32, 64) 0 bn_conv1[0][0]
__________________________________________________________________________________________________
max_pooling2d_3 (MaxPooling2D) (None, 15, 15, 64) 0 activation_96[0][0]
__________________________________________________________________________________________________
res2a_branch2a (Conv2D) (None, 15, 15, 64) 4160 max_pooling2d_3[0][0]
__________________________________________________________________________________________________
bn2a_branch2a (BatchNormalizati (None, 15, 15, 64) 256 res2a_branch2a[0][0]
__________________________________________________________________________________________________
activation_97 (Activation) (None, 15, 15, 64) 0 bn2a_branch2a[0][0]
__________________________________________________________________________________________________
res2a_branch2b (Conv2D) (None, 15, 15, 64) 36928 activation_97[0][0]
__________________________________________________________________________________________________
bn2a_branch2b (BatchNormalizati (None, 15, 15, 64) 256 res2a_branch2b[0][0]
__________________________________________________________________________________________________
activation_98 (Activation) (None, 15, 15, 64) 0 bn2a_branch2b[0][0]
__________________________________________________________________________________________________
res2a_branch2c (Conv2D) (None, 15, 15, 256) 16640 activation_98[0][0]
__________________________________________________________________________________________________
res2a_branch1 (Conv2D) (None, 15, 15, 256) 16640 max_pooling2d_3[0][0]
__________________________________________________________________________________________________
bn2a_branch2c (BatchNormalizati (None, 15, 15, 256) 1024 res2a_branch2c[0][0]
__________________________________________________________________________________________________
bn2a_branch1 (BatchNormalizatio (None, 15, 15, 256) 1024 res2a_branch1[0][0]
__________________________________________________________________________________________________
add_32 (Add) (None, 15, 15, 256) 0 bn2a_branch2c[0][0]
bn2a_branch1[0][0]
__________________________________________________________________________________________________
activation_99 (Activation) (None, 15, 15, 256) 0 add_32[0][0]
__________________________________________________________________________________________________
res2b_branch2a (Conv2D) (None, 15, 15, 64) 16448 activation_99[0][0]
__________________________________________________________________________________________________
bn2b_branch2a (BatchNormalizati (None, 15, 15, 64) 256 res2b_branch2a[0][0]
__________________________________________________________________________________________________
activation_100 (Activation) (None, 15, 15, 64) 0 bn2b_branch2a[0][0]
__________________________________________________________________________________________________
res2b_branch2b (Conv2D) (None, 15, 15, 64) 36928 activation_100[0][0]
__________________________________________________________________________________________________
bn2b_branch2b (BatchNormalizati (None, 15, 15, 64) 256 res2b_branch2b[0][0]
__________________________________________________________________________________________________
activation_101 (Activation) (None, 15, 15, 64) 0 bn2b_branch2b[0][0]
__________________________________________________________________________________________________
res2b_branch2c (Conv2D) (None, 15, 15, 256) 16640 activation_101[0][0]
__________________________________________________________________________________________________
bn2b_branch2c (BatchNormalizati (None, 15, 15, 256) 1024 res2b_branch2c[0][0]
__________________________________________________________________________________________________
add_33 (Add) (None, 15, 15, 256) 0 bn2b_branch2c[0][0]
activation_99[0][0]
__________________________________________________________________________________________________
activation_102 (Activation) (None, 15, 15, 256) 0 add_33[0][0]
__________________________________________________________________________________________________
res2c_branch2a (Conv2D) (None, 15, 15, 64) 16448 activation_102[0][0]
__________________________________________________________________________________________________
bn2c_branch2a (BatchNormalizati (None, 15, 15, 64) 256 res2c_branch2a[0][0]
__________________________________________________________________________________________________
activation_103 (Activation) (None, 15, 15, 64) 0 bn2c_branch2a[0][0]
__________________________________________________________________________________________________
res2c_branch2b (Conv2D) (None, 15, 15, 64) 36928 activation_103[0][0]
__________________________________________________________________________________________________
bn2c_branch2b (BatchNormalizati (None, 15, 15, 64) 256 res2c_branch2b[0][0]
__________________________________________________________________________________________________
activation_104 (Activation) (None, 15, 15, 64) 0 bn2c_branch2b[0][0]
__________________________________________________________________________________________________
res2c_branch2c (Conv2D) (None, 15, 15, 256) 16640 activation_104[0][0]
__________________________________________________________________________________________________
bn2c_branch2c (BatchNormalizati (None, 15, 15, 256) 1024 res2c_branch2c[0][0]
__________________________________________________________________________________________________
add_34 (Add) (None, 15, 15, 256) 0 bn2c_branch2c[0][0]
activation_102[0][0]
__________________________________________________________________________________________________
activation_105 (Activation) (None, 15, 15, 256) 0 add_34[0][0]
__________________________________________________________________________________________________
res3a_branch2a (Conv2D) (None, 15, 15, 128) 32896 activation_105[0][0]
__________________________________________________________________________________________________
bn3a_branch2a (BatchNormalizati (None, 15, 15, 128) 512 res3a_branch2a[0][0]
__________________________________________________________________________________________________
activation_106 (Activation) (None, 15, 15, 128) 0 bn3a_branch2a[0][0]
__________________________________________________________________________________________________
res3a_branch2b (Conv2D) (None, 15, 15, 128) 147584 activation_106[0][0]
__________________________________________________________________________________________________
bn3a_branch2b (BatchNormalizati (None, 15, 15, 128) 512 res3a_branch2b[0][0]
__________________________________________________________________________________________________
activation_107 (Activation) (None, 15, 15, 128) 0 bn3a_branch2b[0][0]
__________________________________________________________________________________________________
res3a_branch2c (Conv2D) (None, 15, 15, 512) 66048 activation_107[0][0]
__________________________________________________________________________________________________
res3a_branch1 (Conv2D) (None, 15, 15, 512) 131584 activation_105[0][0]
__________________________________________________________________________________________________
bn3a_branch2c (BatchNormalizati (None, 15, 15, 512) 2048 res3a_branch2c[0][0]
__________________________________________________________________________________________________
bn3a_branch1 (BatchNormalizatio (None, 15, 15, 512) 2048 res3a_branch1[0][0]
__________________________________________________________________________________________________
add_35 (Add) (None, 15, 15, 512) 0 bn3a_branch2c[0][0]
bn3a_branch1[0][0]
__________________________________________________________________________________________________
activation_108 (Activation) (None, 15, 15, 512) 0 add_35[0][0]
__________________________________________________________________________________________________
res3b_branch2a (Conv2D) (None, 15, 15, 128) 65664 activation_108[0][0]
__________________________________________________________________________________________________
bn3b_branch2a (BatchNormalizati (None, 15, 15, 128) 512 res3b_branch2a[0][0]
__________________________________________________________________________________________________
activation_109 (Activation) (None, 15, 15, 128) 0 bn3b_branch2a[0][0]
__________________________________________________________________________________________________
res3b_branch2b (Conv2D) (None, 15, 15, 128) 147584 activation_109[0][0]
__________________________________________________________________________________________________
bn3b_branch2b (BatchNormalizati (None, 15, 15, 128) 512 res3b_branch2b[0][0]
__________________________________________________________________________________________________
activation_110 (Activation) (None, 15, 15, 128) 0 bn3b_branch2b[0][0]
__________________________________________________________________________________________________
res3b_branch2c (Conv2D) (None, 15, 15, 512) 66048 activation_110[0][0]
__________________________________________________________________________________________________
bn3b_branch2c (BatchNormalizati (None, 15, 15, 512) 2048 res3b_branch2c[0][0]
__________________________________________________________________________________________________
add_36 (Add) (None, 15, 15, 512) 0 bn3b_branch2c[0][0]
activation_108[0][0]
__________________________________________________________________________________________________
activation_111 (Activation) (None, 15, 15, 512) 0 add_36[0][0]
__________________________________________________________________________________________________
res3c_branch2a (Conv2D) (None, 15, 15, 128) 65664 activation_111[0][0]
__________________________________________________________________________________________________
bn3c_branch2a (BatchNormalizati (None, 15, 15, 128) 512 res3c_branch2a[0][0]
__________________________________________________________________________________________________
activation_112 (Activation) (None, 15, 15, 128) 0 bn3c_branch2a[0][0]
__________________________________________________________________________________________________
res3c_branch2b (Conv2D) (None, 15, 15, 128) 147584 activation_112[0][0]
__________________________________________________________________________________________________
bn3c_branch2b (BatchNormalizati (None, 15, 15, 128) 512 res3c_branch2b[0][0]
__________________________________________________________________________________________________
activation_113 (Activation) (None, 15, 15, 128) 0 bn3c_branch2b[0][0]
__________________________________________________________________________________________________
res3c_branch2c (Conv2D) (None, 15, 15, 512) 66048 activation_113[0][0]
__________________________________________________________________________________________________
bn3c_branch2c (BatchNormalizati (None, 15, 15, 512) 2048 res3c_branch2c[0][0]
__________________________________________________________________________________________________
add_37 (Add) (None, 15, 15, 512) 0 bn3c_branch2c[0][0]
activation_111[0][0]
__________________________________________________________________________________________________
activation_114 (Activation) (None, 15, 15, 512) 0 add_37[0][0]
__________________________________________________________________________________________________
res3d_branch2a (Conv2D) (None, 15, 15, 128) 65664 activation_114[0][0]
__________________________________________________________________________________________________
bn3d_branch2a (BatchNormalizati (None, 15, 15, 128) 512 res3d_branch2a[0][0]
__________________________________________________________________________________________________
activation_115 (Activation) (None, 15, 15, 128) 0 bn3d_branch2a[0][0]
__________________________________________________________________________________________________
res3d_branch2b (Conv2D) (None, 15, 15, 128) 147584 activation_115[0][0]
__________________________________________________________________________________________________
bn3d_branch2b (BatchNormalizati (None, 15, 15, 128) 512 res3d_branch2b[0][0]
__________________________________________________________________________________________________
activation_116 (Activation) (None, 15, 15, 128) 0 bn3d_branch2b[0][0]
__________________________________________________________________________________________________
res3d_branch2c (Conv2D) (None, 15, 15, 512) 66048 activation_116[0][0]
__________________________________________________________________________________________________
bn3d_branch2c (BatchNormalizati (None, 15, 15, 512) 2048 res3d_branch2c[0][0]
__________________________________________________________________________________________________
add_38 (Add) (None, 15, 15, 512) 0 bn3d_branch2c[0][0]
activation_114[0][0]
__________________________________________________________________________________________________
activation_117 (Activation) (None, 15, 15, 512) 0 add_38[0][0]
__________________________________________________________________________________________________
res4a_branch2a (Conv2D) (None, 8, 8, 256) 131328 activation_117[0][0]
__________________________________________________________________________________________________
bn4a_branch2a (BatchNormalizati (None, 8, 8, 256) 1024 res4a_branch2a[0][0]
__________________________________________________________________________________________________
activation_118 (Activation) (None, 8, 8, 256) 0 bn4a_branch2a[0][0]
__________________________________________________________________________________________________
res4a_branch2b (Conv2D) (None, 8, 8, 256) 590080 activation_118[0][0]
__________________________________________________________________________________________________
bn4a_branch2b (BatchNormalizati (None, 8, 8, 256) 1024 res4a_branch2b[0][0]
__________________________________________________________________________________________________
activation_119 (Activation) (None, 8, 8, 256) 0 bn4a_branch2b[0][0]
__________________________________________________________________________________________________
res4a_branch2c (Conv2D) (None, 8, 8, 1024) 263168 activation_119[0][0]
__________________________________________________________________________________________________
res4a_branch1 (Conv2D) (None, 8, 8, 1024) 525312 activation_117[0][0]
__________________________________________________________________________________________________
bn4a_branch2c (BatchNormalizati (None, 8, 8, 1024) 4096 res4a_branch2c[0][0]
__________________________________________________________________________________________________
bn4a_branch1 (BatchNormalizatio (None, 8, 8, 1024) 4096 res4a_branch1[0][0]
__________________________________________________________________________________________________
add_39 (Add) (None, 8, 8, 1024) 0 bn4a_branch2c[0][0]
bn4a_branch1[0][0]
__________________________________________________________________________________________________
activation_120 (Activation) (None, 8, 8, 1024) 0 add_39[0][0]
__________________________________________________________________________________________________
res4b_branch2a (Conv2D) (None, 8, 8, 256) 262400 activation_120[0][0]
__________________________________________________________________________________________________
bn4b_branch2a (BatchNormalizati (None, 8, 8, 256) 1024 res4b_branch2a[0][0]
__________________________________________________________________________________________________
activation_121 (Activation) (None, 8, 8, 256) 0 bn4b_branch2a[0][0]
__________________________________________________________________________________________________
res4b_branch2b (Conv2D) (None, 8, 8, 256) 590080 activation_121[0][0]
__________________________________________________________________________________________________
bn4b_branch2b (BatchNormalizati (None, 8, 8, 256) 1024 res4b_branch2b[0][0]
__________________________________________________________________________________________________
activation_122 (Activation) (None, 8, 8, 256) 0 bn4b_branch2b[0][0]
__________________________________________________________________________________________________
res4b_branch2c (Conv2D) (None, 8, 8, 1024) 263168 activation_122[0][0]
__________________________________________________________________________________________________
bn4b_branch2c (BatchNormalizati (None, 8, 8, 1024) 4096 res4b_branch2c[0][0]
__________________________________________________________________________________________________
add_40 (Add) (None, 8, 8, 1024) 0 bn4b_branch2c[0][0]
activation_120[0][0]
__________________________________________________________________________________________________
activation_123 (Activation) (None, 8, 8, 1024) 0 add_40[0][0]
__________________________________________________________________________________________________
res4c_branch2a (Conv2D) (None, 8, 8, 256) 262400 activation_123[0][0]
__________________________________________________________________________________________________
bn4c_branch2a (BatchNormalizati (None, 8, 8, 256) 1024 res4c_branch2a[0][0]
__________________________________________________________________________________________________
activation_124 (Activation) (None, 8, 8, 256) 0 bn4c_branch2a[0][0]
__________________________________________________________________________________________________
res4c_branch2b (Conv2D) (None, 8, 8, 256) 590080 activation_124[0][0]
__________________________________________________________________________________________________
bn4c_branch2b (BatchNormalizati (None, 8, 8, 256) 1024 res4c_branch2b[0][0]
__________________________________________________________________________________________________
activation_125 (Activation) (None, 8, 8, 256) 0 bn4c_branch2b[0][0]
__________________________________________________________________________________________________
res4c_branch2c (Conv2D) (None, 8, 8, 1024) 263168 activation_125[0][0]
__________________________________________________________________________________________________
bn4c_branch2c (BatchNormalizati (None, 8, 8, 1024) 4096 res4c_branch2c[0][0]
__________________________________________________________________________________________________
add_41 (Add) (None, 8, 8, 1024) 0 bn4c_branch2c[0][0]
activation_123[0][0]
__________________________________________________________________________________________________
activation_126 (Activation) (None, 8, 8, 1024) 0 add_41[0][0]
__________________________________________________________________________________________________
res4d_branch2a (Conv2D) (None, 8, 8, 256) 262400 activation_126[0][0]
__________________________________________________________________________________________________
bn4d_branch2a (BatchNormalizati (None, 8, 8, 256) 1024 res4d_branch2a[0][0]
__________________________________________________________________________________________________
activation_127 (Activation) (None, 8, 8, 256) 0 bn4d_branch2a[0][0]
__________________________________________________________________________________________________
res4d_branch2b (Conv2D) (None, 8, 8, 256) 590080 activation_127[0][0]
__________________________________________________________________________________________________
bn4d_branch2b (BatchNormalizati (None, 8, 8, 256) 1024 res4d_branch2b[0][0]
__________________________________________________________________________________________________
activation_128 (Activation) (None, 8, 8, 256) 0 bn4d_branch2b[0][0]
__________________________________________________________________________________________________
res4d_branch2c (Conv2D) (None, 8, 8, 1024) 263168 activation_128[0][0]
__________________________________________________________________________________________________
bn4d_branch2c (BatchNormalizati (None, 8, 8, 1024) 4096 res4d_branch2c[0][0]
__________________________________________________________________________________________________
add_42 (Add) (None, 8, 8, 1024) 0 bn4d_branch2c[0][0]
activation_126[0][0]
__________________________________________________________________________________________________
activation_129 (Activation) (None, 8, 8, 1024) 0 add_42[0][0]
__________________________________________________________________________________________________
res4e_branch2a (Conv2D) (None, 8, 8, 256) 262400 activation_129[0][0]
__________________________________________________________________________________________________
bn4e_branch2a (BatchNormalizati (None, 8, 8, 256) 1024 res4e_branch2a[0][0]
__________________________________________________________________________________________________
activation_130 (Activation) (None, 8, 8, 256) 0 bn4e_branch2a[0][0]
__________________________________________________________________________________________________
res4e_branch2b (Conv2D) (None, 8, 8, 256) 590080 activation_130[0][0]
__________________________________________________________________________________________________
bn4e_branch2b (BatchNormalizati (None, 8, 8, 256) 1024 res4e_branch2b[0][0]
__________________________________________________________________________________________________
activation_131 (Activation) (None, 8, 8, 256) 0 bn4e_branch2b[0][0]
__________________________________________________________________________________________________
res4e_branch2c (Conv2D) (None, 8, 8, 1024) 263168 activation_131[0][0]
__________________________________________________________________________________________________
bn4e_branch2c (BatchNormalizati (None, 8, 8, 1024) 4096 res4e_branch2c[0][0]
__________________________________________________________________________________________________
add_43 (Add) (None, 8, 8, 1024) 0 bn4e_branch2c[0][0]
activation_129[0][0]
__________________________________________________________________________________________________
activation_132 (Activation) (None, 8, 8, 1024) 0 add_43[0][0]
__________________________________________________________________________________________________
res5a_branch2a (Conv2D) (None, 4, 4, 512) 524800 activation_132[0][0]
__________________________________________________________________________________________________
bn5a_branch2a (BatchNormalizati (None, 4, 4, 512) 2048 res5a_branch2a[0][0]
__________________________________________________________________________________________________
activation_133 (Activation) (None, 4, 4, 512) 0 bn5a_branch2a[0][0]
__________________________________________________________________________________________________
res5a_branch2b (Conv2D) (None, 4, 4, 512) 2359808 activation_133[0][0]
__________________________________________________________________________________________________
bn5a_branch2b (BatchNormalizati (None, 4, 4, 512) 2048 res5a_branch2b[0][0]
__________________________________________________________________________________________________
activation_134 (Activation) (None, 4, 4, 512) 0 bn5a_branch2b[0][0]
__________________________________________________________________________________________________
res5a_branch2c (Conv2D) (None, 4, 4, 2048) 1050624 activation_134[0][0]
__________________________________________________________________________________________________
res5a_branch1 (Conv2D) (None, 4, 4, 2048) 2099200 activation_132[0][0]
__________________________________________________________________________________________________
bn5a_branch2c (BatchNormalizati (None, 4, 4, 2048) 8192 res5a_branch2c[0][0]
__________________________________________________________________________________________________
bn5a_branch1 (BatchNormalizatio (None, 4, 4, 2048) 8192 res5a_branch1[0][0]
__________________________________________________________________________________________________
add_44 (Add) (None, 4, 4, 2048) 0 bn5a_branch2c[0][0]
bn5a_branch1[0][0]
__________________________________________________________________________________________________
activation_135 (Activation) (None, 4, 4, 2048) 0 add_44[0][0]
__________________________________________________________________________________________________
res5b_branch2a (Conv2D) (None, 4, 4, 512) 1049088 activation_135[0][0]
__________________________________________________________________________________________________
bn5b_branch2a (BatchNormalizati (None, 4, 4, 512) 2048 res5b_branch2a[0][0]
__________________________________________________________________________________________________
activation_136 (Activation) (None, 4, 4, 512) 0 bn5b_branch2a[0][0]
__________________________________________________________________________________________________
res5b_branch2b (Conv2D) (None, 4, 4, 512) 2359808 activation_136[0][0]
__________________________________________________________________________________________________
bn5b_branch2b (BatchNormalizati (None, 4, 4, 512) 2048 res5b_branch2b[0][0]
__________________________________________________________________________________________________
activation_137 (Activation) (None, 4, 4, 512) 0 bn5b_branch2b[0][0]
__________________________________________________________________________________________________
res5b_branch2c (Conv2D) (None, 4, 4, 2048) 1050624 activation_137[0][0]
__________________________________________________________________________________________________
bn5b_branch2c (BatchNormalizati (None, 4, 4, 2048) 8192 res5b_branch2c[0][0]
__________________________________________________________________________________________________
add_45 (Add) (None, 4, 4, 2048) 0 bn5b_branch2c[0][0]
activation_135[0][0]
__________________________________________________________________________________________________
activation_138 (Activation) (None, 4, 4, 2048) 0 add_45[0][0]
__________________________________________________________________________________________________
res5c_branch2a (Conv2D) (None, 4, 4, 512) 1049088 activation_138[0][0]
__________________________________________________________________________________________________
bn5c_branch2a (BatchNormalizati (None, 4, 4, 512) 2048 res5c_branch2a[0][0]
__________________________________________________________________________________________________
activation_139 (Activation) (None, 4, 4, 512) 0 bn5c_branch2a[0][0]
__________________________________________________________________________________________________
res5c_branch2b (Conv2D) (None, 4, 4, 512) 2359808 activation_139[0][0]
__________________________________________________________________________________________________
bn5c_branch2b (BatchNormalizati (None, 4, 4, 512) 2048 res5c_branch2b[0][0]
__________________________________________________________________________________________________
activation_140 (Activation) (None, 4, 4, 512) 0 bn5c_branch2b[0][0]
__________________________________________________________________________________________________
res5c_branch2c (Conv2D) (None, 4, 4, 2048) 1050624 activation_140[0][0]
__________________________________________________________________________________________________
bn5c_branch2c (BatchNormalizati (None, 4, 4, 2048) 8192 res5c_branch2c[0][0]
__________________________________________________________________________________________________
add_46 (Add) (None, 4, 4, 2048) 0 bn5c_branch2c[0][0]
activation_138[0][0]
__________________________________________________________________________________________________
activation_141 (Activation) (None, 4, 4, 2048) 0 add_46[0][0]
__________________________________________________________________________________________________
average_pooling2d_3 (AveragePoo (None, 2, 2, 2048) 0 activation_141[0][0]
__________________________________________________________________________________________________
flatten_3 (Flatten) (None, 8192) 0 average_pooling2d_3[0][0]
__________________________________________________________________________________________________
fc6 (Dense) (None, 6) 49158 flatten_3[0][0]
==================================================================================================
Total params: 22,515,078
Trainable params: 22,465,030
Non-trainable params: 50,048
__________________________________________________________________________________________________参考：
- Deep Residual Learning for Image Recognition
- 吴恩达深度学习系列视频：卷积神经网络
- 卷积神经网络 第三周作业：Residual Networks± v1
