caffe模型在训练完成后，会生成一个*.caffemodel的文件，在运行的时候，直接调用caffe就可以读取其中的相应权值参数。但是如果用一个第三方软件打开这个，却是不可以可视化的二值乱码。
将模型中的参数导出，可编辑化后能有哪些好处呢，
（1）方便进行fpga平台的移植
（2）可以基于别人训练好的模型，0数据训练自己的模型，使用自己的模型拟合别人模型的权值分布，达到用模型训模型的目的。
（3）可以对网络进行剪支，加速等操作。
将模型中的特征图和权值可视化有哪些好处呢，
（1）方便对卷积网络的特征有所了解，训练好的特征总是有规则的特征图，可以侧面辅助训练过程。
这里分析lenet5这样的网络结构，所有其他网络都通用。
核心程序：
（1）只导出weights，不进行显示
void parse_caffemodel(string caffemodel, string outtxt) { printf("%s\n", caffemodel.c_str()); NetParameter net; bool success = loadCaffemodel(caffemodel.c_str(), &net); if (!success){ printf("读取错误啦:%s\n", caffemodel.c_str()); return; } FILE* fmodel = fopen(outtxt.c_str(), "wb"); for (int i = 0; i < net.layer_size(); ++i){ LayerParameter& param = *net.mutable_layer(i); int n = param.mutable_blobs()->size(); if (n){ const BlobProto& blob = param.blobs(0); printf("layer: %s weight(%d)", param.name().c_str(), blob.data_size()); fprintf(fmodel, "\nlayer: %s weight(%d)\n", param.name().c_str(), blob.data_size()); writeData(fmodel, blob.data().data(), blob.data_size()); if (n > 1){ const BlobProto& bais = param.blobs(1); printf(" bais(%d)", bais.data_size()); fprintf(fmodel, "\nlayer: %s bais(%d)\n", param.name().c_str(), bais.data_size()); writeData(fmodel, bais.data().data(), bais.data_size()); } printf("\n"); } } fclose(fmodel); }
（2）weights可视化
cv::Mat visualize_weights(string prototxt, string caffemodel, int weights_layer_num) { ::google::InitGoogleLogging("0"); #ifdef CPU_ONLY Caffe::set_mode(Caffe::CPU); #else Caffe::set_mode(Caffe::GPU); #endif Net<float> net(prototxt, TEST); net.CopyTrainedLayersFrom(caffemodel); vector<shared_ptr<Blob<float> > > params = net.params(); std::cout << "各层参数的维度信息为：\n"; for (int i = 0; i<params.size(); ++i) std::cout << params[i]->shape_string() << std::endl; int width = params[weights_layer_num]->shape(3); //宽度 int height = params[weights_layer_num]->shape(2); //高度 int channel = params[weights_layer_num]->shape(1); //通道数 int num = params[weights_layer_num]->shape(0); //个数 int imgHeight = (int)(1 + sqrt(num))*height; int imgWidth = (int)(1 + sqrt(num))*width; Mat img(imgHeight, imgWidth, CV_8UC3, Scalar(0, 0, 0)); float maxValue = -1000, minValue = 10000; const float* tmpValue = params[weights_layer_num]->cpu_data(); for (int i = 0; i<params[weights_layer_num]->count(); i++){ maxValue = std::max(maxValue, tmpValue[i]); minValue = std::min(minValue, tmpValue[i]); } int kk = 0; for (int y = 0; y<imgHeight; y += height){ for (int x = 0; x<imgWidth; x += width){ if (kk >= num) continue; Mat roi = img(Rect(x, y, width, height)); for (int i = 0; i<height; i++){ for (int j = 0; j<width; j++){ for (int k = 0; k<channel; k++){ float value = params[weights_layer_num]->data_at(kk, k, i, j); roi.at<Vec3b>(i, j)[k] = (value - minValue) / (maxValue - minValue) * 255; } } } ++kk; } } return img; }
（3）featuremap可视化
cv::Mat Classifier::visualize_featuremap(const cv::Mat& img,string layer_name) { Blob<float>* input_layer = net_->input_blobs()[0]; input_layer->Reshape(1, num_channels_, input_geometry_.height, input_geometry_.width); net_->Reshape(); std::vector<cv::Mat> input_channels; WrapInputLayer(&input_channels); Preprocess(img, &input_channels); net_->Forward(); std::cout << "网络中的Blobs名称为：\n"; vector<shared_ptr<Blob<float> > > blobs = net_->blobs(); vector<string> blob_names = net_->blob_names(); std::cout << blobs.size() << " " << blob_names.size() << std::endl; for (int i = 0; i<blobs.size(); i++){ std::cout << blob_names[i] << " " << blobs[i]->shape_string() << std::endl; } std::cout << std::endl; assert(net_->has_blob(layer_name)); shared_ptr<Blob<float> > conv1Blob = net_->blob_by_name(layer_name); std::cout << "测试图片的特征响应图的形状信息为：" << conv1Blob->shape_string() << std::endl; float maxValue = -10000000, minValue = 10000000; const float* tmpValue = conv1Blob->cpu_data(); for (int i = 0; i<conv1Blob->count(); i++){ maxValue = std::max(maxValue, tmpValue[i]); minValue = std::min(minValue, tmpValue[i]); } int width = conv1Blob->shape(3); //响应图的高度 int height = conv1Blob->shape(2); //响应图的宽度 int channel = conv1Blob->shape(1); //通道数 int num = conv1Blob->shape(0); //个数 int imgHeight = (int)(1 + sqrt(channel))*height; int imgWidth = (int)(1 + sqrt(channel))*width; cv::Mat img(imgHeight, imgWidth, CV_8UC1, cv::Scalar(0)); int kk = 0; for (int x = 0; x<imgHeight; x += height){ for (int y = 0; y<imgWidth; y += width){ if (kk >= channel) continue; cv::Mat roi = img(cv::Rect(y, x, width, height)); for (int i = 0; i<height; i++){ for (int j = 0; j<width; j++){ float value = conv1Blob->data_at(0, kk, i, j); roi.at<uchar>(i, j) = (value - minValue) / (maxValue - minValue) * 255; } } kk++; } } return img; }
运行结果：
（1）
string caffemodel = "lenet_iter_10000.caffemodel";; string outtxt = "lenet.txt"; parse_caffemodel(caffemodel, outtxt);
（2）
string prototxt = "lenet.prototxt"; string caffemodel = "lenet_iter_10000.caffemodel"; int weights_layer_num = 0; Mat image=visualize_weights(prototxt, caffemodel, weights_layer_num); imshow("weights", image); waitKey(0);
（3）
::google::InitGoogleLogging(argv[0]); string model_file = "lenet.prototxt"; string trained_file = "lenet_iter_10000.caffemodel"; Classifier classifier(model_file, trained_file); string file = "5.jpg"; cv::Mat img = cv::imread(file, -1); CHECK(!img.empty()) << "Unable to decode image " << file; cv::Mat feature_map = classifier.visualize_featuremap(img,"conv2"); imshow("feature_map", feature_map); cv::waitKey(0);
将权值导入matlab中，可以看到权值基本都是服从均值为0，方差很小的分布。
完整程序下载链接：http://download.csdn.net/detail/qq_14845119/9895412
