CN110472533A - A kind of face identification method based on semi-supervised training - Google Patents
A kind of face identification method based on semi-supervised training Download PDFInfo
- Publication number
- CN110472533A CN110472533A CN201910698243.XA CN201910698243A CN110472533A CN 110472533 A CN110472533 A CN 110472533A CN 201910698243 A CN201910698243 A CN 201910698243A CN 110472533 A CN110472533 A CN 110472533A
- Authority
- CN
- China
- Prior art keywords
- label
- picture
- face
- loss function
- loss
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- 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
- G06V40/168—Feature extraction; Face representation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of face identification methods based on semi-supervised training, belong to computer vision field;Use recognition of face data set as there is label data first, face picture is crawled from internet as no label data and obtains training data by Face datection, alignment for there is label data and without label data;The loss function based on no label picture is introduced, semi-supervised training is carried out together with the loss function for having label picture;Introducing task balance factor α and data balancing factor-beta, balance have the relationship between monitor task and unsupervised task.The face identification method without label data is used compared to other, the present invention without label picture without that will cluster, and the mode using no label picture is more efficient, and the performance of model is more preferable;The method of the present invention achieves good performance boost on multiple recognition of face test sets, has good universality.
Description
Technical field
The present invention relates to a kind of face identification methods based on semi-supervised training, more particularly to one kind to be based on having label picture
The face identification method that semi-supervised training is carried out with no label picture, belongs to technical field of computer vision.
Background technique
With the development of deep learning, the precision of human face recognition model has also obtained great promotion.Face recognition technology
It has been widely used in intelligent security guard, the fields such as financial payment, gate inhibition check card, has there is high commercial value.Currently, face
The scale of identification data set alreadys exceed ten million picture, 100,000 face classifications.Further collecting more has label data
Need to expend a large amount of manpower and material resources.And there is largely without label data, there is no by reasonable utilization in internet.
Currently, existing research personnel use without label data optimize human face recognition model, but these methods require by
No label data is clustered, and the accuracy of clustering algorithm itself is lower, and ultra-large cluster needs in a large amount of
It deposits and the time.Data after cluster, every other quantity of type is inconsistent, leads to serious class imbalance.
Summary of the invention
The purpose of the present invention is overcome the deficiencies in the prior art, propose a kind of recognition of face side based on semi-supervised training
Method can more reasonably use no label data, further increase the performance of human face recognition model.
The purpose of the present invention is what is be achieved through the following technical solutions.
A kind of face identification method based on semi-supervised training: include:
Step 1: pretreatment training data;
For there is label data and without label data, carry out Face datection be aligned, make the dimension of picture m after being aligned ×
N meets the input dimension of picture of network structure model requirement;
Preferably, the m=n=112.
Preferably, using recognition of face data set as there is label data, face picture conduct is crawled from internet
Without label data.
Step 2: planned network structural model;
Network structure model includes two parts: core network and loss function altogether;Wherein, core network is responsible for feature
Extract, loss function determine network optimization content, loss function be divided into based on have label picture have supervision loss function with
And the unsupervised loss function based on no label picture;
Step 3: the loss function designed for training network;
It is combined using a variety of loss functions, the loss function of network includes two parts: including supervision loss Llabel、
Unsupervised loss Lunlabel;Whole loss function LtotalAre as follows:
Ltotal=Llabel+αLunlabel
Wherein, α controls the weight of unsupervised loss;
Preferably, described have supervision loss LlabelUsing entropy loss is intersected, it is expressed as follows:
Wherein, N indicates the number of exemplar picture, and M represents face class number, and f is face picture by trunk
The activation vector obtained after network, dimension M;yiIt is the label of picture i;fj, j=1,2,3 ..., M are that face picture is corresponding
In the activation value of classification j;fyiIt is the activation value that face picture corresponds to label.
Preferably, the unsupervised loss LunlabelIt is lost, is expressed as follows using Euclid:
Wherein, it is the number of no label picture, indicates the similarity of face classification in no label picture and training set;
Step 4: having label training data, have supervision loss function, the network of training step 2 in step 3 with step 1
Structural model obtains model parameter Paramslabel;
Step 5: there are label training data and the whole loss function in no label training data, step 3 with step 1,
The network structure model of further training step 2;Wherein, the parameter of network uses the model parameter that training obtains in step 4
ParamslabelIt is initialized;
Preferably, this step is realized by following procedure:
Step1: the model parameter Params in step 4 is usedlabelInitialize network structure model;
Step2: will have label picture and obtain the feature of picture by batch input core network without label picture, each
There are label picture and the number ratio without label picture to be determined by data balancing factor-beta in the picture of batch;
Step3: the feature of picture is passed through into full articulamentum, obtains the activation value of every picture, and activation value is divided into has
Tag activation value and no tag activation value;
Step4: there will be tag activation value to obtain probability distribution by softmax function, and connect cross entropy loss function
Llabel;
Step5: Euclid's loss function L will be met without tag activation valueunlabel, and multiplied by the weight a of unsupervised loss;
Step6: by the whole loss function L of step 3totalIt carries out that final loss is calculated, then backpropagation meter
Gradient is calculated, and updates the parameter value of core network Yu full articulamentum;
Step7: Step2-Step6 is repeated, until whole loss tends towards stability.
Step 6: carrying out face alignment application, judge whether two people are the same person: by two by pretreated
Face picture passes through core network respectively and extracts feature, and calculates the similarity of two features, when the value of similarity is greater than threshold value
When, it is believed that it is the same person;Otherwise it is assumed that not being the same person.
Beneficial effect
The method of the present invention has the advantages that compared with prior art
Invention introduces unsupervised loss functions, without that will cluster without label picture, because what is used has label data
The overwhelming majority is the picture of famous person, and what is crawled is plain people's picture without label data, i.e., the two is not same people.Therefore directly most
Smallization every is concentrated with the other similarity of tag class without label picture and training.By this method, it is able to use more
Remove Optimized model without label data, to keep the performance of model more preferable;Due to not will receive clustering algorithm and bringing without cluster
Defect, noise and data imbalance problem are such as introduced, when the noise of training data increases, data nonbalance problem is tighter
When weight, trained model performance can be reduced.
Invention introduces semi-supervised training methods, first using there is label data training pattern, in the base of this model
On plinth, the semi-supervised training of progress will be combined with there is label picture without label picture, it is flat by introducing task balance factor α and data
Factor-beta weigh to adjust the relationship between monitor task and unsupervised task, which being capable of further lift scheme
Performance.
The present invention achieves good performance boost on multiple recognition of face test sets, has good universality.
Detailed description of the invention
Fig. 1 is the work flow diagram of the method for the present invention;
Fig. 2 is the training data pretreatment process figure of the method for the present invention;
Fig. 3 is the overall network structure architecture diagram of the method for the present invention.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings
Embodiment
The present embodiment is the overall flow and network structure using the human face recognition model of semi-supervised training.
A kind of face identification method based on semi-supervised training, as shown in Figure 1, comprising the following steps:
Step 1: obtaining and pre-process training data;
As shown in Fig. 2, crawl face picture from internet for no label data, using MTCNN human-face detector into
Row Face datection obtains face rectangle frame and five key points;According to face rectangle frame and five key points, use OpenCV's
WarpAffine function carries out face alignment, and the dimension of picture after alignment is 112 × 112;For there is label data, select
Somebody's face identify data set, as Microsoft issue MS1M, as no label data carry out Face datection be aligned, after alignment
Dimension of picture be 112 × 112.
Step 2: planned network structural model;
As shown in figure 3, network structure model includes two parts: core network and loss function.Wherein, core network
It is responsible for feature extraction, can be common network model in current deep learning, such as ResNet, MobileNet, this implementation
Example selection ResNet50.The optimization content of loss function decision network.Wherein, loss function is divided into based on there is having for label picture
Supervise loss function and the unsupervised loss function based on no label picture.
Step 3: the loss function designed for training network;
It is combined using two kinds of loss functions, the loss function of network includes supervision loss Llabel, unsupervised loss
Lunlabel;Whole loss function LtotalAre as follows:
Ltotal=Llabel+αLunlabel
Wherein a controls the weight of unsupervised loss, such as α=0.1.
There is supervision loss LlabelUsing entropy loss is intersected, it is expressed as follows:
Wherein, N indicates the number of exemplar picture, and M represents face class number, and f is face picture by trunk
The activation vector obtained after network, dimension M.yiIt is the label of picture i.fJ, j=1,2,3 ..., MIt is that face picture corresponds to classification
The activation value of j.fyiIt is the activation value that face picture corresponds to label.
Unsupervised loss LunlabelIt is lost, is expressed as follows using Euclid:
Wherein, it is the number of no label picture, indicates the similarity of face classification in no label picture and training set;
Certainly, one skilled in the art will appreciate that there is supervision loss LlabelIt is not limited to using intersection entropy loss, it is possible to use its
Its loss function, such as triple loss, spaced intersection entropy loss;Unsupervised loss LunlabelIt is not limited to using Euclid
Loss, it is possible to use other loss functions, such as L1 loss, SmoothL1 loss.
Step 4: having label training data, have supervision loss function, the network of training step 2 in step 3 with step 1
Model.Wherein, the initial value of learning rate is 0.1, in the 100000th, 160,000,220,000,240,000 iteration divided by 10;Weight declines
It is kept to 5e-4;Momentum is 0.9;Optimization method is stochastic gradient descent;Batch size on every piece of video card is 128, totally 4 pieces of P40
Video card.Obtain model parameter Paramslabel。
Step 5: there are label training data and the whole loss function in no label training data, step 3 with step 1,
The network model of further training step 2;
Preferably, this step is realized by following procedure:
Step1: the model parameter Params in step 4 is usedlabelInitialization model, using 500 Wan Zhangyou label pictures,
230 Wan Zhangwu label pictures train network.Wherein, initial learning rate is 0.0001, in 30,000,60,000 iteration divided by 10;Power
5e-4 is decayed to again;Momentum is 0.9;Optimization method is stochastic gradient descent;Batch size on every piece of video card is 128, totally 4 pieces
P40 video card.
Step2: both including label picture in the picture of each batch, also comprising no label picture.The ratio of the two
It is determined by data balancing factor-beta, such as without label picture: having label picture is 1: 3.All pictures pass through core network, obtain
The feature of picture;
Step3: the feature of picture is passed through into full articulamentum, obtains the activation value of every picture, and activation value is divided into has
Tag activation value and no tag activation value;
Step4: there will be tag activation value to obtain probability distribution by softmax function, and connect cross entropy loss function;
Step5: Euclid's loss function will be connect without tag activation value, and multiplied by the weight α of unsupervised loss;
Step6: by the whole loss function L of step 3totalIt is calculated, obtains final loss function, it is then reversed to pass
Calculating gradient is broadcast, and updates the parameter value of core network Yu full articulamentum;
Step7: Step2-Step6 is repeated, until loss function tends towards stability.
Step 6: carrying out face recognition application, two are mentioned by core network respectively by pretreated face picture
Feature is taken, and judge whether it is the same person using the cosine similarity of two features.
The model of the method for the present invention training achieves good performance boost on multiple recognition of face test sets, has
Good universality.Specifically, on IJB-C test set, when false positive rate (False Positive Rate) is 1e-6, this
The kidney-Yang rate (True Positive Rate) for inventing the semi-supervised model realized is 88.01%, has number of tags with Jin Shiyong
It is compared according to trained model and improves about 5%;On IQVID test set, when false positive rate (False Positive Rate) is
When 1e-4, the kidney-Yang rate (True Positive Rate) of semi-supervised model that the present invention is realized is 48.3%, and is used only
There is the model of label data training to compare and improves about 8%.
In order to illustrate the contents of the present invention and implementation method, this specification gives above-mentioned specific embodiment.But ability
Field technique personnel should be understood that the present invention is not limited to above-mentioned preferred forms, anyone can obtain under the inspiration of the present invention
Other various forms of products out, however, make any variation in its shape or structure, it is all have it is same as the present application or
Similar technical solution, is within the scope of the present invention.
Claims (4)
1. a kind of face identification method based on semi-supervised training, which comprises the following steps:
Step 1: pretreatment training data;
Training data includes label data and without label data, has label data and the no label data for described, into
Row Face datection be aligned;
Step 2: planned network structural model;
Network structure model includes two parts: core network and loss function altogether;Wherein, core network is responsible for feature and is mentioned
It takes, loss function determines the optimization content of network, and the loss function, which is divided into, has supervision loss function based on have a label picture
And the unsupervised loss function based on no label picture;
Step 3: the loss function designed for training network;
It is combined using a variety of loss functions, the loss function of network includes two parts: including supervision loss Llabel, it is unsupervised
Lose Lunlabel;Whole loss function LtotalAre as follows:
Ltotal=Llabel+αLunlabel
Wherein, α is the weight of unsupervised loss;
Step 4: having label training data, have supervision loss function, the network mould of training step 2 in step 3 with step 1
Type;Obtain model parameter Paramslabel;
Step 5: having label training data and the whole loss function in no label training data, step 3 with step 1, into one
Walk the network model of training step 2;Wherein, the parameter of network uses ParamslabelIt is initialized;
Step 6: carrying out face alignment application, judge whether two people are the same person: passing through pretreated face for two
The core network that picture passes through the network model of step 5 training respectively extracts feature, and calculates the similarity of two features, works as phase
Like degree value be greater than threshold value when, it is believed that be the same person;Otherwise it is assumed that not being the same person.
2. the method according to claim 1, wherein described have supervision loss LlabelUse intersection entropy loss, table
Show as follows:
Wherein, N indicates the number of exemplar picture, and f is the activation vector that face picture obtains later by core network,
Dimension is M, fyiIt is the activation value for the classification that picture i belongs to label yi, M is the class number of face in training set, fjIt is picture i
Belong to the activation value of classification j.
3. the method according to claim 1, wherein the unsupervised loss LunlabelIt is lost using Euclid,
It is expressed as follows:
Wherein, U is the number of no label picture, and M is the class number of face in training set, sijIndicate no label picture i and instruction
Practice the similarity for concentrating face classification j.
4. method according to claim 1 to 3, which is characterized in that the step 5 is realized by following procedure:
Step1: Params is usedlabelInitialize network structure model;
Step2: having label picture and obtains the feature of picture by batch input core network without label picture, each batch
There are label picture and the number ratio without label picture to be determined by data balancing factor-beta in picture;
Step3: the feature of picture is passed through into full articulamentum, obtains the activation value of every picture, and activation value is divided into has label
Activation value and no tag activation value;
Step4: there will be tag activation value to obtain probability distribution by softmax function, and meet cross entropy loss function Llabel;
Step5: Euclid's loss function will be connect without tag activation value, and multiplied by the weight α of unsupervised loss;
Step6: by the whole loss function L of step 3totalIt carries out that final loss is calculated, then backpropagation calculates ladder
Degree, and update the parameter value of core network Yu full articulamentum;
Step7: Step2-Step6 is repeated, until whole loss tends towards stability.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910698243.XA CN110472533B (en) | 2019-07-31 | 2019-07-31 | Face recognition method based on semi-supervised training |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910698243.XA CN110472533B (en) | 2019-07-31 | 2019-07-31 | Face recognition method based on semi-supervised training |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110472533A true CN110472533A (en) | 2019-11-19 |
CN110472533B CN110472533B (en) | 2021-11-09 |
Family
ID=68509246
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910698243.XA Active CN110472533B (en) | 2019-07-31 | 2019-07-31 | Face recognition method based on semi-supervised training |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110472533B (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111222648A (en) * | 2020-01-15 | 2020-06-02 | 深圳前海微众银行股份有限公司 | Semi-supervised machine learning optimization method, device, equipment and storage medium |
CN111461002A (en) * | 2020-03-31 | 2020-07-28 | 华南理工大学 | Sample processing method for thermal imaging pedestrian detection |
CN111522958A (en) * | 2020-05-28 | 2020-08-11 | 泰康保险集团股份有限公司 | Text classification method and device |
CN111553267A (en) * | 2020-04-27 | 2020-08-18 | 腾讯科技(深圳)有限公司 | Image processing method, image processing model training method and device |
CN111723756A (en) * | 2020-06-24 | 2020-09-29 | 中国科学技术大学 | Facial feature point tracking method based on self-supervision and semi-supervision learning |
CN111797935A (en) * | 2020-07-13 | 2020-10-20 | 扬州大学 | Semi-supervised deep network picture classification method based on group intelligence |
CN111860669A (en) * | 2020-07-27 | 2020-10-30 | 平安科技(深圳)有限公司 | Training method and device of OCR recognition model and computer equipment |
CN112329735A (en) * | 2020-11-30 | 2021-02-05 | 姜培生 | Training method of face recognition model and online education system |
CN112417986A (en) * | 2020-10-30 | 2021-02-26 | 四川天翼网络服务有限公司 | Semi-supervised online face recognition method and system based on deep neural network model |
CN113128620A (en) * | 2021-05-11 | 2021-07-16 | 北京理工大学 | Semi-supervised domain self-adaptive picture classification method based on hierarchical relationship |
CN113591914A (en) * | 2021-06-28 | 2021-11-02 | 中国平安人寿保险股份有限公司 | Data classification method and device, computer equipment and storage medium |
CN113627366A (en) * | 2021-08-16 | 2021-11-09 | 电子科技大学 | Face recognition method based on incremental clustering |
CN114329003A (en) * | 2021-12-27 | 2022-04-12 | 北京达佳互联信息技术有限公司 | Media resource data processing method and device, electronic equipment and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102332094A (en) * | 2011-10-24 | 2012-01-25 | 西安电子科技大学 | Semi-supervised online study face detection method |
US8571272B2 (en) * | 2006-03-12 | 2013-10-29 | Google Inc. | Techniques for enabling or establishing the use of face recognition algorithms |
US20140067738A1 (en) * | 2012-08-28 | 2014-03-06 | International Business Machines Corporation | Training Deep Neural Network Acoustic Models Using Distributed Hessian-Free Optimization |
CN106845336A (en) * | 2016-12-02 | 2017-06-13 | 厦门理工学院 | A kind of semi-supervised face identification method based on local message and group sparse constraint |
CN109829433A (en) * | 2019-01-31 | 2019-05-31 | 北京市商汤科技开发有限公司 | Facial image recognition method, device, electronic equipment and storage medium |
CN110046583A (en) * | 2019-04-18 | 2019-07-23 | 南京信息工程大学 | Color face recognition method based on semi-supervised multiple view increment dictionary learning |
-
2019
- 2019-07-31 CN CN201910698243.XA patent/CN110472533B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8571272B2 (en) * | 2006-03-12 | 2013-10-29 | Google Inc. | Techniques for enabling or establishing the use of face recognition algorithms |
CN102332094A (en) * | 2011-10-24 | 2012-01-25 | 西安电子科技大学 | Semi-supervised online study face detection method |
US20140067738A1 (en) * | 2012-08-28 | 2014-03-06 | International Business Machines Corporation | Training Deep Neural Network Acoustic Models Using Distributed Hessian-Free Optimization |
CN106845336A (en) * | 2016-12-02 | 2017-06-13 | 厦门理工学院 | A kind of semi-supervised face identification method based on local message and group sparse constraint |
CN109829433A (en) * | 2019-01-31 | 2019-05-31 | 北京市商汤科技开发有限公司 | Facial image recognition method, device, electronic equipment and storage medium |
CN110046583A (en) * | 2019-04-18 | 2019-07-23 | 南京信息工程大学 | Color face recognition method based on semi-supervised multiple view increment dictionary learning |
Non-Patent Citations (4)
Title |
---|
JIWEN L.等: "Cost-Sensitive Semi-Supervised Discriminant Analysis for Face Recognition", 《IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY》 * |
KE L.等: "A novel semi-supervised face recognition for video", 《2010 INTERNATIONAL CONFERENCE ON INTELLIGENT CONTROL AND INFORMATION PROCESSING》 * |
卢小玲: "基于半监督学习的人脸识别算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
李凯 等: "半监督学习算法的收敛性及其在人脸识别中的应用", 《河北大学(自然科学版)》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111222648A (en) * | 2020-01-15 | 2020-06-02 | 深圳前海微众银行股份有限公司 | Semi-supervised machine learning optimization method, device, equipment and storage medium |
CN111222648B (en) * | 2020-01-15 | 2023-09-26 | 深圳前海微众银行股份有限公司 | Semi-supervised machine learning optimization method, device, equipment and storage medium |
CN111461002A (en) * | 2020-03-31 | 2020-07-28 | 华南理工大学 | Sample processing method for thermal imaging pedestrian detection |
CN111461002B (en) * | 2020-03-31 | 2023-05-26 | 华南理工大学 | Sample processing method for thermal imaging pedestrian detection |
CN111553267A (en) * | 2020-04-27 | 2020-08-18 | 腾讯科技(深圳)有限公司 | Image processing method, image processing model training method and device |
CN111553267B (en) * | 2020-04-27 | 2023-12-01 | 腾讯科技(深圳)有限公司 | Image processing method, image processing model training method and device |
CN111522958A (en) * | 2020-05-28 | 2020-08-11 | 泰康保险集团股份有限公司 | Text classification method and device |
CN111723756A (en) * | 2020-06-24 | 2020-09-29 | 中国科学技术大学 | Facial feature point tracking method based on self-supervision and semi-supervision learning |
CN111723756B (en) * | 2020-06-24 | 2022-09-06 | 中国科学技术大学 | Facial feature point tracking method based on self-supervision and semi-supervision learning |
CN111797935B (en) * | 2020-07-13 | 2023-10-31 | 扬州大学 | Semi-supervised depth network picture classification method based on group intelligence |
CN111797935A (en) * | 2020-07-13 | 2020-10-20 | 扬州大学 | Semi-supervised deep network picture classification method based on group intelligence |
WO2021139342A1 (en) * | 2020-07-27 | 2021-07-15 | 平安科技(深圳)有限公司 | Training method and apparatus for ocr recognition model, and computer device |
CN111860669B (en) * | 2020-07-27 | 2024-05-07 | 平安科技(深圳)有限公司 | Training method and device for OCR (optical character recognition) model and computer equipment |
CN111860669A (en) * | 2020-07-27 | 2020-10-30 | 平安科技(深圳)有限公司 | Training method and device of OCR recognition model and computer equipment |
CN112417986A (en) * | 2020-10-30 | 2021-02-26 | 四川天翼网络服务有限公司 | Semi-supervised online face recognition method and system based on deep neural network model |
CN112417986B (en) * | 2020-10-30 | 2023-03-10 | 四川天翼网络股份有限公司 | Semi-supervised online face recognition method and system based on deep neural network model |
CN112329735A (en) * | 2020-11-30 | 2021-02-05 | 姜培生 | Training method of face recognition model and online education system |
CN113128620A (en) * | 2021-05-11 | 2021-07-16 | 北京理工大学 | Semi-supervised domain self-adaptive picture classification method based on hierarchical relationship |
CN113128620B (en) * | 2021-05-11 | 2022-10-21 | 北京理工大学 | Semi-supervised domain self-adaptive picture classification method based on hierarchical relationship |
CN113591914A (en) * | 2021-06-28 | 2021-11-02 | 中国平安人寿保险股份有限公司 | Data classification method and device, computer equipment and storage medium |
CN113627366B (en) * | 2021-08-16 | 2023-04-07 | 电子科技大学 | Face recognition method based on incremental clustering |
CN113627366A (en) * | 2021-08-16 | 2021-11-09 | 电子科技大学 | Face recognition method based on incremental clustering |
CN114329003A (en) * | 2021-12-27 | 2022-04-12 | 北京达佳互联信息技术有限公司 | Media resource data processing method and device, electronic equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN110472533B (en) | 2021-11-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110472533A (en) | A kind of face identification method based on semi-supervised training | |
CN109165566A (en) | A kind of recognition of face convolutional neural networks training method based on novel loss function | |
CN102521656B (en) | Integrated transfer learning method for classification of unbalance samples | |
CN107944874B (en) | Wind control method, device and system based on transfer learning | |
CN109829430A (en) | Cross-module state pedestrian based on isomery stratification attention mechanism recognition methods and system again | |
WO2021088499A1 (en) | False invoice issuing identification method and system based on dynamic network representation | |
CN107871100A (en) | The training method and device of faceform, face authentication method and device | |
CN105608471A (en) | Robust transductive label estimation and data classification method and system | |
CN107729919A (en) | In-depth based on big data technology is complained and penetrates analysis method | |
CN104933428B (en) | A kind of face identification method and device based on tensor description | |
CN109359515A (en) | A kind of method and device that the attributive character for target object is identified | |
CN109993100A (en) | The implementation method of facial expression recognition based on further feature cluster | |
CN111292195A (en) | Risk account identification method and device | |
CN101315663A (en) | Nature scene image classification method based on area dormant semantic characteristic | |
CN109214263A (en) | A kind of face identification method based on feature multiplexing | |
CN105335756A (en) | Robust learning model and image classification system | |
CN109102157A (en) | A kind of bank's work order worksheet processing method and system based on deep learning | |
CN109002926A (en) | The photovoltaic power generation quantity prediction model and its construction method of a kind of high accuracy and application | |
CN109117872A (en) | A kind of user power utilization behavior analysis method based on automatic Optimal Clustering | |
CN109726715A (en) | A kind of character image serializing identification, structural data output method | |
CN116306864A (en) | Method for training deep learning model of power system | |
CN109272058A (en) | Integrated power load curve clustering method | |
CN106529604A (en) | Adaptive image tag robust prediction method and system | |
CN111178946A (en) | Method and system for representing user behaviors | |
CN107423818A (en) | A kind of method and system of the test data set generation of power information acquisition system unified interface |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |