CN108197561B - Face recognition model optimization control method, device, equipment and storage medium - Google Patents
Face recognition model optimization control method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN108197561B CN108197561B CN201711472112.7A CN201711472112A CN108197561B CN 108197561 B CN108197561 B CN 108197561B CN 201711472112 A CN201711472112 A CN 201711472112A CN 108197561 B CN108197561 B CN 108197561B
- Authority
- CN
- China
- Prior art keywords
- gradient
- model
- face recognition
- loss function
- optimization control
- 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.)
- Active
Links
- 238000005457 optimization Methods 0.000 title claims abstract description 55
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000004364 calculation method Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 17
- 230000006870 function Effects 0.000 claims description 76
- 238000010606 normalization Methods 0.000 claims description 15
- 238000013136 deep learning model Methods 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 4
- 239000000126 substance Substances 0.000 claims description 3
- 230000009471 action Effects 0.000 abstract description 5
- 230000000694 effects Effects 0.000 abstract description 4
- 230000001427 coherent effect Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
Images
Classifications
-
- 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
-
- 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
Abstract
The invention discloses a face recognition model optimization control method, a device, equipment and a storage medium, wherein the optimization control method comprises the following steps: performing loss calculation on the normalized features by using a triplet loss function to generate a first feedback gradient; performing loss calculation on the characteristics output after the full connection of the plurality of characteristic values by using a softmax loss function and generating a second pass-back gradient; and optimizing the parameters of the deep network model by utilizing the first return gradient and the second return gradient. According to the invention, the softmax loss function is introduced when the triplet loss function is used for optimizing the model, and the softmax loss function has a faster convergence effect, so that the training speed and the recognition accuracy of the face recognition model can be effectively improved through the auxiliary action of the softmax loss function.
Description
Technical Field
The present invention relates to the field of face recognition, and in particular, to a method, an apparatus, a device, and a storage medium for optimizing and controlling a face recognition model.
Background
With the progress of science and technology, more and more automatic algorithms and devices are applied to our lives, and the face recognition algorithm is developed rapidly in recent years because the face recognition algorithm can automatically authenticate the identity of a user. At present, the face recognition product is widely applied to enterprises and public institutions such as finance, judicial, army, public security, side inspection and the like, and is widely accepted by the public.
The face recognition algorithm has been developed for decades, and the recognition accuracy has been developed dramatically from the early stage of using manually designed face features to the present stage of using deep learning method to extract features. In 2006 Raia Hadsell et al proposed to optimize the model using a contrast method (coherent) that yielded better classification by limiting the intra-class and inter-class distances. In 2015, Florian Schroff et al proposed a metric learning method using triplets (triplets) to optimize a deep learning model, which is similar to the method proposed by Raia Hadsell et al, and the method also optimizes by limiting the intra-class and inter-class distances, but it does not need to limit all intra-class distances to be smaller than and all inter-class distances to be larger than a certain threshold, and it only needs to compare a pair of distances, in which the difference between the intra-class distance and the inter-class distance is larger than an interval, and also omits the setting of the threshold.
The method for optimizing by using the coherent cost function needs to set thresholds applied to all intra-class distances and inter-class distances, and because the application scenes are complex and changeable, a uniform threshold value is usually difficult to find and is applicable to all intra-class and inter-class distances. Although the method for optimizing the depth model by using the triplet loss function overcomes the defects, the method for optimizing the depth learning model is difficult to converge, the optimization speed is slow, and a satisfactory optimization result is difficult to obtain.
Disclosure of Invention
The invention provides a face recognition model optimization control method, device, equipment and storage medium, and aims to solve the technical problems that when the face recognition model is optimized by using the conventional coherent cost function or triple loss function, the convergence speed of the model is low and even the convergence cannot be realized.
The technical scheme adopted by the invention is as follows:
the invention provides a face recognition model optimization control method, which is used for optimizing parameters of a deep learning model for face recognition, wherein the deep learning model comprises a deep network model for receiving a plurality of training images and outputting a plurality of characteristic values and a normalization model for normalizing the plurality of characteristic values, and the optimization control method comprises the following steps:
performing loss calculation on the normalized features by using a triplet loss function to generate a first feedback gradient;
performing loss calculation on the characteristics output after the full connection of the plurality of characteristic values by using a softmax loss function and generating a second pass-back gradient;
and optimizing the parameters of the deep network model by utilizing the first return gradient and the second return gradient.
Further, optimizing the parameters of the deep network model using the first backhaul gradient and the second backhaul gradient includes:
setting a first weight value for the first return gradient, and setting a second weight value for the second return gradient, wherein the first weight value is greater than the second weight value;
and updating the parameters of the depth network model by utilizing the linear weighting of the first return gradient and the second return gradient.
Further, performing loss calculation on the feature output after the full connection of the plurality of feature values by using the softmax loss function and generating a second pass back gradient includes:
performing full-connection mapping on the plurality of characteristic values by adopting a full-connection layer to obtain a plurality of output characteristics;
and sending the output characteristics into a softmax loss function to calculate a second pass-back gradient.
Further, the softmax loss function is:
wherein the content of the first and second substances,
tkrepresenting the kth dimension of the fully-connected-layer input feature t, N being the total dimension of the feature, ykThe classification labels are M types, the values of the M types are all 0 or 1, the class label of the image is corresponding to 1, and the rest are 0.
Further, the face recognition model optimization control method of the invention further comprises:
and updating the parameters of the full connection layer by utilizing the second backhaul gradient.
Further, the triplet loss function is:
wherein x isa、xp、xnRespectively representing three images marked as a, p and n, and T () is a deep neural network including a deep network model and a normalization model, [ x ]]+＝max(0,x)；
As the distance between the images a and p of the same person,
the distance between different human images a and n is defined, and alpha is the distance between ap and an; when d isapAnd danIf the distance between the first and second electrodes is greater than 0 and less than α, a loss occurs and a first return gradient is generated.
According to another aspect of the present invention, there is also provided a face recognition model optimization control apparatus for optimizing parameters of a deep learning model for face recognition, the deep learning model including a deep network model for receiving a plurality of training images and outputting a plurality of feature values, and a normalization model for performing normalization processing on the plurality of feature values, the optimization control apparatus including:
the first operation module is used for performing loss calculation on the characteristics of the normalization processing by using a triplet loss function and generating a first echo gradient;
the second operation module is used for performing loss calculation on the characteristics output after the full connection of the plurality of characteristic values by using a softmax loss function and generating a second pass-back gradient;
and the parameter optimization module is used for optimizing the parameters of the deep network model by utilizing the first return gradient and the second return gradient.
Further, the parameter optimization module comprises:
the weight setting unit is used for setting a first weight value for the first return gradient and setting a second weight value for the second return gradient, wherein the first weight value is greater than the second weight value;
and the joint optimization unit is used for updating the parameters of the deep network model by utilizing the linear weighting of the first return gradient and the second return gradient.
According to another aspect of the present invention, there is also provided a face recognition model optimization control apparatus, including a processor, where the processor is configured to execute a program, and the program executes the face recognition model optimization control method according to the present invention.
According to another aspect of the present invention, a storage medium is further provided, where the storage medium includes a stored program, and the program controls, when running, an apparatus on which the storage medium is located to execute the face recognition model optimization control method of the present invention.
The invention has the following beneficial effects:
according to the face recognition model optimization control method, device, equipment and storage medium, the softmax loss function is introduced when the triplet loss function is utilized to optimize the model, and the softmax loss function has a faster convergence effect, so that the training speed and the recognition accuracy of the face recognition model can be effectively improved through the auxiliary action of the softmax loss function.
In addition to the objects, features and advantages described above, other objects, features and advantages of the present invention are also provided. The present invention will be described in further detail below with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the invention and, together with the description, serve to explain the invention and not to limit the invention. In the drawings:
FIG. 1 is a schematic diagram illustrating steps of a face recognition model optimization control method according to a preferred embodiment of the present invention;
FIG. 2 is a flow chart of a face recognition model optimization control method according to a preferred embodiment of the present invention;
fig. 3 is a schematic block diagram of a face recognition model optimization control device according to a preferred embodiment of the present invention.
Detailed Description
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present invention will be described in detail below with reference to the embodiments with reference to the attached drawings.
Referring to fig. 1 and fig. 2, a preferred embodiment of the present invention provides a face recognition model optimization control method, configured to optimize parameters of a deep learning model for face recognition, where in this embodiment, the deep learning model includes a deep network model for receiving a plurality of training images and outputting a plurality of feature values, and a normalization model for performing normalization processing on the plurality of feature values, and the optimization control method in this embodiment includes:
step S100, loss calculation is carried out on the normalized features by using a triplet loss function, and a first echo gradient is generated;
step S200, loss calculation is carried out on the characteristics output after the characteristic values are fully connected by utilizing a softmax loss function, and a second pass-back gradient is generated;
step S300, optimizing parameters of the deep network model by using the first backhaul gradient and the second backhaul gradient.
When the face recognition model is optimized by simply utilizing the coherent loss function or the triplet loss function, the problem that the convergence speed of the model is slow or even the convergence cannot be realized exists. In the embodiment, a softmax loss function which is easier to optimize is introduced when the model is optimized, the model convergence speed is effectively accelerated by using the auxiliary action of the softmax function, and the model identification precision is improved through the mutual promotion between the two loss functions.
In this embodiment, the deep network model has an input interface for receiving a training image, when the face recognition model is trained, the training image is input into the deep network model, the features f output by the deep network model are linearly mapped to N-dimensional features t through a full connection layer (N is the number of face image classes used for training, i.e., the number of training people, when the network is trained), and the features t are sent to a softmax loss function to complete model loss calculation and gradient pass-back.
In this embodiment, the triple method needs three images for each training, and after three images are selected each time (two similar images, one inter-class image), the three images are sent to the depth network model to output the features f1, f2, and f3 of the three images in the same way as the above method. And (4) normalizing the three features and then sending the normalized three features into a triplet loss function to complete model loss calculation and gradient return.
In the gradient returning process, the gradient values returned by the two loss functions are mutually superposed and jointly act on parameters in the deep network model, so that the network can quickly complete the optimization of the network.
In the embodiment, the softmax loss function is introduced when the triplet loss function is used for optimizing the model, and the softmax loss function has a faster convergence effect, so that the training speed and the recognition accuracy of the face recognition model can be effectively improved through the auxiliary action of the softmax loss function.
Preferably, the optimizing the parameter of the deep network model by using the first backhaul gradient and the second backhaul gradient includes:
setting a first weight value for the first return gradient, and setting a second weight value for the second return gradient, wherein the first weight value is greater than the second weight value;
and updating the parameters of the depth network model by utilizing the linear weighting of the first return gradient and the second return gradient.
In this embodiment, performing loss calculation on the feature output after the full connection of the plurality of feature values by using the softmax loss function and generating the second pass-back gradient includes:
performing full-connection mapping on the plurality of characteristic values by adopting a full-connection layer to obtain a plurality of output characteristics;
and sending the output characteristics into a softmax loss function to calculate a second pass-back gradient.
Preferably, the softmax loss function is:
wherein the content of the first and second substances,
tkrepresenting the kth dimension of the fully-connected-layer input feature t, N being the total dimension of the feature, ykThe classification labels are M types, the values of the M types are all 0 or 1, the class label of the image is corresponding to 1, and the rest are 0.
In this embodiment, the face recognition model optimization control method further includes:
and updating the parameters of the full connection layer by utilizing the second backhaul gradient.
Preferably, the triplet loss function is:
wherein x isa、xp、xnRespectively representing three images marked as a, p and n, and T () is a deep neural network including a deep network model and a normalization model, [ x ]]+＝max(0,x)；
As the distance between the images a and p of the same person,
the distance between different human images a and n is defined, and alpha is the distance between ap and an; when d isapAnd danIf the distance between the first and second electrodes is greater than 0 and less than α, a loss occurs and a first return gradient is generated.
In a preferred embodiment, referring to fig. 2, the face recognition model optimization control method includes the following steps:
1. optimizing a network model using a triplet loss function
The deep learning network model is optimized by utilizing the triplet loss function, three images are required to be input each time, and the three images are respectively marked as a, p and n, wherein a and p belong to the image of the same person, and n belongs to the image of different persons. the optimization of the triplet loss function makes the image characteristics of the same person more similar and the image characteristics of different persons more different. the triplet loss function is as follows:
wherein x isa、xp、xnRespectively representing three images marked as a, p and n, and T () is a deep neural network including a deep network model and a normalized model in FIG. 2, [ x ]]+Max (0, x). In the loss function
As the distance between the images a and p of the same person,
alpha is the distance between the set ap and an, which is the distance between the different human images a and n. When d isapAnd danWhen the distance between ap and an is greater than 0 and less than alpha, loss is generated and the gradient is returned, and the image characteristics can be optimized through the arrangement, so that the difference between ap and an is as large as possible.
In implementation, as shown in fig. 2, the images 1, 2 and 3 are respectively transmitted into the depth network model to obtain the feature f1、f2、f3And respectively normalizing the features, sending the normalized features into a triplet loss function, and finally returning the gradient.
2. Optimizing a network model using a softmax loss function
The network model optimized by the softmax loss function only needs to input one image at a time, and the images 1, 2 and 3 are simultaneously input for matching the use of the triplet function. The Softmax loss function is as follows:
wherein
tkRepresenting the kth dimension of the input feature t of the full connection layer, N being the total dimension of the feature (i.e. the number of corresponding persons in the training sample, one class for each person), ykThe classification labels are M types, the values of the M types are all 0 or 1, the class label of the image is corresponding to 1, and the rest are 0.
As shown in fig. 2, after a feature f of each image is obtained through a depth network model, a full connection layer output feature t is transmitted, and the feature t is transmitted into a softmax loss function to calculate loss and then transmit back a gradient.
3. Deep neural network joint optimization
As shown in the neural network joint optimization flow shown in fig. 2, the parameters updated in the optimization process are gathered in the fully connected layer of the depth network model and the softmax loss function of the front end. The deep network model is a part shared by two loss functions, and the full connection layer only belongs to the softmax loss function part. The update of the corresponding parameter of the full connectivity layer is completely determined by the second backhaul gradient looped back by the softmax penalty function. For the shared part, the respective weights of the two loss functions need to be set, since the triplet loss function is the main cost function for network optimization, a higher weight is set, and the softmax function is an auxiliary function, a lower weight is set, for example, the feedback gradient of the triplet loss function may be set to correspond to the weight of 0.7, and the feedback gradient of the softmax loss function may be set to correspond to the weight of 0.3. The parameter updating variable corresponding to the depth network model is linear weighting of the return gradient of the two loss functions.
The face recognition model optimization control method has the following two advantages:
1) and adding a softmax loss function to jointly optimize the parameters of the deep neural network. Because the triplet loss function of the common metric learning method is slow in convergence or cannot be converged, the model parameters of the deep neural network are optimized by combining the softmax loss function, the training speed of the network model is accelerated, and the model is easy to converge.
2) And introducing a joint loss function, and improving the accuracy of the network model to the face recognition task. In the method, a softmax loss function for face classification and a triplet loss function for face recognition are optimized in a combined mode, and two loss functions are used for simultaneously restricting updating of parameters of a depth network model, so that the network model has higher accuracy for a face recognition task.
According to another aspect of the present invention, there is also provided a face recognition model optimization control apparatus for optimizing parameters of a deep learning model for face recognition, where the deep learning model includes a deep network model for receiving a plurality of training images and outputting a plurality of feature values, and a normalization model for performing normalization processing on the plurality of feature values, and referring to fig. 3, the optimization control apparatus of this embodiment includes:
a first operation module 100, configured to perform loss calculation on the normalized feature by using a triplet loss function and generate a first echo gradient;
the second operation module 200 is configured to perform loss calculation on the features output after the full connection of the multiple feature values by using a softmax loss function, and generate a second pass-back gradient;
the parameter optimization module 300 is configured to optimize parameters of the deep network model by using the first backhaul gradient and the second backhaul gradient.
Preferably, in this embodiment, the parameter optimization module 300 includes:
the weight setting unit is used for setting a first weight value for the first return gradient and setting a second weight value for the second return gradient, wherein the first weight value is greater than the second weight value;
and the joint optimization unit is used for updating the parameters of the deep network model by utilizing the linear weighting of the first return gradient and the second return gradient.
The face recognition model optimization control device of the embodiment corresponds to the method embodiment, and the specific control process can refer to the method embodiment.
According to another aspect of the present invention, there is also provided a face recognition model optimization control apparatus, including a processor, where the processor is configured to execute a program, and the program executes the face recognition model optimization control method according to the foregoing embodiment of the present invention.
According to another aspect of the present invention, a storage medium is further provided, where the storage medium includes a stored program, and the program controls, when running, an apparatus on which the storage medium is located to execute the face recognition model optimization control method according to the foregoing embodiment of the present invention.
According to the face recognition model optimization control method, device, equipment and storage medium, the softmax loss function is introduced when the triplet loss function is used for optimizing the model, and the softmax loss function has a faster convergence effect, so that the training speed and the recognition accuracy of the face recognition model can be effectively improved under the auxiliary action of the softmax loss function.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer-executable instructions and that, although a logical order is illustrated in the flowcharts, in some cases, the steps illustrated or described may be performed in an order different than presented herein.
The functions described in the method of the present embodiment, if implemented in the form of software functional units and sold or used as independent products, may be stored in one or more storage media readable by a computing device. Based on such understanding, part of the contribution of the embodiments of the present invention to the prior art or part of the technical solution may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computing device (which may be a personal computer, a server, a mobile computing device, a network device, or the like) to execute all or part of the steps of the method described in the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same or similar parts among the embodiments are referred to each other.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. A face recognition model optimization control method is used for optimizing parameters of a deep learning model for face recognition, wherein the deep learning model comprises a deep network model for receiving a plurality of training images and outputting a plurality of characteristic values, a normalization model for performing normalization processing on the characteristic values, and a full connection layer for performing full connection mapping on the characteristic values to obtain a plurality of output characteristics, and the optimization control method comprises the following steps:
performing loss calculation on the normalized features by using a triplet loss function to generate a first feedback gradient;
performing loss calculation on the features output after the full connection of the plurality of feature values by using a softmax loss function, and generating a second pass-back gradient;
updating the parameters corresponding to the full connection layer according to the second return gradient, and optimizing the parameters of the deep network model by using the first return gradient and the second return gradient;
the optimizing the parameters of the deep network model using the first backhaul gradient and the second backhaul gradient includes:
setting a first weight value for the first return gradient, and setting a second weight value for the second return gradient, wherein the first weight value is greater than the second weight value;
and updating the parameters of the deep network model by utilizing the linear weighting of the first return gradient and the second return gradient.
2. The face recognition model optimization control method of claim 1,
the softmax loss function is:
wherein the content of the first and second substances,
tkrepresenting the kth dimension of the fully-connected-layer input feature t, N being the total dimension of the feature, ykFor the classification label, the values of all M classes are 0 or 1.
3. The face recognition model optimization control method of claim 1,
the triplet loss function is:
wherein x isa、xp、xnRespectively representing three images marked as a, p, n, T () being a deep neural network comprising said deep network model and said normalized model, [ x ]]+＝max(0,x)；
As the distance between the images a and p of the same person,
the distance between different human images a and n is defined, and alpha is the distance between ap and an; when d isapAnd danIf the distance between the first and second electrodes is greater than 0 and less than α, a loss occurs and a first return gradient is generated.
4. A face recognition model optimization control device is used for optimizing parameters of a deep learning model for face recognition, wherein the deep learning model comprises a deep network model for receiving a plurality of training images and outputting a plurality of characteristic values, a normalization model for performing normalization processing on the characteristic values, and a full connection layer for performing full connection mapping on the characteristic values to obtain a plurality of output characteristics, and the optimization control device comprises:
the first operation module is used for performing loss calculation on the characteristics of the normalization processing by using a triplet loss function and generating a first echo gradient;
the second operation module is used for performing loss calculation on the features output after the full connection of the plurality of feature values by using a softmax loss function and generating a second pass-back gradient;
the parameter optimization module is used for updating the parameters corresponding to the full connection layer according to the second return gradient and optimizing the parameters of the deep network model by using the first return gradient and the second return gradient;
the parameter optimization module comprises:
the weight setting unit is used for setting a first weight value for the first return gradient and setting a second weight value for the second return gradient, wherein the first weight value is greater than the second weight value;
and the joint optimization unit is used for updating the parameters of the deep network model by using the linear weighting of the first return gradient and the second return gradient.
5. A face recognition model optimization control apparatus comprising a processor for executing a program, characterized in that the program executes to perform the face recognition model optimization control method according to any one of claims 1 to 3.
6. A storage medium comprising a stored program, wherein the program when executed controls a device on which the storage medium is located to perform the face recognition model optimization control method according to any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711472112.7A CN108197561B (en) | 2017-12-29 | 2017-12-29 | Face recognition model optimization control method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711472112.7A CN108197561B (en) | 2017-12-29 | 2017-12-29 | Face recognition model optimization control method, device, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108197561A CN108197561A (en) | 2018-06-22 |
| CN108197561B true CN108197561B (en) | 2020-11-03 |
Family
ID=62586343
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711472112.7A Active CN108197561B (en) | 2017-12-29 | 2017-12-29 | Face recognition model optimization control method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108197561B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112912887A (en) * | 2018-11-08 | 2021-06-04 | 北京比特大陆科技有限公司 | Processing method, device and equipment based on face recognition and readable storage medium |
| CN109598385A (en) * | 2018-12-07 | 2019-04-09 | 深圳前海微众银行股份有限公司 | Anti money washing combination learning method, apparatus, equipment, system and storage medium |
| CN113096023B (en) * | 2020-01-08 | 2023-10-27 | 字节跳动有限公司 | Training method, image processing method and device for neural network and storage medium |
| CN111723709B (en) * | 2020-06-09 | 2023-07-11 | 大连海事大学 | Fly face recognition method based on deep convolutional neural network |
| CN114255354B (en) * | 2021-12-31 | 2023-04-07 | 智慧眼科技股份有限公司 | Face recognition model training method, face recognition device and related equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105574215A (en) * | 2016-03-04 | 2016-05-11 | 哈尔滨工业大学深圳研究生院 | Instance-level image search method based on multiple layers of feature representations |
| CN106845330A (en) * | 2016-11-17 | 2017-06-13 | 北京品恩科技股份有限公司 | A kind of training method of the two-dimension human face identification model based on depth convolutional neural networks |
| CN106919951A (en) * | 2017-01-24 | 2017-07-04 | 杭州电子科技大学 | A kind of Weakly supervised bilinearity deep learning method merged with vision based on click |
| CN107247949A (en) * | 2017-08-02 | 2017-10-13 | 北京智慧眼科技股份有限公司 | Face identification method, device and electronic equipment based on deep learning |
| CN107423690A (en) * | 2017-06-26 | 2017-12-01 | 广东工业大学 | A kind of face identification method and device |
-
2017
- 2017-12-29 CN CN201711472112.7A patent/CN108197561B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105574215A (en) * | 2016-03-04 | 2016-05-11 | 哈尔滨工业大学深圳研究生院 | Instance-level image search method based on multiple layers of feature representations |
| CN106845330A (en) * | 2016-11-17 | 2017-06-13 | 北京品恩科技股份有限公司 | A kind of training method of the two-dimension human face identification model based on depth convolutional neural networks |
| CN106919951A (en) * | 2017-01-24 | 2017-07-04 | 杭州电子科技大学 | A kind of Weakly supervised bilinearity deep learning method merged with vision based on click |
| CN107423690A (en) * | 2017-06-26 | 2017-12-01 | 广东工业大学 | A kind of face identification method and device |
| CN107247949A (en) * | 2017-08-02 | 2017-10-13 | 北京智慧眼科技股份有限公司 | Face identification method, device and electronic equipment based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108197561A (en) | 2018-06-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108197561B (en) | Face recognition model optimization control method, device, equipment and storage medium | |
| US10983841B2 (en) | Systems and methods for removing identifiable information | |
| CN106096535B (en) | Face verification method based on bilinear joint CNN | |
| CN111461226A (en) | Countermeasure sample generation method, device, terminal and readable storage medium | |
| CN111444826B (en) | Video detection method, device, storage medium and computer equipment | |
| CN106778910B (en) | Deep learning system and method based on local training | |
| WO2022105118A1 (en) | Image-based health status identification method and apparatus, device and storage medium | |
| CN110929836B (en) | Neural network training and image processing method and device, electronic equipment and medium | |
| JP2022141931A (en) | Method and device for training living body detection model, method and apparatus for living body detection, electronic apparatus, storage medium, and computer program | |
| CN111597779A (en) | Text generation method, device, equipment and storage medium | |
| CN110647916A (en) | Pornographic picture identification method and device based on convolutional neural network | |
| CN112600794A (en) | Method for detecting GAN attack in combined deep learning | |
| CN112749558A (en) | Target content acquisition method and device, computer equipment and storage medium | |
| CN114282059A (en) | Video retrieval method, device, equipment and storage medium | |
| CN115984930A (en) | Micro expression recognition method and device and micro expression recognition model training method | |
| Zhu et al. | A novel simple visual tracking algorithm based on hashing and deep learning | |
| CN111428701A (en) | Small-area fingerprint image feature extraction method, system, terminal and storage medium | |
| He et al. | Finger vein image deblurring using neighbors-based binary-gan (nb-gan) | |
| CN114155572A (en) | Facial expression recognition method and system | |
| Liu | Human face expression recognition based on deep learning-deep convolutional neural network | |
| CN106529490A (en) | System and method for realizing handwriting identification based on sparse auto-encoding codebook | |
| CN113269719A (en) | Model training method, image processing method, device, equipment and storage medium | |
| CN113361346A (en) | Scale parameter self-adaptive face recognition method for replacing adjustment parameters | |
| CN109447240B (en) | Training method of graphic image replication model, storage medium and computing device | |
| KR102393759B1 (en) | Method and system for generating an image processing artificial nerual network model operating in a 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 | ||
| CB02 | Change of applicant information |
Address after: 100097 Beijing Haidian District Kunming Hunan Road 51 C block two floor 207. Applicant after: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD. Address before: 100193 4, 403, block A, 14 building, 10 East North Road, Haidian District, Beijing. Applicant before: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD. |
|
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 410205 14 Changsha Zhongdian Software Park Phase I, 39 Jianshan Road, Changsha High-tech Development Zone, Yuelu District, Changsha City, Hunan Province Applicant after: Wisdom Eye Technology Co.,Ltd. Address before: 100097 2nd Floor 207, Block C, 51 Hunan Road, Kunming, Haidian District, Beijing Applicant before: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD. |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: Optimization control method, device, equipment and storage medium of face recognition model Effective date of registration: 20221205 Granted publication date: 20201103 Pledgee: Agricultural Bank of China Limited Hunan Xiangjiang New Area Branch Pledgor: Wisdom Eye Technology Co.,Ltd. Registration number: Y2022430000107 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Date of cancellation: 20231220 Granted publication date: 20201103 Pledgee: Agricultural Bank of China Limited Hunan Xiangjiang New Area Branch Pledgor: Wisdom Eye Technology Co.,Ltd. Registration number: Y2022430000107 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| CP03 | Change of name, title or address |
Address after: No. 205, Building B1, Huigu Science and Technology Industrial Park, No. 336 Bachelor Road, Bachelor Street, Yuelu District, Changsha City, Hunan Province, 410000 Patentee after: Wisdom Eye Technology Co.,Ltd. Address before: 410205 building 14, phase I, Changsha Zhongdian Software Park, No. 39, Jianshan Road, Changsha high tech Development Zone, Yuelu District, Changsha City, Hunan Province Patentee before: Wisdom Eye Technology Co.,Ltd. |
|
| CP03 | Change of name, title or address |