各位同学好，今天和大家分享一下如何使用 Tensorflow 复现轻量化神经网络 ShuffleNetV2。
为了能将神经网络模型用于移动端（手机）和终端（安防监控、无人驾驶）的实时计算，通常这些设备计算能力有限，因此我们需要减少模型参数量、减小计算量、更少的内存访问量、更少的能耗。MobileNet、ShuffleNet 等轻量化网络就非常适合于一些计算能力受限的设备，牺牲少量准确率换取更快的运算速度。
在之前的章节中，我详细介绍了 MobileNetV1 和 MobileNetV2 网络模型，今天介绍的 ShuffleNet 网络也会使用到 MobileNet 的深度可分离卷积。建议先看一下我的前两篇文章：
MobileNetV1：https://blog.csdn.net/dgvv4/article/details/123415708
MobileNetV2：https://blog.csdn.net/dgvv4/article/details/123417739
左图是各个神经网络模型的准确率--参数量散点图，越靠近左上角代表模型参数量少而且准确率高。右图是单位参数量对准确率的贡献，ShuffleNet 的计算量和参数量很少，但是计算的效率很高，计算速度很快。
???
1. ShuffleNetV1 创新点
在讲 ShuffleNetV2 之前我们需要对 ShuffleNetV1 有一定的了解，这里主要介绍ShuffleNetV1的两个核心内容。（1）分组1x1卷积；（2）通道重排。
1.1 分组卷积（Group?Convolution）
（1）在标准卷积中，输入有多少个通道，卷积核就有多少个通道。
举个例子：若输入的图像shape为5x5x6，一个卷积核的shape为3x3x6，使用3个卷积核，得到的特征图shape为3x3x3。参数量 = 5x5x6x3 = 450
（2）在分组卷积中，每个卷积核只处理部分通道。如下图，红色的卷积核只处理输入图像中红色的两个通道，绿色的卷积核只处理输入图像中间的两个绿色的通道，第三个卷积核只处理黄色的两个通道。此时，每个卷积核有两个通道，每个卷积核生成一个特征图。
举个例子：若输入的图像shape为5x5x6，一个分组卷积核的shape为3x3x2，使用3个分组卷积核，得到的特征图shape为3x3x3。参数量 = 5x5x(6/3)x(3/3)x3?= 5x5x2x1x3 = 150 。可见，分成三组，参数量为原来的三分之一。
因此，分组卷积能够有效地降低参数量和计算量。
（3）深度可分离卷积是分组卷积的一个特例。随着分组的组数不断增加，输入图像有多少个通道就分几组，每个卷积核处理一个通道的特征，每个卷积核生成一个特征图，输入图像有多少通道就生成多少张特征图，再堆叠起来。
深度可分离卷积详细内容，看我前一篇MobileNetV1的文章。
1.2 通道重排（Channel Shuffle）
分组卷积存在一个问题，各个分组之间相互独立，没有特征融合。通道重排方法实现跨组的信息交融。
如下图（a）所示，卷积核分三组，生成特征图也是三组，每组只在内部进行信息交互，组与组之间没有任何信息交融。
如图（b, c）所示，将每个组的第一份，收集起来作为下一组；每组的第二份收集起来作为下一组....这样就实现了跨组的信息交流。
举个例子来说，如下图。分组卷积生成的三组特征图，第一组1~4；第二组5~8；第三组9~12。先将特征图重塑，为三行N列的矩形。然后进行转置，变成N行三列。最后压平，从二维tensor变成一维tensor，每一组的特征图交叉组合在一起。实现各组之间的信息交融。
2. ShuffleNetV2 模型
2.1 四条轻量化网络的设计原则
（1）输入输出通道数相同时，内存访问量MAC最小
输出通道数对应的是卷积核的数量。比如输入图像的通道数是256个，那么卷积核个数最好也是256个。
（2）分组数过大的分组卷积会增加MAC
ShuffleNetV1使用的分组1x1卷积，减少了参数量、计算量，增大网络表达能力。但是增加了内存占用量。
（3）碎片化操作对并行加速（GPU）不友好
碎片化操作指多分枝、多通路，如Inception系列网络的过多的分支
（4）逐元素操作（残差网络求和、relu激活等）带来的内存和耗时不可忽略
如：避免使用残差求和，将原来的 layers.Add() 变成 layers.Concat()
2.2 网络主干结构
作者根据这四条原则设计出来 ShuffleNetV2 轻量化网络。如下图。
Channel Spilt 模块将输入图像的通道数平均分成两份，一份用于残差连接，一份用于特征提取。
Channel Shuffle 模块将堆叠的特征图的通道重新排序，实现各分组之间的特征融合。
在基本模块中，特征图size不变，通道数不变。在下采样模块中，特征图的长宽减半，通道数加倍。
3. ShuffleNetV2 代码复现
根据论文中给出的网络构架，搭建模型。其中 Stage模块 由 一个ShuffleNetv2下采样模块 和 若干个ShuffleNetv2基本模块 组成。输入图像的shape为 224x224x3。output channel列的几个宽度超参数，一般只用超参数=1那一列。FLOPs代表乘-加浮点运算次数
3.1 构建各个模块
（1）标准卷积模块
标准卷积模块是由 普通卷积+批标准化BN层+ReLU激活函数 构成
#（1）标准卷积块
def conv_block(input_tensor, filters, kernel_size, stride=1):
# 卷积+批标准化+relu激活
x = layers.Conv2D(filters, kernel_size,
strides = stride, # 步长
padding = 'same', # strides=1卷积过程中特征图size不变,strides=2卷积过程中size减半
use_bias = False)(input_tensor) # 有BN层就不需要偏置
x = layers.BatchNormalization()(x) # 批标准化
x = layers.ReLU()(x) # relu激活
return x # 返回一次标准卷积后的tensor（2）深度可分离卷积模块
由2.2中的网络结构图可知，深度可分离卷积块只需要包含 3x3的深度卷积 和 批标准化BN层
#（2）深度可分离卷积块
def depthwise_conv_block(input_tensor, stride=1):
# 深度可分离卷积+批标准化
# 不需要传入卷积核个数，输入有几个通道，就有几个卷积核，每个卷积核负责一个通道
x = layers.DepthwiseConv2D(kernel_size = (3,3), # 深度卷积核size默认3*3
strides = stride, # 步长
padding = 'same', # strides=1卷积过程中特征图size不变,strides=2卷积过程中size减半
use_bias = False)(input_tensor) # 有BN层就不需要偏置
x = layers.BatchNormalization()(x) # 批标准化
return x # 返回深度可分离卷积后的tensor（3）通道重排模块
Channel Shuffle 模块用于对两组堆叠在一起的特征图进行重新排列，实现各组之间的信息融合。
#（3）通道重排，跨组信息交融
def channel_shuffle(input_tensor, num=2): # 默认时2组特征：shortcut和卷积后的x
# 先得到输入特征图的shape，b:batch size，h,w:一张图的size，c:通道数
b, h, w, c = input_tensor.shape
# 确定shape = [b, h, w, num, c//num]。通道维度原来是一个长为c的一维tensor，变成2行n列的矩阵
# 在通道维度上将特征图reshape为2行n列的矩阵。
x_reshaped = tf.reshape(input_tensor, [-1, h, w, num, c//num])
# 确定转置的矩形的shape = [b, h, w, c//num, num]
# 矩阵转置，最后两个维度从2行n列变成n行2列
x_transposed = tf.transpose(x_reshaped, [0,1,2,4,3])
# 重新排列，shotcut和x的通道像素交叉排列，通道维度重新变成一维tensor
output = tf.reshape(x_transposed, [-1, h, w, c])
return output # 返回通道维度交叉排序后的tensor（4）ShuffleNetV2 基本模块
当步长为1时，先将输入特征图在通道维度上平均分成两份。如2.2中的结构图，左分支输入等于输出，右分支特征传递，将两个分支的结果堆叠在一起，通道数变成原始大小。
#（4）步长=1时的卷积块
def shufflent_unit_1(input_tensor, filters):
# 首先将输入特征图在通道维度上平均分成两份：一部分用于残差连接，一部分卷积提取特征
shortcut, x = tf.split(input_tensor, 2, axis=-1) # axis指定轴
# 现在shotcut和x的通道数都只有原来的二分之一
# 1*1卷积+3*3深度卷积+1*1卷积
x = conv_block(x, filters//2, kernel_size=(1,1), stride=1) # 1*1卷积，通道数保持不变
x = depthwise_conv_block(x, stride=1) # 3*3深度卷积
x = conv_block(x, filters//2, kernel_size=(1,1), stride=1) # 1*1卷积跨通道信息融合
# 堆叠shoutcut和x，要求两个tensor的size相同
x = tf.concat([shortcut, x], axis=-1) # 在通道维度上堆叠
# 将堆叠后2组特征图，在通道维度上重新排列
x = channel_shuffle(x)
return x # 返回步长为1时的卷积块输出的tensor（5）ShuffleNetV2 下采样模块
当步长为2时，输入图像不平分通道数。左分支先3*3深度卷积(步长=2)，然后1*1卷积，输出特征图的长宽减半，通道数不变。右分支先1*1卷积降维，然后3*3深度卷积(步长=2)，再1*1卷积升维。
这里要注意，左分支输出特征图数量+右分支输出特征图数量=下采样模块输出特征图数量。
已知左分支输出特征图的通道数保持不变(in_channel)，下采样模块输出特征图数量(out_channel)已给出，可计算得到右分支输出特征图的通道数，将它作为右分支最后一个1*1卷积的卷积核个数。
#（5）步长=2时（下采样）的卷积块
def shufflenet_unit_2(input_tensor, out_channel):
# 输入特征图的通道数
in_channel = input_tensor.shape[-1]
# 首先将输入特征图复制一份，分别用于左右两个分支的卷积
shortcut = input_tensor
# ① 左分支的卷积部分==深度卷积+逐点卷积，输出特征图通道数等于原通道数
shortcut = depthwise_conv_block(shortcut, stride=2) # 特征图size减半
shortcut = conv_block(shortcut, filters=in_channel, kernel_size=(1,1), stride=1) # 输出特征图个数不变
# ② 右分支==1*1卷积下降通道数+3*3深度卷积+1*1卷积上升通道数
x = conv_block(input_tensor, in_channel//2, kernel_size=(1,1), stride=1)
x = depthwise_conv_block(x, stride=2)
# 右分支的通道数和左分支的通道数叠加==输出特征图的通道数out_channel
x = conv_block(x, out_channel-in_channel, kernel_size=(1,1), stride=1)
# ③ 左右分支的输出特征图在通道维度上堆叠，并且output.shape[-1]==out_channel
output = tf.concat([shortcut, x], axis=-1)
# ④ 堆叠后的2组特征在通道维度上重新排列
output = channel_shuffle(output)
return output # 返回步长=2时的输出结果（6）Stage 模块
一个Stage模块是由一个下采样模块，和多个基本模块组合而成的
#（6）构建shufflenet卷积块
# 一个shuffle卷积块是由一个shufflenet_unit_2下采样单元，和若干个shufflenet_unit_1特征传递单元构成
def stage(input_tensor, filters, n): # filters代表输出通道数
# 下采样单元
x = shufflenet_unit_2(input_tensor, out_channel=filters)
# 特征传递单元循环n次
for i in range(n):
x = shufflent_unit_1(x, filters=filters)
return x # 返回一个shufflenet卷积结果3.2 完整代码展示
结合上文的分析，完整代码如下，代码中每行都有注释，有疑问的在评论区留言
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers, Model
#（1）标准卷积块
def conv_block(input_tensor, filters, kernel_size, stride=1):
# 卷积+批标准化+relu激活
x = layers.Conv2D(filters, kernel_size,
strides = stride, # 步长
padding = 'same', # strides=1卷积过程中特征图size不变,strides=2卷积过程中size减半
use_bias = False)(input_tensor) # 有BN层就不需要偏置
x = layers.BatchNormalization()(x) # 批标准化
x = layers.ReLU()(x) # relu激活
return x # 返回一次标准卷积后的tensor
#（2）深度可分离卷积块
def depthwise_conv_block(input_tensor, stride=1):
# 深度可分离卷积+批标准化
# 不需要传入卷积核个数，输入有几个通道，就有几个卷积核，每个卷积核负责一个通道
x = layers.DepthwiseConv2D(kernel_size = (3,3), # 深度卷积核size默认3*3
strides = stride, # 步长
padding = 'same', # strides=1卷积过程中特征图size不变,strides=2卷积过程中size减半
use_bias = False)(input_tensor) # 有BN层就不需要偏置
x = layers.BatchNormalization()(x) # 批标准化
return x # 返回深度可分离卷积后的tensor
#（3）通道重排，跨组信息交融
def channel_shuffle(input_tensor, num=2): # 默认时2组特征：shortcut和卷积后的x
# 先得到输入特征图的shape，b:batch size，h,w:一张图的size，c:通道数
b, h, w, c = input_tensor.shape
# 确定shape = [b, h, w, num, c//num]。通道维度原来是一个长为c的一维tensor，变成2行n列的矩阵
# 在通道维度上将特征图reshape为2行n列的矩阵。
x_reshaped = tf.reshape(input_tensor, [-1, h, w, num, c//num])
# 确定转置的矩形的shape = [b, h, w, c//num, num]
# 矩阵转置，最后两个维度从2行n列变成n行2列
x_transposed = tf.transpose(x_reshaped, [0,1,2,4,3])
# 重新排列，shotcut和x的通道像素交叉排列，通道维度重新变成一维tensor
output = tf.reshape(x_transposed, [-1, h, w, c])
return output # 返回通道维度交叉排序后的tensor
#（4）步长=1时的卷积块
def shufflent_unit_1(input_tensor, filters):
# 首先将输入特征图在通道维度上平均分成两份：一部分用于残差连接，一部分卷积提取特征
shortcut, x = tf.split(input_tensor, 2, axis=-1) # axis指定轴
# 现在shotcut和x的通道数都只有原来的二分之一
# 1*1卷积+3*3深度卷积+1*1卷积
x = conv_block(x, filters//2, kernel_size=(1,1), stride=1) # 1*1卷积，通道数保持不变
x = depthwise_conv_block(x, stride=1) # 3*3深度卷积
x = conv_block(x, filters//2, kernel_size=(1,1), stride=1) # 1*1卷积跨通道信息融合
# 堆叠shoutcut和x，要求两个tensor的size相同
x = tf.concat([shortcut, x], axis=-1) # 在通道维度上堆叠
# 将堆叠后2组特征图，在通道维度上重新排列
x = channel_shuffle(x)
return x # 返回步长为1时的卷积块输出的tensor
#（5）步长=2时（下采样）的卷积块
def shufflenet_unit_2(input_tensor, out_channel):
# 输入特征图的通道数
in_channel = input_tensor.shape[-1]
# 首先将输入特征图复制一份，分别用于左右两个分支的卷积
shortcut = input_tensor
# ① 左分支的卷积部分==深度卷积+逐点卷积，输出特征图通道数等于原通道数
shortcut = depthwise_conv_block(shortcut, stride=2) # 特征图size减半
shortcut = conv_block(shortcut, filters=in_channel, kernel_size=(1,1), stride=1) # 输出特征图个数不变
# ② 右分支==1*1卷积下降通道数+3*3深度卷积+1*1卷积上升通道数
x = conv_block(input_tensor, in_channel//2, kernel_size=(1,1), stride=1)
x = depthwise_conv_block(x, stride=2)
# 右分支的通道数和左分支的通道数叠加==输出特征图的通道数out_channel
x = conv_block(x, out_channel-in_channel, kernel_size=(1,1), stride=1)
# ③ 左右分支的输出特征图在通道维度上堆叠，并且output.shape[-1]==out_channel
output = tf.concat([shortcut, x], axis=-1)
# ④ 堆叠后的2组特征在通道维度上重新排列
output = channel_shuffle(output)
return output # 返回步长=2时的输出结果
#（6）构建shufflenet卷积块
# 一个shuffle卷积块是由一个shufflenet_unit_2下采样单元，和若干个shufflenet_unit_1特征传递单元构成
def stage(input_tensor, filters, n): # filters代表输出通道数
# 下采样单元
x = shufflenet_unit_2(input_tensor, out_channel=filters)
# 特征传递单元循环n次
for i in range(n):
x = shufflent_unit_1(x, filters=filters)
return x # 返回一个shufflenet卷积结果
#（7）构建网络模型
def ShuffleNet(input_shape, classes):
# 构建网络输入的tensor
inputs = keras.Input(shape=input_shape)
# [224,224,3]==>[112,112,24]
x = layers.Conv2D(filters=24, kernel_size=(3,3), strides=2, padding='same')(inputs) # 普通卷积
# [112,112,24]==>[56,56,24]
x = layers.MaxPooling2D(pool_size=(3,3), strides=2, padding='same')(x) # 最大池化
# [56,56,24]==>[28,28,116]
x = stage(x, filters=116, n=3)
# [28,28,116]==>[14,14,232]
x = stage(x, filters=232, n=7)
# [14,14,232]==>[7,7,464]
x = stage(x, filters=464, n=3)
# [7,7,464]==>[7,7,1024]
x = layers.Conv2D(filters=1024, kernel_size=(1,1), strides=1, padding='same')(x) # 1*1普通卷积
# [7,7,1024]==>[None,1024]
x = layers.GlobalAveragePooling2D()(x) # 在通道维度上全局平均池化
# 按论文输出层使用全连接层，也可改为卷积层再Reshape
logits = layers.Dense(classes)(x) # 为了网络稳定，训练时再使用Softmax函数
# 完成网络架构
model = Model(inputs, logits)
return model # 返回网络模型
#（8）接收网络模型
if __name__ == '__main__':
model = ShuffleNet(input_shape=[224,224,3], # 输入图像的shape
classes=1000) # 图像分类类别
model.summary() # 查看网络结构3.3 查看网络构架
通过 model.summary() 查看网络结构，可见 ShuffleNetV2 的参数量只有两百多万，对比 MobileNetV2 的三百多万的参数量，已经非常轻量化了。
Model: "model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 224, 224, 3) 0
__________________________________________________________________________________________________
conv2d (Conv2D) (None, 112, 112, 24) 672 input_1[0][0]
__________________________________________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 56, 56, 24) 0 conv2d[0][0]
__________________________________________________________________________________________________
conv2d_2 (Conv2D) (None, 56, 56, 12) 288 max_pooling2d[0][0]
__________________________________________________________________________________________________
batch_normalization_2 (BatchNor (None, 56, 56, 12) 48 conv2d_2[0][0]
__________________________________________________________________________________________________
re_lu_1 (ReLU) (None, 56, 56, 12) 0 batch_normalization_2[0][0]
__________________________________________________________________________________________________
depthwise_conv2d (DepthwiseConv (None, 28, 28, 24) 216 max_pooling2d[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_1 (DepthwiseCo (None, 28, 28, 12) 108 re_lu_1[0][0]
__________________________________________________________________________________________________
batch_normalization (BatchNorma (None, 28, 28, 24) 96 depthwise_conv2d[0][0]
__________________________________________________________________________________________________
batch_normalization_3 (BatchNor (None, 28, 28, 12) 48 depthwise_conv2d_1[0][0]
__________________________________________________________________________________________________
conv2d_1 (Conv2D) (None, 28, 28, 24) 576 batch_normalization[0][0]
__________________________________________________________________________________________________
conv2d_3 (Conv2D) (None, 28, 28, 92) 1104 batch_normalization_3[0][0]
__________________________________________________________________________________________________
batch_normalization_1 (BatchNor (None, 28, 28, 24) 96 conv2d_1[0][0]
__________________________________________________________________________________________________
batch_normalization_4 (BatchNor (None, 28, 28, 92) 368 conv2d_3[0][0]
__________________________________________________________________________________________________
re_lu (ReLU) (None, 28, 28, 24) 0 batch_normalization_1[0][0]
__________________________________________________________________________________________________
re_lu_2 (ReLU) (None, 28, 28, 92) 0 batch_normalization_4[0][0]
__________________________________________________________________________________________________
tf.concat (TFOpLambda) (None, 28, 28, 116) 0 re_lu[0][0]
re_lu_2[0][0]
__________________________________________________________________________________________________
tf.reshape (TFOpLambda) (None, 28, 28, 2, 58 0 tf.concat[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose (TFOpLam (None, 28, 28, 58, 2 0 tf.reshape[0][0]
__________________________________________________________________________________________________
tf.reshape_1 (TFOpLambda) (None, 28, 28, 116) 0 tf.compat.v1.transpose[0][0]
__________________________________________________________________________________________________
tf.split (TFOpLambda) [(None, 28, 28, 58), 0 tf.reshape_1[0][0]
__________________________________________________________________________________________________
conv2d_4 (Conv2D) (None, 28, 28, 58) 3364 tf.split[0][1]
__________________________________________________________________________________________________
batch_normalization_5 (BatchNor (None, 28, 28, 58) 232 conv2d_4[0][0]
__________________________________________________________________________________________________
re_lu_3 (ReLU) (None, 28, 28, 58) 0 batch_normalization_5[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_2 (DepthwiseCo (None, 28, 28, 58) 522 re_lu_3[0][0]
__________________________________________________________________________________________________
batch_normalization_6 (BatchNor (None, 28, 28, 58) 232 depthwise_conv2d_2[0][0]
__________________________________________________________________________________________________
conv2d_5 (Conv2D) (None, 28, 28, 58) 3364 batch_normalization_6[0][0]
__________________________________________________________________________________________________
batch_normalization_7 (BatchNor (None, 28, 28, 58) 232 conv2d_5[0][0]
__________________________________________________________________________________________________
re_lu_4 (ReLU) (None, 28, 28, 58) 0 batch_normalization_7[0][0]
__________________________________________________________________________________________________
tf.concat_1 (TFOpLambda) (None, 28, 28, 116) 0 tf.split[0][0]
re_lu_4[0][0]
__________________________________________________________________________________________________
tf.reshape_2 (TFOpLambda) (None, 28, 28, 2, 58 0 tf.concat_1[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_1 (TFOpL (None, 28, 28, 58, 2 0 tf.reshape_2[0][0]
__________________________________________________________________________________________________
tf.reshape_3 (TFOpLambda) (None, 28, 28, 116) 0 tf.compat.v1.transpose_1[0][0]
__________________________________________________________________________________________________
tf.split_1 (TFOpLambda) [(None, 28, 28, 58), 0 tf.reshape_3[0][0]
__________________________________________________________________________________________________
conv2d_6 (Conv2D) (None, 28, 28, 58) 3364 tf.split_1[0][1]
__________________________________________________________________________________________________
batch_normalization_8 (BatchNor (None, 28, 28, 58) 232 conv2d_6[0][0]
__________________________________________________________________________________________________
re_lu_5 (ReLU) (None, 28, 28, 58) 0 batch_normalization_8[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_3 (DepthwiseCo (None, 28, 28, 58) 522 re_lu_5[0][0]
__________________________________________________________________________________________________
batch_normalization_9 (BatchNor (None, 28, 28, 58) 232 depthwise_conv2d_3[0][0]
__________________________________________________________________________________________________
conv2d_7 (Conv2D) (None, 28, 28, 58) 3364 batch_normalization_9[0][0]
__________________________________________________________________________________________________
batch_normalization_10 (BatchNo (None, 28, 28, 58) 232 conv2d_7[0][0]
__________________________________________________________________________________________________
re_lu_6 (ReLU) (None, 28, 28, 58) 0 batch_normalization_10[0][0]
__________________________________________________________________________________________________
tf.concat_2 (TFOpLambda) (None, 28, 28, 116) 0 tf.split_1[0][0]
re_lu_6[0][0]
__________________________________________________________________________________________________
tf.reshape_4 (TFOpLambda) (None, 28, 28, 2, 58 0 tf.concat_2[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_2 (TFOpL (None, 28, 28, 58, 2 0 tf.reshape_4[0][0]
__________________________________________________________________________________________________
tf.reshape_5 (TFOpLambda) (None, 28, 28, 116) 0 tf.compat.v1.transpose_2[0][0]
__________________________________________________________________________________________________
tf.split_2 (TFOpLambda) [(None, 28, 28, 58), 0 tf.reshape_5[0][0]
__________________________________________________________________________________________________
conv2d_8 (Conv2D) (None, 28, 28, 58) 3364 tf.split_2[0][1]
__________________________________________________________________________________________________
batch_normalization_11 (BatchNo (None, 28, 28, 58) 232 conv2d_8[0][0]
__________________________________________________________________________________________________
re_lu_7 (ReLU) (None, 28, 28, 58) 0 batch_normalization_11[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_4 (DepthwiseCo (None, 28, 28, 58) 522 re_lu_7[0][0]
__________________________________________________________________________________________________
batch_normalization_12 (BatchNo (None, 28, 28, 58) 232 depthwise_conv2d_4[0][0]
__________________________________________________________________________________________________
conv2d_9 (Conv2D) (None, 28, 28, 58) 3364 batch_normalization_12[0][0]
__________________________________________________________________________________________________
batch_normalization_13 (BatchNo (None, 28, 28, 58) 232 conv2d_9[0][0]
__________________________________________________________________________________________________
re_lu_8 (ReLU) (None, 28, 28, 58) 0 batch_normalization_13[0][0]
__________________________________________________________________________________________________
tf.concat_3 (TFOpLambda) (None, 28, 28, 116) 0 tf.split_2[0][0]
re_lu_8[0][0]
__________________________________________________________________________________________________
tf.reshape_6 (TFOpLambda) (None, 28, 28, 2, 58 0 tf.concat_3[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_3 (TFOpL (None, 28, 28, 58, 2 0 tf.reshape_6[0][0]
__________________________________________________________________________________________________
tf.reshape_7 (TFOpLambda) (None, 28, 28, 116) 0 tf.compat.v1.transpose_3[0][0]
__________________________________________________________________________________________________
conv2d_11 (Conv2D) (None, 28, 28, 58) 6728 tf.reshape_7[0][0]
__________________________________________________________________________________________________
batch_normalization_16 (BatchNo (None, 28, 28, 58) 232 conv2d_11[0][0]
__________________________________________________________________________________________________
re_lu_10 (ReLU) (None, 28, 28, 58) 0 batch_normalization_16[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_5 (DepthwiseCo (None, 14, 14, 116) 1044 tf.reshape_7[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_6 (DepthwiseCo (None, 14, 14, 58) 522 re_lu_10[0][0]
__________________________________________________________________________________________________
batch_normalization_14 (BatchNo (None, 14, 14, 116) 464 depthwise_conv2d_5[0][0]
__________________________________________________________________________________________________
batch_normalization_17 (BatchNo (None, 14, 14, 58) 232 depthwise_conv2d_6[0][0]
__________________________________________________________________________________________________
conv2d_10 (Conv2D) (None, 14, 14, 116) 13456 batch_normalization_14[0][0]
__________________________________________________________________________________________________
conv2d_12 (Conv2D) (None, 14, 14, 116) 6728 batch_normalization_17[0][0]
__________________________________________________________________________________________________
batch_normalization_15 (BatchNo (None, 14, 14, 116) 464 conv2d_10[0][0]
__________________________________________________________________________________________________
batch_normalization_18 (BatchNo (None, 14, 14, 116) 464 conv2d_12[0][0]
__________________________________________________________________________________________________
re_lu_9 (ReLU) (None, 14, 14, 116) 0 batch_normalization_15[0][0]
__________________________________________________________________________________________________
re_lu_11 (ReLU) (None, 14, 14, 116) 0 batch_normalization_18[0][0]
__________________________________________________________________________________________________
tf.concat_4 (TFOpLambda) (None, 14, 14, 232) 0 re_lu_9[0][0]
re_lu_11[0][0]
__________________________________________________________________________________________________
tf.reshape_8 (TFOpLambda) (None, 14, 14, 2, 11 0 tf.concat_4[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_4 (TFOpL (None, 14, 14, 116, 0 tf.reshape_8[0][0]
__________________________________________________________________________________________________
tf.reshape_9 (TFOpLambda) (None, 14, 14, 232) 0 tf.compat.v1.transpose_4[0][0]
__________________________________________________________________________________________________
tf.split_3 (TFOpLambda) [(None, 14, 14, 116) 0 tf.reshape_9[0][0]
__________________________________________________________________________________________________
conv2d_13 (Conv2D) (None, 14, 14, 116) 13456 tf.split_3[0][1]
__________________________________________________________________________________________________
batch_normalization_19 (BatchNo (None, 14, 14, 116) 464 conv2d_13[0][0]
__________________________________________________________________________________________________
re_lu_12 (ReLU) (None, 14, 14, 116) 0 batch_normalization_19[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_7 (DepthwiseCo (None, 14, 14, 116) 1044 re_lu_12[0][0]
__________________________________________________________________________________________________
batch_normalization_20 (BatchNo (None, 14, 14, 116) 464 depthwise_conv2d_7[0][0]
__________________________________________________________________________________________________
conv2d_14 (Conv2D) (None, 14, 14, 116) 13456 batch_normalization_20[0][0]
__________________________________________________________________________________________________
batch_normalization_21 (BatchNo (None, 14, 14, 116) 464 conv2d_14[0][0]
__________________________________________________________________________________________________
re_lu_13 (ReLU) (None, 14, 14, 116) 0 batch_normalization_21[0][0]
__________________________________________________________________________________________________
tf.concat_5 (TFOpLambda) (None, 14, 14, 232) 0 tf.split_3[0][0]
re_lu_13[0][0]
__________________________________________________________________________________________________
tf.reshape_10 (TFOpLambda) (None, 14, 14, 2, 11 0 tf.concat_5[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_5 (TFOpL (None, 14, 14, 116, 0 tf.reshape_10[0][0]
__________________________________________________________________________________________________
tf.reshape_11 (TFOpLambda) (None, 14, 14, 232) 0 tf.compat.v1.transpose_5[0][0]
__________________________________________________________________________________________________
tf.split_4 (TFOpLambda) [(None, 14, 14, 116) 0 tf.reshape_11[0][0]
__________________________________________________________________________________________________
conv2d_15 (Conv2D) (None, 14, 14, 116) 13456 tf.split_4[0][1]
__________________________________________________________________________________________________
batch_normalization_22 (BatchNo (None, 14, 14, 116) 464 conv2d_15[0][0]
__________________________________________________________________________________________________
re_lu_14 (ReLU) (None, 14, 14, 116) 0 batch_normalization_22[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_8 (DepthwiseCo (None, 14, 14, 116) 1044 re_lu_14[0][0]
__________________________________________________________________________________________________
batch_normalization_23 (BatchNo (None, 14, 14, 116) 464 depthwise_conv2d_8[0][0]
__________________________________________________________________________________________________
conv2d_16 (Conv2D) (None, 14, 14, 116) 13456 batch_normalization_23[0][0]
__________________________________________________________________________________________________
batch_normalization_24 (BatchNo (None, 14, 14, 116) 464 conv2d_16[0][0]
__________________________________________________________________________________________________
re_lu_15 (ReLU) (None, 14, 14, 116) 0 batch_normalization_24[0][0]
__________________________________________________________________________________________________
tf.concat_6 (TFOpLambda) (None, 14, 14, 232) 0 tf.split_4[0][0]
re_lu_15[0][0]
__________________________________________________________________________________________________
tf.reshape_12 (TFOpLambda) (None, 14, 14, 2, 11 0 tf.concat_6[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_6 (TFOpL (None, 14, 14, 116, 0 tf.reshape_12[0][0]
__________________________________________________________________________________________________
tf.reshape_13 (TFOpLambda) (None, 14, 14, 232) 0 tf.compat.v1.transpose_6[0][0]
__________________________________________________________________________________________________
tf.split_5 (TFOpLambda) [(None, 14, 14, 116) 0 tf.reshape_13[0][0]
__________________________________________________________________________________________________
conv2d_17 (Conv2D) (None, 14, 14, 116) 13456 tf.split_5[0][1]
__________________________________________________________________________________________________
batch_normalization_25 (BatchNo (None, 14, 14, 116) 464 conv2d_17[0][0]
__________________________________________________________________________________________________
re_lu_16 (ReLU) (None, 14, 14, 116) 0 batch_normalization_25[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_9 (DepthwiseCo (None, 14, 14, 116) 1044 re_lu_16[0][0]
__________________________________________________________________________________________________
batch_normalization_26 (BatchNo (None, 14, 14, 116) 464 depthwise_conv2d_9[0][0]
__________________________________________________________________________________________________
conv2d_18 (Conv2D) (None, 14, 14, 116) 13456 batch_normalization_26[0][0]
__________________________________________________________________________________________________
batch_normalization_27 (BatchNo (None, 14, 14, 116) 464 conv2d_18[0][0]
__________________________________________________________________________________________________
re_lu_17 (ReLU) (None, 14, 14, 116) 0 batch_normalization_27[0][0]
__________________________________________________________________________________________________
tf.concat_7 (TFOpLambda) (None, 14, 14, 232) 0 tf.split_5[0][0]
re_lu_17[0][0]
__________________________________________________________________________________________________
tf.reshape_14 (TFOpLambda) (None, 14, 14, 2, 11 0 tf.concat_7[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_7 (TFOpL (None, 14, 14, 116, 0 tf.reshape_14[0][0]
__________________________________________________________________________________________________
tf.reshape_15 (TFOpLambda) (None, 14, 14, 232) 0 tf.compat.v1.transpose_7[0][0]
__________________________________________________________________________________________________
tf.split_6 (TFOpLambda) [(None, 14, 14, 116) 0 tf.reshape_15[0][0]
__________________________________________________________________________________________________
conv2d_19 (Conv2D) (None, 14, 14, 116) 13456 tf.split_6[0][1]
__________________________________________________________________________________________________
batch_normalization_28 (BatchNo (None, 14, 14, 116) 464 conv2d_19[0][0]
__________________________________________________________________________________________________
re_lu_18 (ReLU) (None, 14, 14, 116) 0 batch_normalization_28[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_10 (DepthwiseC (None, 14, 14, 116) 1044 re_lu_18[0][0]
__________________________________________________________________________________________________
batch_normalization_29 (BatchNo (None, 14, 14, 116) 464 depthwise_conv2d_10[0][0]
__________________________________________________________________________________________________
conv2d_20 (Conv2D) (None, 14, 14, 116) 13456 batch_normalization_29[0][0]
__________________________________________________________________________________________________
batch_normalization_30 (BatchNo (None, 14, 14, 116) 464 conv2d_20[0][0]
__________________________________________________________________________________________________
re_lu_19 (ReLU) (None, 14, 14, 116) 0 batch_normalization_30[0][0]
__________________________________________________________________________________________________
tf.concat_8 (TFOpLambda) (None, 14, 14, 232) 0 tf.split_6[0][0]
re_lu_19[0][0]
__________________________________________________________________________________________________
tf.reshape_16 (TFOpLambda) (None, 14, 14, 2, 11 0 tf.concat_8[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_8 (TFOpL (None, 14, 14, 116, 0 tf.reshape_16[0][0]
__________________________________________________________________________________________________
tf.reshape_17 (TFOpLambda) (None, 14, 14, 232) 0 tf.compat.v1.transpose_8[0][0]
__________________________________________________________________________________________________
tf.split_7 (TFOpLambda) [(None, 14, 14, 116) 0 tf.reshape_17[0][0]
__________________________________________________________________________________________________
conv2d_21 (Conv2D) (None, 14, 14, 116) 13456 tf.split_7[0][1]
__________________________________________________________________________________________________
batch_normalization_31 (BatchNo (None, 14, 14, 116) 464 conv2d_21[0][0]
__________________________________________________________________________________________________
re_lu_20 (ReLU) (None, 14, 14, 116) 0 batch_normalization_31[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_11 (DepthwiseC (None, 14, 14, 116) 1044 re_lu_20[0][0]
__________________________________________________________________________________________________
batch_normalization_32 (BatchNo (None, 14, 14, 116) 464 depthwise_conv2d_11[0][0]
__________________________________________________________________________________________________
conv2d_22 (Conv2D) (None, 14, 14, 116) 13456 batch_normalization_32[0][0]
__________________________________________________________________________________________________
batch_normalization_33 (BatchNo (None, 14, 14, 116) 464 conv2d_22[0][0]
__________________________________________________________________________________________________
re_lu_21 (ReLU) (None, 14, 14, 116) 0 batch_normalization_33[0][0]
__________________________________________________________________________________________________
tf.concat_9 (TFOpLambda) (None, 14, 14, 232) 0 tf.split_7[0][0]
re_lu_21[0][0]
__________________________________________________________________________________________________
tf.reshape_18 (TFOpLambda) (None, 14, 14, 2, 11 0 tf.concat_9[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_9 (TFOpL (None, 14, 14, 116, 0 tf.reshape_18[0][0]
__________________________________________________________________________________________________
tf.reshape_19 (TFOpLambda) (None, 14, 14, 232) 0 tf.compat.v1.transpose_9[0][0]
__________________________________________________________________________________________________
tf.split_8 (TFOpLambda) [(None, 14, 14, 116) 0 tf.reshape_19[0][0]
__________________________________________________________________________________________________
conv2d_23 (Conv2D) (None, 14, 14, 116) 13456 tf.split_8[0][1]
__________________________________________________________________________________________________
batch_normalization_34 (BatchNo (None, 14, 14, 116) 464 conv2d_23[0][0]
__________________________________________________________________________________________________
re_lu_22 (ReLU) (None, 14, 14, 116) 0 batch_normalization_34[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_12 (DepthwiseC (None, 14, 14, 116) 1044 re_lu_22[0][0]
__________________________________________________________________________________________________
batch_normalization_35 (BatchNo (None, 14, 14, 116) 464 depthwise_conv2d_12[0][0]
__________________________________________________________________________________________________
conv2d_24 (Conv2D) (None, 14, 14, 116) 13456 batch_normalization_35[0][0]
__________________________________________________________________________________________________
batch_normalization_36 (BatchNo (None, 14, 14, 116) 464 conv2d_24[0][0]
__________________________________________________________________________________________________
re_lu_23 (ReLU) (None, 14, 14, 116) 0 batch_normalization_36[0][0]
__________________________________________________________________________________________________
tf.concat_10 (TFOpLambda) (None, 14, 14, 232) 0 tf.split_8[0][0]
re_lu_23[0][0]
__________________________________________________________________________________________________
tf.reshape_20 (TFOpLambda) (None, 14, 14, 2, 11 0 tf.concat_10[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_10 (TFOp (None, 14, 14, 116, 0 tf.reshape_20[0][0]
__________________________________________________________________________________________________
tf.reshape_21 (TFOpLambda) (None, 14, 14, 232) 0 tf.compat.v1.transpose_10[0][0]
__________________________________________________________________________________________________
tf.split_9 (TFOpLambda) [(None, 14, 14, 116) 0 tf.reshape_21[0][0]
__________________________________________________________________________________________________
conv2d_25 (Conv2D) (None, 14, 14, 116) 13456 tf.split_9[0][1]
__________________________________________________________________________________________________
batch_normalization_37 (BatchNo (None, 14, 14, 116) 464 conv2d_25[0][0]
__________________________________________________________________________________________________
re_lu_24 (ReLU) (None, 14, 14, 116) 0 batch_normalization_37[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_13 (DepthwiseC (None, 14, 14, 116) 1044 re_lu_24[0][0]
__________________________________________________________________________________________________
batch_normalization_38 (BatchNo (None, 14, 14, 116) 464 depthwise_conv2d_13[0][0]
__________________________________________________________________________________________________
conv2d_26 (Conv2D) (None, 14, 14, 116) 13456 batch_normalization_38[0][0]
__________________________________________________________________________________________________
batch_normalization_39 (BatchNo (None, 14, 14, 116) 464 conv2d_26[0][0]
__________________________________________________________________________________________________
re_lu_25 (ReLU) (None, 14, 14, 116) 0 batch_normalization_39[0][0]
__________________________________________________________________________________________________
tf.concat_11 (TFOpLambda) (None, 14, 14, 232) 0 tf.split_9[0][0]
re_lu_25[0][0]
__________________________________________________________________________________________________
tf.reshape_22 (TFOpLambda) (None, 14, 14, 2, 11 0 tf.concat_11[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_11 (TFOp (None, 14, 14, 116, 0 tf.reshape_22[0][0]
__________________________________________________________________________________________________
tf.reshape_23 (TFOpLambda) (None, 14, 14, 232) 0 tf.compat.v1.transpose_11[0][0]
__________________________________________________________________________________________________
conv2d_28 (Conv2D) (None, 14, 14, 116) 26912 tf.reshape_23[0][0]
__________________________________________________________________________________________________
batch_normalization_42 (BatchNo (None, 14, 14, 116) 464 conv2d_28[0][0]
__________________________________________________________________________________________________
re_lu_27 (ReLU) (None, 14, 14, 116) 0 batch_normalization_42[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_14 (DepthwiseC (None, 7, 7, 232) 2088 tf.reshape_23[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_15 (DepthwiseC (None, 7, 7, 116) 1044 re_lu_27[0][0]
__________________________________________________________________________________________________
batch_normalization_40 (BatchNo (None, 7, 7, 232) 928 depthwise_conv2d_14[0][0]
__________________________________________________________________________________________________
batch_normalization_43 (BatchNo (None, 7, 7, 116) 464 depthwise_conv2d_15[0][0]
__________________________________________________________________________________________________
conv2d_27 (Conv2D) (None, 7, 7, 232) 53824 batch_normalization_40[0][0]
__________________________________________________________________________________________________
conv2d_29 (Conv2D) (None, 7, 7, 232) 26912 batch_normalization_43[0][0]
__________________________________________________________________________________________________
batch_normalization_41 (BatchNo (None, 7, 7, 232) 928 conv2d_27[0][0]
__________________________________________________________________________________________________
batch_normalization_44 (BatchNo (None, 7, 7, 232) 928 conv2d_29[0][0]
__________________________________________________________________________________________________
re_lu_26 (ReLU) (None, 7, 7, 232) 0 batch_normalization_41[0][0]
__________________________________________________________________________________________________
re_lu_28 (ReLU) (None, 7, 7, 232) 0 batch_normalization_44[0][0]
__________________________________________________________________________________________________
tf.concat_12 (TFOpLambda) (None, 7, 7, 464) 0 re_lu_26[0][0]
re_lu_28[0][0]
__________________________________________________________________________________________________
tf.reshape_24 (TFOpLambda) (None, 7, 7, 2, 232) 0 tf.concat_12[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_12 (TFOp (None, 7, 7, 232, 2) 0 tf.reshape_24[0][0]
__________________________________________________________________________________________________
tf.reshape_25 (TFOpLambda) (None, 7, 7, 464) 0 tf.compat.v1.transpose_12[0][0]
__________________________________________________________________________________________________
tf.split_10 (TFOpLambda) [(None, 7, 7, 232), 0 tf.reshape_25[0][0]
__________________________________________________________________________________________________
conv2d_30 (Conv2D) (None, 7, 7, 232) 53824 tf.split_10[0][1]
__________________________________________________________________________________________________
batch_normalization_45 (BatchNo (None, 7, 7, 232) 928 conv2d_30[0][0]
__________________________________________________________________________________________________
re_lu_29 (ReLU) (None, 7, 7, 232) 0 batch_normalization_45[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_16 (DepthwiseC (None, 7, 7, 232) 2088 re_lu_29[0][0]
__________________________________________________________________________________________________
batch_normalization_46 (BatchNo (None, 7, 7, 232) 928 depthwise_conv2d_16[0][0]
__________________________________________________________________________________________________
conv2d_31 (Conv2D) (None, 7, 7, 232) 53824 batch_normalization_46[0][0]
__________________________________________________________________________________________________
batch_normalization_47 (BatchNo (None, 7, 7, 232) 928 conv2d_31[0][0]
__________________________________________________________________________________________________
re_lu_30 (ReLU) (None, 7, 7, 232) 0 batch_normalization_47[0][0]
__________________________________________________________________________________________________
tf.concat_13 (TFOpLambda) (None, 7, 7, 464) 0 tf.split_10[0][0]
re_lu_30[0][0]
__________________________________________________________________________________________________
tf.reshape_26 (TFOpLambda) (None, 7, 7, 2, 232) 0 tf.concat_13[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_13 (TFOp (None, 7, 7, 232, 2) 0 tf.reshape_26[0][0]
__________________________________________________________________________________________________
tf.reshape_27 (TFOpLambda) (None, 7, 7, 464) 0 tf.compat.v1.transpose_13[0][0]
__________________________________________________________________________________________________
tf.split_11 (TFOpLambda) [(None, 7, 7, 232), 0 tf.reshape_27[0][0]
__________________________________________________________________________________________________
conv2d_32 (Conv2D) (None, 7, 7, 232) 53824 tf.split_11[0][1]
__________________________________________________________________________________________________
batch_normalization_48 (BatchNo (None, 7, 7, 232) 928 conv2d_32[0][0]
__________________________________________________________________________________________________
re_lu_31 (ReLU) (None, 7, 7, 232) 0 batch_normalization_48[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_17 (DepthwiseC (None, 7, 7, 232) 2088 re_lu_31[0][0]
__________________________________________________________________________________________________
batch_normalization_49 (BatchNo (None, 7, 7, 232) 928 depthwise_conv2d_17[0][0]
__________________________________________________________________________________________________
conv2d_33 (Conv2D) (None, 7, 7, 232) 53824 batch_normalization_49[0][0]
__________________________________________________________________________________________________
batch_normalization_50 (BatchNo (None, 7, 7, 232) 928 conv2d_33[0][0]
__________________________________________________________________________________________________
re_lu_32 (ReLU) (None, 7, 7, 232) 0 batch_normalization_50[0][0]
__________________________________________________________________________________________________
tf.concat_14 (TFOpLambda) (None, 7, 7, 464) 0 tf.split_11[0][0]
re_lu_32[0][0]
__________________________________________________________________________________________________
tf.reshape_28 (TFOpLambda) (None, 7, 7, 2, 232) 0 tf.concat_14[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_14 (TFOp (None, 7, 7, 232, 2) 0 tf.reshape_28[0][0]
__________________________________________________________________________________________________
tf.reshape_29 (TFOpLambda) (None, 7, 7, 464) 0 tf.compat.v1.transpose_14[0][0]
__________________________________________________________________________________________________
tf.split_12 (TFOpLambda) [(None, 7, 7, 232), 0 tf.reshape_29[0][0]
__________________________________________________________________________________________________
conv2d_34 (Conv2D) (None, 7, 7, 232) 53824 tf.split_12[0][1]
__________________________________________________________________________________________________
batch_normalization_51 (BatchNo (None, 7, 7, 232) 928 conv2d_34[0][0]
__________________________________________________________________________________________________
re_lu_33 (ReLU) (None, 7, 7, 232) 0 batch_normalization_51[0][0]
__________________________________________________________________________________________________
depthwise_conv2d_18 (DepthwiseC (None, 7, 7, 232) 2088 re_lu_33[0][0]
__________________________________________________________________________________________________
batch_normalization_52 (BatchNo (None, 7, 7, 232) 928 depthwise_conv2d_18[0][0]
__________________________________________________________________________________________________
conv2d_35 (Conv2D) (None, 7, 7, 232) 53824 batch_normalization_52[0][0]
__________________________________________________________________________________________________
batch_normalization_53 (BatchNo (None, 7, 7, 232) 928 conv2d_35[0][0]
__________________________________________________________________________________________________
re_lu_34 (ReLU) (None, 7, 7, 232) 0 batch_normalization_53[0][0]
__________________________________________________________________________________________________
tf.concat_15 (TFOpLambda) (None, 7, 7, 464) 0 tf.split_12[0][0]
re_lu_34[0][0]
__________________________________________________________________________________________________
tf.reshape_30 (TFOpLambda) (None, 7, 7, 2, 232) 0 tf.concat_15[0][0]
__________________________________________________________________________________________________
tf.compat.v1.transpose_15 (TFOp (None, 7, 7, 232, 2) 0 tf.reshape_30[0][0]
__________________________________________________________________________________________________
tf.reshape_31 (TFOpLambda) (None, 7, 7, 464) 0 tf.compat.v1.transpose_15[0][0]
__________________________________________________________________________________________________
conv2d_36 (Conv2D) (None, 7, 7, 1024) 476160 tf.reshape_31[0][0]
__________________________________________________________________________________________________
global_average_pooling2d (Globa (None, 1024) 0 conv2d_36[0][0]
__________________________________________________________________________________________________
dense (Dense) (None, 1000) 1025000 global_average_pooling2d[0][0]
==================================================================================================
Total params: 2,216,440
Trainable params: 2,203,236
Non-trainable params: 13,204
__________________________________________________________________________________________________