CN109801228A - A kind of jewelry picture beautification algorithm based on deep learning - Google Patents
A kind of jewelry picture beautification algorithm based on deep learning Download PDFInfo
- Publication number
- CN109801228A CN109801228A CN201811554180.2A CN201811554180A CN109801228A CN 109801228 A CN109801228 A CN 109801228A CN 201811554180 A CN201811554180 A CN 201811554180A CN 109801228 A CN109801228 A CN 109801228A
- Authority
- CN
- China
- Prior art keywords
- picture
- beautification
- network
- jewelry
- loss function
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 18
- 238000010586 diagram Methods 0.000 claims abstract description 37
- 230000006870 function Effects 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000013480 data collection Methods 0.000 claims abstract description 4
- 230000004913 activation Effects 0.000 claims description 12
- 230000004069 differentiation Effects 0.000 claims description 6
- 238000005315 distribution function Methods 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 3
- 230000001537 neural effect Effects 0.000 claims description 2
- 230000009977 dual effect Effects 0.000 abstract description 4
- 230000000694 effects Effects 0.000 abstract description 4
- 238000012545 processing Methods 0.000 abstract description 4
- 238000000034 method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003796 beauty Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of, and the jewelry picture based on deep learning beautifies algorithm, is related to field of image processing.The present invention includes the following steps: S01, data collection: a large amount of pictorial diagrams and a large amount of beautification pictorial diagrams are collected from network；S02, data scaling: several pictorial diagrams are extracted from the pictorial diagram of collection at random and carry out landscaping treatment, form mutual corresponding calibration picture；S03, it builds network model: pictorial diagram, beautification pictorial diagram and calibration picture input model is trained；S04, it defines loss function: calculating the loss function of training iteration every time, obtain optimal models；S05, training picture: pictorial diagram, beautification figure, mark picture input input model are obtained into optimal models.The present invention beautifies jewelry picture by deep learning, the picture of sub-fraction is demarcated, is exercised supervision while learn, non-nominal data is added and carries out dual training, beautification picture that can directly after output edit after training study, promotes the landscaping effect and treatment effeciency of jewelry picture.
Description
Technical field
The invention belongs to field of image processings, beautify algorithm more particularly to a kind of jewelry picture based on deep learning.
Background technique
With the development of internet and the digital products such as mobile phone it is universal, the acquisition of picture becomes now compared with before
It must be more easier.To but also the demand to the intelligent editing of picture is also very urgent.The softwares such as Photoshop are suitable for special
Industry personnel provide the tool that complicated editor can be carried out to picture.Also to user, more stringent requirements are proposed simultaneously.User
The training that must carry out professional software, could operate software.For ordinary people, it is badly in need of a kind of foolproof mode
Come the function of being edited to picture.This tool does not need the knowledge that user has profession, and therefore, the upper hand of software is not yet
Need it is additional give training, reduce the threshold used, while being also able to satisfy the most demand of user.
With the development of deep learning, this foolproof tool is possibly realized.If there is the picture of sufficient amount, and this
Internet era is not problem.A suitable neural network is built, and has the picture of sufficient amount to exercise supervision study, then
After training study, user inputs any one picture, which can picture directly after output edit.But training
Picture needs are concentrated manually to be demarcated, this needs to expend a large amount of manpower and carries out artificial calibration picture, and needs stringent
Calibration assessment system calibration result is assessed.Therefore, the acquisition of training data, the i.e. artificial calibration of data set, become
Limitation training result, the i.e. bottleneck of system quality.
Therefore, the invention proposes one kind based on small data set training, but is added and does not mark sample progress dual training
Method, the landscaping effect of Lai Tisheng picture can effectively solve the problem that above-mentioned bottleneck problem.
Summary of the invention
The purpose of the present invention is to provide a kind of, and the jewelry picture based on deep learning beautifies algorithm, passes through deep learning pair
Jewelry picture is beautified, and is demarcated to the picture of sub-fraction, is exercised supervision while learn, be added non-nominal data into
Row dual training solves the problems, such as that existing picture processing needs artificial calibration low efficiency, picture landscaping effect bad.
In order to solve the above technical problems, the present invention is achieved by the following technical solutions:
The present invention is a kind of jewelry picture beautification algorithm based on deep learning, is included the following steps:
Step S01, a large amount of jewelry pictorial diagrams and a large amount of beautification jewelry pictorial diagrams data collection: are collected from network；
Step S02, data scaling: several pictorial diagrams are extracted from the jewelry pictorial diagram of collection at random and carry out beautification place
Reason, forms mutual corresponding calibration picture；
Step S03, build network model: by jewelry pictorial diagram, beautification jewelry pictorial diagram and calibration picture input model into
Row training；
Step S04, it defines loss function: calculating the loss function of training iteration every time, obtain optimal models；
Step S05, pictorial diagram, beautification figure, mark picture input input model training picture: are obtained into optimal models；
Wherein, in step S03, model includes beautification network and differentiation network；The beautification network is one 23 layers of convolution mind
Through network；It is 2 that convolution, which is broadly divided into step-length, in the beautification network, and convolution that core is 3*3, step-length 1, core are the convolution of 3*3
It is 1/2 with step-length, core is the convolution of 3*3；The differentiation network is one 7 layers of convolutional neural networks and the last layer is to connect entirely
Layer；The convolution for differentiating network first tier and using 5*5, the convolution that other layers are 3*3 using core.
Preferably, in the step S04, loss function is divided into three classes when inputting picture:
(1) when picture of the input for mark, loss function are as follows:
(2) when input picture is the diagram in kind not marked, loss function are as follows:
(3) when inputting picture is the beautification figure not marked, loss function are as follows:
Wherein, x is pictorial diagram, and y is beautification figure, y*For the beautification figure of mark, ρyFor the true distribution function of y, ρx,yFor x, y
Joint distribution function, S be beautification network, D be differentiate network, L (S (x), y*)=| | S (x)-y*||2, α, β be weight and α=
β=8 × 10-5。
Preferably, in the step S03, beautification network first tier uses core for the convolution of 5*5, and zero padding is the volume of 2*2
Product, the last layer use sigmoid activation primitive, remaining beautification network is all made of the zero padding and ReLU activation primitive of 1*1.
Preferably, in the step S03, differentiate that network is one 7 layers of convolutional neural networks and the last layer is full articulamentum；
The convolution for differentiating network first tier and using 5*5, the convolution that other layers are 3*3 using core, and latter two volume base, every layer
50% dropout layer is added, full articulamentum uses sigmoid activation primitive, and other layers use ReLU activation primitive.
Preferably, the sum of three classes loss function when the loss function of training iteration is input picture every time, i.e.,
Preferably, in the step S05, picture after training, after all volume bases use batch
Normalization layers, optimization uses ADADELTA algorithm.
The invention has the following advantages:
The present invention beautifies jewelry picture by deep learning, demarcates, is supervised to the picture of sub-fraction
While educational inspector practises, non-nominal data is added and carries out dual training, it being capable of beautification figure directly after output edit after training study
Piece promotes the landscaping effect and treatment effeciency of jewelry picture.
Certainly, it implements any of the products of the present invention and does not necessarily require achieving all the advantages described above at the same time.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, will be described below to embodiment required
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for ability
For the those of ordinary skill of domain, without creative efforts, it can also be obtained according to these attached drawings other attached
Figure.
Fig. 1 is that the step of a kind of jewelry picture based on deep learning of the invention beautifies algorithm is schemed.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts all other
Embodiment shall fall within the protection scope of the present invention.
Refering to Figure 1, the present invention is a kind of jewelry picture beautification algorithm based on deep learning, including walk as follows
It is rapid:
Step S01, data collection: collecting 1000 jewelry pictorial diagrams and 500 beautification jewelry pictorial diagrams from network,
It requires to include most of common jewelry type and style in the picture of collection, pictorial diagram and beautification figure have not required corresponding pass
System；
Step S02, data scaling: 500 pictorial diagrams are extracted in the jewelry pictorial diagram collected at random from 1000 and carry out beauty
Change processing, forms mutual corresponding 500 calibration picture；
Step S03, build network model: by jewelry pictorial diagram, beautification jewelry pictorial diagram and calibration picture input model into
Row training；
Step S04, it defines loss function: calculating the loss function of training iteration every time, obtain optimal models；
Step S05, pictorial diagram, beautification figure, mark picture input input model training picture: are obtained into optimal models；
Wherein, in step S03, model includes beautification network and differentiation network；Beautification network is one 23 layers of convolutional Neural net
Network；It is 2 that convolution, which is broadly divided into step-length, in beautification network, and core is the convolution of 3*3, i.e. I type of convolution, step-length 1, and core is the volume of 3*3
Product, i.e. II type of convolution and step-length are 1/2, and core is the convolution of 3*3, i.e. III type of convolution；Differentiation network is one 7 layers of convolutional neural networks
And the last layer is full articulamentum；Differentiate the convolution that network first tier uses the convolution of 5*5, other layers are 3*3 using core.
As shown in the table, before network 7 layers resolution ratio is reduced to 1/8th of original image, include 3 I types of convolution, later
There are 7 II types of convolution, last 9 layers include 3 III types of convolution, and resolution ratio is restored to the resolution ratio of original image.
Wherein, in step S04, loss function is divided into three classes when inputting picture:
(1) when picture of the input for mark, loss function are as follows:
(2) when input picture is the diagram in kind not marked, loss function are as follows:
(3) when inputting picture is the beautification figure not marked, loss function are as follows:
Wherein, x is pictorial diagram, and y is beautification figure, y*For the beautification figure of mark, ρyFor the true distribution function of y, ρx,yFor x, y
Joint distribution function, S be beautification network, D be differentiate network, L (S (x), y*)=| | S (x)-y*||2, α, β be weight and α=
β=8 × 10-5。
Wherein, in step S03, beautification network first tier uses core for the convolution of 5*5, and zero padding is the convolution of 2*2, finally
One layer uses sigmoid activation primitive, remaining beautification network is all made of the zero padding and ReLU activation primitive of 1*1.
Wherein, in step S03, differentiate that network is one 7 layers of convolutional neural networks and the last layer is full articulamentum；Differentiate net
Network first layer uses the convolution of 5*5, the convolution that other layers are 3*3 using core, and latter two volume base, and every layer is added 50%
Dropout layers, full articulamentum uses sigmoid activation primitive, and other layers use ReLU activation primitive, and following table is to differentiate network:
。
Wherein, the sum of three classes loss function when the loss function of training iteration is input picture every time, i.e.,
Loss function can be good at the gap of reaction model and real data, and loss function can be more preferably to subsequent optimization
Tool is analyzed and is understood, loss function is smaller, then shows that model is better.
Wherein, in step S05, picture after training, after all volume bases use batch
Normalization layers, optimization uses ADADELTA algorithm.
Data include 500 pairs of mark pictures, 500 pictorial diagrams and 500 beautification figures without mark without mark.Instruction
Practice iteration 150000 times.In iteration each time, for 16 pairs of mark pictures, 16 pictorial diagrams, 16 beautification figures, random cropping
The region of 384*384 size, as input picture.
It is worth noting that, included each unit is only drawn according to function logic in the above system embodiment
Point, but be not limited to the above division, as long as corresponding functions can be realized；In addition, each functional unit is specific
Title is also only for convenience of distinguishing each other, the protection scope being not intended to restrict the invention.
In addition, those of ordinary skill in the art will appreciate that realizing all or part of the steps in the various embodiments described above method
It is that relevant hardware can be instructed to complete by program, corresponding program can store to be situated between in a computer-readable storage
In matter.
Present invention disclosed above preferred embodiment is only intended to help to illustrate the present invention.There is no detailed for preferred embodiment
All details are described, are not limited the invention to the specific embodiments described.Obviously, according to the content of this specification,
It can make many modifications and variations.These embodiments are chosen and specifically described to this specification, is in order to better explain the present invention
Principle and practical application, so that skilled artisan be enable to better understand and utilize the present invention.The present invention is only
It is limited by claims and its full scope and equivalent.
Claims (6)
1. a kind of jewelry picture based on deep learning beautifies algorithm, which comprises the steps of:
Step S01, a large amount of jewelry pictorial diagrams and a large amount of beautification jewelry pictorial diagrams data collection: are collected from network；
Step S02, data scaling: several pictorial diagrams are extracted from the jewelry pictorial diagram of collection at random and carry out landscaping treatment, shape
At mutual corresponding calibration picture；
Step S03, it builds network model: jewelry pictorial diagram, beautification jewelry pictorial diagram and calibration picture input model is instructed
Practice；
Step S04, it defines loss function: calculating the loss function of training iteration every time, obtain optimal models；
Step S05, pictorial diagram, beautification figure, mark picture input input model training picture: are obtained into optimal models；
Wherein, in step S03, model includes beautification network and differentiation network；The beautification network is one 23 layers of convolutional Neural net
Network；It is 2 that convolution, which is broadly divided into step-length, in the beautification network, and convolution that core is 3*3, step-length 1, the convolution sum that core is 3*3 walk
A length of 1/2, core is the convolution of 3*3.
2. a kind of jewelry picture based on deep learning according to claim 1 beautifies algorithm, which is characterized in that the step
In rapid S04, loss function is divided into three classes when inputting picture:
(1) when picture of the input for mark, loss function are as follows:
(2) when input picture is the diagram in kind not marked, loss function are as follows:
(3) when inputting picture is the beautification figure not marked, loss function are as follows:
Wherein, x is pictorial diagram, and y is beautification figure, y*For the beautification figure of mark, ρyFor the true distribution function of y, ρx,yFor x, the connection of y
Distribution function is closed, S is beautification network, and D is to differentiate network, L (S (x), y*)=| | S (x)-y*||2, α, β are weight and α=β=8
×10-5。
3. a kind of jewelry picture based on deep learning according to claim 1 beautifies algorithm, which is characterized in that the step
In rapid S03, beautification network first tier uses core for the convolution of 5*5, and zero padding is the convolution of 2*2, and the last layer uses sigmoid
Activation primitive, remaining beautification network are all made of the zero padding and ReLU activation primitive of 1*1.
4. a kind of jewelry picture based on deep learning according to claim 1 beautifies algorithm, which is characterized in that the step
In rapid S03, differentiate that network is one 7 layers of convolutional neural networks and the last layer is full articulamentum；The differentiation network first tier makes
With the convolution of 5*5, the convolution that other layers are 3*3 using cores, and latter two volume base, every layer be added 50% dropout layer,
Full articulamentum uses sigmoid activation primitive, and other layers use ReLU activation primitive.
5. a kind of jewelry picture based on deep learning according to claim 1 beautifies algorithm, which is characterized in that instruction every time
The sum of three classes loss function when the loss function for practicing iteration is input picture, i.e.,
6. a kind of jewelry picture based on deep learning according to claim 1 beautifies algorithm, which is characterized in that the step
In rapid S05, picture after training, batch normalization layer is used after all volume bases, optimization use
ADADELTA algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811554180.2A CN109801228A (en) | 2018-12-18 | 2018-12-18 | A kind of jewelry picture beautification algorithm based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811554180.2A CN109801228A (en) | 2018-12-18 | 2018-12-18 | A kind of jewelry picture beautification algorithm based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109801228A true CN109801228A (en) | 2019-05-24 |
Family
ID=66557158
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811554180.2A Pending CN109801228A (en) | 2018-12-18 | 2018-12-18 | A kind of jewelry picture beautification algorithm based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109801228A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111145107A (en) * | 2019-12-06 | 2020-05-12 | 深圳市雄帝科技股份有限公司 | Image beautifying method, device, medium and equipment |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107239766A (en) * | 2017-06-08 | 2017-10-10 | 深圳市唯特视科技有限公司 | A kind of utilization resists network and the significantly face of three-dimensional configuration model ajusts method |
| AU2017101166A4 (en) * | 2017-08-25 | 2017-11-02 | Lai, Haodong MR | A Method For Real-Time Image Style Transfer Based On Conditional Generative Adversarial Networks |
| US20180075581A1 (en) * | 2016-09-15 | 2018-03-15 | Twitter, Inc. | Super resolution using a generative adversarial network |
| CN107977629A (en) * | 2017-12-04 | 2018-05-01 | 电子科技大学 | A kind of facial image aging synthetic method of feature based separation confrontation network |
| US20180122048A1 (en) * | 2016-03-18 | 2018-05-03 | Magic Pony Technology Limited | Generative methods of super resolution |
| CN108038823A (en) * | 2017-12-06 | 2018-05-15 | 厦门美图之家科技有限公司 | Image-type becomes the training method of network model, image-type becomes method and computing device |
| CN108171770A (en) * | 2018-01-18 | 2018-06-15 | 中科视拓（北京）科技有限公司 | A kind of human face expression edit methods based on production confrontation network |
| CN108711138A (en) * | 2018-06-06 | 2018-10-26 | 北京印刷学院 | A kind of gray scale picture colorization method based on generation confrontation network |
| US20180336471A1 (en) * | 2017-05-19 | 2018-11-22 | Mehdi Rezagholizadeh | Semi-supervised regression with generative adversarial networks |
-
2018
- 2018-12-18 CN CN201811554180.2A patent/CN109801228A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180122048A1 (en) * | 2016-03-18 | 2018-05-03 | Magic Pony Technology Limited | Generative methods of super resolution |
| US20180075581A1 (en) * | 2016-09-15 | 2018-03-15 | Twitter, Inc. | Super resolution using a generative adversarial network |
| US20180336471A1 (en) * | 2017-05-19 | 2018-11-22 | Mehdi Rezagholizadeh | Semi-supervised regression with generative adversarial networks |
| CN107239766A (en) * | 2017-06-08 | 2017-10-10 | 深圳市唯特视科技有限公司 | A kind of utilization resists network and the significantly face of three-dimensional configuration model ajusts method |
| AU2017101166A4 (en) * | 2017-08-25 | 2017-11-02 | Lai, Haodong MR | A Method For Real-Time Image Style Transfer Based On Conditional Generative Adversarial Networks |
| CN107977629A (en) * | 2017-12-04 | 2018-05-01 | 电子科技大学 | A kind of facial image aging synthetic method of feature based separation confrontation network |
| CN108038823A (en) * | 2017-12-06 | 2018-05-15 | 厦门美图之家科技有限公司 | Image-type becomes the training method of network model, image-type becomes method and computing device |
| CN108171770A (en) * | 2018-01-18 | 2018-06-15 | 中科视拓（北京）科技有限公司 | A kind of human face expression edit methods based on production confrontation network |
| CN108711138A (en) * | 2018-06-06 | 2018-10-26 | 北京印刷学院 | A kind of gray scale picture colorization method based on generation confrontation network |
Non-Patent Citations (2)
| Title |
|---|
| KUMAR SRICHARAN ET.AL: "Semi-supervised Conditional GANs", 《ARXIV:1708.05789V1》 * |
| TIM SALIMANS ET.AL: "Improved Techniques for Training GANs", 《ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 29(NIPS 2016)》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111145107A (en) * | 2019-12-06 | 2020-05-12 | 深圳市雄帝科技股份有限公司 | Image beautifying method, device, medium and equipment |
| CN111145107B (en) * | 2019-12-06 | 2023-03-10 | 深圳市雄帝科技股份有限公司 | Image beautifying method, device, medium and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107403141B (en) | Face detection method and device, computer readable storage medium and equipment | |
| CN105374007B (en) | Merge the pencil drawing generation method and device of skeleton stroke and textural characteristics | |
| CN110473141A (en) | Image processing method, device, storage medium and electronic equipment | |
| CN108961245A (en) | Picture quality classification method based on binary channels depth parallel-convolution network | |
| CN111626918B (en) | Method and system for carrying out style change on digital image based on semantic segmentation network technology | |
| CN103544685B (en) | A kind of image composition beautification method adjusted based on main body and system | |
| CN110335350B (en) | Virtual terrain generation method based on terrain features | |
| JP2020187727A (en) | Line drawing generation | |
| CN113362422B (en) | A shadow-robust makeup transfer system and method based on decoupling representation | |
| CN106650795A (en) | Sorting method of hotel room type images | |
| CN112967178A (en) | Image conversion method, device, equipment and storage medium | |
| CN106383912A (en) | Picture retrieval method and apparatus | |
| CN110335340A (en) | Object threedimensional model generation method, system and storage medium based on super-pixel | |
| CN113822830A (en) | Multi-exposure image fusion method based on depth perception enhancement | |
| CN107784360A (en) | Step-by-step movement convolutional neural networks beta pruning compression method | |
| CN109740529A (en) | A kind of virtual fit method drawn based on neural network | |
| US11734389B2 (en) | Method for generating human-computer interactive abstract image | |
| CN114820292A (en) | Image synthesis method, device, device and storage medium | |
| CN105447573A (en) | Method and system for interactively completing the process of solving mathematical problems | |
| CN108305311A (en) | A kind of digitized image wash painting style technology | |
| CN109801228A (en) | A kind of jewelry picture beautification algorithm based on deep learning | |
| CN116957866A (en) | A personalized teaching device for digital human teachers | |
| CN111325212A (en) | Model training method and device, electronic equipment and computer readable storage medium | |
| CN110990589A (en) | Knowledge graph automatic generation method based on deep reinforcement learning | |
| CN114917583A (en) | Animation style game background generation method and platform based on generation confrontation network |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20210111 Address after: Room 902, 9 / F, building F, Tianhe Software Park, Tianhe District, Guangzhou 510000 Applicant after: GUANGDONG 3VJIA INFORMATION TECHNOLOGY Co.,Ltd. Address before: 230000 room 323, building E1, phase II, innovation industrial park, 2800 innovation Avenue, high tech Zone, Hefei City, Anhui Province Applicant before: HEFEI ABACI SCIENCE & TECHNOLOGY Co.,Ltd. |
|
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190524 |