Resources for a link click here
-
Chapter 1 Course Guidance
This chapter introduces the main content of the course, the core knowledge curriculum related to the application case, the depth of learning algorithm design general process, adapt to the crowd, learning pre-conditions of this course, after learning to achieve results, to help everyone from the whole on understand the overall context of this course.
- 1-1 Course Guidance Look
-
Chapter 2 deep learning basic Crosstalk (the necessary theoretical knowledge)
Describes the basics of the depth of learning, including the development process of the depth of learning, the basic concept (forward operation, back-propagation, parameter optimization), deep learning mathematical basis (derivative directional derivative, partial derivative, gradient) and the like, by depth learning the basics to help you from macro and micro angles grasp the basic concepts of deep learning, lay the foundation for subsequent learning courses. ...
- Basic Concepts 2-1 convolution neural network
- 2-2 before the operation
- 2-3 reverse the spread of basic concepts
- 2-4 back-propagation iterative process and parameter optimization concept
- 2-5 backpropagation derivative concept, the directional derivative, partial derivative, gradient
- 2-6 gradient descent algorithm of back-propagation
- Reasons for the rapid development of deep learning 2-7
-
Chapter 3 convolutional neural network infrastructure Crosstalk (the necessary theoretical knowledge and skills)
It introduces the basics of convolution neural network, including the development, network infrastructure, relatively convolution neural network computing performance, network structure is, network design, small convolution kernel Magical, 1 * 1 convolution kernels use , lightweight convolutional neural network design and typical network structure of a convolutional neural network in the Attention mechanism (SENET etc.), multi-branched convolutional neural network (Siamese, TripletNet etc.), a convolutional neural network compression method, by introducing ...
- 3-1 convolutional neural network are summarized as
- Basic unit 3-2
- 3-3 define the convolution operation
- Important parameters 3-4 convolution and convolution kernels
- 3-5 share the local connection weights
- 3-6 convolution kernel and receptive field
- And step 3-7 Pad
- 3-8 describes the use of defined convolution (Tensorflow and Caffe)
- 3-9 pooling layer
- 3-10 active layer
- 3-11 BN
- 3-12 fully connected layer
- 3-13 dropout
- 3-14 loss layer (1)
- 3-15 loss layer (2)
- 3-16 convolutional neural network development history
- How to 3-17 LeNet and AlexNet- convolutional neural network and reduce the amount of computation parameters
- How to 3-18 ZFNet and VggNet- convolutional neural network and reduce the amount of computation parameters
- 3-19 Inception Series - How to reduce the convolutional neural network parameters and computation
- 3-20 convolution think from the perspective of how to reduce the amount of network computing?
- 3-21 resnet series network (1)
- 3-22 resnet series network (2)
- Network performance comparison calculation 3-23
- 3-24 Lightweight convolutional neural network -SqueezeNet
- 3-25 Lightweight convolutional neural network -MobileNet
- 3-26 Lightweight convolution neural network -ShuffleNet V1
- 3-27 Lightweight convolution neural network -ShuffleNet V2
- 3-28 multi-branch convolution neural network
- 3-29 convolutional neural network of Attention
- Compression 3-30 convolutional neural network
-
Chapter 4 Tensorflow basis Crosstalk (TF framework necessary knowledge and practical operation)
TF describes the use of foundation, including the basic concepts (graph, session, tensor, operation, feed, fetch, etc.), the core API interface, high-level API interface, data acquisition and programming, TFRecord data format and packaged programming, Cifar10 data analysis and program implementation, tensorboard debugging techniques, TF data enhancement, to help you understand through specific Cifar-10 image classification task of how to build a real Tensorflow deep ...
- 4-1 TensorFlow Concepts -Graph
- 4-2 Session-Tensor-Operation-Feed-Fetch介绍
- 4-3 TensorFlow core API interface
- 4-4 TensorFlow API method and data reading mechanism
- 4-5 Cifar10 data analysis program Case
- 4-6 Tensorflow in TFRecord data packaging program Case Look
- 4-7 How to use the sample list tf.train.slice_input_producer read the file
- 4-8 How to use the sample list tf.train.string_input_producer read the file
- 4-9 how to parse the data had been packaged by TF
- Advanced API interface 4-10 TF in
- Data 4-11 TF enhancement
- 4-12 Tensorboard debugging techniques
-
Chapter 5 Tensorflow challenge Cifar-10 image classification task
It describes how to use Tensorflow complete cifar-10 image classification, including the Cifar-10 data sets introduced, data download, data packaging, Tensorflow training framework structures, network structures, classified network training, data enhancement implementation, data analysis, Tensorflow classification model test and model optimization, through specific Cifar-10 image classification tasks to help you learn how to build a real Tensorflow deep learning network, and model training, testing ...
- 5-1 TF challenge cifar10
- 5-2 Cifar10 data read and data enhancement
- 5-3 TensorFlow + Slim build a network structure
- 5-4 Loss, Optimal, Learning Rate, BN and other definitions
- 5-5 Train write some code
- 5-6 Test portion of the coding
- 5-7 Tensorboard+tf.summary
- 5-8 recovery model and storage model
- 5-9 network structure optimization model -resnet
- 5-10 TF official version of the classification task training Cifar10
-
Chapter 6 Face Detection combat operations
It describes structures SSD face detection model, including the service description (business scenario, evaluation, etc.), Tensorflow-SSD model introduced environmental structures, Reading frame, TF-record face detection data packing, to achieve different backbone network SSD configured to adjust the different output sizes (256VS300) parameters, SSD framework of training parameters and interpret the meaning of an important parameter adjustment (learning rate, step length, aspect ratio, etc.), TensorBoard debugging, viewing images, LOS ...
- 6-1 face real business scenarios
- 6-2 Business Description Face detection and face tagging method
- 6-3 Face Detection performance evaluation
- 6-4 based on the traditional face detection method
- 6-5 Face detection
- 6-6 Face Detection problems faced by the villain face problems
- 6-7 SSD model introduced backbone network with multi-scale Feature map
- 6-8 SSD model introduces the principle (Anchor and Default box)
- 6-9 SSD model introduces the principle (Prior box, loss of function, sample configuration, data enhancement)
- 6-10 TensorFlow-ssd environmental structures (1)
- 6-11 TensorFlow-ssd build environment (2)
- 6-12 data cleaning and data packing - theory to explain (1)
- 6-13 data cleaning and data packing - Theoretical explanation (2)
- 6-14 data cleaning and data packing - practical operation (1)
- 6-15 data cleaning and data packing - practical operation (2)
- 6-16 data cleaning and data packing - practical operation (3)
- 6-17 Frame Model Training TensorFlow-ssd Interpretation (1)
- 6-18 Frame Model Training TensorFlow-ssd Explanation (2)
- 6-19 TensorFlow-ssd model training - practical operation (1)
- 6-20 TensorFlow-ssd model training - the practical operation (2)
- 6-21 TensorFlow-ssd model training - the practical operation (3)
- 6-22 How to trained model into pb file
- 6-23 TensorFlow-ssd model test
-
Chapter 7 Flask package face detection web service model
To complete the face detection web interface package by building Flask services, and to guide students to complete specific practical operation through specific programming cases.
- 7-1 Flask Introduction
- A 7-2 start Flask Case
- 7-3 in conjunction with face detection Flask define a web interface
-
Chapter 8 Web service interface calls and face detection module development
Introduces Process Intelligence applet developed to retrieve a video stream data, communication data of the front and rear end, via a web service call to the good depth learning model has been encapsulated, to build small programs related to the development of face detection functions. This chapter is mainly for face detection function modules, each subsequent chapter will gradually improve other functional modules development (face match, live detection, face attribute analysis). ...
- Introduction 8-1 micro-channel applet
- 8-2 Creating applets project
- 8-3 small face detection program to explain the project directory
- 8-4 capture of a facial image
- 8-5 Upload facial image
- With the result 8-6 draw interface calls
- 8-7 Face Detection applet demonstrates Look
-
Chapter 9 Face Matching combat operations
The main matching model TripletNet introducer face, particularly the face matching model comprises TripletNet introduced, introduction face verification database and downloading, TF-Recoder packed face matching training data, interpret the source frame, different parameters define the face matching network frame model, frame Description skill set, model train, TensorBoard debug view LOSS other information, compare performance under different parameters and backbone network, based on similarity measure TripletNet complete face ...
- 9-1 Face Matching Business
- 9-2 facial features indicate a problem (1)
- 9-3 facial features indicate a problem (2)
- 9-4 measure learning
- 9-5 facenet Introduction Principles
- 9-6 facenet environment to build
- 9-7 facenet data preparation - introduction and explanation data set
- 9-8 facenet data preparation -LFW-MTCNN
- 9-9 facenet data preparation -Dlib CASIA-Face processing and presentation CELEBA
- 9-10 facenet model training
- 9-11 facenet Source Code Reading and optimization (1)
- 9-12 facenet Source Code Reading and optimization (2)
- 9-13 facenet model test
- 9-14 pb file transfer training model, model cured
- 9-15 web interface package of the business process described face matching
- 9-16 facenet web interface package (1)
- 9-17 facenet web interface package (2)
- 9-18 end face registration applet programming
- 9-19 face registration flask server-side programming
- 9-20 Face-side programming applets sign-on implementation
- 9-21 Face Login flask server-side programming
- 9-22 Face login process review and the threshold determination
-
Chapter 10 face a 68-point positioning service real critical point
Introduces face key point positioning model 68 points, the specific content including data collection download, environment building, model building, model training, model testing, model optimization, TensorBoard debug view LOSS other information, the backbone network optimization, parameter tuning the key point to predict the results of visualization, through Tensorflow real problems people face key points to help you understand if the problem solving point return in specific engineering business, combined with the specific optimization strategies to help large ...
- 10-1 Face aligned basic concepts
- 10-2 face alignment algorithm evaluation
- Face alignment 10-3 - conventional method (1)
- 10-4 face alignment - traditional methods (2)
- 10-5 face alignment algorithm depth learning algorithm and 3D face problems (1)
- 10-6 face alignment algorithm depth learning algorithm and 3D face problems (2)
- 10-7 face alignment algorithm commonly used data sets
- Face alignment algorithm 10-8 Common Problems and Solutions
- 10-9 Tensorflow-SENet model details
- 10-10 Data Preparation and environmental parameters
- 10-11 Face Key data package (1)
- 10-12 Face Key data package (2)
- 10-13 face critical point model training program instance (1)
- 10-14 face critical point model training programming examples (2)
- 10-15 face critical point model training program instances (3)
- 10-16 face critical point Pb model exported file (model cured)
- 10-17 face key test model
- 10-18 key face model Flaskweb interface package (1)
- 10-19 key face model Flaskweb interface package (2)
- 10-20 face critical point-side programming model applets combat
-
Chapter 11 combat vivo testing business
Describes the in vivo detection model, the specific content including the call key point model, the data set is ready, open or closed eye detection algorithm, Zhang shut detection algorithms, and algorithm tuning. This section of Python3 + by model further key point of use, to complete the task in vivo detection to help people learn more about the key usage scenarios in the actual project, and by way of real, hands-on exercise everyone's practical operation ability. ...
-
Chapter 12 face real property business
To complete the predicted facial attributes and the return by building Tensorflow + multi-tasking network, specific tasks, including the face attribute Business, facial attributes multitasking network dataset description, download, TF-Recoder facial attribute data collection package, face attributes defined multi-tasking network, multi-tasking network model training, multi-tasking network backbone network optimization model, the overall model parameter optimization, model test comparison (parameter amount, computation, inference time, accuracy ...