CN109410131B - Face beautifying method and system based on condition generation antagonistic neural network - Google Patents

Face beautifying method and system based on condition generation antagonistic neural network Download PDF

Info

Publication number
CN109410131B
CN109410131B CN201811140959.XA CN201811140959A CN109410131B CN 109410131 B CN109410131 B CN 109410131B CN 201811140959 A CN201811140959 A CN 201811140959A CN 109410131 B CN109410131 B CN 109410131B
Authority
CN
China
Prior art keywords
beauty
face image
sample data
beautifying
generating
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
Application number
CN201811140959.XA
Other languages
Chinese (zh)
Other versions
CN109410131A (en
Inventor
陈继
谢衍涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Gexiang Technology Co ltd
Original Assignee
Hangzhou Gexiang Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hangzhou Gexiang Technology Co ltd filed Critical Hangzhou Gexiang Technology Co ltd
Priority to CN201811140959.XA priority Critical patent/CN109410131B/en
Publication of CN109410131A publication Critical patent/CN109410131A/en
Application granted granted Critical
Publication of CN109410131B publication Critical patent/CN109410131B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/77—Retouching; Inpainting; Scratch removal
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person
    • G06T2207/30201—Face

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)
  • Processing Or Creating Images (AREA)

Abstract

The invention discloses a face beautifying method and system based on a condition generation antagonistic neural network, wherein the method comprises the following steps: acquiring initial face image sample data and face image sample data after beautifying; the facial image sample data after beautifying comprises more than two beautifying styles; generating an antagonistic neural network based on conditions according to initial face image sample data and face image sample data after beautifying to construct a beautifying prediction model; acquiring a face image to be processed; and inputting the face image to be processed and the target beauty style vector into a beauty prediction model to generate a beautified face image. By utilizing the invention, all the beautifying operations are completed by adopting a uniform frame, and different beautifying styles can be flexibly generated.

Description

Face beautifying method and system based on condition generation antagonistic neural network
Technical Field
The invention relates to the technical field of image processing, in particular to a face beautifying method and system based on condition generation and antagonistic neural network.
Background
The face beautifying algorithm is to perform operations of whitening, removing freckles, ruddy and the like on a face image, so that the face looks more beautiful/handsome.
Early attempts were made by professionals using software to interactively modify the skin, such as skin reddening, skin lightening and spot removal, sometimes including face thinning, eye enlargement, etc., which was rather burdensome. With the improvement of the computing power of mobile devices such as digital cameras and mobile phones, the automatic face beautifying algorithm is developed greatly, the operations have corresponding automatic algorithms, and the face beautifying operation of portrait photos can be automatically completed without any intervention of users by combining one algorithm. However, this approach has the following disadvantages: 1. each operation requires the design of a special algorithm; 2. once set, the style of the algorithm cannot be changed, and generally only the degree can be adjusted.
Disclosure of Invention
The invention aims to provide a face beautifying method and system based on condition generation and antagonistic neural network, which solve the problems in the prior art and provide more beautifying styles.
The invention provides a face beautifying method based on a condition generation antagonistic neural network, which comprises the following steps:
acquiring initial face image sample data and face image sample data after beautifying; the facial image sample data after beautifying comprises more than two facial styles;
generating an antagonistic neural network based on conditions according to initial face image sample data and face image sample data after beautifying to construct a beautifying prediction model;
acquiring a face image to be processed;
and inputting the face image to be processed and the target beauty style vector into the beauty prediction model to generate a beautified face image.
Preferably, the generating a confrontation neural network based on the condition according to the initial face image sample data and the beautified face image sample data to construct the beautification prediction model includes:
obtaining optimal weights of a generator and a judger based on a gradient back propagation algorithm according to initial face image sample data and face image sample data after beautifying;
and constructing a beauty prediction model according to the optimal weight values of the generator and the judger.
Preferably, the cost function of the gradient back propagation algorithm is:
Figure BDA0001815798070000021
Figure BDA0001815798070000022
wherein the content of the first and second substances,
Figure BDA0001815798070000023
Figure BDA0001815798070000024
Figure BDA0001815798070000025
Figure BDA0001815798070000026
θGgenerating weights, θ, for the producers in the antagonistic neural network for the conditionsDGenerating a weight of a decision maker in the antagonistic neural network for the condition;
λstyleand λsimIs a constant;
x is a face image to be processed;
y is face image sample data after beautifying;
c is the target beauty style vector.
Preferably, λstyle＝1；λsim＝10。
Preferably, inputting the face image to be processed and the target beauty style vector into the beauty prediction model, and generating a beautified face image includes:
converting the target beauty style vector into a beauty style channel;
connecting the beauty style channel with the RGB channel of the face image to be processed along a depth dimension to form a synthesized tensor;
and inputting the synthesized tensor into the beauty prediction model to generate a face image after beauty.
The invention also provides a face beautifying system based on the condition generation antagonistic neural network, which comprises:
the first acquisition unit is used for acquiring initial face image sample data and face image sample data after beautifying; the facial image sample data after beautifying comprises more than two facial styles;
the model construction unit is used for generating an antagonistic neural network based on conditions to construct a beauty prediction model according to the initial human face image sample data and the facial image sample data after beauty;
the second acquisition unit is used for acquiring a face image to be processed;
and the image generation unit is used for inputting the face image to be processed and the target beauty style vector into the beauty prediction model and generating a face image after beauty.
Preferably, the model construction unit is specifically configured to: obtaining optimal weights of a generator and a judger based on a gradient back propagation algorithm according to initial face image sample data and face image sample data after beautifying; and constructing a beauty prediction model according to the optimal weight values of the generator and the judger.
Preferably, the cost function of the gradient back propagation algorithm is:
Figure BDA0001815798070000031
Figure BDA0001815798070000032
wherein the content of the first and second substances,
Figure BDA0001815798070000033
Figure BDA0001815798070000034
Figure BDA0001815798070000035
Figure BDA0001815798070000036
θGgenerating weights, θ, for the producers in the antagonistic neural network for the conditionsDGenerating a weight of a decision maker in the antagonistic neural network for the condition;
λstyleand λsimIs a constant;
x is a face image to be processed;
y is face image sample data after beautifying;
c is the target beauty style vector.
Preferably, λstyle＝1；λsim＝10。
Preferably, the image generation unit includes:
a conversion subunit, configured to convert the target beauty style vector into a beauty style channel;
the synthesis subunit is used for connecting the beauty style channel with the RGB channel of the face image to be processed along a depth dimension to form a synthesized tensor;
and the generating subunit is used for inputting the synthesized tensor into the beauty prediction model and generating a face image after beauty.
The face beautifying method and the system based on the condition generation antagonistic neural network provided by the invention generate the antagonistic neural network based on the condition to construct a beautifying prediction model, and then input the face image to be processed and the target beautifying style vector into the beautifying prediction model to generate the face image after beautifying. By utilizing the invention, all the beautifying operations are completed by adopting a uniform frame, and different beautifying styles can be flexibly generated.
Drawings
Fig. 1 is a flowchart of a face beautifying method for generating an anti-neural network based on conditions according to an embodiment of the present invention;
FIG. 2 is an architecture diagram of a conditionally generated antagonistic neural network;
FIG. 3 is a network architecture diagram of a generator;
FIG. 4 is a network architecture diagram of a decider;
fig. 5 is a block diagram of a face beautifying system for generating an anti-neural network based on conditions according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative only and should not be construed as limiting the invention.
Fig. 1 is a flowchart of a face beautifying method for generating an antagonistic neural network based on a condition according to an embodiment of the present invention, and as shown in fig. 1, the embodiment of the present invention provides a face beautifying method for generating an antagonistic neural network based on a condition, including the following steps:
s101, acquiring initial face image sample data and face image sample data after face beautifying. The facial image sample data after facial beautification comprises more than two facial beautification styles.
S102, according to the initial face image sample data and the face image sample data after the face is beautified, an antagonistic neural network is generated based on conditions to construct a face-beautifying prediction model.
The method is used for constructing a beauty prediction model based on a conditional generation antagonistic neural network (cGAN for short), wherein the network consists of two parts, one part is a generator and is represented by G, and the other part is a decision device and is represented by D. In the training phase, G and D learn by inputting pictures from sets a and B. In the prediction stage, a face image after the face is beautified is generated by inputting a face image to be processed and a target beautification style vector. The training set a is a plurality of original facial image samples, i.e. original facial image sample data, without beauty. The training set B is composed of beautified pictures which are all provided with beautified style labels c. That is, the training set B is sample data of the beautified face image, which includes more than two beautification styles.
Preferably, the step S102 specifically includes:
obtaining optimal weights of a generator and a judger based on a gradient back propagation algorithm according to initial face image sample data and face image sample data after beautifying;
and constructing a beauty prediction model according to the optimal weight values of the generator and the judger.
Conditional Generation of an antagonistic neural network in an embodiment of the present invention is illustrated in FIG. 2, where the weights of the generator sub-networks are represented by θGWeights of decider subnetsDenoted by thetaD. x is a face image to be processed; y is face image sample data after beautifying; c is the target beauty style vector.
The output of the generator G is an image generated in accordance with the set beauty style, and is denoted by G (x, c). The input to the decider is either G (x, c) or y. If the decision device D outputs True, namely Real, the image of the input comes from the training set, and outputs False, namely Fake, the input is generated by G; another output of the decider is the beauty style c' of the input image.
Fig. 3 is a network architecture diagram of a generator, fig. 4 is a network architecture diagram of a determiner, and parameter settings of the generator and the determiner refer to tables 1 and 2.
TABLE 1
Figure BDA0001815798070000051
Figure BDA0001815798070000061
TABLE 2
Figure BDA0001815798070000062
The data transferred between each layer of the network is a tensor, where the tensor is of 3-order, i.e., C × H × W, for example, a color image is a tensor with C equal to 3, where H × W represents the height and width of the image, and C is the number of channels.
Convolutional layers are standard artificial neural network layers, which are not described here, but the parameters of these layers need to be set, which are shown in the layer parameter columns in the table, and the meanings of the symbols therein are described below:
k3x3 indicates that the Convolution (Convolution) kernel size is 3 × 3, and similarly K7x7 indicates that the Convolution kernel size is 7 × 7, and so on.
Re L U indicates that the transfer function of the layer is a Re L U function:
Figure BDA0001815798070000063
l initial indicates that the transfer function of the layer is a linear function f (x) x.
Tanh indicates that the transfer function of the layer is:
Figure BDA0001815798070000064
the generator G and the decision device D both have a large number of weights, which are reflected in the weights of convolution kernels of all convolution layers, the weights need to find the optimal values through learning, and the optimal weights are found through inputting source data and target data pairs.
The method provided by the embodiment of the invention is established on a standard back propagation algorithm, and the training stage is based on the weight theta of a training data set A and a training data set B learning generator and a decision deviceGAnd thetaG。
The learning is to search the optimal weight value based on the gradient back propagation algorithm
Figure BDA0001815798070000071
And
Figure BDA0001815798070000072
Figure BDA0001815798070000073
Figure BDA0001815798070000074
the training process is carried out alternately in two stages, and the parameter theta is fixed firstlyGUpdate thetaDMinimize equation 1 and fix the parameter θDUpdate thetaGMinimizing equation 2.
The cost function is as follows:
Figure BDA0001815798070000075
Figure BDA0001815798070000076
wherein the content of the first and second substances,
Figure BDA0001815798070000077
Figure BDA0001815798070000078
Figure BDA0001815798070000079
Figure BDA00018157980700000710
λstyleand λsimAre constants which are respectively used to control the importance of the beauty style and the importance similar to the original image. In the present embodiment, λ is preferablestyle＝1；λsim＝10。
And S103, acquiring a face image to be processed.
And S104, inputting the face image to be processed and the target beauty style vector into the beauty prediction model to generate a face image with beautiful appearance.
Step S104 preferably specifically includes:
converting the target beauty style vector into a beauty style channel;
connecting the beauty style channel with the RGB channel of the face image to be processed along a depth dimension to form a synthesized tensor;
and inputting the synthesized tensor into the beauty prediction model to generate a face image after beauty.
For a beautified picture, several labels can be marked, such as male/female, ruddy, freckle-removing, whitening and the like, for a binarized label, a digit is used for representation, for a classified label, a one-hot vector is used for representation, and finally the digits or vectors are connected end to form a new vector, namely the target beautification style vector c. Here, the one-hot vector is, for example, assumed that the beauty types include four types of the ramen, the japanese korean, the indonesian, and the european style, and the ramen can be represented by [1,0,0,0], the japanese-korean by [0,1,0,0], and so on.
Given a beauty style vector, a beauty style channel may be constructed, see specifically Algorithm 1.
Figure BDA0001815798070000081
The connection operation is to connect a plurality of input tensors along the direction of the specified dimension to form a new tensor, in the embodiment of the invention, the RGB channel and a plurality of beauty style channels T are connected along the 1 st dimensioncAre connected together. Is provided with n input tensors, in turn
Figure BDA0001815798070000082
The resolution of these tensors is equal, i.e. H × W, and the length C of dimension 1iMay be different, a new tensor T is obtained after the Concat operation
Figure BDA0001815798070000083
Figure BDA0001815798070000084
And so on.
Fig. 5 is a block diagram of a face beautifying system for generating an antagonistic neural network based on a condition according to an embodiment of the present invention, and as shown in fig. 5, the embodiment of the present invention further provides a face beautifying system for generating an antagonistic neural network based on a condition, which includes a first obtaining unit, a model constructing unit, a second obtaining unit, and an image generating unit.
The first acquisition unit is used for acquiring initial face image sample data and face image sample data after beautifying; the facial image sample data after beautifying comprises more than two facial styles. The model construction unit is used for generating an antagonistic neural network based on conditions to construct a beauty prediction model according to the initial human face image sample data and the beautified human face image sample data. The second acquisition unit is used for acquiring a face image to be processed. And the image generation unit is used for inputting the face image to be processed and the target beauty style vector into the beauty prediction model and generating a face image after beauty.
Preferably, the model construction unit is specifically configured to: obtaining optimal weights of a generator and a judger based on a gradient back propagation algorithm according to initial face image sample data and face image sample data after beautifying; and constructing a beauty prediction model according to the optimal weight values of the generator and the judger.
The cost function of the gradient back propagation algorithm is as follows:
Figure BDA0001815798070000091
Figure BDA0001815798070000092
wherein the content of the first and second substances,
Figure BDA0001815798070000093
Figure BDA0001815798070000094
Figure BDA0001815798070000095
Figure BDA0001815798070000096
θGgenerating weights, θ, for the producers in the antagonistic neural network for the conditionsDGenerating a weight of a decision maker in the antagonistic neural network for the condition;
λstyleand λsimAre constant and are respectively used for controlling the emphasis of beauty styleImportance and importance similar to the original, i.e. if λsim＞λstyleIt is shown that the beauty effect will retain more of the original image features, and the larger the size, the more biased the original image. Otherwise, the target beautifying effect is more biased. Preferably in this embodiment, λstyle＝1；λsim＝10。
x is a face image to be processed;
y is face image sample data after beautifying;
c is the target beauty style vector.
Further, the image generation unit includes a conversion subunit, a synthesis subunit, and a generation subunit.
And the conversion subunit is used for converting the target beauty style vector into a beauty style channel. And the synthesis subunit is used for connecting the beauty style channel with the RGB channel of the face image to be processed along the depth dimension to form a synthesized tensor. And the generation subunit is used for inputting the synthesized tensor into the beauty prediction model and generating a face image after beauty.
The face beautifying method and the system based on the condition generation antagonistic neural network provided by the embodiment of the invention construct a beautifying prediction model based on the condition generation antagonistic neural network, and then input the face image to be processed and the target beautifying style vector into the beautifying prediction model to generate the face image after beautifying. By utilizing the invention, all the beautifying operations are completed by adopting a uniform frame, and different beautifying styles can be flexibly generated.
The construction, features and functions of the present invention are described in detail in the embodiments illustrated in the drawings, which are only preferred embodiments of the present invention, but the present invention is not limited by the drawings, and all equivalent embodiments modified or changed according to the idea of the present invention should fall within the protection scope of the present invention without departing from the spirit of the present invention covered by the description and the drawings.

Claims (8)

1. A face beautifying method for generating an antagonistic neural network based on conditions is characterized by comprising the following steps:
acquiring initial face image sample data and face image sample data after beautifying; the facial image sample data after beautifying comprises more than two facial styles;
generating an antagonistic neural network based on conditions according to initial face image sample data and face image sample data after beautifying to construct a beautifying prediction model;
acquiring a face image to be processed;
inputting the face image to be processed and the target beauty style vector into the beauty prediction model to generate a face image with beautiful appearance;
inputting the face image to be processed and the target beauty style vector into the beauty prediction model, and generating a beautified face image comprises:
converting the target beauty style vector into a beauty style channel;
connecting the beauty style channel with the RGB channel of the face image to be processed along a depth dimension to form a synthesized tensor;
and inputting the synthesized tensor into the beauty prediction model to generate a face image after beauty.
2. The method of claim 1, wherein the generating an antagonistic neural network based on the condition to construct the beauty prediction model according to the initial face image sample data and the beautified face image sample data comprises:
obtaining optimal weights of a generator and a judger based on a gradient back propagation algorithm according to initial face image sample data and face image sample data after beautifying;
and constructing a beauty prediction model according to the optimal weight values of the generator and the judger.
3. The method of claim 2, wherein the cost function of the gradient backpropagation algorithm is:
Figure FDA0002542357520000011
Figure FDA0002542357520000012
wherein the content of the first and second substances,
Figure FDA0002542357520000013
is a loss function of the decider;
Figure FDA0002542357520000021
a loss function for the generator;
Figure FDA0002542357520000022
a decision loss function for sample authenticity;
Figure FDA0002542357520000023
a decision loss function that is a true sample beauty condition;
Figure FDA0002542357520000024
a decision loss function that is a pseudo sample beauty condition;
Figure FDA0002542357520000025
generating a similarity measure function of the sample and the true sample;
θGgenerating weights, θ, for the producers in the antagonistic neural network for the conditionsDGenerating a weight of a decision maker in the antagonistic neural network for the condition;
Figure FDA0002542357520000026
for a mathematical expectation that y is a random variable, a function of y,
Figure FDA0002542357520000027
for the mathematical expectation of a function of x and c with x and c as random variables, Ds(y；θD) Given thetaDIn the case of (1), the probability that the face image sample data y after face beautification given by the decider is a true sample, Ds(G(x，c；θG)；θD) Given thetaDAnd thetaGIn the case of (2), the determiner judges that the beauty image G (x, c; theta) is generatedG) Probability of being a true sample, Dc(c|x；θD) Given by thetaDIn the case of (1), the decision maker judges the probability that the face image x to be processed is generated according to the beauty condition c, Dc(c|G(x，c；θG)；θD) Given thetaDAnd thetaGIn the case of (2), the determiner judges that the beauty image G (x, c; theta) is generatedG) The probability is generated according to the target beauty style vector c, G (x, c) is an input image x and a beauty condition c, and a beauty image is generated by the generator;
λstyleand λsimIs a constant;
x is a face image to be processed;
y is face image sample data after beautifying;
c is the target beauty style vector.
4. Method according to claim 3, characterized in that λstyle＝1；λsim＝10。
5. A face beautification system for generating an antagonistic neural network based on a condition, comprising:
the first acquisition unit is used for acquiring initial face image sample data and face image sample data after beautifying; the facial image sample data after beautifying comprises more than two facial styles;
the model construction unit is used for generating an antagonistic neural network based on conditions to construct a beauty prediction model according to the initial human face image sample data and the facial image sample data after beauty;
the second acquisition unit is used for acquiring a face image to be processed;
the image generation unit is used for inputting the face image to be processed and the target beauty style vector into the beauty prediction model and generating a face image after beauty;
the image generation unit includes:
a conversion subunit, configured to convert the target beauty style vector into a beauty style channel;
the synthesis subunit is used for connecting the beauty style channel with the RGB channel of the face image to be processed along a depth dimension to form a synthesized tensor;
and the generating subunit is used for inputting the synthesized tensor into the beauty prediction model and generating a face image after beauty.
6. The system according to claim 5, characterized in that the model construction unit is specifically configured to: obtaining optimal weights of a generator and a judger based on a gradient back propagation algorithm according to initial face image sample data and face image sample data after beautifying; and constructing a beauty prediction model according to the optimal weight values of the generator and the judger.
7. The system of claim 6, wherein the cost function of the gradient backpropagation algorithm is:
Figure FDA0002542357520000031
Figure FDA0002542357520000032
wherein the content of the first and second substances,
Figure FDA0002542357520000033
is a loss function of the decider;
Figure FDA0002542357520000034
a loss function for the generator;
Figure FDA0002542357520000035
a decision loss function for sample authenticity;
Figure FDA0002542357520000036
a decision loss function that is a true sample beauty condition;
Figure FDA0002542357520000037
a decision loss function that is a pseudo sample beauty condition;
Figure FDA0002542357520000038
generating a similarity measure function of the sample and the true sample;
θGgenerating weights, θ, for the producers in the antagonistic neural network for the conditionsDGenerating a weight of a decision maker in the antagonistic neural network for the condition;
Figure FDA0002542357520000041
for a mathematical expectation that y is a random variable, a function of y,
Figure FDA0002542357520000042
for the mathematical expectation of a function of x and c with x and c as random variables, Ds(y；θD) Given thetaDIn the case of (1), the probability that the face image sample data y after face beautification given by the decider is a true sample, Ds(G(x，c；θG)；θD) Given thetaDAnd thetaGIn the case of (2), the determiner judges that the beauty image G (x, c; theta) is generatedG) Probability of being a true sample, Dc(c|x；θD) To giveTheta to thetaDIn the case of (1), the decision maker judges the probability that the face image x to be processed is generated according to the beauty condition c, Dc(c|G(x，c；θG)；θD) Given thetaDAnd thetaGIn the case of (2), the determiner judges that the beauty image G (x, c; theta) is generatedG) The probability is generated according to the target beauty style vector c, G (x, c) is an input image x and a beauty condition c, and a beauty image is generated by the generator;
λstyleand λsimIs a constant;
x is a face image to be processed;
y is face image sample data after beautifying;
c is the target beauty style vector.
8. The system of claim 7, wherein λstyle＝1；λsim＝10。
CN201811140959.XA 2018-09-28 2018-09-28 Face beautifying method and system based on condition generation antagonistic neural network Active CN109410131B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811140959.XA CN109410131B (en) 2018-09-28 2018-09-28 Face beautifying method and system based on condition generation antagonistic neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811140959.XA CN109410131B (en) 2018-09-28 2018-09-28 Face beautifying method and system based on condition generation antagonistic neural network

Publications (2)

Publication Number Publication Date
CN109410131A CN109410131A (en) 2019-03-01
CN109410131B true CN109410131B (en) 2020-08-04

Family

ID=65465507

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811140959.XA Active CN109410131B (en) 2018-09-28 2018-09-28 Face beautifying method and system based on condition generation antagonistic neural network

Country Status (1)

Country Link
CN (1) CN109410131B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113542759B (en) * 2020-04-15 2024-05-10 辉达公司 Generating an antagonistic neural network assisted video reconstruction
CN111784611B (en) * 2020-07-03 2023-11-03 厦门美图之家科技有限公司 Portrait whitening method, device, electronic equipment and readable storage medium
CN112991358A (en) * 2020-09-30 2021-06-18 北京字节跳动网络技术有限公司 Method for generating style image, method, device, equipment and medium for training model
CN112967180B (en) * 2021-03-17 2023-12-22 福建库克智能科技有限公司 Training method for generating countermeasure network, image style conversion method and device
CN113222841A (en) * 2021-05-08 2021-08-06 北京字跳网络技术有限公司 Image processing method, device, equipment and medium
CN113256513B (en) * 2021-05-10 2022-07-01 杭州格像科技有限公司 Face beautifying method and system based on antagonistic neural network
CN113822245B (en) * 2021-11-22 2022-03-04 杭州魔点科技有限公司 Face recognition method, electronic device, and medium
CN115908119B (en) * 2023-01-05 2023-06-06 广州佰锐网络科技有限公司 Face image beautifying processing method and system based on artificial intelligence

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127702A (en) * 2016-06-17 2016-11-16 兰州理工大学 A kind of image mist elimination algorithm based on degree of depth study
CN108268638A (en) * 2018-01-18 2018-07-10 浙江工业大学 A kind of generation confrontation network distribution type implementation method based on Spark frames
CN108520503A (en) * 2018-04-13 2018-09-11 湘潭大学 A method of based on self-encoding encoder and generating confrontation network restoration face Incomplete image
CN108520986A (en) * 2018-05-10 2018-09-11 杭州电子科技大学 A kind of power battery method for group matching based on generation confrontation network

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8193020B2 (en) * 2006-11-15 2012-06-05 The Regents Of The University Of California Method for heteroepitaxial growth of high-quality N-face GaN, InN, and AlN and their alloys by metal organic chemical vapor deposition

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127702A (en) * 2016-06-17 2016-11-16 兰州理工大学 A kind of image mist elimination algorithm based on degree of depth study
CN108268638A (en) * 2018-01-18 2018-07-10 浙江工业大学 A kind of generation confrontation network distribution type implementation method based on Spark frames
CN108520503A (en) * 2018-04-13 2018-09-11 湘潭大学 A method of based on self-encoding encoder and generating confrontation network restoration face Incomplete image
CN108520986A (en) * 2018-05-10 2018-09-11 杭州电子科技大学 A kind of power battery method for group matching based on generation confrontation network

Also Published As

Publication number Publication date
CN109410131A (en) 2019-03-01

Similar Documents

Publication Publication Date Title
CN109410131B (en) Face beautifying method and system based on condition generation antagonistic neural network
Suárez et al. Infrared image colorization based on a triplet dcgan architecture
Svoboda et al. Two-stage peer-regularized feature recombination for arbitrary image style transfer
CN110188239B (en) Double-current video classification method and device based on cross-mode attention mechanism
CN110599395B (en) Target image generation method, device, server and storage medium
US10991150B2 (en) View generation from a single image using fully convolutional neural networks
US11501563B2 (en) Image processing method and system
CN104811684B (en) A kind of three-dimensional U.S. face method and device of image
CN109361934B (en) Image processing method, device, equipment and storage medium
US20200193578A1 (en) Method and system for image enhancement
JP2022522564A (en) Image processing methods and their devices, computer equipment and computer programs
KR20200140713A (en) Method and apparatus for training neural network model for enhancing image detail
JP2016085579A (en) Image processing apparatus and method for interactive device, and the interactive device
CN115424088A (en) Image processing model training method and device
CN115239857B (en) Image generation method and electronic device
CN113034355A (en) Portrait image double-chin removing method based on deep learning
CN113052783A (en) Face image fusion method based on face key points
JP2020091852A (en) Method for simulating realistic rendering of makeup product
JP2023082065A (en) Method of discriminating objet in image having biometric characteristics of user to verify id of the user by separating portion of image with biometric characteristic from other portion
Wang et al. Exposure fusion using a relative generative adversarial network
CN116434224A (en) Fine-granularity image recognition method and system
Goyal et al. Deep neural network ensemble by data augmentation and bagging for skin lesion classification
CN114627293A (en) Image matting method based on multi-task learning
RU2720361C1 (en) Multi-frame training of realistic neural models of speakers heads
CN113793252A (en) Image processing method, device, chip and module equipment thereof

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