caffe c++示例（mnist 多层感知机c++训练，测试）

雨点打透心脏的1/2处 2022-05-09 12:03 383阅读 0赞

`caffe`训练网络模型一般直接使用的`caffe.bin`: `caffe train -solver solver.prototxt`，其实这个命令的本质也是调用c++的`Solver`.

本文给出使用纯`c++`代码，使用`mnist数据`\+`多层感知机网络`，训练数字分类问题。然后用`C++`调用训练好的模型测试分类。

`solver.prototxt`文件内容如下：

net: "/home/xy/caffe_analysis/my_caffe/example/lenet.prototxt"
    test_iter: 1000
    test_interval: 100
    base_lr: 0.001
    lr_policy: "step"
    gamma: 0.1
    stepsize: 5000
    display: 1000
    max_iter: 20000
    momentum: 0.9
    weight_decay: 0.0005
    snapshot: 10000
    snapshot_prefix: "model"
    solver_mode: CPU

注：网络名称虽然为lenet.prototxt，实际是多层感知机。

训练网络如下：

![在这里插入图片描述][70]

name: "LeNet"
    layer {
      name: "mnist"
      #name: "data"
      type: "Data"
      top: "data"
      top: "label"
      include {
        phase: TRAIN
      }
      transform_param {
        scale: 0.00390625
      }
      data_param {
        source: "/home/xy/caffe-master/examples/mnist/mnist_train_lmdb"
        batch_size: 64
        backend: LMDB
      }
    }
    
    layer {
      name: "mnist"
      type: "Data"
      top: "data"
      top: "label"
      include {
        phase: TEST
      }
      transform_param {
        scale: 0.00390625
      }
      data_param {
        source: "/home/xy/caffe-master/examples/mnist/mnist_test_lmdb"
        batch_size: 100
        backend: LMDB
      }
    }
    
    
    layer {
      name: "fc1"
      type: "InnerProduct"
      # learning rate and decay multipliers for the weights
      param { lr_mult: 1 decay_mult: 1 }
      # learning rate and decay multipliers for the biases
      param { lr_mult: 2 decay_mult: 0 }
      inner_product_param {
        num_output: 300
        weight_filler {
          type: "xavier"
        }
        bias_filler {
          type: "constant"
          value: 0
        }
      }
      bottom: "data"
      top: "fc1"
    }
    
    layer {
      name: "relu1"
      type: "ReLU"
      bottom: "fc1"
      top: "fc1"
    }
    
    layer {
      name: "ip2"
      type: "InnerProduct"
      bottom: "fc1"
      top: "fc2"
      param {
        lr_mult: 1
      }
      param {
        lr_mult: 2
      }
      inner_product_param {
        num_output: 10
        weight_filler {
          type: "xavier"
        }
        bias_filler {
          type: "constant"
        }
      }
    }
    
    layer {
      name: "accuracy"
      type: "Accuracy"
      bottom: "fc2"
      bottom: "label"
      top: "accuracy"
      include {
        phase: TEST
      }
    }
    layer {
      name: "loss"
      type: "SoftmaxWithLoss"
      bottom: "fc2"
      bottom: "label"
      top: "loss"
    }

训练代码如下：

google::InitGoogleLogging("test_solver");
    FLAGS_alsologtostderr = 1;
    
    string base_dir = "/home/xy/caffe_analysis/my_caffe/example/";
    string path = base_dir + "mlp_solver.prototxt";
    SGDSolver<float> solver(path);
    
    // resuming from trained weights, 也可以加载已经训练好的模型
    string trained_weight_path = base_dir + "model_iter_20000.caffemodel";
    solver.net()->CopyTrainedLayersFrom(trained_weight_path);
    // 开始优化
    solver.Solve();

训练过程如下：

![在这里插入图片描述][70 1]

接下来测试训练模型的效果的，代码如下：

void test_net(){
        Caffe::set_mode(Caffe::CPU);
        string base_dir = "/home/xy/caffe_analysis/my_caffe/example/";
        string net_path = base_dir + "lenet_deploy.prototxt";
        string trained_path = base_dir + "model_iter_20000.caffemodel";
    
        Net<float> net(net_path, Phase::TEST);
        net.CopyTrainedLayersFrom(trained_path);
    
        CHECK_EQ(net.num_inputs(), 1) << "Network should have exactly one input.";
        CHECK_EQ(net.num_outputs(), 1) << "Network should have exactly one output.";
    
        // read Datum from file
        string datum_path = base_dir + "0.proto";
        Datum datum;
        ReadProtoFromBinaryFileOrDie(datum_path, &datum);
    
        // Datum convert
        TransformationParameter parameter;
        parameter.set_scale(0.00390625);
        DataTransformer<float> dataTransformer(parameter, Phase::TEST);
    
        Blob<float>* input_layer = net.input_blobs()[0];
    //    input_layer->Reshape(1, 1, 28, 28);//no need
        dataTransformer.Transform(datum, input_layer);
    
    //    net.Reshape();
        net.Forward();
        Blob<float>* output_layer = net.output_blobs()[0];
    
        const float* begin = output_layer->cpu_data();
        const float* end = begin + output_layer->channels();
    
        std::vector<float> ret(begin, end);
    
        for(auto item:ret)cout<<item<<" ";
        cout<<endl;
    
    }

运行输出如下：可以看到0的概率最大。

![在这里插入图片描述][70 2]

测试inference的网络定义如下：

![在这里插入图片描述][70 3]

name: "LeNet"
    
    layer {
      name: "data"
      type: "Input"
      top: "data"
      input_param { shape: { dim: 1 dim: 1 dim: 28 dim: 28 } }
    }
    
    layer {
      name: "fc1"
      type: "InnerProduct"
      # learning rate and decay multipliers for the weights
      param { lr_mult: 1 decay_mult: 1 }
      # learning rate and decay multipliers for the biases
      param { lr_mult: 2 decay_mult: 0 }
      inner_product_param {
        num_output: 300
        weight_filler {
          type: "xavier"
        }
        bias_filler {
          type: "constant"
          value: 0
        }
      }
      bottom: "data"
      top: "fc1"
    }
    
    layer {
      name: "relu1"
      type: "ReLU"
      bottom: "fc1"
      top: "fc1"
    }
    
    layer {
      name: "ip2"
      type: "InnerProduct"
      bottom: "fc1"
      top: "fc2"
      param {
        lr_mult: 1
      }
      param {
        lr_mult: 2
      }
      inner_product_param {
        num_output: 10
        weight_filler {
          type: "xavier"
        }
        bias_filler {
          type: "constant"
        }
      }
    }
    
    layer {
      name: "prob"
      type: "Softmax"
      bottom: "fc2"
      top: "prob"
    }

--------------------

### `caffe`系列源码分析介绍 ###

本系列深度学习框架`caffe` 源码分析主要内容如下：

##### 1. [caffe源码分析-cmake 工程构建][caffe_-cmake]: #####

[caffe源码分析-cmake 工程构建][caffe_-cmake]主要内容:

> 自己从头构建一遍工程，这样能让我更好的了解大型的项目的构建。当然原始的caffe的构建感觉还是比较复杂(主要是cmake)，我这里仅仅使用cmake构建，而且简化点，当然最重要的是支持CLion直接运行调试（如果需要这个工程可以评论留下你的邮箱，我给你发送过去）。

![这里写图片描述][70 4]

##### 2. caffe的数据内存分配类`SyncedMemory`， 以及类`Blob`数据传输的媒介. #####

主要内容：  
[caffe源码分析-SyncedMemory][caffe_-SyncedMemory]  
[caffe源码分析-Blob][caffe_-Blob]  
其中`Blob`分析给出了其直接与opencv的图片相互转化以及操作，可以使得我们更好的理解`Blob`.

##### 3. caffe `layer`的源码分析，包括从整体上说明了`layer`类别以及其proto定义与核心函数. #####

内容如下：  
[caffe源码分析-layer][caffe_-layer]  
[caffe源码分析-ReLULayer][caffe_-ReLULayer]  
[caffe源码分析-inner\_product\_layer][caffe_-inner_product_layer]  
[caffe源码分析-layer\_factory][caffe_-layer_factory]

首先分析了最简单的`layer` `Relu`，然后在是`inner_product_layer全连接层`， 最后是`layer_factory`caffe中 以此工厂模式create各种Layer.

##### 4. 数据输入层，主要是多线程+BlockingQueue的方式读取数据训练： #####

内容如下：  
[caffe源码分析-BlockingQueue][caffe_-BlockingQueue]  
[caffe源码分析-InternalThread][caffe_-InternalThread]  
[caffe源码分析-DataReader][caffe_-DataReader]

##### 5. IO处理例如读取proto文件转化为网络，以及网络参数的序列化 #####

内容如下：  
[caffe源码分析-DataTransformer][caffe_-DataTransformer]  
[caffe源码分析-db, io][caffe_-db_ io]

##### 6. 最后给出了使用纯C++结合多层感知机网络训练mnist的示例 #####

内容如下：

[caffe c++示例（mnist 多层感知机c++训练，测试）][caffe c_mnist _c]

类似与`caffe`一样按照`layer、solver、loss、net`等模块构建的神经网络实现可以见下面这篇blog，相信看懂了这个python的代码理解caffe框架会更简单点.

[神经网络python实现][python]

--------------------

**最后如果需要`cmake` \+ `CLion`直接运行调试`caffe`的代码工程，可以评论留下你的邮箱，我给你发送过去.**

[70]: /images/20220509/92035050885a4c33a052b3e4b3d09714.png
[70 1]: /images/20220509/319d0f426bc44a008bab1388caefed5c.png
[70 2]: /images/20220509/f2c4b5469ae7428aa9373267f1f7f072.png
[70 3]: /images/20220509/60437859104f48c4ad00d3fdc3cb06dc.png
[caffe_-cmake]: https://blog.csdn.net/haluoluo211/article/details/82668188
[70 4]: /images/20220509/a122bc87a1164668a802990f4ddb86a3.png
[caffe_-SyncedMemory]: https://blog.csdn.net/haluoluo211/article/details/82668358
[caffe_-Blob]: https://blog.csdn.net/haluoluo211/article/details/82668678
[caffe_-layer]: https://blog.csdn.net/haluoluo211/article/details/82956105
[caffe_-ReLULayer]: https://blog.csdn.net/haluoluo211/article/details/82956157
[caffe_-inner_product_layer]: https://blog.csdn.net/haluoluo211/article/details/82956259
[caffe_-layer_factory]: https://blog.csdn.net/haluoluo211/article/details/82956378
[caffe_-BlockingQueue]: https://blog.csdn.net/haluoluo211/article/details/82956464
[caffe_-InternalThread]: https://blog.csdn.net/haluoluo211/article/details/82956589
[caffe_-DataReader]: https://blog.csdn.net/haluoluo211/article/details/82956624
[caffe_-DataTransformer]: https://blog.csdn.net/haluoluo211/article/details/82956728
[caffe_-db_ io]: https://blog.csdn.net/haluoluo211/article/details/82956800
[caffe c_mnist _c]: https://blog.csdn.net/haluoluo211/article/details/82958604
[python]: https://blog.csdn.net/haluoluo211/article/details/81158209