caffe blob 数据存储 blob.ccp

最新推荐文章于 2019-01-23 11:11:11 发布

原创最新推荐文章于 2019-01-23 11:11:11 发布 · 4k 阅读

2 ·

CC 4.0 BY-SA版权

菜鸟从零开始学习Deep learning 专栏收录该内容

87 篇文章

订阅专栏

本文详细介绍了在Caffe中使用Blob类进行数据的存储和读取过程，包括头文件导入、函数定义及具体实现步骤。同时，展示了在Matlab中如何通过特定函数读取存储的数据。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

blob。h函数修改：

void save_to_file(const std::string& file_name);

blob。cpp 的修改：

头文件

#include <fstream>
#include <iostream>
using namespace std;

template <typename Dtype>

void Blob<Dtype>::save_to_file( const std::string& file_name)
{
FILE* fp = fopen(file_name.c_str(), "wb");
CHECK(fp);
fwrite(&num_, sizeof(int), 1, fp);
fwrite(&channels_, sizeof(int), 1, fp);
fwrite(&height_, sizeof(int), 1, fp);
fwrite(&width_, sizeof(int), 1, fp);

fwrite(data_->cpu_data(), sizeof(Dtype), count_, fp); //for save the forward data

//fwrite(diff_->cpu_data(), sizeof(Dtype), count_, fp); //for save the backward gradient data

// fwrite(mydata, sizeof(Dtype), 10000, fp); //cpudata
fclose(fp);

// ofstream fout(file_name.c_str());
//fout << num_ <<" " ;
//fout << channels_ <<" " ;
//fout << height_ <<" " ;
//fout << width_ <<" " ;
//const Dtype* data = this->cpu_data();
//for (int i = 0; i < count_; ++i) {
// fout << data[i] <<" " ;
// //LOG(INFO) << "CPU_data"<<data_->cpu_data()[i];
//}
////fout.write(data_->cpu_data(), strlen(data_->cpu_data));
// fout.close();

}

caffe中如何调用并存储数据？

(*top)[0]->save_to_file("top" + this->layer_param_.name());

bottom[0]->save_to_file("bottom" + this->layer_param_.name());

matlab中如何查看数据？*********************************************************************************************************

function y = read_blob_data(file_name)

fp = fopen(file_name);

num_images = fread(fp, 1, 'int32');
num_channels = fread(fp, 1, 'int32');
num_height = fread(fp, 1, 'int32');
num_width = fread(fp, 1, 'int32');

num_count = num_images * num_channels * num_height * num_width;

data = fread(fp, num_count, 'float');

data = reshape(data, num_width, num_height, num_channels, num_images);
y = zeros(num_height, num_width, num_channels, num_images);
for i = 1 : num_images
for j = 1 : num_channels
z = data(:, :, j, i);
y(:, :, j, i) = z';
end
end