CN110322005A - Neural network model training method and device, face identification method - Google Patents
Neural network model training method and device, face identification method Download PDFInfo
- Publication number
- CN110322005A CN110322005A CN201910564227.1A CN201910564227A CN110322005A CN 110322005 A CN110322005 A CN 110322005A CN 201910564227 A CN201910564227 A CN 201910564227A CN 110322005 A CN110322005 A CN 110322005A
- Authority
- CN
- China
- Prior art keywords
- training
- data
- model
- face
- performance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000003062 neural network model Methods 0.000 title claims abstract description 35
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 53
- 238000003745 diagnosis Methods 0.000 claims abstract description 50
- 230000008569 process Effects 0.000 claims abstract description 13
- 238000009877 rendering Methods 0.000 claims description 12
- 241001269238 Data Species 0.000 claims description 8
- 230000001172 regenerating effect Effects 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 9
- 230000000694 effects Effects 0.000 description 3
- 238000011084 recovery Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 239000000654 additive Substances 0.000 description 2
- 230000000996 additive effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000001815 facial effect Effects 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 239000011800 void material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
This application discloses a kind of neural network model training method and devices, face identification method.This method includes acquisition image data；It generates three-dimensional data and saves the training data in render process；According to training set and diagnosis collection, training obtains convolutional neural networks model；The convolutional neural networks model obtained to training diagnoses, it assesses performance of the model under each dimension and obtains the corresponding relationship of variable and performance, and three-dimensional data is regenerated according to the corresponding index of the variable when the performance is unable to satisfy preset need；Three-dimensional data is regenerated according to described, then carries out model training until diagnostic result reaches expected requirement.Present application addresses the not high technical problems of associated scenario human face accuracy of identification.Optimize the training of recognition of face neural network model by the application, improves accuracy of identification of the model in actual scene.
Description
Technical field
This application involves recognitions of face, model training field, in particular to a kind of neural network model training method
And device, face identification method.
Background technique
Using the face identification method model of convolutional neural networks, can be widely applied for mobile phone unlock, gate inhibition, attendance,
In the scenes such as remote authentication.
Inventors have found that in the application scenarios of the non-cooperation of some users, since the image captured from camera shows
Wide-angle, fuzzy, resolution ratio is low or the disturbing factors such as uneven illumination is even.The precision of recognition of face is bad.For example, security protection dynamic
It identifies in scene, accuracy of identification still has very big room for promotion.
For the not high problem of associated scenario human face accuracy of identification in the related technology, effective solution is not yet proposed at present
Scheme.
Summary of the invention
The main purpose of the application is to provide a kind of neural network model training method and device, recognition of face side
Method, to solve the problems, such as that associated scenario human face accuracy of identification is not high.
To achieve the goals above, according to the one aspect of the application, a kind of neural network model training method is provided.
Neural network model training method according to the application includes: acquisition image data, wherein described image data are
One multiple face image datas；It generates three-dimensional data and saves the training data in render process, wherein the training
Data are the two-dimensional image datas after rendering；According to training set and diagnosis collection, training obtains convolutional neural networks model,
In, the diagnosis collection generates data at least one set of two dimension acquisition data and three-dimensional；The convolution mind that training is obtained
It is diagnosed through network model, assesses performance of the model under each dimension and obtain the corresponding relationship of variable and performance, and
And three-dimensional data is regenerated according to the corresponding index of the variable when the performance is unable to satisfy preset need；According to described heavy
Newly-generated three-dimensional data, then model training is carried out until diagnostic result reaches expected requirement.
Further, collect according to training set and diagnosis, it includes: building convolutional Neural that training, which obtains convolutional neural networks model,
Network structure includes at least: convolutional layer and full articulamentum；By the acquisition image data, the generation three-dimensional data point
For training set and diagnosis collection；It determines loss function, carries out convolutional neural networks model training.
Further, the convolutional neural networks model obtained to training diagnoses, and assesses the model at each
It includes: to carry out to the convolutional neural networks model that training obtains that performance under dimension, which obtains variable and the corresponding relationship of performance,
Diagnosis, performance of the assessment models under each dimension draw variable-performance curve.
Further, three-dimensional is regenerated according to the corresponding index of the variable when the performance is unable to satisfy preset need
Data include: to re-start three-dimensional data according to the corresponding index of the variable when the performance is unable to satisfy preset need
Generation, to increase the data sample under the conditions of the variable respective value.
To achieve the goals above, according to the another aspect of the application, a kind of face identification method is provided.
According to the face identification method of the application, comprising: establish Face Image Database, included at least in the Face Image Database
The model of one face；The convolutional neural networks model established using the Face Image Database and through the above steps, treats knowledge
Other face picture is identified, belongs to a kind of face mark of certain in Face Image Database with the determination face to be identified.
To achieve the goals above, according to the another aspect of the application, a kind of neural network model training device is provided.
It include: acquisition module according to the neural network model training device of the application, for acquiring image data, wherein
Described image data are one multiple face image datas；Three-dimensional data generation module, for generating three-dimensional data and protecting
Deposit the training data in render process, wherein the training data is the two-dimensional image data after rendering；Training module,
For collecting according to training set and diagnosis, training obtains convolutional neural networks model, wherein the diagnosis collection is at least one set of two dimension
It acquires data and three-dimensional generates data；Model Diagnosis module, the convolutional neural networks model for being obtained to training
It is diagnosed, assesses performance of the model under each dimension and obtain the corresponding relationship of variable and performance, and in the property
Three-dimensional data is regenerated according to the corresponding index of the variable when can be unable to satisfy preset need；Generation module, for according to institute
It states and regenerates three-dimensional data, then carry out model training until diagnostic result reaches expected requirement.
Further, the training module includes: network structure elements, for constructing convolutional neural networks structure, at least
It include: convolutional layer and full articulamentum；Training and diagnosis collection unit, for the acquisition image data, the generation is three-dimensional empty
Quasi- data are divided into training set and diagnosis collection；Loss function unit carries out convolutional neural networks model instruction for determining loss function
Practice.
Further, the Model Diagnosis module includes: diagnosis unit, the convolutional Neural net for obtaining to training
Network model is diagnosed, and performance of the assessment models under each dimension draws variable-performance curve.
Further, the Model Diagnosis module includes: iteration unit, for being unable to satisfy preset need in the performance
When, the generation of three-dimensional data is re-started according to the corresponding index of the variable, thus under the conditions of increasing the variable respective value
Data sample.
To achieve the goals above, according to the application's in another aspect, providing a kind of face identification device.
According to the face identification device of the application, comprising: faceform's module, for establishing Face Image Database, the people
The model of a face is included at least in face model library；Model training module, for using the Face Image Database and passing through
The convolutional neural networks model for stating step foundation, identifies face picture to be identified, with the determination face category to be identified
Certain a kind of face mark in Face Image Database.
Neural network model training method and device, face identification method in the embodiment of the present application are schemed using acquisition
As data, generates three-dimensional data and save the mode of the training data in render process, by according to training set and diagnosis
Collection, training obtain convolutional neural networks model, and the data that have reached three-dimensional and truthful data set generate and training
Purpose to realize the technical effect of optimization recognition of face neural network model training method, and then solves raising model
The not high technical problem of precision in actual scene.
Detailed description of the invention
The attached drawing constituted part of this application is used to provide further understanding of the present application, so that the application's is other
Feature, objects and advantages become more apparent upon.The illustrative examples attached drawing and its explanation of the application is for explaining the application, not
Constitute the improper restriction to the application.In the accompanying drawings:
Fig. 1 is the neural network model training method flow diagram according to the application first embodiment；
Fig. 2 is the neural network model training method flow diagram according to the application second embodiment；
Fig. 3 is the neural network model training method flow diagram according to the application 3rd embodiment；
Fig. 4 is the neural network model training method flow diagram according to the application fourth embodiment；
Fig. 5 is the neural network model training device structural schematic diagram according to the application first embodiment；
Fig. 6 is the neural network model training device structural schematic diagram according to the application second embodiment；
Fig. 7 is the neural network model training device structural schematic diagram according to the application 3rd embodiment；
Fig. 8 is the application realization principle schematic diagram；
Fig. 9 is variable-performance curve schematic diagram.
Specific embodiment
In order to make those skilled in the art more fully understand application scheme, below in conjunction in the embodiment of the present application
Attached drawing, the technical scheme in the embodiment of the application is clearly and completely described, it is clear that described embodiment is only
The embodiment of the application a part, instead of all the embodiments.Based on the embodiment in the application, ordinary skill people
Member's every other embodiment obtained without making creative work, all should belong to the model of the application protection
It encloses.
It should be noted that the description and claims of this application and term " first " in above-mentioned attached drawing, "
Two " etc. be to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should be understood that using in this way
Data be interchangeable under appropriate circumstances, so as to embodiments herein described herein.In addition, term " includes " and " tool
Have " and their any deformation, it is intended that cover it is non-exclusive include, for example, containing a series of steps or units
Process, method, system, product or equipment those of are not necessarily limited to be clearly listed step or unit, but may include without clear
Other step or units listing to Chu or intrinsic for these process, methods, product or equipment.
In this application, term " on ", "lower", "left", "right", "front", "rear", "top", "bottom", "inner", "outside",
" in ", "vertical", "horizontal", " transverse direction ", the orientation or positional relationship of the instructions such as " longitudinal direction " be orientation based on the figure or
Positional relationship.These terms are not intended to limit indicated dress primarily to better describe the application and embodiment
Set, element or component must have particular orientation, or constructed and operated with particular orientation.
Also, above-mentioned part term is other than it can be used to indicate that orientation or positional relationship, it is also possible to for indicating it
His meaning, such as term " on " also are likely used for indicating certain relations of dependence or connection relationship in some cases.For ability
For the those of ordinary skill of domain, the concrete meaning of these terms in this application can be understood as the case may be.
In addition, term " installation ", " setting ", " being equipped with ", " connection ", " connected ", " socket " shall be understood in a broad sense.For example,
It may be a fixed connection, be detachably connected or monolithic construction；It can be mechanical connection, or electrical connection；It can be direct phase
It even, or indirectly connected through an intermediary, or is two connections internal between device, element or component.
For those of ordinary skills, the concrete meaning of above-mentioned term in this application can be understood as the case may be.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.The application is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
As shown in Figure 1, this method includes the following steps, namely S102 to step S110:
Step S102 acquires image data,
Described image data are one multiple face image datas.
Specifically, one multiple faces can be acquired according to traditional human face recognition model collecting training data method
Image data.
Step S104 generates three-dimensional data and saves the training data in render process,
The training data is the two-dimensional image data after rendering.
Specifically, three-dimensional face can be restored from individual facial image by 3DMM method, then uses three-dimensional rendering
Engine renders the three-dimensional face of recovery.To shine in three-dimensional perspective, three-dimensional light, block, the dimensions progress such as three-dimensional expression
Rendering, and the two dimensional image after rendering is saved as training data.
It should be noted that above-mentioned 3DMM method is only used as the preferred embodiment in the embodiment of the present application, this field skill
Art personnel generate according to actual use situation the selection of three-dimensional data mode, in this application and without tool
Body limits.
Step S106 collects according to training set and diagnosis, and training obtains convolutional neural networks model,
The diagnosis collection generates data at least one set of two dimension acquisition data and three-dimensional.
Specifically, the acquisition of two dimension described in one group data, three-dimensional generate data as a training set, the training volume
Product neural network model.By two dimension acquisition real data and three-dimensional data be divided into training set and diagnosis collection, and diagnose collection with
The difference of traditional verifying collection is: the data for diagnosing collection are made of acquisition data and the virtual data that generate, and distribution is controllable.
Step S108, the convolutional neural networks model obtained to training diagnose, and assess the model at each
Performance under dimension obtains the corresponding relationship of variable and performance, and when the performance is unable to satisfy preset need according to the change
It measures corresponding index and regenerates three-dimensional data；
Specifically, sampling diagnostic data set diagnoses the convolutional neural networks model, and assessment models are at each
Performance under dimension, and draw variable-performance curve.If diagnosis obtains under certain variable, performance is bad or is not able to satisfy
The information of application scenarios demand then re-starts the generation of three-dimensional data according to the corresponding index of the variable, and increases the variable
Data sample under the conditions of respective value.
Step S110 regenerates three-dimensional data according to described, then carries out model training until diagnostic result reaches expected
It is required that.
Specifically, model training can be carried out again according to newly-generated data, until diagnostic result reaches expected requirement, thus
Realize data iteration.After the diagnostic result of Model Diagnosis meets the expected requirements, trained model is deployed in application system,
Realize the application of the face recognition application field recognition of face high to required precision.
It can be seen from the above description that the application realizes following technical effect:
In the embodiment of the present application, using acquisition image data, three-dimensional data is generated and are saved in render process
The mode of training data, by collecting according to training set and diagnosis, training obtains convolutional neural networks model, has reached three-dimensional
The purpose for generating and training with the data of truthful data set, to realize optimization recognition of face neural network model training
The technical effect of mode, and then solve and improve the not high technical problem of precision of the model in actual scene.
According to the embodiment of the present application, as preferred in the present embodiment, as shown in Fig. 2, collecting according to training set and diagnosis, instruction
Getting convolutional neural networks model includes:
Step S202 constructs convolutional neural networks structure, includes at least: convolutional layer and full articulamentum；
The acquisition image data, the generation three-dimensional data are divided into training set and diagnosis collect by step S204；
Step S206 determines loss function, carries out convolutional neural networks model training.
It specifically, include convolutional layer and full articulamentum when constructing the structure of convolutional neural networks.The diagnosis collection
Data are made of acquisition 2-D data and the virtual three-dimensional data that generates, and distribution is controllable.
For example, angle is from -70 degree to+70 degree, between being with 1 degree for the individual human face image acquired under different condition
Every every 1 degree has a photo as present sample, forms diagnostic data set.
For another example, center-loss Additive Angular Margin can be used using loss function
Loss function carries out convolutional neural networks model training.It should be noted that not to specific damage in embodiments herein
It loses function to be defined, as long as meeting the training requirement of convolutional neural networks model.
According to the embodiment of the present application, as preferred in the present embodiment, as shown in figure 3, the convolution obtained to training
Neural network model is diagnosed, and is assessed performance of the model under each dimension and is obtained the corresponding relationship packet of variable and performance
It includes:
Step S302, the convolutional neural networks model obtained to training diagnose, and assessment models are in each dimension
Performance under degree draws variable-performance curve.
Specifically, the curved line relation between performance and variable is established.
According to the embodiment of the present application, as preferred in the present embodiment, as shown in figure 3, being unable to satisfy in the performance pre-
Include: if regenerating three-dimensional data according to the corresponding index of the variable when demand
Step S304 is re-started when the performance is unable to satisfy preset need according to the corresponding index of the variable
The generation of three-dimensional data, to increase the data sample under the conditions of the variable respective value.
Specifically, the process that iteration is further comprised in above-mentioned steps re-starts three according to the corresponding index of different variables
The generation of dimension data increases the data sample under the conditions of the variable respective value.
In another embodiment of the application, a kind of face identification method is provided, as shown in figure 4, this method comprises:
Step S402 establishes Face Image Database, the model of a face is included at least in the Face Image Database；
Step S404 uses the convolutional neural networks mould of any foundation in the Face Image Database and Claims 1-4
Type identifies face picture to be identified, belongs to a kind of face of certain in Face Image Database with the determination face to be identified
Mark.
Specifically, using Face Image Database and the convolution mind obtained by the training of above-mentioned neural network model training method
Through network model, identification face picture can be identified, certain in Face Image Database is belonged to the determination face to be identified
A kind of face mark.
It should be noted that being deployed in application system can be in client deployment or disposes in server end, this
Field technical staff can select according to outdoor scene usage scenario, in embodiments herein and without specifically limiting.
It should be noted that step shown in the flowchart of the accompanying drawings can be in such as a group of computer-executable instructions
It is executed in computer system, although also, logical order is shown in flow charts, and it in some cases, can be with not
The sequence being same as herein executes shown or described step.
According to the embodiment of the present application, additionally provide it is a kind of for implementing the neural network model training device of the above method,
As shown in figure 5, the device includes: acquisition module 10, for acquiring image data, wherein described image data be one multiple
Face image data；Three-dimensional data generation module 20, for generating three-dimensional data and saving the training number in render process
According to, wherein the training data is the two-dimensional image data after rendering；Training module 30, for according to training set and examining
Disconnected collection, training obtain convolutional neural networks model, wherein the diagnosis collection is at least one set of two dimension acquisition data and three-dimensional
Generate data；Model Diagnosis module 40, the convolutional neural networks model for obtaining to training diagnose, and assess the mould
Performance of the type under each dimension obtains the corresponding relationship of variable and performance, and is unable to satisfy preset need in the performance
When three-dimensional data regenerated according to the corresponding index of the variable；Generation module 50, for regenerating three dimensions according to
According to, then model training is carried out until diagnostic result reaches expected requirement.
In the acquisition module 10 of the embodiment of the present application specifically, it can be adopted according to traditional human face recognition model training data
Set method acquires one multiple face image datas.
It in the three-dimensional data generation module 20 of the embodiment of the present application specifically, can be by 3DMM method, from individual face
Restore three-dimensional face in image, then the three-dimensional face of recovery is rendered using 3 d rendering engine.To in three dimensional angular
The dimensions such as degree, three-dimensional light shine, block, three-dimensional expression are rendered, and save the two dimensional image after rendering as training data.
It should be noted that above-mentioned 3DMM method is only used as the preferred embodiment in the embodiment of the present application, this field skill
Art personnel generate according to actual use situation the selection of three-dimensional data mode, in this application and without tool
Body limits.
In the training module 30 of the embodiment of the present application specifically, the acquisition of two dimension described in one group data, three-dimensional generate number
According to as a training set, the training convolutional neural networks model.By two dimension acquisition real data and three-dimensional data point
Collect for training set and diagnosis, and the difference for diagnosing collection and traditional verifying collection is: the data of collection are diagnosed by acquisition data and void
Quasi- to generate data composition, distribution is controllable.
In the Model Diagnosis module 40 of the embodiment of the present application specifically, sampling diagnostic data set is to the convolutional neural networks
Model is diagnosed, performance of the assessment models under each dimension, and draws variable-performance curve.If diagnosis obtains
Under certain variable, performance is bad or is not able to satisfy the information of application scenarios demand, then according to the corresponding index of the variable again into
The generation of row three-dimensional data, and increase the data sample under the conditions of the variable respective value.
In the generation module 50 of the embodiment of the present application specifically, model training can be carried out again according to newly-generated data, directly
Reach expected requirement to diagnostic result, to realize data iteration.After the diagnostic result of Model Diagnosis meets the expected requirements, it will instruct
The model perfected is deployed in application system, realizes the application of the face recognition application field recognition of face high to required precision.
According to the embodiment of the present application, as preferred in the present embodiment, as shown in fig. 6, the training module 30 includes:
Network structure elements 301 include at least: convolutional layer and full articulamentum for constructing convolutional neural networks structure；Training and diagnosis
Collect unit 302, for the acquisition image data, the generation three-dimensional data to be divided into training set and diagnosis collection；Loss
Function unit 303 carries out convolutional neural networks model training for determining loss function.
It in the embodiment of the present application specifically, include convolutional layer and full connection when constructing the structure of convolutional neural networks
Layer.The data of the diagnosis collection are made of acquisition 2-D data and the virtual three-dimensional data that generates, and distribution is controllable.
For example, angle is from -70 degree to+70 degree, between being with 1 degree for the individual human face image acquired under different condition
Every every 1 degree has a photo as present sample, forms diagnostic data set.
For another example, center-loss Additive Angular Margin can be used using loss function
Loss function carries out convolutional neural networks model training.It should be noted that not to specific damage in embodiments herein
It loses function to be defined, as long as meeting the training requirement of convolutional neural networks model.
According to the embodiment of the present application, as preferred in the present embodiment, as shown in fig. 6, the Model Diagnosis module 40 is wrapped
Include: diagnosis unit 401, the convolutional neural networks model for obtaining to training diagnose, and assessment models are at each
Performance under dimension draws variable-performance curve.
In the embodiment of the present application specifically, as shown in figure 9, establishing the curved line relation between performance and variable.
According to the embodiment of the present application, as preferred in the present embodiment, as shown in fig. 6, the Model Diagnosis module 40 is wrapped
Include: iteration unit 402, for when the performance is unable to satisfy preset need, according to the corresponding index of the variable again into
The generation of row three-dimensional data, to increase the data sample under the conditions of the variable respective value.
In the embodiment of the present application specifically, the process that iteration is further comprised in above-mentioned steps, it is corresponding according to different variables
Index re-starts the generation of three-dimensional data, increases the data sample under the conditions of the variable respective value.
According to another embodiment of the application, as preferred in the present embodiment, as shown in fig. 7, face identification device, packet
Include: faceform's module 60 includes at least the model of a face for establishing Face Image Database in the Face Image Database；
Model training module 70, the convolutional neural networks model for using the Face Image Database and above-mentioned training method to establish are right
Face picture to be identified is identified, belongs to a kind of face mark of certain in Face Image Database with the determination face to be identified.
In embodiments herein specifically, using Face Image Database and pass through above-mentioned neural network model training side
The convolutional neural networks model that method training obtains can identify identification face picture, with the determination face to be identified
Belong to a kind of face mark of certain in Face Image Database.
It should be noted that being deployed in application system can be in client deployment or disposes in server end, this
Field technical staff can select according to outdoor scene usage scenario, in embodiments herein and without specifically limiting.
Obviously, those skilled in the art should be understood that each module of above-mentioned the application or each step can be with general
Computing device realize that they can be concentrated on a single computing device, or be distributed in multiple computing devices and formed
Network on, optionally, they can be realized with the program code that computing device can perform, it is thus possible to which they are stored
Be performed by computing device in the storage device, perhaps they are fabricated to each integrated circuit modules or by they
In multiple modules or step be fabricated to single integrated circuit module to realize.In this way, the application be not limited to it is any specific
Hardware and software combines.
As shown in figure 8, being the neural network model training method realization principle in the application:
Step S1 is acquired data under single many condition；
According to traditional human face recognition model collecting training data method, one multiple face image datas are acquired.
Step S2, virtual data generate example；
By 3DMM method, restore three-dimensional face from individual facial image, then using 3 d rendering engine to recovery
Three-dimensional face rendered, shine in three-dimensional perspective, three-dimensional light, block, the dimensions such as three-dimensional expression are rendered, then saving wash with watercolours
Two dimensional image after dye is as training data.
Step S3, model training；
Convolutional neural networks structure, including convolutional layer and full articulamentum are constructed, it will be by the number of step 1 and step 2 generation
Collect according to being divided into training set and diagnosing.
Step S4, Model Diagnosis
The model that sampling diagnostic data set obtains step 3 training diagnoses, and assessment models are under each dimension
Performance draws variable-performance curve, and diagnosis obtains under certain variable, and performance is bad or is not able to satisfy application scenarios demand
Information re-starts the generation of three-dimensional data according to the corresponding index of the variable, increases the data under the conditions of the variable respective value
Sample.
Step S5, step iteration；
Model training is carried out again according to newly-generated data, until diagnostic result reaches expected requirement.
Step S6, model application.
Trained model is deployed in application system, realizes the application of recognition of face.
The foregoing is merely preferred embodiment of the present application, are not intended to limit this application, for the skill of this field
For art personnel, various changes and changes are possible in this application.Within the spirit and principles of this application, made any to repair
Change, equivalent replacement, improvement etc., should be included within the scope of protection of this application.
Claims (10)
1. a kind of neural network model training method characterized by comprising
Acquire image data, wherein described image data are one multiple face image datas；
It generates three-dimensional data and saves the training data in render process, wherein the training data is after rendering
Two-dimensional image data；
According to training set and diagnosis collection, training obtains convolutional neural networks model, wherein the diagnosis collection is at least one set of two dimension
It acquires data and three-dimensional generates data；
The convolutional neural networks model obtained to training diagnoses, and assesses performance of the model under each dimension and obtains
To the corresponding relationship of variable and performance, and when the performance is unable to satisfy preset need according to the corresponding index weight of the variable
Newly-generated three-dimensional data；
Three-dimensional data is regenerated according to described, then carries out model training until diagnostic result reaches expected requirement.
2. neural network model training method according to claim 1, which is characterized in that collect according to training set and diagnosis,
Training obtains convolutional neural networks model and includes:
Convolutional neural networks structure is constructed, is included at least: convolutional layer and full articulamentum；
The acquisition image data, the generation three-dimensional data are divided into training set and diagnosis collection；
It determines loss function, carries out convolutional neural networks model training.
3. neural network model training method according to claim 1, which is characterized in that the convolution obtained to training
Neural network model is diagnosed, and is assessed performance of the model under each dimension and is obtained the corresponding relationship packet of variable and performance
It includes:
The convolutional neural networks model obtained to training diagnoses, and performance of the assessment models under each dimension is drawn
Variable-performance curve.
4. neural network model training method according to claim 3, which is characterized in that be unable to satisfy in the performance pre-
Include: if regenerating three-dimensional data according to the corresponding index of the variable when demand
When the performance is unable to satisfy preset need, the life of three-dimensional data is re-started according to the corresponding index of the variable
At to increase the data sample under the conditions of the variable respective value.
5. a kind of face identification method characterized by comprising
Face Image Database is established, the model of a face is included at least in the Face Image Database；
Using the convolutional neural networks model of any foundation in the Face Image Database and Claims 1-4, to face to be identified
Picture is identified, belongs to a kind of face mark of certain in Face Image Database with the determination face to be identified.
6. a kind of neural network model training device characterized by comprising
Acquisition module, for acquiring image data, wherein described image data are one multiple face image datas；
Three-dimensional data generation module, for generating three-dimensional data and saving the training data in render process, wherein described
Training data is the two-dimensional image data after rendering；
Training module, for collecting according to training set and diagnosis, training obtains convolutional neural networks model, wherein the diagnosis collection
Data are generated at least one set of two dimension acquisition data and three-dimensional；
Model Diagnosis module, the convolutional neural networks model for obtaining to training diagnose, and assess the model every
Performance under one dimension obtains the corresponding relationship of variable and performance, and when the performance is unable to satisfy preset need according to
The corresponding index of the variable regenerates three-dimensional data；
Generation module for regenerating three-dimensional data according to, then carries out model training until diagnostic result reaches expected
It is required that.
7. neural network model training device according to claim 6, which is characterized in that the training module includes:
Network structure elements include at least: convolutional layer and full articulamentum for constructing convolutional neural networks structure；
Training and diagnosis collection unit, for by the acquisition image data, the generation three-dimensional data be divided into training set and
Diagnosis collection；
Loss function unit carries out convolutional neural networks model training for determining loss function.
8. neural network model training device according to claim 6, which is characterized in that the Model Diagnosis module packet
It includes:
Diagnosis unit, the convolutional neural networks model for obtaining to training diagnose, and assessment models are in each dimension
Performance under degree draws variable-performance curve.
9. neural network model training device according to claim 6, which is characterized in that the Model Diagnosis module packet
It includes:
Iteration unit, for being re-started according to the corresponding index of the variable when the performance is unable to satisfy preset need
The generation of three-dimensional data, to increase the data sample under the conditions of the variable respective value.
10. a kind of face identification device characterized by comprising
Faceform's module includes at least the model of a face for establishing Face Image Database in the Face Image Database；
Model training module, for using the convolutional neural networks of any foundation in the Face Image Database and Claims 1-4
Model identifies face picture to be identified, belongs to a kind of people of certain in Face Image Database with the determination face to be identified
Face mark.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910564227.1A CN110322005A (en) | 2019-06-25 | 2019-06-25 | Neural network model training method and device, face identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910564227.1A CN110322005A (en) | 2019-06-25 | 2019-06-25 | Neural network model training method and device, face identification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110322005A true CN110322005A (en) | 2019-10-11 |
Family
ID=68120505
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910564227.1A Pending CN110322005A (en) | 2019-06-25 | 2019-06-25 | Neural network model training method and device, face identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110322005A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112215357A (en) * | 2020-09-29 | 2021-01-12 | 三一专用汽车有限责任公司 | Model optimization method, device, equipment and computer readable storage medium |
| CN112858260A (en) * | 2021-02-05 | 2021-05-28 | 广东海洋大学 | Portable reagent quantitative detection system and method based on information platform |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE10316381A1 (en) * | 2003-04-10 | 2004-10-28 | Bayer Technology Services Gmbh | Procedure for training neural networks |
| CN107886064A (en) * | 2017-11-06 | 2018-04-06 | 安徽大学 | A kind of method that recognition of face scene based on convolutional neural networks adapts to |
| CN108805977A (en) * | 2018-06-06 | 2018-11-13 | 浙江大学 | A kind of face three-dimensional rebuilding method based on end-to-end convolutional neural networks |
| CN109753873A (en) * | 2018-11-23 | 2019-05-14 | 合刃科技（武汉）有限公司 | Image processing method and relevant apparatus |
-
2019
- 2019-06-25 CN CN201910564227.1A patent/CN110322005A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE10316381A1 (en) * | 2003-04-10 | 2004-10-28 | Bayer Technology Services Gmbh | Procedure for training neural networks |
| CN107886064A (en) * | 2017-11-06 | 2018-04-06 | 安徽大学 | A kind of method that recognition of face scene based on convolutional neural networks adapts to |
| CN108805977A (en) * | 2018-06-06 | 2018-11-13 | 浙江大学 | A kind of face three-dimensional rebuilding method based on end-to-end convolutional neural networks |
| CN109753873A (en) * | 2018-11-23 | 2019-05-14 | 合刃科技（武汉）有限公司 | Image processing method and relevant apparatus |
Non-Patent Citations (2)
| Title |
|---|
| 吴丹, 南京：河海大学出版社 * |
| 崔国勤等: "基于双层虚拟视图和支持向量的人脸识别方法", 《计算机学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112215357A (en) * | 2020-09-29 | 2021-01-12 | 三一专用汽车有限责任公司 | Model optimization method, device, equipment and computer readable storage medium |
| CN112858260A (en) * | 2021-02-05 | 2021-05-28 | 广东海洋大学 | Portable reagent quantitative detection system and method based on information platform |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109218619A (en) | Image acquiring method, device and system | |
| CN108961272B (en) | Method for generating skin disease image based on deep convolution countermeasure generation network | |
| CN106796716B (en) | For providing the device and method of super-resolution for low-resolution image | |
| CN109253888A (en) | Detection method and system for vehicle vehicle condition | |
| CN108182265B (en) | Multilayer iterative screening method and device for relational network | |
| CN107341443B (en) | Method for processing video frequency, device and storage medium | |
| CN112330795B (en) | Human body three-dimensional reconstruction method and system based on single RGBD image | |
| CN109614935A (en) | Car damage identification method and device, storage medium and electronic equipment | |
| CN108234927A (en) | Video frequency tracking method and system | |
| CN106778645A (en) | A kind of image processing method and device | |
| CN109559310A (en) | Power transmission and transformation inspection image quality evaluating method and system based on conspicuousness detection | |
| CN110322005A (en) | Neural network model training method and device, face identification method | |
| CN111382808A (en) | Vehicle detection processing method and device | |
| CN108921038A (en) | A kind of classroom based on deep learning face recognition technology is quickly called the roll method of registering | |
| CN105022999A (en) | Man code company real-time acquisition system | |
| CN109948439A (en) | A kind of biopsy method, system and terminal device | |
| CN110517500A (en) | A kind of people's vehicle association process method and device | |
| CN109784230A (en) | A kind of facial video image quality optimization method, system and equipment | |
| CN109977832A (en) | A kind of image processing method, device and storage medium | |
| CN112435337A (en) | Landscape visual field analysis method and system | |
| CN109671055A (en) | Pulmonary nodule detection method and device | |
| CN114187221A (en) | Infrared and visible light image fusion method based on adaptive weight learning | |
| CN106851199A (en) | A kind of method for processing video frequency and device | |
| CN107368202B (en) | Method and device for recognizing behavior of driver using mobile phone and computing equipment | |
| CN110163201B (en) | Image testing method and device, storage medium and electronic device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191011 |
|
| RJ01 | Rejection of invention patent application after publication |