通过ARM Compute Library写的VGG16的模型

PS：之前那一篇写的太简单了，可能很多操作大家都不懂，这次修改很很多，完全写好了详细一点，顺便说说我用的arm compute library是17.05版的，前几天出了17.06版应该修好了挺多bug，例如17.05版softmax会有overflow的情况，还有一些typo的问题，大家可以尝试用用17.06版，估计过几个月也会有新的版本。而且arm compute library暂时不能train，所以train可能要在tensorflow或者caffe上跑，然后把数据加载到这边，可能arm compute library暂时只优化了验证这边吧。至于如何通过arm compute library写模型就看PO主这篇薄荷！！！传送们在这！！！

http://blog.csdn.net/u014432647/article/details/74738947

正文开始啦！！！

不是通过tensorflow和caffe配置的，是通过ARM Compute Library写的。测试是手机CPU。

首先放出ARM Compute Library的github：

https://github.com/ARM-software/ComputeLibrary

具体的操作方法和基本的examples实现上面都有，这个的确很麻烦，操作了很久。

首先先看看研究一下上面那个东东。PO主放出自己的github：

https://github.com/ppplinday/vgg16-by-ARM-Compute-Library

这个PO主是已经下载好了arm compute library，把写好的文件和权重都放进去啦！本文基本都是按照PO主这个github操作的！但由于之前没有好好将一下详细过程，所以这次讲一讲吧！当然想偷懒的小伙伴也可以直接跑PO主的脚本！下面总结也会说到的！

第一步就要开始build这个库：

scons Werror=1 -j8 debug=0 asserts=1 neon=1 opencl=1 embed_kernels=1 os=android arch=arm64-v8a

在arm compute library中的Documentation的neon是=0，由于是cpu所以这边也是neon=1。而且PO主是用64位滴，大伙注意啦！

用这个建好库之后会多出一个build的文件夹 内面有很多.so这些库。然后再建一个空的opencl库。。。。

aarch64-linux-android-gcc -o libOpenCL.so -Iinclude -shared opencl-1.2-stubs/opencl_stubs.c -fPIC -shared

就会多出一个libOpenCL.so

PO主的主要cpp是放在examples中的vgg16_model_arm_compute_library_NEON.cpp

接着是编译：（编译是要aarch64-linux-android-g++这些编译器，自己下载啦！）

aarch64-linux-android-g++ examples/vgg16_model_arm_compute_library_NEON.cpp test_helpers/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute-static -lOpenCL -L/home/zhoupeilin/vgg16-by-ARM-Compute-Library/build/arm_compute -L. -o main -static-libstdc++ -pie

编译好了的文件是main。由于PO主是用android手机跑，所以要按转NDK。NDK的安装和环境变量设置，这些就不是本文要说的啦！自行google啦！

接着就要跑一下PO主的脚本，把权重都放入手机上面：

sh pushweight.sh

权证文件放在vgg16_weight中大约有1.8g的csv文件。。。。。。

最后可以把刚刚编译好的程序push到手机上接着就可以跑啦！！！

adb push main /data/local/tmp/
adb shell /data/local/tmp/main

PO主的手机是红米Note 4X，cpu是高通骁龙625，内存3g可用大约2g，这个程序的手机可用内存最好是2G以上，否测可能会因分配内存不够而死机。

想偷懒的童鞋可以下载好PO主的github然后直接跑两个脚本文件就行啦！记得要插上手机！

sh pushweight.sh
sh main.sh

最后VGG16就不用多说吧，深度学习的都懂。我是在Linux系统交叉编译然后在Android手机跑的程序。

放出我的vgg的代码！！！PO主会在下一篇博客会详细介绍怎样用arm compute lirbrary写模型！！！

代码来啦！！！

#include "arm_compute/runtime/NEON/NEFunctions.h"
#include "arm_compute/runtime/CL/CLFunctions.h"

#include "arm_compute/core/Types.h"
#include "test_helpers/Utils.h"
#include <iostream>
#include <sstream>
#include <fstream>
#include <ostream>
#include <sys/time.h>
#include <map>

using namespace arm_compute;
using namespace test_helpers;

float StringToFloat(const std::string & str){
std::istringstream iss(str);
float number;
iss >> number;
return number;
}

void main_neon_vgg16(int argc, const char **argv)
{
/*----------------------------------[init_model_vgg16]-----------------------------------*/

/*----------------------------------BEGIN:[init_Tensor]----------------------------------*/
//init_input_tensor
Tensor input;

//init_conv_1_tensor
Tensor weights_1_1;
Tensor biases_1_1;
Tensor out_1_1;
Tensor act_1_1;

Tensor weights_1_2;
Tensor biases_1_2;
Tensor out_1_2;
Tensor act_1_2;

Tensor pool_1;

//init_conv_2_tensor
Tensor weights_2_1;
Tensor biases_2_1;
Tensor out_2_1;
Tensor act_2_1;

Tensor weights_2_2;
Tensor biases_2_2;
Tensor out_2_2;
Tensor act_2_2;

Tensor pool_2;

//init_conv_3_tensor
Tensor weights_3_1;
Tensor biases_3_1;
Tensor out_3_1;
Tensor act_3_1;

Tensor weights_3_2;
Tensor biases_3_2;
Tensor out_3_2;
Tensor act_3_2;

Tensor weights_3_3;
Tensor biases_3_3;
Tensor out_3_3;
Tensor act_3_3;

Tensor pool_3;

//init_conv_4_tensor
Tensor weights_4_1;
Tensor biases_4_1;
Tensor out_4_1;
Tensor act_4_1;

Tensor weights_4_2;
Tensor biases_4_2;
Tensor out_4_2;
Tensor act_4_2;

Tensor weights_4_3;
Tensor biases_4_3;
Tensor out_4_3;
Tensor act_4_3;

Tensor pool_4;

//init_conv_5_tensor
Tensor weights_5_1;
Tensor biases_5_1;
Tensor out_5_1;
Tensor act_5_1;

Tensor weights_5_2;
Tensor biases_5_2;
Tensor out_5_2;
Tensor act_5_2;

Tensor weights_5_3;
Tensor biases_5_3;
Tensor out_5_3;
Tensor act_5_3;

Tensor pool_5;

//init_fc_6
Tensor weights_6;
Tensor biases_6;
Tensor out_6;
Tensor act_6;

//init_fc_7
Tensor weights_7;
Tensor biases_7;
Tensor out_7;
Tensor act_7;

//init_fc_8
Tensor weights_8;
Tensor biases_8;
Tensor out_8;

Tensor softmax_tensor;

//init_tensor
constexpr unsigned int input_width  = 224;
constexpr unsigned int input_height = 224;
constexpr unsigned int input_fm     = 3;

const TensorShape input_shape(input_width, input_height, input_fm);
input.allocator() -> init(TensorInfo(input_shape, 1, DataType::F32));

//init_conv_1_1
constexpr unsigned int conv_1_1_kernel_x = 3;
constexpr unsigned int conv_1_1_kernel_y = 3;
constexpr unsigned int conv_1_1_fm       = 64;

const TensorShape conv_1_1_weights_shape(conv_1_1_kernel_x, conv_1_1_kernel_y, input_shape.z(), conv_1_1_fm);
const TensorShape conv_1_1_biases_shape(conv_1_1_weights_shape[3]);
const TensorShape conv_1_1_out_shape(input_shape.x(), input_shape.y(), conv_1_1_weights_shape[3]);

weights_1_1.allocator() -> init(TensorInfo(conv_1_1_weights_shape, 1, DataType::F32));
biases_1_1.allocator() -> init(TensorInfo(conv_1_1_biases_shape, 1, DataType::F32));
out_1_1.allocator() -> init(TensorInfo(conv_1_1_out_shape, 1, DataType::F32));

act_1_1.allocator() -> init(TensorInfo(conv_1_1_out_shape, 1, DataType::F32));

//init_conv_1_2
constexpr unsigned int conv_1_2_kernel_x = 3;
constexpr unsigned int conv_1_2_kernel_y = 3;
constexpr unsigned int conv_1_2_fm       = 64;

const TensorShape conv_1_2_weights_shape(conv_1_2_kernel_x, conv_1_2_kernel_y, conv_1_1_out_shape.z(), conv_1_2_fm);
const TensorShape conv_1_2_biases_shape(conv_1_2_weights_shape[3]);
const TensorShape conv_1_2_out_shape(conv_1_1_out_shape.x(), conv_1_1_out_shape.y(), conv_1_1_weights_shape[3]);

weights_1_2.allocator() -> init(TensorInfo(conv_1_2_weights_shape, 1, DataType::F32));
biases_1_2.allocator() -> init(TensorInfo(conv_1_2_biases_shape, 1, DataType::F32));
out_1_2.allocator() -> init(TensorInfo(conv_1_2_out_shape, 1, DataType::F32));

act_1_2.allocator() -> init(TensorInfo(conv_1_2_out_shape, 1, DataType::F32));

TensorShape conv_1_pool = conv_1_2_out_shape;
conv_1_pool.set(0, conv_1_pool.x() / 2);
conv_1_pool.set(1, conv_1_pool.y() / 2);
pool_1.allocator() -> init(TensorInfo(conv_1_pool, 1, DataType::F32));

//init_conv_2_1
constexpr unsigned int conv_2_1_kernel_x = 3;
constexpr unsigned int conv_2_1_kernel_y = 3;
constexpr unsigned int conv_2_1_fm       = 128;

const TensorShape conv_2_1_weights_shape(conv_2_1_kernel_x, conv_2_1_kernel_y, conv_1_pool.z(), conv_2_1_fm);
const TensorShape conv_2_1_biases_shape(conv_2_1_weights_shape[3]);
const TensorShape conv_2_1_out_shape(conv_1_pool.x(), conv_1_pool.y(), conv_2_1_weights_shape[3]);

weights_2_1.allocator() -> init(TensorInfo(conv_2_1_weights_shape, 1, DataType::F32));
biases_2_1.allocator() -> init(TensorInfo(conv_2_1_biases_shape, 1, DataType::F32));
out_2_1.allocator() -> init(TensorInfo(conv_2_1_out_shape, 1, DataType::F32));

act_2_1.allocator() -> init(TensorInfo(conv_2_1_out_shape, 1, DataType::F32));

//init_conv_2_2
constexpr unsigned int conv_2_2_kernel_x = 3;
constexpr unsigned int conv_2_2_kernel_y = 3;
constexpr unsigned int conv_2_2_fm       = 128;

const TensorShape conv_2_2_weights_shape(conv_2_2_kernel_x, conv_2_2_kernel_y, conv_2_1_out_shape.z(), conv_2_2_fm);
const TensorShape conv_2_2_biases_shape(conv_2_2_weights_shape[3]);
const TensorShape conv_2_2_out_shape(conv_2_1_out_shape.x(), conv_2_1_out_shape.y(), conv_2_2_weights_shape[3]);

weights_2_2.allocator() -> init(TensorInfo(conv_2_2_weights_shape, 1, DataType::F32));
biases_2_2.allocator() -> init(TensorInfo(conv_2_2_biases_shape, 1, DataType::F32));
out_2_2.allocator() -> init(TensorInfo(conv_2_2_out_shape, 1, DataType::F32));

act_2_2.allocator() -> init(TensorInfo(conv_2_2_out_shape, 1, DataType::F32));

TensorShape conv_2_pool = conv_2_2_out_shape;
conv_2_pool.set(0, conv_2_pool.x() / 2);
conv_2_pool.set(1, conv_2_pool.y() / 2);
pool_2.allocator() -> init(TensorInfo(conv_2_pool, 1, DataType::F32));

//init_conv_3_1
constexpr unsigned int conv_3_1_kernel_x = 3;
constexpr unsigned int conv_3_1_kernel_y = 3;
constexpr unsigned int conv_3_1_fm       = 256;

const TensorShape conv_3_1_weights_shape(conv_3_1_kernel_x, conv_3_1_kernel_y, conv_2_pool.z(), conv_3_1_fm);
const TensorShape conv_3_1_biases_shape(conv_3_1_weights_shape[3]);
const TensorShape conv_3_1_out_shape(conv_2_pool.x(), conv_2_pool.y(), conv_3_1_weights_shape[3]);

weights_3_1.allocator() -> init(TensorInfo(conv_3_1_weights_shape, 1, DataType::F32));
biases_3_1.allocator() -> init(TensorInfo(conv_3_1_biases_shape, 1, DataType::F32));
out_3_1.allocator() -> init(TensorInfo(conv_3_1_out_shape, 1, DataType::F32));

act_3_1.allocator() -> init(TensorInfo(conv_3_1_out_shape, 1, DataType::F32));

//init_conv_3_2
constexpr unsigned int conv_3_2_kernel_x = 3;
constexpr unsigned int conv_3_2_kernel_y = 3;
constexpr unsigned int conv_3_2_fm       = 256;

const TensorShape conv_3_2_weights_shape(conv_3_2_kernel_x, conv_3_2_kernel_y, conv_3_1_out_shape.z(), conv_3_2_fm);
const TensorShape conv_3_2_biases_shape(conv_3_2_weights_shape[3]);
const TensorShape conv_3_2_out_shape(conv_3_1_out_shape.x(), conv_3_1_out_shape.y(), conv_3_2_weights_shape[3]);

weights_3_2.allocator() -> init(TensorInfo(conv_3_2_weights_shape, 1, DataType::F32));
biases_3_2.allocator() -> init(TensorInfo(conv_3_2_biases_shape, 1, DataType::F32));
out_3_2.allocator() -> init(TensorInfo(conv_3_2_out_shape, 1, DataType::F32));

act_3_2.allocator() -> init(TensorInfo(conv_3_2_out_shape, 1, DataType::F32));

//init_conv_3_3
constexpr unsigned int conv_3_3_kernel_x = 3;
constexpr unsigned int conv_3_3_kernel_y = 3;
constexpr unsigned int conv_3_3_fm       = 256;

const TensorShape conv_3_3_weights_shape(conv_3_3_kernel_x, conv_3_3_kernel_y, conv_3_2_out_shape.z(), conv_3_3_fm);
const TensorShape conv_3_3_biases_shape(conv_3_3_weights_shape[3]);
const TensorShape conv_3_3_out_shape(conv_3_2_out_shape.x(), conv_3_2_out_shape.y(), conv_3_3_weights_shape[3]);

weights_3_3.allocator() -> init(TensorInfo(conv_3_3_weights_shape, 1, DataType::F32));
biases_3_3.allocator() -> init(TensorInfo(conv_3_3_biases_shape, 1, DataType::F32));
out_3_3.allocator() -> init(TensorInfo(conv_3_3_out_shape, 1, DataType::F32));

act_3_3.allocator() -> init(TensorInfo(conv_3_3_out_shape, 1, DataType::F32));

TensorShape conv_3_pool = conv_3_3_out_shape;
conv_3_pool.set(0, conv_3_pool.x() / 2);
conv_3_pool.set(1, conv_3_pool.y() / 2);
pool_3.allocator() -> init(TensorInfo(conv_3_pool, 1, DataType::F32));

//init_conv_4_1
constexpr unsigned int conv_4_1_kernel_x = 3;
constexpr unsigned int conv_4_1_kernel_y = 3;
constexpr unsigned int conv_4_1_fm       = 512;

const TensorShape conv_4_1_weights_shape(conv_4_1_kernel_x, conv_4_1_kernel_y, conv_3_pool.z(), conv_4_1_fm);
const TensorShape conv_4_1_biases_shape(conv_4_1_weights_shape[3]);
const TensorShape conv_4_1_out_shape(conv_3_pool.x(), conv_3_pool.y(), conv_4_1_weights_shape[3]);

weights_4_1.allocator() -> init(TensorInfo(conv_4_1_weights_shape, 1, DataType::F32));
biases_4_1.allocator() -> init(TensorInfo(conv_4_1_biases_shape, 1, DataType::F32));
out_4_1.allocator() -> init(TensorInfo(conv_4_1_out_shape, 1, DataType::F32));

act_4_1.allocator() -> init(TensorInfo(conv_4_1_out_shape, 1, DataType::F32));

//init_conv_4_2
constexpr unsigned int conv_4_2_kernel_x = 3;
constexpr unsigned int conv_4_2_kernel_y = 3;
constexpr unsigned int conv_4_2_fm       = 512;

const TensorShape conv_4_2_weights_shape(conv_4_2_kernel_x, conv_4_2_kernel_y, conv_4_1_out_shape.z(), conv_4_2_fm);
const TensorShape conv_4_2_biases_shape(conv_4_2_weights_shape[3]);
const TensorShape conv_4_2_out_shape(conv_4_1_out_shape.x(), conv_4_1_out_shape.y(), conv_4_2_weights_shape[3]);

weights_4_2.allocator() -> init(TensorInfo(conv_4_2_weights_shape, 1, DataType::F32));
biases_4_2.allocator() -> init(TensorInfo(conv_4_2_biases_shape, 1, DataType::F32));
out_4_2.allocator() -> init(TensorInfo(conv_4_2_out_shape, 1, DataType::F32));

act_4_2.allocator() -> init(TensorInfo(conv_4_2_out_shape, 1, DataType::F32));

//init_conv_4_3
constexpr unsigned int conv_4_3_kernel_x = 3;
constexpr unsigned int conv_4_3_kernel_y = 3;
constexpr unsigned int conv_4_3_fm       = 512;

const TensorShape conv_4_3_weights_shape(conv_4_3_kernel_x, conv_4_3_kernel_y, conv_4_2_out_shape.z(), conv_4_3_fm);
const TensorShape conv_4_3_biases_shape(conv_4_3_weights_shape[3]);
const TensorShape conv_4_3_out_shape(conv_4_2_out_shape.x(), conv_4_2_out_shape.y(), conv_4_3_weights_shape[3]);

weights_4_3.allocator() -> init(TensorInfo(conv_4_3_weights_shape, 1, DataType::F32));
biases_4_3.allocator() -> init(TensorInfo(conv_4_3_biases_shape, 1, DataType::F32));
out_4_3.allocator() -> init(TensorInfo(conv_4_3_out_shape, 1, DataType::F32));

act_4_3.allocator() -> init(TensorInfo(conv_4_3_out_shape, 1, DataType::F32));

TensorShape conv_4_pool = conv_4_3_out_shape;
conv_4_pool.set(0, conv_4_pool.x() / 2);
conv_4_pool.set(1, conv_4_pool.y() / 2);
pool_4.allocator() -> init(TensorInfo(conv_4_pool, 1, DataType::F32));

//init_conv_5_1
constexpr unsigned int conv_5_1_kernel_x = 3;
constexpr unsigned int conv_5_1_kernel_y = 3;
constexpr unsigned int conv_5_1_fm       = 512;

const TensorShape conv_5_1_weights_shape(conv_5_1_kernel_x, conv_5_1_kernel_y, conv_4_pool.z(), conv_5_1_fm);
const TensorShape conv_5_1_biases_shape(conv_5_1_weights_shape[3]);
const TensorShape conv_5_1_out_shape(conv_4_pool.x(), conv_4_pool.y(), conv_5_1_weights_shape[3]);

weights_5_1.allocator() -> init(TensorInfo(conv_5_1_weights_shape, 1, DataType::F32));
biases_5_1.allocator() -> init(TensorInfo(conv_5_1_biases_shape, 1, DataType::F32));
out_5_1.allocator() -> init(TensorInfo(conv_5_1_out_shape, 1, DataType::F32));

act_5_1.allocator() -> init(TensorInfo(conv_5_1_out_shape, 1, DataType::F32));

//init_conv_5_2
constexpr unsigned int conv_5_2_kernel_x = 3;
constexpr unsigned int conv_5_2_kernel_y = 3;
constexpr unsigned int conv_5_2_fm       = 512;

const TensorShape conv_5_2_weights_shape(conv_5_2_kernel_x, conv_5_2_kernel_y, conv_5_1_out_shape.z(), conv_5_2_fm);
const TensorShape conv_5_2_biases_shape(conv_5_2_weights_shape[3]);
const TensorShape conv_5_2_out_shape(conv_5_1_out_shape.x(), conv_5_1_out_shape.y(), conv_5_2_weights_shape[3]);

weights_5_2.allocator() -> init(TensorInfo(conv_5_2_weights_shape, 1, DataType::F32));
biases_5_2.allocator() -> init(TensorInfo(conv_5_2_biases_shape, 1, DataType::F32));
out_5_2.allocator() -> init(TensorInfo(conv_5_2_out_shape, 1, DataType::F32));

act_5_2.allocator() -> init(TensorInfo(conv_5_2_out_shape, 1, DataType::F32));

//init_conv_5_3
constexpr unsigned int conv_5_3_kernel_x = 3;
constexpr unsigned int conv_5_3_kernel_y = 3;
constexpr unsigned int conv_5_3_fm       = 512;

const TensorShape conv_5_3_weights_shape(conv_5_3_kernel_x, conv_5_3_kernel_y, conv_5_2_out_shape.z(), conv_5_3_fm);
const TensorShape conv_5_3_biases_shape(conv_5_3_weights_shape[3]);
const TensorShape conv_5_3_out_shape(conv_5_2_out_shape.x(), conv_5_2_out_shape.y(), conv_5_3_weights_shape[3]);

weights_5_3.allocator() -> init(TensorInfo(conv_5_3_weights_shape, 1, DataType::F32));
biases_5_3.allocator() -> init(TensorInfo(conv_5_3_biases_shape, 1, DataType::F32));
out_5_3.allocator() -> init(TensorInfo(conv_5_3_out_shape, 1, DataType::F32));

act_5_3.allocator() -> init(TensorInfo(conv_5_3_out_shape, 1, DataType::F32));

TensorShape conv_5_pool = conv_5_3_out_shape;
conv_5_pool.set(0, conv_5_pool.x() / 2);
conv_5_pool.set(1, conv_5_pool.y() / 2);
pool_5.allocator() -> init(TensorInfo(conv_5_pool, 1, DataType::F32));

//init_fc_6
constexpr unsigned int fc_6_numoflabel = 4096;

const TensorShape fc_6_weights_shape(conv_5_pool.x() * conv_5_pool.y() * conv_5_pool.z(), fc_6_numoflabel);
const TensorShape fc_6_biases_shape(fc_6_numoflabel);
const TensorShape fc_6_out_shape(fc_6_numoflabel);

weights_6.allocator() -> init(TensorInfo(fc_6_weights_shape, 1, DataType::F32));
biases_6.allocator() -> init(TensorInfo(fc_6_biases_shape, 1, DataType::F32));
out_6.allocator() -> init(TensorInfo(fc_6_out_shape, 1, DataType::F32));

act_6.allocator() -> init(TensorInfo(fc_6_out_shape, 1, DataType::F32));

//init_fc_7
constexpr unsigned int fc_7_numoflabel = 4096;

const TensorShape fc_7_weights_shape(fc_6_out_shape.x(), fc_7_numoflabel);
const TensorShape fc_7_biases_shape(fc_7_numoflabel);
const TensorShape fc_7_out_shape(fc_7_numoflabel);

weights_7.allocator() -> init(TensorInfo(fc_7_weights_shape, 1, DataType::F32));
biases_7.allocator() -> init(TensorInfo(fc_7_biases_shape, 1, DataType::F32));
out_7.allocator() -> init(TensorInfo(fc_7_out_shape, 1, DataType::F32));

act_7.allocator() -> init(TensorInfo(fc_7_out_shape, 1, DataType::F32));

//init_fc_8
constexpr unsigned int fc_8_numoflabel = 1000;

const TensorShape fc_8_weights_shape(fc_7_out_shape.x(), fc_8_numoflabel);
const TensorShape fc_8_biases_shape(fc_8_numoflabel);
const TensorShape fc_8_out_shape(fc_8_numoflabel);

weights_8.allocator() -> init(TensorInfo(fc_8_weights_shape, 1, DataType::F32));
biases_8.allocator() -> init(TensorInfo(fc_8_biases_shape, 1, DataType::F32));
out_8.allocator() -> init(TensorInfo(fc_8_out_shape, 1, DataType::F32));

const TensorShape softmax_shape(fc_8_out_shape.x());
softmax_tensor.allocator() -> init(TensorInfo(softmax_shape, 1, DataType::F32));

/*----------------------------------END:[init_Tensor]----------------------------------*/

/*-----------------------------BEGIN:[Configure Functions]-----------------------------*/
//init_layer
//NEON CPU
NEConvolutionLayer    conv_1_1;
NEConvolutionLayer    conv_1_2;
NEConvolutionLayer    conv_2_1;
NEConvolutionLayer    conv_2_2;
NEConvolutionLayer    conv_3_1;
NEConvolutionLayer    conv_3_2;
NEConvolutionLayer    conv_3_3;
NEConvolutionLayer    conv_4_1;
NEConvolutionLayer    conv_4_2;
NEConvolutionLayer    conv_4_3;
NEConvolutionLayer    conv_5_1;
NEConvolutionLayer    conv_5_2;
NEConvolutionLayer    conv_5_3;
NEActivationLayer     Nact_1_1;
NEActivationLayer     Nact_1_2;
NEActivationLayer     Nact_2_1;
NEActivationLayer     Nact_2_2;
NEActivationLayer     Nact_3_1;
NEActivationLayer     Nact_3_2;
NEActivationLayer     Nact_3_3;
NEActivationLayer     Nact_4_1;
NEActivationLayer     Nact_4_2;
NEActivationLayer     Nact_4_3;
NEActivationLayer     Nact_5_1;
NEActivationLayer     Nact_5_2;
NEActivationLayer     Nact_5_3;
NEActivationLayer     Nact_6;
NEActivationLayer     Nact_7;
NEPoolingLayer        Npool_1;
NEPoolingLayer        Npool_2;
NEPoolingLayer        Npool_3;
NEPoolingLayer        Npool_4;
NEPoolingLayer        Npool_5;
NEFullyConnectedLayer fc_6;
NEFullyConnectedLayer fc_7;
NEFullyConnectedLayer fc_8;
NESoftmaxLayer        softmax;

//conv_1
//in: 224 * 224 * 3, kernel: 3 * 3 * 3 * 64, out: 224 * 224 * 64
conv_1_1.configure(&input, &weights_1_1, &biases_1_1, &out_1_1, PadStrideInfo(1, 1, 1, 1));

//in: 224 * 224 * 64, out: 224 * 224 * 64
Nact_1_1.configure(&out_1_1, &act_1_1, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 224 × 224 × 64, kernel: 3 * 3 * 64 * 64, out: 224 * 224 * 64
conv_1_2.configure(&act_1_1, &weights_1_2, &biases_1_2, &out_1_2, PadStrideInfo(1, 1, 1, 1));

//in: 224 * 224 * 64, out: 224 * 224 * 64
Nact_1_2.configure(&out_1_2, &act_1_2, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 224 * 224 * 64, out: 112 * 112 * 64
Npool_1.configure(&act_1_2, &pool_1, PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2)));

//conv_2
//in: 112 * 112 * 64, kernel: 3 * 3 * 64 * 128. out: 112 * 112 * 128
conv_2_1.configure(&pool_1, &weights_2_1, &biases_2_1, &out_2_1, PadStrideInfo(1, 1, 1, 1));

//in: 112 * 112 * 128, out: 112 * 112 * 128
Nact_2_1.configure(&out_2_1, &act_2_1, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 112 * 112 * 128, kernel: 3 * 3 * 128 * 128, out: 112 * 112 * 128
conv_2_2.configure(&act_2_1, &weights_2_2, &biases_2_2, &out_2_2, PadStrideInfo(1, 1, 1, 1));

//in: 112 * 112 * 128, out: 112 * 112 * 128
Nact_2_2.configure(&out_2_2, &act_2_2, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 112 * 112 * 128, out: 56 * 56 * 128
Npool_2.configure(&act_2_2, &pool_2, PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2)));

//conv_3
//in: 56 * 56 * 128, kernel: 3 * 3 * 128 * 256, out: 56 * 56 * 256
conv_3_1.configure(&pool_2, &weights_3_1, &biases_3_1, &out_3_1, PadStrideInfo(1, 1, 1, 1));

//in: 56 * 56 * 256, out: 56 * 56 * 256
Nact_3_1.configure(&out_3_1, &act_3_1, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 56 * 56 * 256, kernel: 3 * 3 * 256 * 256, out: 56 * 56 * 256
conv_3_2.configure(&act_3_1, &weights_3_2, &biases_3_2, &out_3_2, PadStrideInfo(1, 1, 1, 1));

//in: 56 * 56 * 256, out: 56 * 56 * 256
Nact_3_2.configure(&out_3_2, &act_3_2, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 56 * 56 * 256, kernel: 3 * 3 * 256 * 256, out: 56 * 56 * 256
conv_3_3.configure(&act_3_2, &weights_3_3, &biases_3_3, &out_3_3, PadStrideInfo(1, 1, 1, 1));

//in: 56 * 56 * 256, out: 56 * 56 * 256
Nact_3_3.configure(&out_3_3, &act_3_3, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 56 * 56 * 256, out: 28 * 28 * 256
Npool_3.configure(&act_3_3, &pool_3, PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2)));

//conv_4
//in: 28 * 28 * 256, kernel: 3 * 3 * 256 * 512, out: 28 * 28 * 512
conv_4_1.configure(&pool_3, &weights_4_1, &biases_4_1, &out_4_1, PadStrideInfo(1, 1, 1, 1));

//in: 28 * 28 * 512, out: 28 * 28 * 512
Nact_4_1.configure(&out_4_1, &act_4_1, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 28 * 28 * 512, kernel: 3 * 3 * 512 * 512, out: 28 * 28 * 512
conv_4_2.configure(&act_4_1, &weights_4_2, &biases_4_2, &out_4_2, PadStrideInfo(1, 1, 1, 1));

//in: 28 * 28 * 512, out: 28 * 28 * 512
Nact_4_2.configure(&out_4_2, &act_4_2, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 28 * 28 * 512, kernel: 3 * 3 * 512 * 512, out: 28 * 28 * 512
conv_4_3.configure(&act_4_2, &weights_4_3, &biases_4_3, &out_4_3, PadStrideInfo(1, 1, 1, 1));

//in: 28 * 28 * 512, out: 28 * 28 * 512
Nact_4_3.configure(&out_4_3, &act_4_3, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 28 * 28 * 512, out: 14 * 14 * 512
Npool_4.configure(&act_4_3, &pool_4, PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2)));

//conv_5
//in: 14 * 14 * 512, kernel: 3 * 3 * 512 * 512, out: 14 * 14 * 512
conv_5_1.configure(&pool_4, &weights_5_1, &biases_5_1, &out_5_1, PadStrideInfo(1, 1, 1, 1));

//in: 14 * 14 * 512, out: 14 * 14 * 512
Nact_5_1.configure(&out_5_1, &act_5_1, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 14 * 14 * 512, kernel: 3 * 3 * 512 * 512, out: 14 * 14 * 512
conv_5_2.configure(&act_5_1, &weights_5_2, &biases_5_2, &out_5_2, PadStrideInfo(1, 1, 1, 1));

//in: 14 * 14 * 512, out: 14 * 14 * 512
Nact_5_2.configure(&out_5_2, &act_5_2, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 14 * 14 * 512 kernel: 3 * 3 * 512 * 512, out: 14 * 14 * 512
conv_5_3.configure(&act_5_2, &weights_5_3, &biases_5_3, &out_5_3, PadStrideInfo(1, 1, 1, 1));

//in: 14 * 14 * 512, out: 14 * 14 * 512
Nact_5_3.configure(&out_5_3, &act_5_3, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//in: 14 * 14 * 512, out: 7 * 7 * 512
Npool_5.configure(&act_5_3, &pool_5, PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2)));

//fc_6
//in: 7 * 7 * 512, out: 4096
fc_6.configure(&pool_5, &weights_6, &biases_6, &out_6);

//in: 4096, out: 4096
Nact_6.configure(&out_6, &act_6, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//fc_7
//in: 4096, out: 4096
fc_7.configure(&act_6, &weights_7, &biases_7, &out_7);

//in:4096, out: 4096
Nact_7.configure(&out_7, &act_7, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));

//fc_8
//in: 4096, out: 1000
fc_8.configure(&act_7, &weights_8, &biases_8, &out_8);

//softmax layer: 1000
softmax.configure(&out_8, &softmax_tensor);

/*------------------------------END:[Configure Functions]------------------------------*/

/*------------------------------BEGIN:[Allocate tensors]-------------------------------*/

//input
input.allocator() -> allocate();

//conv_1
weights_1_1.allocator() -> allocate();
biases_1_1.allocator() -> allocate();
out_1_1.allocator() -> allocate();
act_1_1.allocator() -> allocate();

weights_1_2.allocator() -> allocate();
biases_1_2.allocator() -> allocate();
out_1_2.allocator() -> allocate();
act_1_2.allocator() -> allocate();

pool_1.allocator() -> allocate();

//conv_2
weights_2_1.allocator() -> allocate();
biases_2_1.allocator() -> allocate();
out_2_1.allocator() -> allocate();
act_2_1.allocator() -> allocate();

weights_2_2.allocator() -> allocate();
biases_2_2.allocator() -> allocate();
out_2_2.allocator() -> allocate();
act_2_2.allocator() -> allocate();

pool_2.allocator() -> allocate();

//conv_3
weights_3_1.allocator() -> allocate();
biases_3_1.allocator() -> allocate();
out_3_1.allocator() -> allocate();
act_3_1.allocator() -> allocate();

weights_3_2.allocator() -> allocate();
biases_3_2.allocator() -> allocate();
out_3_2.allocator() -> allocate();
act_3_2.allocator() -> allocate();

weights_3_3.allocator() -> allocate();
biases_3_3.allocator() -> allocate();
out_3_3.allocator() -> allocate();
act_3_3.allocator() -> allocate();

pool_3.allocator() -> allocate();

//conv_4
weights_4_1.allocator() -> allocate();
biases_4_1.allocator() -> allocate();
out_4_1.allocator() -> allocate();
act_4_1.allocator() -> allocate();

weights_4_2.allocator() -> allocate();
biases_4_2.allocator() -> allocate();
out_4_2.allocator() -> allocate();
act_4_2.allocator() -> allocate();

weights_4_3.allocator() -> allocate();
biases_4_3.allocator() -> allocate();
out_4_3.allocator() -> allocate();
act_4_3.allocator() -> allocate();

pool_4.allocator() -> allocate();

//conv_5
weights_5_1.allocator() -> allocate();
biases_5_1.allocator() -> allocate();
out_5_1.allocator() -> allocate();
act_5_1.allocator() -> allocate();

weights_5_2.allocator() -> allocate();
biases_5_2.allocator() -> allocate();
out_5_2.allocator() -> allocate();
act_5_2.allocator() -> allocate();

weights_5_3.allocator() -> allocate();
biases_5_3.allocator() -> allocate();
out_5_3.allocator() -> allocate();
act_5_3.allocator() -> allocate();

pool_5.allocator() -> allocate();

//fc_6
weights_6.allocator() -> allocate();
biases_6.allocator() -> allocate();
out_6.allocator() -> allocate();
act_6.allocator() -> allocate();

//fc_7
weights_7.allocator() -> allocate();
biases_7.allocator() -> allocate();
out_7.allocator() -> allocate();
act_7.allocator() -> allocate();

//fc_8
weights_8.allocator() -> allocate();
biases_8.allocator() -> allocate();
out_8.allocator() -> allocate();
softmax_tensor.allocator() -> allocate();

/*------------------------------END:[Allocate tensors]-------------------------------*/

/*-----------------------------BEGIN:[Load the weights]------------------------------*/

//init_pulice_variable
int x, y, z, k;
std::string line;

//conv_1_1
//W: 3 * 3 * 3 * 64
x = y = z = k = 0;
std::ifstream conv_1_1_Wfile("/data/local/tmp/conv1_1_W.csv");
line = "";
while(getline(conv_1_1_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_1_1.buffer() + weights_1_1.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_1_1_weights_shape.x()) ++ y, x = 0;
if(y == conv_1_1_weights_shape.y()) ++ z, y = 0;
if(z == conv_1_1_weights_shape.z()) ++ k, z = 0;
}
}
conv_1_1_Wfile.close();

//b: 64
x = y = z = k = 0;
std::ifstream conv_1_1_bfile("/data/local/tmp/conv1_1_b.csv");
line = "";
while(getline(conv_1_1_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_1_1.buffer() + biases_1_1.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_1_1_bfile.close();

//conv_1_2
//W: 3 * 3 * 64 * 64
x = y = z = k = 0;
std::ifstream conv_1_2_Wfile("/data/local/tmp/conv1_2_W.csv");
line = "";
while(getline(conv_1_2_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_1_2.buffer() + weights_1_2.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_1_2_weights_shape.x()) ++ y, x = 0;
if(y == conv_1_2_weights_shape.y()) ++ z, y = 0;
if(z == conv_1_2_weights_shape.z()) ++ k, z = 0;
}
}
conv_1_2_Wfile.close();

//b: 64
x = y = z = k = 0;
std::ifstream conv_1_2_bfile("/data/local/tmp/conv1_2_b.csv");
line = "";
while(getline(conv_1_2_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_1_2.buffer() + biases_1_2.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_1_2_bfile.close();

//conv_2_1
//W: 3 * 3 * 64 * 128
x = y = z = k = 0;
std::ifstream conv_2_1_Wfile("/data/local/tmp/conv2_1_W.csv");
line = "";
while(getline(conv_2_1_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_2_1.buffer() + weights_2_1.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_2_1_weights_shape.x()) ++ y, x = 0;
if(y == conv_2_1_weights_shape.y()) ++ z, y = 0;
if(z == conv_2_1_weights_shape.z()) ++ k, z = 0;
}
}
conv_2_1_Wfile.close();

//b: 128
x = y = z = k = 0;
std::ifstream conv_2_1_bfile("/data/local/tmp/conv2_1_b.csv");
line = "";
while(getline(conv_2_1_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_2_1.buffer() + biases_2_1.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_2_1_bfile.close();

//conv_2_2
//W: 3 * 3 * 128 * 128
x = y = z = k = 0;
std::ifstream conv_2_2_Wfile("/data/local/tmp/conv2_2_W.csv");
line = "";
while(getline(conv_2_2_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_2_2.buffer() + weights_2_2.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_2_2_weights_shape.x()) ++ y, x = 0;
if(y == conv_2_2_weights_shape.y()) ++ z, y = 0;
if(z == conv_2_2_weights_shape.z()) ++ k, z = 0;
}
}
conv_2_2_Wfile.close();

//b: 128
x = y = z = k = 0;
std::ifstream conv_2_2_bfile("/data/local/tmp/conv2_2_b.csv");
line = "";
while(getline(conv_2_2_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_2_2.buffer() + biases_2_2.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_2_2_bfile.close();

//conv_3_1
//W: 3 * 3 * 128 * 256
x = y = z = k = 0;
std::ifstream conv_3_1_Wfile("/data/local/tmp/conv3_1_W.csv");
line = "";
while(getline(conv_3_1_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_3_1.buffer() + weights_3_1.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_3_1_weights_shape.x()) ++ y, x = 0;
if(y == conv_3_1_weights_shape.y()) ++ z, y = 0;
if(z == conv_3_1_weights_shape.z()) ++ k, z = 0;
}
}
conv_3_1_Wfile.close();

//b: 256
x = y = z = k = 0;
std::ifstream conv_3_1_bfile("/data/local/tmp/conv3_1_b.csv");
line = "";
while(getline(conv_3_1_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_3_1.buffer() + biases_3_1.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_3_1_bfile.close();

//conv_3_2
//W: 3 * 3 * 256 * 256
x = y = z = k = 0;
std::ifstream conv_3_2_Wfile("/data/local/tmp/conv3_2_W.csv");
line = "";
while(getline(conv_3_2_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_3_2.buffer() + weights_3_2.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_3_2_weights_shape.x()) ++ y, x = 0;
if(y == conv_3_2_weights_shape.y()) ++ z, y = 0;
if(z == conv_3_2_weights_shape.z()) ++ k, z = 0;
}
}
conv_3_2_Wfile.close();

//b: 256
x = y = z = k = 0;
std::ifstream conv_3_2_bfile("/data/local/tmp/conv3_2_b.csv");
line = "";
while(getline(conv_3_2_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_3_2.buffer() + biases_3_2.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_3_2_bfile.close();

//conv_3_3
//W: 3 * 3 * 256 * 256
x = y = z = k = 0;
std::ifstream conv_3_3_Wfile("/data/local/tmp/conv3_3_W.csv");
line = "";
while(getline(conv_3_3_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_3_3.buffer() + weights_3_3.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_3_3_weights_shape.x()) ++ y, x = 0;
if(y == conv_3_3_weights_shape.y()) ++ z, y = 0;
if(z == conv_3_3_weights_shape.z()) ++ k, z = 0;
}
}
conv_3_3_Wfile.close();

//b: 256
x = y = z = k = 0;
std::ifstream conv_3_3_bfile("/data/local/tmp/conv3_3_b.csv");
line = "";
while(getline(conv_3_3_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_3_3.buffer() + biases_3_3.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_3_3_bfile.close();

//conv_4_1
//W: 3 * 3 * 256 * 512
x = y = z = k = 0;
std::ifstream conv_4_1_Wfile("/data/local/tmp/conv4_1_W.csv");
line = "";
while(getline(conv_4_1_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_4_1.buffer() + weights_4_1.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_4_1_weights_shape.x()) ++ y, x = 0;
if(y == conv_4_1_weights_shape.y()) ++ z, y = 0;
if(z == conv_4_1_weights_shape.z()) ++ k, z = 0;
}
}
conv_4_1_Wfile.close();

//b: 512
x = y = z = k = 0;
std::ifstream conv_4_1_bfile("/data/local/tmp/conv4_1_b.csv");
line = "";
while(getline(conv_4_1_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_4_1.buffer() + biases_4_1.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_4_1_bfile.close();

//conv_4_2
//W: 3 * 3 * 512 * 512
x = y = z = k = 0;
std::ifstream conv_4_2_Wfile("/data/local/tmp/conv4_2_W.csv");
line = "";
while(getline(conv_4_2_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_4_2.buffer() + weights_4_2.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_4_2_weights_shape.x()) ++ y, x = 0;
if(y == conv_4_2_weights_shape.y()) ++ z, y = 0;
if(z == conv_4_2_weights_shape.z()) ++ k, z = 0;
}
}
conv_4_2_Wfile.close();

//b: 512
x = y = z = k = 0;
std::ifstream conv_4_2_bfile("/data/local/tmp/conv4_2_b.csv");
line = "";
while(getline(conv_4_2_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_4_2.buffer() + biases_4_2.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_4_2_bfile.close();

//conv_4_3
//W: 3 * 3 * 512 * 512
x = y = z = k = 0;
std::ifstream conv_4_3_Wfile("/data/local/tmp/conv4_3_W.csv");
line = "";
while(getline(conv_4_3_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_4_3.buffer() + weights_4_3.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_4_3_weights_shape.x()) ++ y, x = 0;
if(y == conv_4_3_weights_shape.y()) ++ z, y = 0;
if(z == conv_4_3_weights_shape.z()) ++ k, z = 0;
}
}
conv_4_3_Wfile.close();

//b: 512
x = y = z = k = 0;
std::ifstream conv_4_3_bfile("/data/local/tmp/conv4_3_b.csv");
line = "";
while(getline(conv_4_3_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_4_3.buffer() + biases_4_3.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_4_3_bfile.close();

//conv_5_1
//W: 3 * 3 * 512 * 512
x = y = z = k = 0;
std::ifstream conv_5_1_Wfile("/data/local/tmp/conv5_1_W.csv");
line = "";
while(getline(conv_5_1_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_5_1.buffer() + weights_5_1.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_5_1_weights_shape.x()) ++ y, x = 0;
if(y == conv_5_1_weights_shape.y()) ++ z, y = 0;
if(z == conv_5_1_weights_shape.z()) ++ k, z = 0;
}
}
conv_5_1_Wfile.close();

//b: 512
x = y = z = k = 0;
std::ifstream conv_5_1_bfile("/data/local/tmp/conv5_1_b.csv");
line = "";
while(getline(conv_5_1_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_5_1.buffer() + biases_5_1.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_5_1_bfile.close();

//conv_5_2
//W: 3 * 3 * 512 * 512
x = y = z = k = 0;
std::ifstream conv_5_2_Wfile("/data/local/tmp/conv5_2_W.csv");
line = "";
while(getline(conv_5_2_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_5_2.buffer() + weights_5_2.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_5_2_weights_shape.x()) ++ y, x = 0;
if(y == conv_5_2_weights_shape.y()) ++ z, y = 0;
if(z == conv_5_2_weights_shape.z()) ++ k, z = 0;
}
}
conv_5_2_Wfile.close();

//b: 512
x = y = z = k = 0;
std::ifstream conv_5_2_bfile("/data/local/tmp/conv5_2_b.csv");
line = "";
while(getline(conv_5_2_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_5_2.buffer() + biases_5_2.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_5_2_bfile.close();

//conv_5_3
//W: 3 * 3 * 512 * 512
x = y = z = k = 0;
std::ifstream conv_5_3_Wfile("/data/local/tmp/conv5_3_W.csv");
line = "";
while(getline(conv_5_3_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_5_3.buffer() + weights_5_3.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == conv_5_3_weights_shape.x()) ++ y, x = 0;
if(y == conv_5_3_weights_shape.y()) ++ z, y = 0;
if(z == conv_5_3_weights_shape.z()) ++ k, z = 0;
}
}
conv_5_3_Wfile.close();

//b: 512
x = y = z = k = 0;
std::ifstream conv_5_3_bfile("/data/local/tmp/conv5_3_b.csv");
line = "";
while(getline(conv_5_3_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_5_3.buffer() + biases_5_3.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
conv_5_3_bfile.close();

//fc_6
//W: 7 * 7 * 512 * 4096
x = y = z = k = 0;
std::ifstream fc_6_Wfile("/data/local/tmp/fc6_W.csv");
line = "";
while(getline(fc_6_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_6.buffer() + weights_6.info()->offset_element_in_bytes(Coordinates(x, y, z, k))) = temp;
++ x;
if(x == fc_6_weights_shape.x()) ++ y, x = 0;
if(y == fc_6_weights_shape.y()) ++ z, y = 0;
if(z == fc_6_weights_shape.z()) ++ k, z = 0;
}
}
fc_6_Wfile.close();

//b: 4096
x = y = z = k = 0;
std::ifstream fc_6_bfile("/data/local/tmp/fc6_b.csv");
line = "";
while(getline(fc_6_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_6.buffer() + biases_6.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
fc_6_bfile.close();

//fc_7
//W: 4096 * 4096
x = y = z = k = 0;
std::ifstream fc_7_Wfile("/data/local/tmp/fc7_W.csv");
line = "";
while(getline(fc_7_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_7.buffer() + weights_7.info()->offset_element_in_bytes(Coordinates(x, y))) = temp;
++ x;
if(x == fc_7_weights_shape.x()) ++ y, x = 0;
}
}
fc_7_Wfile.close();

//b: 4096
x = y = z = k = 0;
std::ifstream fc_7_bfile("/data/local/tmp/fc7_b.csv");
line = "";
while(getline(fc_7_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_7.buffer() + biases_7.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
fc_7_bfile.close();

//fc_8
//W: 4096 * 1000
x = y = z = k = 0;
std::ifstream fc_8_Wfile("/data/local/tmp/fc8_W.csv");
line = "";
while(getline(fc_8_Wfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(weights_8.buffer() + weights_8.info()->offset_element_in_bytes(Coordinates(x, y))) = temp;
++ x;
if(x == fc_8_weights_shape.x()) ++ y, x = 0;
}
}
fc_8_Wfile.close();

//b: 1000
x = y = z = k = 0;
std::ifstream fc_8_bfile("/data/local/tmp/fc8_b.csv");
line = "";
while(getline(fc_8_bfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(biases_8.buffer() + biases_8.info()->offset_element_in_bytes(Coordinates(x))) = temp;
++ x;
}
}
fc_8_bfile.close();

/*------------------------------END:[Load the weights]-------------------------------*/

/*-----------------------------------BEGIN:[Input]-----------------------------------*/

x = y = z = k = 0;
float mean[3] = {123.68, 116.779, 103.939};
int t = 0;
std::ifstream inputfile("/data/local/tmp/input.csv");
line = "";int number = 0;
while(getline(inputfile, line)){
std::stringstream strstr(line);
std::string word = "";
while(getline(strstr, word, ',')){
float temp = StringToFloat(word);
*reinterpret_cast<float *>(input.buffer() + input.info()->offset_element_in_bytes(Coordinates(x, y, z))) = temp - mean[t];
t = (t + 1) % 3;
++ x;
if(x == input_shape.x()) ++ y, x = 0;
if(y == input_shape.y()) ++ z, y = 0;++number;
}
}
inputfile.close();

/*------------------------------------END:[Input]------------------------------------*/

/*--------------------------BEGIN:[Execute the functions]----------------------------*/

//time
struct timeval start, end;
gettimeofday(&start, NULL);

//conv_1
conv_1_1.run();
Nact_1_1.run();
conv_1_2.run();
Nact_1_2.run();
Npool_1.run();

//conv_2
conv_2_1.run();
Nact_2_1.run();
conv_2_2.run();
Nact_2_2.run();
Npool_2.run();

//conv_3
conv_3_1.run();
Nact_3_1.run();
conv_3_2.run();
Nact_3_2.run();
conv_3_3.run();
Nact_3_3.run();
Npool_3.run();

//conv_4
conv_4_1.run();
Nact_4_1.run();
conv_4_2.run();
Nact_4_2.run();
conv_4_3.run();
Nact_4_3.run();
Npool_4.run();

//conv_5
conv_5_1.run();
Nact_5_1.run();
conv_5_2.run();
Nact_5_2.run();
conv_5_3.run();
Nact_5_3.run();
Npool_5.run();

//fc_6
fc_6.run();
Nact_6.run();

//fc_7
fc_7.run();
Nact_7.run();

//fc_8
fc_8.run();
softmax.run();

gettimeofday(&end, NULL);
/*---------------------------END:[Execute the functions]-----------------------------*/

//test

std::cout << std::endl << std::endl << std::endl;
int timeuse = 1000000 * ( end.tv_sec - start.tv_sec ) + end.tv_usec -start.tv_usec;
printf("time: %d us\n", timeuse);

std::map<float, std::string>ma;
std::ifstream outfile("/data/local/tmp/synset.txt");
line = "";
x = 0;
while(getline(outfile, line)){
//std::cout << line << std::endl;
float temp = *reinterpret_cast<float *>(softmax_tensor.buffer() + softmax_tensor.info()->offset_element_in_bytes(Coordinates(x)));
++ x;
ma[temp] = line;
}
int h = 1;
for(auto it = ma.rbegin(); it != ma.rend(); ++ it){
std::cout << "N0 " << h++ << " " << it->first << " is " << it->second << std::endl;
}

std::cout << "fine! Jason!" << std::endl;
}

/** Main program for convolution test
* @param[in] argc Number of arguments
* @param[in] argv Arguments ( [optional] Path to PPM image to process )
*/
int main(int argc, const char **argv)
{
return test_helpers::run_example(argc, argv, main_neon_vgg16);
}

PO主的系统是Linux ubuntu16。如果不想在手机上跑的想在linux上跑的也可以参考这个流程，基本一样，详细操作可以上arm compute library看！！！

最后的最后有什么问题，错误，意见或者需要楼主的改进请尽情提出哈哈哈哈！！！