Benpian general guiding principle is based on the basic knowledge and character depth study on the use TensorFlow ability to train the actual programming to address issues related to image processing. Full article seeks layman's language, through language and detailed procedures easy to understand analysis, presentation TensorFlow basic usage, design and write high-level model corresponding procedures.
Benpian emphasis on linking theory with practice, focusing on TensorFlow programming to solve the image recognition applications, providing large data sets, and realize the depth learning model in the form of code in order for the reader.
Benpian as artificial neural network learning, deep learning TensorFlow programming and image processing and other related content programming to learn.
** Benpian a total of 22 chapters, including installation and use of Python libraries, TensorFlow basic data structures and use, create and read TensorFlow data sets, artificial neural networks, recurrent neural networks, the theoretical basis of the full convolution neural network the characteristics of deep learning model creation, model, algorithm, ResNet, Slim, GAN and so on.
Since the details of the content is too much, so small series only the part of the knowledge point shots out of the rough introduction, each section has a more granular content, we hope to be able to derive the true meaning! **
Chapter 1 introduces the basic content of the depth of learning, deep learning is applied to the initial introduction of computer vision and direction of development, introduced the use of prospect depth study to solve the problem of computer vision, depth study to illustrate the use of artificial intelligence and computer vision is to achieve future growth direction, but also the inevitable trend.
Chapter 2 describes the Python installation and the most commonly used class libraries. Python language is easy to use very strong language, vision and formulas can easily be expressed in the form of code without having to learn too much programming knowledge. Python library dedicated threading is not common, but to read and generate TensorFlow proprietary format lay the foundation for later data.
Chapter 3 describes the overall machine learning basic classification algorithms and theoretical basis, here we describe the different algorithms, such as concrete implementation and application of regression algorithm and decision tree algorithms. These are the basic theory of deep learning through these structures and application thoroughly and accurately show the depth of learning to the reader, to lay a solid foundation for further study in depth to master the application of computer vision.
Chapter 4 describes how to use the Python language. And by introducing different Python class library to help readers strengthen programming skills in Python, to study the appropriate library. These are the repeated use of the content in later. Meanwhile borrow visual display of skills knowledge learning data. The skills in data analysis, although the basic skills, but has a very important role.
Chapter 5-6 is an introduction to OpenCV library to use them. This book focus on image processing, image data is read, edit, and processing is the most important book. OpenCV is dedicated to the Python library for image processing, through basic and advanced presentation to explain the reader master the use of this important class library. Learn to write to crop the image, the transformation and translation code.
Chapter 5 as an example of the basic contents of the convolution kernel to do a presentation, and to achieve a convolution kernel functions with Python language. Convolution kernel is a very important part of the basis of this book, is a very important part of image processing, to achieve the convolution processing check images through the corresponding programming, grasp and understand: convolution neural networks are very helpful.
Chapter 7 to 8 basis TensorFlow entry, showing the basic application TensorFlow the reader through a an entertainment nature of the site, presentations neural network classification of the fitting process by way of graphic images, while understanding the underlying entertainment Content.
Chapter 9 is a focus of the book is the basis for the content neural network. Feedback algorithm of this chapter is to address the milestone of excessive neural network algorithm to calculate. Through careful to explain in detail, using plain language description of the algorithm, and to achieve this neural network algorithm is the most important content for readers in the form of independent writing code. This chapter looks small, but very important.
Chapter 10 of the data input and output TensorFlow made a detailed introduction. From start to read a CSV file, to make the reader church dedicated TensorFlow data format TFRecord, which in the current market books rarely involved. Use TensorFlow framework for programming, preparation and standardization of data is a top priority, so this chapter is more important chapters.
12 ~ Chapter 11 is a basic tutorial, a convolution neural network learning on TensorFlow framework, through the preparation and presentation of the previous section, a convolution using the basic theory of neural networks to identify handwriting is deep learning basic skills, but also learn a very important base. And in the preparation of the program, the authors demonstrated the important role model parameter adjustment of the test results to the reader. This is the content of the books currently on the market not covered is very important.
- Chapter 1314 is the introduction and application of convolution neural network algorithm. In these two chapters, the author describes in detail a convolutional neural network applications, particularly applications in image recognition, the handwriting recognition simple numerical identification of the development of the display object. Match recognition by the image dataset, using a convolutional neural network model prize in the game, the reader grasp variant convolutional neural network. Theoretical basis convolution neural network is the convolution of the forward and reverse process, the general process forward better understand and learn, but for the reverse operation, basically not involved, yes, it is only copy of the formula and excerpt. Chapter 14 describes the operation in the convolutional neural network calculation and the reverse processes in detail, expressed by numerous examples, the reverse operation is described first convolutional neural network in great detail. This is the lack of content in the books.
Chapter 15 through a complete example demonstrates the use of a convolution process neural network image recognition. Examples of competition from ImageNet image recognition, the highest accuracy of the model employed is race model obtained. A detailed analysis of the project every step of the hand and the church reader how to use convolution neural network image recognition.
Chapter 16 VGGNet composition structure, focusing on the network VGGNet Scheduling and thereafter performing Finetuning ability. The examples in this chapter Chapter 15 multiplexes VGG16 achieve, to provide readers - ideas to solve a problem in a different perspective and different models method.
Chapter 17 for a current depth study employed given - some answers to interview questions, these questions can help recruiters who are high-level analysis of the interviewer, can also help improve their employment of technical concepts and knowledge, identify their own position, paving the way for the future promotion and pay rise.
Chapter 18 describes the depth ResNet learning network model, it uses a large number of residual components of the network as the basic modules in the network, the main role is that the network changes with increasing depth, without causing attenuation weights and the gradient attenuating or disappear these and other issues. In addition to ResNet model, this chapter introduces the new convolution neural models, including SqueezeNet and Xception.
Chapter 19-20 TensorFlow entered the advanced stage of learning, is a highlight of the API Slim, which is used to define. Lightweight class library developed more complex models of training and evaluation. These two chapters not only introduces its use, also produced a MLP MLP, a convolution neural network CNN through it, and finally the use of pre-trained Slim model Finetuning.
Chapter 21 describes the full convolution image segmentation neural network, first explain the rationale and implementation divide, and then gives a full convolution neural network step by step process and programming based image segmentation. Finally, the use of full volume VGG16 image plot network: actual segmentation.
Chapter 22 explains the GAN is a confrontation generation network, while seemingly boring theory in this chapter, but I use a "generator" and a "discriminator" Common kept a "confrontation" in a network metaphor, reduced read more difficult. By using the GAN case ultimately generate handwritten numbers so that readers really learn GAN applications.
In addition, the whole chapter to the current depth study to identify the most popular model image and get the best score made the introduction, these are the focus of current research focus and depth of learning.
Due to space limitations article, so in this small series do too much introduction, and the needs of the application [TensorFlow depth learning practice] small partner ++++++ V x ①⑧①③③⑤③②⑨⑦ document can be acquired!
By TensorFlow image processing, to fully grasp the depth of learning model and its application
In-depth explanation feedback neural network theory and convolution neural network system
Combined with actual cases to achieve the depth of learning to master TensorFlow programming methods and techniques
Depth study focused on practical training application development and problem-solving ability