c++实现lenet推理模型
2024-01-13 本文已影响0人
一路向后
1.tensor.h
#ifndef _CONVNET_TENSOR_H_
#define _CONVNET_TENSOR_H_
#include <vector>
namespace convnet {
class Tensor {
public:
Tensor();
Tensor(int a);
Tensor(int a, int b);
Tensor(int a, int b, int c);
~Tensor();
void resize(int a);
void resize(int a, int b);
void resize(int a, int b, int c);
void relu();
void sigmoid();
void argmax(int &s);
void set(std::vector<double> &data);
int size();
private:
friend class Linear;
friend class Conv2d;
friend class MaxPool2d;
friend class Reshape;
std::vector<int> dim;
std::vector<double> data;
};
}
#endif
2.tensor.cpp
#include <math.h>
#include <iostream>
#include "tensor.h"
using namespace std;
convnet::Tensor::Tensor()
{
//dim.clear();
}
convnet::Tensor::Tensor(int a)
{
dim.resize(1);
dim[0] = a;
if(dim[0] > 0)
{
data.resize(a);
}
}
convnet::Tensor::Tensor(int a, int b)
{
dim.resize(2);
dim[0] = a;
dim[1] = b;
if(a*b > 0)
{
data.resize(a*b);
}
}
convnet::Tensor::Tensor(int a, int b, int c)
{
dim.resize(3);
dim[0] = a;
dim[1] = b;
dim[2] = c;
if(a*b*c > 0)
{
data.resize(a*b*c);
}
}
void convnet::Tensor::resize(int a)
{
dim.resize(1);
dim[0] = a;
if(dim[0] > 0 && a != data.size())
{
data.resize(a);
}
}
void convnet::Tensor::resize(int a, int b)
{
dim.resize(2);
dim[0] = a;
dim[1] = b;
if(a*b > 0 && a*b != data.size())
{
data.resize(a*b);
}
}
void convnet::Tensor::resize(int a, int b, int c)
{
if(dim.size() != 0)
{
dim.clear();
}
if(data.size() != 0)
{
data.clear();
}
dim.resize(3);
dim[0] = a;
dim[1] = b;
dim[2] = c;
if(a*b*c > 0)
{
data.resize(a*b*c);
}
}
convnet::Tensor::~Tensor()
{
dim.clear();
}
void convnet::Tensor::set(std::vector<double> &data)
{
this->data = data;
}
int convnet::Tensor::size()
{
return data.size();
}
void convnet::Tensor::relu()
{
for(int i=0; i<data.size(); i++)
{
if(data[i] < 0)
{
data[i] = 0;
}
}
}
void convnet::Tensor::sigmoid()
{
for(int i=0; i<data.size(); i++)
{
data[i] = 1 / (1+expf(-data[i]));
}
}
void convnet::Tensor::argmax(int &s)
{
if(dim.size() == 1 && dim[0] > 0)
{
int u = 0;
int i = 1;
for(i=1; i<dim[0]; i++)
{
if(data[i] >= data[u])
{
u = i;
}
}
s = u;
}
}
3.conv2d.h
#ifndef _CONVNET_CONV2D_H_
#define _CONVNET_CONV2D_H_
#include "tensor.h"
namespace convnet {
class Conv2d {
public:
Conv2d(int iw, int ih, int ic, int ow, int oh, int oc, int kw, int kh);
Conv2d(int ic, int oc, int kw);
~Conv2d();
void set(Tensor *i, Tensor *o);
void setargs(std::vector<double> &k);
void setargs(std::vector<double> &k, std::vector<double> &b);
void forward();
void print(int type);
private:
int iw;
int ih;
int ic;
int ow;
int oh;
int oc;
int kw;
int kh;
Tensor kernel;
Tensor bias;
Tensor *input;
Tensor *output;
};
}
#endif
4.conv2d.cpp
#include <stdio.h>
#include <iostream>
#include <cassert>
#include "conv2d.h"
using namespace std;
convnet::Conv2d::Conv2d(int iw, int ih, int ic, int ow, int oh, int oc, int kw, int kh)
{
this->iw = iw;
this->ih = ih;
this->ic = ic;
this->ow = ow;
this->oh = oh;
this->oc = oc;
this->kw = kw;
this->kh = kh;
}
convnet::Conv2d::Conv2d(int ic, int oc, int kw)
{
this->iw = 0;
this->ih = 0;
this->ic = ic;
this->ow = 0;
this->oh = 0;
this->oc = oc;
this->kw = kw;
this->kh = kw;
}
convnet::Conv2d::~Conv2d()
{
}
void convnet::Conv2d::set(Tensor *i, Tensor *o)
{
input = i;
output = o;
if(input->size() != iw * ih * ic)
{
if(iw * ih != 0)
{
input->resize(ic, ih, iw);
}
else
{
if(input->dim.size() == 3)
{
iw = input->dim[1];
ih = input->dim[2];
//cout << "iw: " << iw << ", " << "ih: " << ih << endl;
}
}
}
if(output->size() != ow * oh * oc || ow * oh == 0)
{
if(ow * oh == 0)
{
ow = abs(iw - kw + 1);
oh = abs(ih - kh + 1);
//cout << "ow: " << ow << ", " << "oh: " << oh << ", oc: " << oc << ", ic: " << ic << endl;
}
output->resize(oc, oh, ow);
}
}
void convnet::Conv2d::setargs(std::vector<double> &k)
{
if(kw * kh == k.size())
{
kernel.resize(kw * kh);
kernel.set(k);
}
}
void convnet::Conv2d::setargs(std::vector<double> &k, std::vector<double> &b)
{
bias.dim.resize(1);
bias.dim[0] = b.size();
bias.data = b;
assert(k.size() == kw * kh * ic * oc);
assert(b.size() == oc);
kernel.set(k);
}
void convnet::Conv2d::forward()
{
int i, j, k, l, p, q;
for(i=0; i<oc; i++)
{
for(j=0; j<oh; j++)
{
for(k=0; k<ow; k++)
{
output->data[i*oh*ow+j*ow+k] = 0.0;
for(l=0; l<ic; l++)
{
//printf("i = %d, kh = %d, kw = %d\n", i, kh, kw);
for(p=0; p<kh; p++)
{
for(q=0; q<kw; q++)
{
//printf("p = %d, q = %d, u = %d, k = %d\n", p, q, i*kh*kw+p*kw+q, kernel.data[i*kh*kw+p*kw+q]);
output->data[i*oh*ow+j*ow+k] += (kernel.data[(i*ic+l)*kh*kw+p*kw+q] * input->data[l*ih*iw+(j+p)*iw+(k+q)]);
//output->data[i*oh*ow+j*ow+k] += kernel.data[(l)*kh*kw+p*kw+q] * input->data[l*ih*iw+(j+p)*iw+(k+q)];
}
}
//output->data[i*oh*ow+j*ow+k] += bias.data[i];
}
output->data[i*oh*ow+j*ow+k] += bias.data[i];
}
}
}
}
void convnet::Conv2d::print(int type)
{
if(type == 0)
{
int m = iw * ih * ic;
for(int i=0; i<m; i++)
{
printf("%.6lf ", input->data[i]);
}
printf("\n");
}
else if(type == 1)
{
int n = ow * oh * oc;
for(int i=0; i<n; i++)
{
printf("%.6lf ", output->data[i]);
}
printf("\n");
}
}
5.maxpool2d.h
#define _CONVNET_MAXPOOL2D_H_
#include "tensor.h"
namespace convnet {
class MaxPool2d {
public:
MaxPool2d(int iw, int ih, int w, int h, int stride);
MaxPool2d(int w, int h, int stride);
MaxPool2d(int w, int stride);
~MaxPool2d();
void set(Tensor *i, Tensor *o);
void forward();
void print(int type);
private:
int ic;
int iw;
int ih;
int ow;
int oh;
int oc;
int kw;
int kh;
int stride;
Tensor *input;
Tensor *output;
};
}
#endif
6.maxpool2d.cpp
#include <stdio.h>
#include <iostream>
#include "maxpool2d.h"
using namespace std;
convnet::MaxPool2d::MaxPool2d(int iw, int ih, int w, int h, int stride)
{
this->iw = iw;
this->ih = ih;
this->kw = kw;
this->kh = kh;
this->stride = stride;
ow = iw / stride;
oh = ih / stride;
}
convnet::MaxPool2d::MaxPool2d(int w, int h, int stride)
{
this->iw = 0;
this->ih = 0;
this->ow = 0;
this->oh = 0;
this->kw = w;
this->kh = h;
this->stride = stride;
}
convnet::MaxPool2d::MaxPool2d(int w, int stride)
{
this->iw = 0;
this->ih = 0;
this->ic = 0;
this->ow = 0;
this->oh = 0;
this->oc = 0;
this->kw = w;
this->kh = w;
this->stride = stride;
}
convnet::MaxPool2d::~MaxPool2d()
{
}
void convnet::MaxPool2d::set(Tensor *i, Tensor *o)
{
if(iw != 0 && ih != 0)
{
if(iw * ih != i->size())
{
i->resize(iw, ih);
}
}
else
{
if(i->dim.size() == 2)
{
iw = i->dim[0];
ih = i->dim[1];
ic = 1;
oc = 1;
}
else if(i->dim.size() == 3)
{
ic = i->dim[0];
ih = i->dim[1];
iw = i->dim[2];
oc = ic;
//cout << "dddd" << iw << ", " << ih << ", ic: " << ic << endl;
}
int j = 0;
ow = iw / stride;
oh = ih / stride;
}
if(ow * oh * oc != 0 && ow * oh * oc != o->size())
{
o->resize(oc, oh, ow);
//cout << "eee " << ow << ", " << oh << ", oc: " << oc << endl;
}
input = i;
output = o;
}
void convnet::MaxPool2d::forward()
{
int i, j, k;
int p, q;
for(i=0; i<oc; i++)
{
for(j=0; j<oh; j++)
{
for(k=0; k<ow; k++)
{
output->data[i*oh*ow+j*ow+k] = input->data[i*ih*iw+j*stride*iw+k*stride];
for(p=0; p<stride; p++)
{
for(q=0; q<stride; q++)
{
if(input->data[i*ih*iw+(j*stride+p)*iw+(k*stride+q)] > output->data[i*oh*ow+j*ow+k])
{
output->data[i*oh*ow+j*ow+k] = input->data[i*ih*iw+(j*stride+p)*iw+(k*stride+q)];
}
}
}
}
}
}
}
void convnet::MaxPool2d::print(int type)
{
if(type == 0)
{
int m = ic * ih * iw;
for(int i=0; i<m; i++)
{
printf("%.6lf ", input->data[i]);
}
printf("\n");
}
else if(type == 1)
{
int n = oc * oh * ow;
for(int i=0; i<n; i++)
{
printf("%.6lf ", output->data[i]);
}
printf("\n");
}
}
7.reshape.h
#ifndef _CONVNET_RESHAPE_H_
#define _CONVNET_RESHAPE_H_
#include "tensor.h"
namespace convnet {
class Reshape {
public:
Reshape(int a, int b);
~Reshape();
void set(Tensor *input, Tensor *output);
void forward();
void print(int type);
private:
std::vector<int> dim;
Tensor *input;
Tensor *output;
};
}
#endif
8.reshape.cpp
#include <stdio.h>
#include <math.h>
#include "reshape.h"
#include <cassert>
convnet::Reshape::Reshape(int a, int b)
{
dim.resize(2);
dim[0] = a;
dim[1] = b;
}
convnet::Reshape::~Reshape()
{
}
void convnet::Reshape::set(Tensor *input, Tensor *output)
{
if(dim.size() > 0)
{
if(dim[0] == -1 && dim[1] > 0)
{
//printf("size: %d\n", input->dim.size());
//printf("dim: %d\n", input->dim[0] * input->dim[1]);
if(input->dim.size() == 3)
{
assert(input->dim[0] * input->dim[1] * input->dim[2] == dim[1]);
}
else if(input->dim.size() == 2)
{
assert(input->dim[0] * input->dim[1] == dim[1]);
}
output->resize(dim[1]);
}
this->input = input;
this->output = output;
}
}
void convnet::Reshape::forward()
{
output->data = input->data;
}
void convnet::Reshape::print(int type)
{
if(type == 0)
{
int m = abs(dim[0] * dim[1]);
//printf("input size: %p\n", input);
for(int i=0; i<m; i++)
{
printf("%.6lf ", input->data[i]);
}
printf("\n");
}
else if(type == 1)
{
int n = abs(dim[0] * dim[1]);
for(int i=0; i<n; i++)
{
printf("%.6lf ", output->data[i]);
}
printf("\n");
}
}
9.linear.h
#ifndef _CONVNET_LINEAR_H_
#define _CONVNET_LINEAR_H_
#include "tensor.h"
namespace convnet {
class Linear {
public:
Linear(int m, int n);
void set(Tensor *i, Tensor *o);
void forward();
void setargs(std::vector<double> &w, std::vector<double> &b);
void print(int type);
private:
int m;
int n;
Tensor weight;
Tensor bias;
Tensor *input;
Tensor *output;
};
}
#endif
10.linear.cpp
#include <stdio.h>
#include <cassert>
#include "linear.h"
convnet::Linear::Linear(int m, int n)
{
this->m = m;
this->n = n;
weight.resize(m, n);
bias.resize(n);
}
void convnet::Linear::set(Tensor *i, Tensor *o)
{
//printf("m = %d, %d\n", m, i->size());
if(m != i->size())
{
i->resize(m);
}
if(n != o->size())
{
o->resize(n);
}
input = i;
output = o;
}
void convnet::Linear::setargs(std::vector<double> &w, std::vector<double> &b)
{
assert(w.size() == m * n && b.size() == n);
{
weight.data = w;
bias.data = b;
}
}
void convnet::Linear::forward()
{
for(int out=0; out<n; out++)
{
output->data[out] = 0.0;
for(int in=0; in<m; in++)
{
output->data[out] += input->data[in] * weight.data[m*out + in];
}
output->data[out] += bias.data[out];
}
}
void convnet::Linear::print(int type)
{
if(type == 0)
{
for(int i=0; i<m; i++)
{
printf("%.6lf ", input->data[i]);
}
printf("\n");
}
else if(type == 1)
{
for(int i=0; i<n; i++)
{
printf("%.6lf ", output->data[i]);
}
printf("\n");
}
}
11.main.cpp
#include <iostream>
#include <cstring>
#include <cassert>
#include "conv2d.h"
#include "maxpool2d.h"
#include "linear.h"
#include "reshape.h"
#include "pugixml.hpp"
using namespace std;
using namespace convnet;
void get_numbers(const std::string &line, std::vector<int> &s)
{
char num[32];
int len = line.length();
int i = 0;
int j = 0;
int k = 0;
int flag = 0;
while(i < len)
{
flag = 0;
if(line[i] == '+' || line[i] == '-' || (line[i] >= '0' && line[i] <= '9'))
{
flag = 1;
num[j++] = line[i];
}
else if(isspace(line[i]))
{
}
else
{
break;
}
i++;
if(flag == 1)
{
while(line[i] == '+' || line[i] == '-' || (line[i] >= '0' && line[i] <= '9'))
{
num[j++] = line[i];
i++;
}
num[j] = 0x00;
if(i < len && line[i] == ',')
{
i++;
}
k = atoi(num);
j = 0;
s.push_back(k);
}
}
}
void get_numbers(const std::string &line, std::vector<double> &s)
{
char num[32];
int len = line.length();
int i = 0;
int j = 0;
double k = 0;
int flag = 0;
while(i < len)
{
flag = 0;
if(line[i] == '+' || line[i] == '-' || line[i] == 'e' || line[i] == '.' || (line[i] >= '0' && line[i] <= '9'))
{
flag = 1;
num[j++] = line[i];
}
else if(isspace(line[i]))
{
}
else
{
break;
}
i++;
if(flag == 1)
{
while(line[i] == '+' || line[i] == '-' || line[i] == 'e' || line[i] == '.' || (line[i] >= '0' && line[i] <= '9'))
{
num[j++] = line[i];
i++;
}
num[j] = 0x00;
if(i < len && line[i] == ',')
{
i++;
}
char *endptr = NULL;
k = strtod(num, &endptr);
j = 0;
s.push_back(k);
}
}
}
int main()
{
Tensor *input = NULL;
std::vector<Tensor *> outputs;
std::vector<Conv2d *> conv2ds;
std::vector<MaxPool2d *> maxpool2ds;
std::vector<Reshape *> reshapes;
std::vector<Linear *> linears;
std::vector<int> actfuncs;
std::vector<std::vector<int>> args;
std::vector<int> args_brh;
std::vector<std::string> types;
std::vector<std::string> names;
std::vector<int> intypes;
std::vector<double> nums[2];
int k = 0;
int u = 0;
int v = 0;
int w = 0;
int l = 0;
pugi::xml_document doc;
pugi::xml_parse_result result = doc.load_file("lenet.xml");
if(!result)
{
return -1;
}
pugi::xml_node xnodes = doc.child("lenet");
for(pugi::xml_node xnode = xnodes.first_child(); xnode != NULL; xnode = xnode.next_sibling())
{
std::string type = xnode.child("type").text().as_string();
std::string name = xnode.child("name").text().as_string();
//cout << "Type: " << type << endl;
if(type == "input")
{
std::string value = xnode.child("value").text().as_string();
get_numbers(value, args_brh);
if(args_brh.size() == 3)
{
types.push_back(type);
names.push_back(name);
args.push_back(args_brh);
if(xnode.child("actfunc") != NULL)
{
std::string func = xnode.child("actfunc").text().as_string();
if(func == "relu")
{
actfuncs.push_back(1);
}
else
{
actfuncs.push_back(-1);
}
}
else
{
actfuncs.push_back(-1);
}
}
k++;
args_brh.clear();
}
else if(type == "Conv2d")
{
std::string value = xnode.child("value").text().as_string();
get_numbers(value, args_brh);
if(args_brh.size() == 3)
{
types.push_back(type);
names.push_back(name);
args.push_back(args_brh);
if(xnode.child("actfunc") != NULL)
{
std::string func = xnode.child("actfunc").text().as_string();
if(func == "relu")
{
actfuncs.push_back(1);
}
else
{
actfuncs.push_back(-1);
}
}
else
{
actfuncs.push_back(-1);
}
}
k++;
args_brh.clear();
}
else if(type == "MaxPool2d")
{
std::string value = xnode.child("value").text().as_string();
get_numbers(value, args_brh);
if(args_brh.size() == 2)
{
types.push_back(type);
names.push_back(name);
args.push_back(args_brh);
if(xnode.child("actfunc") != NULL)
{
std::string func = xnode.child("actfunc").text().as_string();
if(func == "relu")
{
actfuncs.push_back(1);
}
else
{
actfuncs.push_back(-1);
}
}
else
{
actfuncs.push_back(-1);
}
}
k++;
args_brh.clear();
}
else if(type == "reshape")
{
std::string value = xnode.child("value").text().as_string();
get_numbers(value, args_brh);
if(args_brh.size() == 2)
{
types.push_back(type);
names.push_back(name);
args.push_back(args_brh);
if(xnode.child("actfunc") != NULL)
{
std::string func = xnode.child("actfunc").text().as_string();
if(func == "relu")
{
actfuncs.push_back(1);
}
else
{
actfuncs.push_back(-1);
}
}
else
{
actfuncs.push_back(-1);
}
}
k++;
args_brh.clear();
}
else if(type == "Linear")
{
std::string value = xnode.child("value").text().as_string();
get_numbers(value, args_brh);
if(args_brh.size() == 2)
{
types.push_back(type);
names.push_back(name);
args.push_back(args_brh);
if(xnode.child("actfunc") != NULL)
{
std::string func = xnode.child("actfunc").text().as_string();
if(func == "relu")
{
actfuncs.push_back(1);
}
else
{
actfuncs.push_back(-1);
}
}
else
{
actfuncs.push_back(-1);
}
}
k++;
args_brh.clear();
}
}
for(k=0; k<types.size(); k++)
{
string type = types[k];
//cout << "type: " << type << ", name: " << names[k] << endl;
/*if(actfuncs[k] == 1)
{
cout << "call function relu" << endl;
}*/
if(type == "input")
{
input = new Tensor(args[k][0], args[k][1], args[k][2]);
intypes.push_back(1);
}
else if(type == "Conv2d")
{
conv2ds.push_back(new Conv2d(args[k][0], args[k][1], args[k][2]));
outputs.push_back(new Tensor());
u = conv2ds.size() - 1;
v = outputs.size() - 1;
if(v == 0)
{
conv2ds[u]->set(input, outputs[v]);
}
else
{
conv2ds[u]->set(outputs[v-1], outputs[v]);
}
intypes.push_back(2);
}
else if(type == "MaxPool2d")
{
maxpool2ds.push_back(new MaxPool2d(args[k][0], args[k][1]));
outputs.push_back(new Tensor());
u = maxpool2ds.size() - 1;
v = outputs.size() - 1;
maxpool2ds[u]->set(outputs[v-1], outputs[v]);
intypes.push_back(3);
}
else if(type == "reshape")
{
reshapes.push_back(new Reshape(args[k][0], args[k][1]));
outputs.push_back(new Tensor());
u = reshapes.size() - 1;
v = outputs.size() - 1;
reshapes[u]->set(outputs[v-1], outputs[v]);
intypes.push_back(4);
}
else if(type == "Linear")
{
linears.push_back(new Linear(args[k][0], args[k][1]));
outputs.push_back(new Tensor());
u = linears.size() - 1;
v = outputs.size() - 1;
linears[u]->set(outputs[v-1], outputs[v]);
intypes.push_back(5);
}
}
FILE *fp = fopen("lenet.mod", "r");
char line[128];
int nodes = 0;
if(fp == NULL)
{
return -1;
}
memset(line, 0x00, sizeof(line));
fread(line, 1, 6, fp);
fread(&nodes, 1, 4, fp);
if(strcmp(line, "lenet") != 0)
{
return -1;
}
u = 0;
v = 0;
w = 0;
for(k=0; k<types.size(); k++)
{
if(intypes[k] == 1)
{
//cout << "do nothing" << endl;
}
else if(intypes[k] == 2)
{
//cout << "deal conv2d" << endl;
nums[0].clear();
memset(line, 0x00, sizeof(line));
fread(line, 1, names[k].length()+8, fp);
fread(&nodes, 1, 4, fp);
//printf("line = %s, number = %d\n", line, nodes);
nums[0].resize(nodes);
for(int i=0; i<nodes; i++)
{
fread(&nums[0][i], 1, sizeof(double), fp);
//printf("nums = %.6lf\n", nums[0][i]);
}
nums[1].clear();
memset(line, 0x00, sizeof(line));
fread(line, 1, names[k].length()+6, fp);
fread(&nodes, 1, 4, fp);
//printf("line = %s, number = %d\n", line, nodes);
nums[1].resize(nodes);
for(int i=0; i<nodes; i++)
{
fread(&nums[1][i], 1, sizeof(double), fp);
//printf("nums = %.6lf\n", nums[1][i]);
}
conv2ds[u++]->setargs(nums[0], nums[1]);
}
else if(intypes[k] == 5)
{
//cout << "deal linear" << endl;
nums[0].clear();
memset(line, 0x00, sizeof(line));
fread(line, 1, names[k].length()+8, fp);
fread(&nodes, 1, 4, fp);
//printf("line = %s, number = %d\n", line, nodes);
nums[0].resize(nodes);
for(int i=0; i<nodes; i++)
{
fread(&nums[0][i], 1, sizeof(double), fp);
//printf("nums = %.6lf\n", nums[0][i]);
}
nums[1].clear();
memset(line, 0x00, sizeof(line));
fread(line, 1, names[k].length()+6, fp);
fread(&nodes, 1, 4, fp);
//printf("line = %s, number = %d\n", line, nodes);
nums[1].resize(nodes);
for(int i=0; i<nodes; i++)
{
fread(&nums[1][i], 1, sizeof(double), fp);
//printf("nums = %.6lf\n", nums[1][i]);
}
linears[v++]->setargs(nums[0], nums[1]);
}
}
fclose(fp);
nums[0].clear();
nums[1].clear();
fp = fopen("input.txt", "r");
if(fp == NULL)
{
return -1;
}
while(!feof(fp))
{
memset(line, 0x00, sizeof(line));
fgets(line, 127, fp);
get_numbers(line, nums[0]);
}
fclose(fp);
//cout << "nums = " << nums[0].size() << endl;
assert(nums[0].size() == input->size());
input->set(nums[0]);
nums[0].clear();
u = 0;
v = 0;
w = 0;
l = 0;
//printf("types.size = %d, intypes.size = %d\n", types.size(), intypes.size());
for(k=1; k<types.size(); k++)
{
switch(intypes[k])
{
case 2:
conv2ds[u]->forward();
if(actfuncs[k] == 1)
{
outputs[k-1]->relu();
//cout << "call relu" << endl;
}
//if(u==1)
//conv2ds[u]->print(1);
u++;
break;
case 3:
maxpool2ds[v]->forward();
if(actfuncs[k] == 1)
{
outputs[k-1]->relu();
}
//if(v==1)
//maxpool2ds[v]->print(1);
v++;
break;
case 4:
reshapes[w]->forward();
if(actfuncs[k] == 1)
{
outputs[k-1]->relu();
}
//cout << "reshape" << endl;
//reshapes[w]->print(0);
w++;
break;
case 5:
linears[l]->forward();
if(actfuncs[k] == 1)
{
outputs[k-1]->relu();
}
//linears[l]->print(1);
l++;
break;
default:
break;
}
}
k = types.size() - 2;
u = 0;
outputs[k]->argmax(u);
printf("predict: %d\n", u);
if(input != NULL)
{
delete input;
input = NULL;
}
for(k=0; k<conv2ds.size(); k++)
{
if(conv2ds[k] != NULL)
{
delete conv2ds[k];
conv2ds[k] = NULL;
}
}
conv2ds.clear();
for(k=0; k<outputs.size(); k++)
{
if(outputs[k] != NULL)
{
delete outputs[k];
outputs[k] = NULL;
}
}
outputs.clear();
for(k=0; k<maxpool2ds.size(); k++)
{
if(maxpool2ds[k] != NULL)
{
delete maxpool2ds[k];
maxpool2ds[k] = NULL;
}
}
maxpool2ds.clear();
for(k=0; k<reshapes.size(); k++)
{
if(reshapes[k] != NULL)
{
delete reshapes[k];
reshapes[k] = NULL;
}
}
reshapes.clear();
for(k=0; k<linears.size(); k++)
{
if(linears[k] != NULL)
{
delete linears[k];
linears[k] = NULL;
}
}
linears.clear();
return 0;
}
12.Makefile
CXX=g++
STD=-std=c++11
DEBUG=-g
LDFLAGS=
CXXFLASG=
OBJS=linear.o tensor.o conv2d.o maxpool2d.o pugixml.o reshape.o
lenet: main.cpp $(OBJS)
$(CXX) $(DEBUG) -o lenet main.cpp $(OBJS) $(STD) $(LDFLAGS)
linear.o: linear.cpp
$(CXX) $(DEBUG) -c linear.cpp $(STD) $(CXXLFAGS)
tensor.o: tensor.cpp tensor.h
$(CXX) $(DEBUG) -c tensor.cpp $(STD) $(CXXLFAGS)
conv2d.o: conv2d.cpp conv2d.h
$(CXX) $(DEBUG) -c conv2d.cpp $(STD) $(CXXLFAGS)
maxpool2d.o: maxpool2d.cpp maxpool2d.h
$(CXX) $(DEBUG) -c maxpool2d.cpp $(STD) $(CXXLFAGS)
reshape.o: reshape.cpp reshape.h
$(CXX) $(DEBUG) -c reshape.cpp $(STD) $(CXXLFAGS)
pugixml.o: pugixml.cpp pugixml.hpp
$(CXX) $(DEBUG) -c pugixml.cpp $(STD) $(CXXLFAGS)
clean:
rm -rf lenet
rm -rf $(OBJS)
13.lenet.xml
<?xml version="1.0" encoding="UTF-8"?>
<lenet>
<node>
<type>input</type>
<name>input</name>
<value>1, 28, 28</value>
</node>
<node>
<type>Conv2d</type>
<name>conv1</name>
<value>1, 6, 5</value>
<actfunc>relu</actfunc>
</node>
<node>
<type>MaxPool2d</type>
<name>pool1</name>
<value>2, 2</value>
</node>
<node>
<type>Conv2d</type>
<name>conv2</name>
<value>6, 16, 5</value>
<actfunc>relu</actfunc>
</node>
<node>
<type>MaxPool2d</type>
<name>pool2</name>
<value>2, 2</value>
</node>
<node>
<type>reshape</type>
<name>view</name>
<value>-1, 256</value>
</node>
<node>
<type>Linear</type>
<name>fc1</name>
<value>256, 120</value>
<actfunc>relu</actfunc>
</node>
<node>
<type>Linear</type>
<name>fc2</name>
<value>120, 84</value>
<actfunc>relu</actfunc>
</node>
<node>
<type>Linear</type>
<name>fc3</name>
<value>84, 10</value>
</node>
</lenet>
14.input.txt
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, 0.6450,
1.9305, 1.5996, 1.4978, 0.3395, 0.0340, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, 2.4015,
2.8088, 2.8088, 2.8088, 2.8088, 2.6433, 2.0960, 2.0960,
2.0960, 2.0960, 2.0960, 2.0960, 2.0960, 2.0960, 1.7396,
0.2377, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, 0.4286,
1.0268, 0.4922, 1.0268, 1.6505, 2.4651, 2.8088, 2.4396,
2.8088, 2.8088, 2.8088, 2.7578, 2.4906, 2.8088, 2.8088,
1.3577, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.2078, 0.4159, -0.2460,
0.4286, 0.4286, 0.4286, 0.3268, -0.1569, 2.5797, 2.8088,
0.9250, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, 0.6322, 2.7960, 2.2360,
-0.1951, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.1442, 2.5415, 2.8215, 0.6322,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, 1.2177, 2.8088, 2.6051, 0.1358,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, 0.3268, 2.7451, 2.8088, 0.3649, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, 1.2686, 2.8088, 1.9560, -0.3606, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.3097, 2.1851, 2.7324, 0.3140, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, 1.1795, 2.8088, 1.8923, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
0.5304, 2.7706, 2.6306, 0.3013, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.1824,
2.3887, 2.8088, 1.6887, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.3860, 2.1596,
2.8088, 2.3633, 0.0213, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, 0.0595, 2.8088,
2.8088, 0.5559, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.0296, 2.4269, 2.8088,
1.0395, -0.4115, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, 1.2686, 2.8088, 2.8088,
0.2377, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, 0.3522, 2.6560, 2.8088, 2.8088,
0.2377, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, 1.1159, 2.8088, 2.8088, 2.3633,
0.0849, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, 1.1159, 2.8088, 2.2105, -0.1951,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242,
-0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242, -0.4242
15.编译源码
$ make
16.运行及其结果
$ ./lenet
predict: 7
