tensorflow 笔记
1. matMul
#-*-coding:utf-8 -*-
import tensorflow as tf w1 = tf.Variable(tf.random_normal([2,3], stddev= 1, seed= 1)) w2 = tf.Variable(tf.random_normal([3,1], stddev= 1, seed= 1)) x = tf.constant([[0.7, 0.9]]) # 1×2 a = tf.matmul(x, w1) # 1×3 y = tf.matmul(a, w2) # 1×1 sess = tf.Session() # w1 w2 sess.run(w1.initializer) sess.run(w2.initializer) sess.run(y) print(y) sess.close()
2. eval 函数 作用:
1.eval(): 将字符串string对象转化为有效的表达式参与求值运算返回计算结果
2.eval()也是启动计算的一种方式。基于Tensorflow的基本原理，首先需要定义图，然后计算图，其中计算图的函数常见的有run()函数，如sess.run()。同样eval()也是此类函数，
3.要注意的是，eval()只能用于tf.Tensor类对象，也就是有输出的Operation。对于没有输出的Operation, 可以用.run()或者Session.run()；Session.run()没有这个限制。
import tensorflow as tf x = tf.Variable(3, name="x") y = tf.Variable(4, name="y") z = tf.Variable(4, name="z") w = tf.Variable(4, name="w") f = x * y * z + 3 - w //单个初始化，变量增多，还真是个事 with tf.Session() as sess: x.initializer.run() y.initializer.run() z.initializer.run() w.initializer.run() print(x) print(y) print(z) print(w) //f fuction eval 方式和js的eval一样，作为方法函数执行 result = f.eval() print(result)
3. 矩阵初始化 :
w1 = tf.constant([[1,1,1],[2,2,2]],tf.float32) w2 = tf.constant([[3],[3],[3]],tf.float32)
4. 使用GPU计算
g = tf.Graph() with g.device('/gpu:0'): a = tf.matmul(x, w1) # 1×3 y = tf.matmul(a, w2) # 1×1
5.变量
weights = tf.Variable(tf.random_normal([2,3], stddev = 2))// 产生2行3列 标准差为2 的变量
biases = tf.Variable(tf.zeros([3]) //初值为0 , 含有3个元素的变量
//使用w2变量初值来初始化w3
w2 = tf.Variable(weights.initialized_value())
w3 = tf.Varibale(weights.initialized_value() * 2.0)
符号变量
#定义‘符号’变量，也称为占位符 a = tf.placeholder("float") b = tf.placeholder("float")
6 tensorflow 训练神经网络
# 定义损失函数 cross_entropy = -tf.reduce_mean(y_ * tf/log(tf.clip_by_value(y, 1e-10, 1.0))) // -y_ln(y) # 定义学习率 learning_rate = 0.01 #定义反响传播算法 train_step = tf.train.AdamOptimizer(learning_rate).minimize(cross_entropy) //使损失函数最小
7. 完整的神经网络训练程序
# -*- coding: utf-8 -*- """ Created on Thu Apr 12 15:58:38 2018 @author: 无尾君 """ import tensorflow as tf from numpy.random import RandomState batch_size = 8 w1 = tf.Variable(tf.random_normal([2,3], stddev= 1, seed= 1)) w2 = tf.Variable(tf.random_normal([3,1], stddev= 1, seed= 1)) x = tf.placeholder(tf.float32, shape= (None,2), name= 'x-input') y_ = tf.placeholder(tf.float32, shape= (None,1), name= 'y-input') a = tf.matmul(x, w1) y = tf.matmul(a, w2) cross_entropy = -tf.reduce_mean(y_ * tf.log(tf.clip_by_value(y, 1e-10, 1.0))) train_step = tf.train.AdamOptimizer(0.001).minimize(cross_entropy) rdm = RandomState(1) dataset_size = 128 X = rdm.rand(dataset_size, 2) Y = [[int(x1+ x2 < 1)] for (x1, x2) in X] with tf.Session() as sess: tf.global_variables_initializer().run() print(sess.run(w1)) print(sess.run(w2)) STEPS = 5000 for i in range(STEPS): start = (i * batch_size) % dataset_size end = min(start + batch_size, dataset_size) sess.run(train_step, feed_dict = {x: X[start:end], y_: Y[start:end]}) if i % 1000 ==0: total_cross_entropy = sess.run(cross_entropy, feed_dict = {x: X, y_: Y}) print("After %d training step(s), cross entropy on all data is %g" % (i, total_cross_entropy))
8 expanddims :
您可以使用expand_dims（image，0）使其成为1个图像，这将使形状[1，高度，宽度，通道]。
9 运行会话
#运行会话，输入数据，并计算节点，同时打印结果 sess.run(y, feed_dict={a: 3, b: 3})
10 算术
| 操作 | 描述 |
|---|---|
| tf.add(x, y, name=None) | 求和 |
| tf.sub(x, y, name=None) | 减法 |
| tf.mul(x, y, name=None) | 乘法 |
| tf.div(x, y, name=None) | 除法 |
| tf.mod(x, y, name=None) | 取模 |
| tf.abs(x, name=None) | 求绝对值 |
| tf.neg(x, name=None) | 取负 (y = -x). |
| tf.sign(x, name=None) | 返回符号 y = sign(x) = -1 if x < 0; 0 if x == 0; 1 if x > 0. |
| tf.inv(x, name=None) | 取反 |
| tf.square(x, name=None) | 计算平方 (y = x * x = x^2). |
| tf.round(x, name=None) | 舍入最接近的整数 # ‘a’ is [0.9, 2.5, 2.3, -4.4] tf.round(a) ==> [ 1.0, 3.0, 2.0, -4.0 ] |
| tf.sqrt(x, name=None) | 开根号 (y = \sqrt{x} = x^{1/2}). |
| tf.pow(x, y, name=None) | 幂次方 # tensor ‘x’ is [[2, 2], [3, 3]] # tensor ‘y’ is [[8, 16], [2, 3]] tf.pow(x, y) ==> [[256, 65536], [9, 27]] |
| tf.exp(x, name=None) | 计算e的次方 |
| tf.log(x, name=None) | 计算log，一个输入计算e的ln，两输入以第二输入为底 |
| tf.maximum(x, y, name=None) | 返回最大值 (x > y ? x : y) |
| tf.minimum(x, y, name=None) | 返回最小值 (x < y ? x : y) |
| tf.cos(x, name=None) | 三角函数cosine |
| tf.sin(x, name=None) | 三角函数sine |
| tf.tan(x, name=None) | 三角函数tan |
| tf.atan(x, name=None) | 三角函数ctan |
11. transpose
tf.transpose(input, [dimension_1, dimenaion_2,..,dimension_n]):这个函数主要适用于交换输入张量的不同维度用的，如果输入张量是二维，就相当是转置。dimension_n是整数，如果张量是三维，就是用0,1,2来表示。这个列表里的每个数对应相应的维度。如果是[2,1,0]，就把输入张量的第三维度和第一维度交换。
Transpose(root.WithOpName("transpose"), dived, { 0,2,1,3 });
12: tensorflow C++ API 示例
这个貌似有问题 运行崩溃
#include <iostream> #include <map> #include "tensorflow/cc/ops/const_op.h" #include "tensorflow/cc/ops/image_ops.h" #include "tensorflow/cc/ops/standard_ops.h" #include "tensorflow/core/framework/graph.pb.h" #include "tensorflow/core/framework/tensor.h" #include "tensorflow/core/graph/default_device.h" #include "tensorflow/core/graph/graph_def_builder.h" #include "tensorflow/core/lib/core/errors.h" #include "tensorflow/core/lib/core/stringpiece.h" #include "tensorflow/core/lib/core/threadpool.h" #include "tensorflow/core/lib/io/path.h" #include "tensorflow/core/lib/strings/stringprintf.h" #include "tensorflow/core/platform/init_main.h" #include "tensorflow/core/platform/logging.h" #include "tensorflow/core/platform/types.h" #include "tensorflow/core/public/session.h" #include "tensorflow/core/util/command_line_flags.h" using namespace std ; using namespace tensorflow; using tensorflow::Flag; using tensorflow::Tensor; using tensorflow::Status; using tensorflow::string; using tensorflow::int32; map<int,string> int2char; string s = "KDA0123456789 "; for(int i=0;i<s.size();i++){ int2char[i]=s[i]; } //从文件名中读取数据 Status ReadTensorFromImageFile(string file_name, const int input_height, const int input_width, vector<Tensor>* out_tensors) { auto root = Scope::NewRootScope(); using namespace ops; auto file_reader = ops::ReadFile(root.WithOpName("file_reader"),file_name); const int wanted_channels = 1; Output image_reader; std::size_t found = file_name.find(".png"); //判断文件格式 if (found!=std::string::npos) { image_reader = DecodePng(root.WithOpName("png_reader"), file_reader,DecodePng::Channels(wanted_channels)); } else { image_reader = DecodeJpeg(root.WithOpName("jpeg_reader"), file_reader,DecodeJpeg::Channels(wanted_channels)); } // 下面几步是读取图片并处理 auto float_caster =Cast(root.WithOpName("float_caster"), image_reader, DT_FLOAT); auto dims_expander = ExpandDims(root, float_caster, 0); auto resized = ResizeBilinear(root, dims_expander,Const(root.WithOpName("resize"), {input_height, input_width})); // Div(root.WithOpName(output_name), Sub(root, resized, {input_mean}),{input_std}); Transpose(root.WithOpName("transpose"),resized,{0,2,1,3}); GraphDef graph; root.ToGraphDef(&graph); unique_ptr<Session> session(NewSession(SessionOptions())); session->Create(graph); session->Run({}, {"transpose"}, {}, out_tensors);//Run，获取图片数据保存到Tensor中 return Status::OK(); } int main(int argc, char* argv[]) { string graph_path = "aov_crnn.pb"; GraphDef graph_def; //读取模型文件 if (!ReadBinaryProto(Env::Default(), graph_path, &graph_def).ok()) { cout << "Read model .pb failed"<<endl; return -1; } //新建session unique_ptr<Session> session; SessionOptions sess_opt; sess_opt.config.mutable_gpu_options()->set_allow_growth(true); (&session)->reset(NewSession(sess_opt)); if (!session->Create(graph_def).ok()) { cout<<"Create graph failed"<<endl; return -1; } //读取图像到inputs中 int input_height = 40; int input_width = 240; vector<Tensor> inputs; // string image_path(argv[1]); string image_path("test.jpg"); if (!ReadTensorFromImageFile(image_path, input_height, input_width,&inputs).ok()) { cout<<"Read image file failed"<<endl; return -1; } vector<Tensor> outputs; string input = "inputs_sq"; string output = "results_sq";//graph中的输入节点和输出节点，需要预先知道 pair<string,Tensor>img(input,inputs[0]); Status status = session->Run({img},{output}, {}, &outputs);//Run,得到运行结果，存到outputs中 if (!status.ok()) { cout<<"Running model failed"<<endl; cout<<status.ToString()<<endl; return -1; } //得到模型运行结果 Tensor t = outputs[0]; auto tmap = t.tensor<int64, 2>(); int output_dim = t.shape().dim_size(1); //预测结果解码为字符串 string res=""; for (int j = 0; j < output_dim; j++) { res+=int2char[tmap(0,j)]; } cout<<res<<endl; return 0; }
这个才是对的
#include <fstream> #include <utility> #include <vector> // #include <Eigen/Core> // #include <Eigen/Dense> #include "tensorflow/cc/ops/const_op.h" #include "tensorflow/cc/ops/image_ops.h" #include "tensorflow/cc/ops/standard_ops.h" #include "tensorflow/core/framework/graph.pb.h" #include "tensorflow/core/framework/tensor.h" #include "tensorflow/core/graph/default_device.h" #include "tensorflow/core/graph/graph_def_builder.h" #include "tensorflow/core/lib/core/errors.h" #include "tensorflow/core/lib/core/stringpiece.h" #include "tensorflow/core/lib/core/threadpool.h" #include "tensorflow/core/lib/io/path.h" #include "tensorflow/core/lib/strings/stringprintf.h" #include "tensorflow/core/platform/env.h" #include "tensorflow/core/platform/init_main.h" #include "tensorflow/core/platform/logging.h" #include "tensorflow/core/platform/types.h" #include "tensorflow/core/public/session.h" #include "tensorflow/core/util/command_line_flags.h" using tensorflow::Flag; using tensorflow::Tensor; using tensorflow::Status; using tensorflow::string; using tensorflow::int32; string model_path = "/work/dl/mosaic/keras/mobilenetv2/trans_model/mosaic.mobilenet_v2_20190425_448X448_DSPD_afew.20.pb"; const std::string class_name[2] = { "dummy", "kit fox" }; // Given an image file name, read in the data, try to decode it as an image, // resize it to the requested size, and then scale the values as desired. Status ReadTensorFromImageFile(string file_name, const int input_height, const int input_width, const float input_mean, const float input_std, std::vector<Tensor>* out_tensors) { auto root = tensorflow::Scope::NewRootScope(); using namespace ::tensorflow::ops; // NOLINT(build/namespaces) string input_name = "file_reader"; string output_name = "normalized"; auto file_reader = tensorflow::ops::ReadFile(root.WithOpName(input_name), file_name); // Now try to figure out what kind of file it is and decode it. const int wanted_channels = 3; tensorflow::Output image_reader; //if (tensorflow::StringPiece(file_name).ends_with(".png")) { // image_reader = DecodePng(root.WithOpName("png_reader"), file_reader, // DecodePng::Channels(wanted_channels)); //} //else if (tensorflow::StringPiece(file_name).ends_with(".gif")) { // image_reader = DecodeGif(root.WithOpName("gif_reader"), file_reader); //} //else { // Assume if it's neither a PNG nor a GIF then it must be a JPEG. image_reader = DecodeJpeg(root.WithOpName("jpeg_reader"), file_reader, DecodeJpeg::Channels(wanted_channels)); } // Now cast the image data to float so we can do normal math on it. auto float_caster = Cast(root.WithOpName("float_caster"), image_reader, tensorflow::DT_FLOAT); // The convention for image ops in TensorFlow is that all images are expected // to be in batches, so that they're four-dimensional arrays with indices of // [batch, height, width, channel]. Because we only have a single image, we // have to add a batch dimension of 1 to the start with ExpandDims(). auto dims_expander = ExpandDims(root, float_caster, 0); // Bilinearly resize the image to fit the required dimensions. auto resized = ResizeBilinear( root, dims_expander, Const(root.WithOpName("size"), { input_height, input_width })); // Subtract the mean and divide by the scale. Div(root.WithOpName(output_name), Sub(root, resized, { input_mean }), { input_std }); // This runs the GraphDef network definition that we've just constructed, and // returns the results in the output tensor. tensorflow::GraphDef graph; TF_RETURN_IF_ERROR(root.ToGraphDef(&graph)); std::unique_ptr<tensorflow::Session> session( tensorflow::NewSession(tensorflow::SessionOptions())); TF_RETURN_IF_ERROR(session->Create(graph)); TF_RETURN_IF_ERROR(session->Run({}, { output_name }, {}, out_tensors)); return Status::OK(); } int main(int argc, char* argv[]) { string graph_path = model_path; tensorflow::port::InitMain(argv[0], &argc, &argv); tensorflow::GraphDef graph_def; if (!ReadBinaryProto(tensorflow::Env::Default(), graph_path, &graph_def).ok()) { LOG(ERROR) << "Read proto"; return -1; } std::unique_ptr<tensorflow::Session> session; tensorflow::SessionOptions sess_opt; sess_opt.config.mutable_gpu_options()->set_allow_growth(true); (&session)->reset(tensorflow::NewSession(sess_opt)); if (!session->Create(graph_def).ok()) { LOG(ERROR) << "Create graph"; return -1; } const int batch_size = argc - 1; if (batch_size != 1) { LOG(ERROR) << "Batch mode for the pretrained inception-v3 is unsupported"; LOG(ERROR) << " - https://github.com/tensorflow/tensorflow/issues/554"; return -1; } int32 input_dim = 448; float input_mean = 128; float input_std = 128; std::vector<Tensor> inputs; std::string image_path(argv[1]); if (!ReadTensorFromImageFile(image_path, input_dim, input_dim, input_mean, input_std, &inputs).ok()) { LOG(ERROR) << "Load image"; return -1; } std::vector<Tensor> outputs; //string input_layer = "Mul"; //string output_layer = "softmax"; string input_layer = "input_1"; string output_layer = "dense_2/Softmax"; if (!session->Run({ { input_layer, inputs[0] } }, { output_layer }, {}, &outputs).ok()) { LOG(ERROR) << "Running model failed"; return -1; } printf("Is predicting ... \n"); //得到模型运行结果 Tensor t = outputs[0]; auto tmap = t.tensor<float, 2>(); for (int i = 0; i != 2; ++i) { printf("p[%d]=%.2f\n", i, tmap(0, i)); } int output_dim = t.shape().dim_size(1); for (int j = 0; j < output_dim; j++) { tmap(0, j); } printf("predicting OK \n"); //Eigen::Map<Eigen::VectorXf> pred(outputs[0].flat<float>().data(), // outputs[0].NumElements()); //int maxIndex; float maxValue = pred.maxCoeff(&maxIndex); //LOG(INFO) << "P( " << class_name[maxIndex] << " | image ) = " << maxValue; return 0; }
编译 :
g++ -g -D_GLIBCXX_USE_CXX11_ABI=0 tf_predict.cpp -o tf_predict -I /usr/include/eigen3 -I /usr/local/include/tf -L/usr/local/lib/ `pkg-config --cflags --libs protobuf` -ltensorflow_cc -ltensorflow_framework
使用Opencv 读取tensor
tensorflow::Tensor readTensor(string filename){ tensorflow::Tensor input_tensor(DT_FLOAT,TensorShape({1,240,40,1})); Mat src=imread(filename,0); Mat dst; resize(src,dst,Size(240,40));//resize Mat dst_transpose=dst.t();//transpose auto tmap=input_tensor.tensor<float,4>(); for(int i=0;i<240;i++){//Mat复制到Tensor for(int j=0;j<40;j++){ tmap(0,i,j,0)=dst_transpose.at<uchar>(i,j); } } return input_tensor; }
13 tensorflow 取指针
float* p = input_tensor.flat<float>().data()
