CN110428476A - A kind of image conversion method and device based on multi-cycle production confrontation network - Google Patents
A kind of image conversion method and device based on multi-cycle production confrontation network Download PDFInfo
- Publication number
- CN110428476A CN110428476A CN201910607839.4A CN201910607839A CN110428476A CN 110428476 A CN110428476 A CN 110428476A CN 201910607839 A CN201910607839 A CN 201910607839A CN 110428476 A CN110428476 A CN 110428476A
- Authority
- CN
- China
- Prior art keywords
- image
- classification
- category
- network
- production
- 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
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 78
- 238000006243 chemical reaction Methods 0.000 title claims abstract description 58
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000012549 training Methods 0.000 claims description 16
- 230000004048 modification Effects 0.000 claims description 13
- 238000012986 modification Methods 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 53
- 230000008569 process Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000037361 pathway Effects 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 230000026683 transduction Effects 0.000 description 2
- 238000010361 transduction Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 150000001875 compounds Chemical group 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000010422 painting Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
- G06T9/002—Image coding using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention discloses a kind of image conversion methods based on multi-cycle production confrontation network, and step includes: the image for obtaining and needing to convert, and described image is divided into different the first image category and the second image category;Go out at least one intermediate image classification according to the image configuration of the image of the first image classification and second image category;The production including several generators and several arbiters, which is constructed, according to the first image classification, second image category and the intermediate image classification fights network;Wherein, the first image classification is corresponding generates the first arbiter, and second image category the second arbiter of corresponding generation and the intermediate image classification, which correspond to, generates third arbiter;Network, which is fought, by the production carries out image conversion.The present invention can be improved image transfer capability to complete more complicated and difficult convert task.
Description
Technical field
The present invention relates to technical field of image processing, more particularly, to a kind of figure based on multi-cycle production confrontation network
As conversion method and device.
Background technique
Production confrontation network based on convolutional neural networks is current image newest method into image conversion art,
Its conversion effect achieves significant raising compared to conventional method.Production confrontation network is recycled by its outstanding conversion effect
Fruit and the advantage for not needing the data set matched two-by-two become current common image conversion method.
Circulation production confrontation network major defect be to be difficult to handle the image convert task of complex difficulty, when
When image difference in circulation production confrontation network between the first image category and the second image category is larger, neural network
It can lead to image convert failed because that can not restrain for trained algorithm.Referring to Figure 1, Fig. 1 is that the circulation of the prior art is raw
An accepted way of doing sth fights network image conversion method, and there are two image category, classification 01 and classifications 02 in figure, and when between image category
When image difference is larger, the algorithm for training that will lead to each circulation production confrontation network can not restrain and image conversion
Failure.
Summary of the invention
In view of the above technical problems, the present invention provides a kind of image conversion sides based on multi-cycle production confrontation network
Method and device can be improved image transfer capability to complete more complicated and difficult convert task.The technical solution is as follows:
In a first aspect, the embodiment of the invention provides a kind of image conversion sides based on multi-cycle production confrontation network
Method, step include:
The image for needing to convert is obtained, and described image is divided into different the first image category and the second image class
Not;
Gone out among at least one according to the image configuration of the image of the first image classification and second image category
Image category;
It include several lifes according to the building of the first image classification, second image category and the intermediate image classification
It grows up to be a useful person and fights network with the production of several arbiters;Wherein, the first arbiter of the corresponding generation of the first image classification, it is described
Second image category is corresponding to generate the second arbiter and the corresponding generation third arbiter of the intermediate image classification;
Network, which is fought, by the production carries out image conversion.
In a first possible implementation of the first aspect of the invention, described according to the first image classification, institute
It states the second image category and intermediate image classification building includes the production confrontation net of several generators and several arbiters
Network, comprising:
Generate the first generator, the second generator, third generator and the 4th generator;Wherein, first generator
For the image of the first image classification to be converted into the image of the intermediate image classification;Second generator is used for will
The image of the intermediate image classification is converted into the image of second image category;The third generator is used for described the
The image of two image categories is converted into the image of the intermediate image classification;4th generator is used for the intermediate image
The image of classification is converted into the image of the first image classification.
In a second possible implementation of the first aspect of the invention, described that network is fought based on multi-cycle production
Image conversion method further include to the production confrontation network be based on loss function training method be trained.
It is described that production confrontation network is based in the third possible implementation of first aspect present invention
Loss function training method is trained, specific steps are as follows:
The included generation confrontation loss function of network is fought according to production and that extracts in production confrontation network follows
Ring consistency loss function constructs total losses function;
The parameter for being constructed by modification generator keeps the total losses function minimum while parameter by modification arbiter
Make the maximum generic function of total losses function;
The generic function is solved.
It is described that the generic function is solved in the 4th kind of possible implementation of first aspect present invention, specifically:
The generic function is solved using the stochastic gradient descent algorithm of standard.
It is described to need the image converted into CT image in the 5th kind of possible implementation of first aspect present invention.
Second aspect, the embodiment of the invention provides a kind of image converting means based on multi-cycle production confrontation network
It sets, comprising:
Described image for obtaining the image for needing to convert, and is divided into the first different images by image collection module
Classification and the second image category;
Intermediate image category construction module, for according to the first image classification image and second image category
Image configuration go out at least one intermediate image classification;
Network generation module, for according to the first image classification, second image category and the intermediate image
Classification constructs production and fights network;Wherein, corresponding first arbiter of the first image classification, second image category pair
The second arbiter and the intermediate image classification is answered to correspond to third arbiter;
Image conversion module carries out image conversion for fighting network by the production.
In a first possible implementation of the second aspect of the invention, described that network is fought based on multi-cycle production
Image conversion apparatus further include network training module, for the production confrontation network be based on loss function training method
It is trained.
It is described that network is fought based on multi-cycle production in second of possible implementation of second aspect of the present invention
Image conversion apparatus further include:
Total losses function constructs module, for fighting the included generation confrontation loss function of network according to production and in life
The circulation consistency loss function building total losses function extracted in accepted way of doing sth confrontation network;
Generic function constructs module, and the parameter for being constructed by modification generator keeps the total losses function minimum while logical
The parameter for crossing modification arbiter makes the maximum generic function of total losses function;
Function solves module, for solving to the generic function.
Compared with the prior art, the embodiment of the present invention has the following beneficial effects:
The present invention possesses more than two classifications and can not handle longer for existing multi-cycle production confrontation network
The shortcomings that information is converted proposes a kind of image conversion method based on multi-cycle production confrontation network, is keeping the first figure
As the convert task between classification and the second image category it is constant in the case where, by with the first image category and the second image class
Other image pattern is automatically synthesized the intermediate image classification between the first image category and the second image category, can be one
The middle category of a intermediate image classification or multiple intermediate image classifications as the first image category and the second image category, makes
Image pattern is obtained to convert by middle category.In the case where keeping original conversion task constant, more difficult image is converted
It is decomposed into multiple relatively simple steps to carry out, the simplification of intermediate image classification bring additional information and single convert task,
So that the transfer capability of multi-cycle production confrontation network increases than generating confrontation network with the circulation of script, and then can
Complete more complicated and difficult convert task.
Detailed description of the invention
Fig. 1 is the schematic diagram of the circulation production confrontation network image conversion method of the prior art;
Fig. 2 is the process of image conversion method of one of the embodiment of the present invention based on multi-cycle production confrontation network
Figure;
Fig. 3 is the generation of image conversion method of one of the embodiment of the present invention based on multi-cycle production confrontation network
The structure chart of formula confrontation network;
Fig. 4 is the principle of image conversion method of one of the embodiment of the present invention based on multi-cycle production confrontation network
Figure;
Fig. 5 is that one of embodiment of the present invention is applied to based on the image conversion method of multi-cycle production confrontation network
The schematic diagram of CT image conversion;
Fig. 6 is the structure of image conversion apparatus of one of the embodiment of the present invention based on multi-cycle production confrontation network
Figure.
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 every other
Embodiment shall fall within the protection scope of the present invention.
Please referring to Fig. 2, it illustrates one kind that an illustrative embodiment of the invention provides to be fought based on multi-cycle production
The image conversion method of network, step include:
S101, the image for needing to convert is obtained, and described image is divided into different the first image category and the second figure
As classification;
S102, at least one is gone out according to the image of the first image classification and the image configuration of second image category
Intermediate image classification;
Specifically, when being transformed into the task of the second image category from the first image category is image denoising, there are many not
Middle category can be synthesized with method.For example, generating centre to the noise of the image addition varying strength of the first image category
Image category simultaneously combines the noise pattern extracted from the image of the second image category.
S103, include according to the building of the first image classification, second image category and the intermediate image classification
The production of several generators and several arbiters fights network;Wherein, the first image classification is corresponding generates the first differentiation
Device, second image category is corresponding to generate the second arbiter and the corresponding generation third arbiter of the intermediate image classification;
S104, network progress image conversion is fought by the production.
In the present embodiment, when using generation confrontation network is recycled, the image for needing to convert includes but is not limited to camera
The CT image that the photo of shooting, electronic painting image, medical treatment use can be a classification by increasing intermediate image classification
Or middle category of multiple classifications as the first image category and the second image category, keeping original conversion task constant
In the case of, more difficult image conversion is decomposed into multiple relatively simple steps and is carried out, intermediate image classification bring is additional
The simplification of information and single convert task, so that multi-cycle generates the transfer capability of confrontation network than generating pair with the circulation of script
Anti- network greatly improves, and then can complete more complicated and difficult convert task, due to the figure of the first image classification
As the image of, second image category, and the image of intermediate image classification generated has also been included into consideration, reaches increase
Information content is to improve the purpose of model performance.
Please referring to Fig. 3, it illustrates one kind that an illustrative embodiment of the invention provides to be fought based on multi-cycle production
The image conversion method of network, production confrontation network include the first generator, the second generator, third generator and the
Four generators;Wherein, first generator is used to the image of the first image classification being converted into the intermediate image class
Other image;Second generator is used to for the image of the intermediate image classification being converted into the figure of second image category
Picture;The third generator is used to for the image of second image category being converted into the image of the intermediate image classification;Institute
The 4th generator is stated for the image of the intermediate image classification to be converted into the image of the first image classification.
Wherein, X, Y, Z are respectively the first image category, the second image category and third image category;DX、DY、DZRespectively
First arbiter, the second arbiter and third arbiter, three arbiters and three image categories correspond;GX->Z、GZ->Y、
GY->Z、GZ->XRespectively the first generator, the second generator, third generator and the 4th generator are deposited between every two adjacent category
The image conversion of opposite direction is carried out respectively in two generators.
Fig. 4 is referred to, the image of the first image classification is converted into the middle category by first generator
Image;The image of the middle category is converted into the image of second image category by second generator;It is described
The image of second image category is converted into the image of the middle category by the third generator;The figure of the middle category
Image as being converted into the first image classification by the 4th generator;
Wherein, between image category 1 and image category 2 be added an image category 3, due to keep image category 1 with
Convert task between image category 2 is constant, and the sample of image category 3 is by the sample of image category 1 and image category 2 institute structure
It creates.Thus the task of script is split into the part of two couplings.Dot 20 indicates the image for belonging to image category 2;
Dot 10 respectively indicates the original image for belonging to image category 1, merely through the figure for converting back image category 1 after image category 3
Picture and the image that image category 1 is converted back after image category 3 and image category 2;Dot 30 is respectively indicated by image category 1
Image be transformed into the image of image category 3, the original image for belonging to image category 3, the image for belonging to image category 3 only passes through
The image of image category 3 is converted back after crossing image category 2 and belongs to the image of image category 1 by image category 3 and image class
The image of image category 3 is converted back after other 2.This figure has only drawn sample conversion cycle from left to right, but in practical applications
Sample conversion cycle from right to left is also required to calculate.Arrow still indicates conversion direction, although may have multiple arrows, not
Only one generator of direction between generic, that is, such arrow all indicate a same generator;Each
Image category has a corresponding arbiter, and corresponding arbiter is not drawn.
In the present embodiment, the image of the first image classification needs just be transformed by intermediate image classification described
Second image category is constrained by increasing the long range between non-adjacent classification, can help the unified direction converted, so that
The direction that each classification is definitely converted, thus the reliability of the image conversion process improved, while between non-adjacent classification
Constraint can provide more information for arbiter and the generator of each classification.
Preferably, loss function training method is based on to production confrontation network to be trained.
Preferably, described that production confrontation network is trained based on loss function training method, specific steps
Are as follows:
The included generation confrontation loss function of network is fought according to production and that extracts in production confrontation network follows
Ring consistency loss function constructs total losses function;
The parameter for being constructed by modification generator keeps the total losses function minimum while parameter by modification arbiter
Make the maximum generic function of total losses function;
The generic function is solved.
Wherein, picture of the confrontation loss function for training arbiter to identify that all picture transduction pathway generate is generated simultaneously
The picture that all generators of training generate it can not other arbiter identification, the included generation confrontation of the production confrontation network
Loss function specifically:
In formula (1), DYFor the arbiter of image category Y;Y any image classification;Ii→ Y indicates a picture transduction pathway;
GYFor the compound function of generator;DY(y) indicate that the picture y of arbiter judgement is the probability of true picture for a scalar;Arbiter D is indicated for a scalarYThe picture that the picture of judgement obtains after the conversion of the generator on Y is true
The probability of picture;IiFor path IiClassification where the starting point of → Y;Pdata(Y) distribution of the true picture on classification Y is indicated;WithArbiter is respectively indicated to sentence true picture and generation picture
Disconnected accuracy.
Wherein, the circulation consistency loss function extracted in production confrontation network specifically:
In formula (2), x indicates authentic specimen,It indicates that x is transformed into after other classifications by different generators to turn again
Gain the result after the classification of script.
It is understood that circulation consistency, which is one, belongs to the image of image category one by image category one to figure
It, should can be again by from image category two to the life of image category one after being transformed into image category two as the generator of classification two
Reconvert of growing up to be a useful person returns image category one, and the result for converting back image category one should be as identical as possible as original image.It follows
The loss of ring consistency is exactly to be used to measure the original image for belonging to image category one and the figure for finally converting back image category one
The difference of picture.
Wherein, the total loss of the total losses function representation pairs of damage-retardation of making a living is become estranged the circulation consistency being proportionally added into
Loss, the total losses function specifically:
In formula (3), λ is adjustable hyper parameter.
Wherein, the generic function specifically:
It is preferably, described that the generic function is solved, specifically:
The generic function is solved using the stochastic gradient descent algorithm of standard.Wherein, the generic function is solved
Method include being solved using the stochastic gradient descent algorithm of standard or its deformation.
In the present embodiment, by fighting the repetition training of network to production, be conducive to the serious forgiveness for improving network;Together
When by setting mapped image data and the function of image conversion process, and calculated automatically using computer to reduce artificial calculate
Workload, and ensure the accuracy rate of calculated result.
Please referring to Fig. 5, it illustrates one kind that an illustrative embodiment of the invention provides to be fought based on multi-cycle production
The image conversion method of network, it is described to need the image converted into CT image.
As shown, three boxes respectively indicate image category X, Z, Y from left to right, it is respectively least clearly in this example
CT image, secondary clearly CT image and clearest CT image.Secondary clearly CT image is by least clearly CT image and most
Clearly CT image is combined into.A generator in a direction, is indicated by trapezoid block, is marked by G between different classes of.
Image in dotted line frame indicates image original in each classification, and the expression of other images is generated via other classification images to such
Other image.The corresponding arbiter of each classification is not drawn.Four arrows 201 constitute respectively from Z and Y two
Circulation between a ZY.Four arrows 101 constitute respectively from the circulation between two ZX of Z and X.Arrow 301 is constituted
The global loops of two ZXY of X- > Z- > Y- > Z- > X and Y- > Z- > X- > Z- > Y.Due to the flow path switch of the expression of arrow 301 and its
His arrow has repetition, and repeating part does not need to compute repeatedly, and repeating part is still indicated with its original number.
In the practical application based on CT image denoising, with non-adjacent long range constraint model and direct splicing two
Circulation generates the model of confrontation network, and by taking three samples as an example, the difference of performance is as shown in table 1.
Table 1 is the table of comparisons of the different samples of the practical application based on CT image denoising
Wherein, mean value is closer to the mean value of original image, while standard deviation is smaller, and performance is better.
Please referring to Fig. 6, it illustrates one kind that an illustrative embodiment of the invention provides to be fought based on multi-cycle production
The image conversion apparatus of network, comprising:
Described image for obtaining the image for needing to convert, and is divided into the first different figures by image collection module 210
As classification and the second image category;
Intermediate image category construction module 220, for the image and second image according to the first image classification
The image configuration of classification goes out at least one intermediate image classification;
Network generation module 230, for according to the first image classification, second image category and the middle graph
As classification building production fights network;Wherein, corresponding first arbiter of the first image classification, second image category
Corresponding second arbiter and the intermediate image classification correspond to third arbiter;
Image conversion module 240 carries out image conversion for fighting network by the production.
Preferably, the image conversion apparatus based on multi-cycle production confrontation network, further includes network training mould
Block is trained for being based on loss function training method to production confrontation network.
Preferably, the image conversion apparatus based on multi-cycle production confrontation network, further includes:
Total losses function constructs module, for fighting the included generation confrontation loss function of network according to production and in life
The circulation consistency loss function building total losses function extracted in accepted way of doing sth confrontation network;
Generic function constructs module, and the parameter for being constructed by modification generator keeps the total losses function minimum while logical
The parameter for crossing modification arbiter makes the maximum generic function of total losses function;
Function solves module, for solving to the generic function.
The above is a preferred embodiment of the present invention, it is noted that for those skilled in the art
For, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also considered as
Protection scope of the present invention.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage medium
In, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic
Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random Access
Memory, RAM) etc..
Claims (9)
1. a kind of image conversion method based on multi-cycle production confrontation network, which is characterized in that step includes:
The image for needing to convert is obtained, and described image is divided into different the first image category and the second image category;
Go out at least one intermediate image according to the image configuration of the image of the first image classification and second image category
Classification;
It include several generators according to the building of the first image classification, second image category and the intermediate image classification
Network is fought with the production of several arbiters;Wherein, the first arbiter of the corresponding generation of the first image classification, described second
Image category is corresponding to generate the second arbiter and the corresponding generation third arbiter of the intermediate image classification;
Network, which is fought, by the production carries out image conversion.
2. the image conversion method as described in claim 1 based on multi-cycle production confrontation network, which is characterized in that described
If according to the first image classification, second image category and the intermediate image classification building include several generators and
The production of dry arbiter fights network, comprising:
Generate the first generator, the second generator, third generator and the 4th generator;Wherein, first generator is used for
The image of the first image classification is converted into the image of the intermediate image classification;Second generator is used for will be described
The image of intermediate image classification is converted into the image of second image category;The third generator is used for second figure
As the image of classification is converted into the image of the intermediate image classification;4th generator is used for the intermediate image classification
Image be converted into the image of the first image classification.
3. the image conversion method as claimed in claim 1 or 2 based on multi-cycle production confrontation network, which is characterized in that
Further include:
Loss function training method is based on to production confrontation network to be trained.
4. the image conversion method as claimed in claim 3 based on multi-cycle production confrontation network, which is characterized in that described
It is based on loss function training method to production confrontation network to be trained, specific steps are as follows:
The circulation one fighting the included generation confrontation loss function of network according to production and being extracted in production confrontation network
Cause property loss function constructs total losses function;
The parameter for being constructed by modification generator keeps the total losses function minimum while making institute by the parameter of modification arbiter
State the maximum generic function of total losses function;
The generic function is solved.
5. the image conversion method as claimed in claim 4 based on multi-cycle production confrontation network, which is characterized in that described
The generic function is solved, specifically:
The generic function is solved using the stochastic gradient descent algorithm of standard.
6. the image conversion method as described in claim 1 based on multi-cycle production confrontation network, which is characterized in that described
The image for needing to convert is CT image.
7. a kind of image conversion apparatus based on multi-cycle production confrontation network characterized by comprising
Described image for obtaining the image for needing to convert, and is divided into the first different image categories by image collection module
With the second image category;
Intermediate image category construction module, for according to the image of the first image classification and the figure of second image category
As constructing at least one intermediate image classification;
Network generation module, for according to the first image classification, second image category and the intermediate image classification
It constructs production and fights network;Wherein, corresponding first arbiter of the first image classification, second image category corresponding the
Two arbiters and the intermediate image classification correspond to third arbiter;
Image conversion module carries out image conversion for fighting network by the production.
8. the image conversion apparatus as claimed in claim 7 based on multi-cycle production confrontation network, which is characterized in that also wrap
It includes:
Network training module is trained for being based on loss function training method to production confrontation network.
9. the image conversion apparatus as claimed in claim 8 based on multi-cycle production confrontation network, which is characterized in that also wrap
It includes:
Total losses function constructs module, for fighting the included generation confrontation loss function of network according to production and in production
The circulation consistency loss function building total losses function extracted in confrontation network;
Generic function constructs module, and the parameter for being constructed by modification generator keeps the total losses function minimum while by repairing
The parameter for changing arbiter makes the maximum generic function of total losses function;
Function solves module, for solving to the generic function.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910607839.4A CN110428476A (en) | 2019-07-05 | 2019-07-05 | A kind of image conversion method and device based on multi-cycle production confrontation network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910607839.4A CN110428476A (en) | 2019-07-05 | 2019-07-05 | A kind of image conversion method and device based on multi-cycle production confrontation network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110428476A true CN110428476A (en) | 2019-11-08 |
Family
ID=68410293
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910607839.4A Pending CN110428476A (en) | 2019-07-05 | 2019-07-05 | A kind of image conversion method and device based on multi-cycle production confrontation network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110428476A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111292270A (en) * | 2020-02-18 | 2020-06-16 | 广州柏视医疗科技有限公司 | Three-dimensional image blood vessel enhancement method based on deep learning network |
CN112508175A (en) * | 2020-12-10 | 2021-03-16 | 深圳先进技术研究院 | Multi-task learning type generation type confrontation network generation method and system for low-dose PET reconstruction |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107590518A (en) * | 2017-08-14 | 2018-01-16 | 华南理工大学 | A kind of confrontation network training method of multiple features study |
CN107610195A (en) * | 2017-07-28 | 2018-01-19 | 上海联影医疗科技有限公司 | The system and method for image conversion |
CN108038818A (en) * | 2017-12-06 | 2018-05-15 | 电子科技大学 | A kind of generation confrontation type network image style transfer method based on Multiple Cycle uniformity |
CN108416752A (en) * | 2018-03-12 | 2018-08-17 | 中山大学 | A method of image is carried out based on production confrontation network and removes motion blur |
CN108615073A (en) * | 2018-04-28 | 2018-10-02 | 北京京东金融科技控股有限公司 | Image processing method and device, computer readable storage medium, electronic equipment |
CN109325989A (en) * | 2018-08-27 | 2019-02-12 | 平安科技(深圳)有限公司 | License plate image generation method, device, equipment and medium |
CN109635774A (en) * | 2018-12-21 | 2019-04-16 | 中山大学 | A kind of human face synthesizing method based on generation confrontation network |
CN109872278A (en) * | 2018-12-18 | 2019-06-11 | 深圳先进技术研究院 | Image cloud layer removing method based on U-shape network and generation confrontation network |
CN109886970A (en) * | 2019-01-18 | 2019-06-14 | 南京航空航天大学 | The detection dividing method and computer storage medium of target object in terahertz image |
CN109934117A (en) * | 2019-02-18 | 2019-06-25 | 北京联合大学 | Based on the pedestrian's weight recognition detection method for generating confrontation network |
-
2019
- 2019-07-05 CN CN201910607839.4A patent/CN110428476A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107610195A (en) * | 2017-07-28 | 2018-01-19 | 上海联影医疗科技有限公司 | The system and method for image conversion |
CN107590518A (en) * | 2017-08-14 | 2018-01-16 | 华南理工大学 | A kind of confrontation network training method of multiple features study |
CN108038818A (en) * | 2017-12-06 | 2018-05-15 | 电子科技大学 | A kind of generation confrontation type network image style transfer method based on Multiple Cycle uniformity |
CN108416752A (en) * | 2018-03-12 | 2018-08-17 | 中山大学 | A method of image is carried out based on production confrontation network and removes motion blur |
CN108615073A (en) * | 2018-04-28 | 2018-10-02 | 北京京东金融科技控股有限公司 | Image processing method and device, computer readable storage medium, electronic equipment |
CN109325989A (en) * | 2018-08-27 | 2019-02-12 | 平安科技(深圳)有限公司 | License plate image generation method, device, equipment and medium |
CN109872278A (en) * | 2018-12-18 | 2019-06-11 | 深圳先进技术研究院 | Image cloud layer removing method based on U-shape network and generation confrontation network |
CN109635774A (en) * | 2018-12-21 | 2019-04-16 | 中山大学 | A kind of human face synthesizing method based on generation confrontation network |
CN109886970A (en) * | 2019-01-18 | 2019-06-14 | 南京航空航天大学 | The detection dividing method and computer storage medium of target object in terahertz image |
CN109934117A (en) * | 2019-02-18 | 2019-06-25 | 北京联合大学 | Based on the pedestrian's weight recognition detection method for generating confrontation network |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111292270A (en) * | 2020-02-18 | 2020-06-16 | 广州柏视医疗科技有限公司 | Three-dimensional image blood vessel enhancement method based on deep learning network |
CN112508175A (en) * | 2020-12-10 | 2021-03-16 | 深圳先进技术研究院 | Multi-task learning type generation type confrontation network generation method and system for low-dose PET reconstruction |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109685116A (en) | Description information of image generation method and device and electronic device | |
Szabó et al. | Structural transitions in scale-free networks | |
Gu et al. | GAN-based model for residential load generation considering typical consumption patterns | |
CN107016406A (en) | The pest and disease damage image generating method of network is resisted based on production | |
CN106709511A (en) | Urban rail transit panoramic monitoring video fault detection method based on depth learning | |
CN108389192A (en) | Stereo-picture Comfort Evaluation method based on convolutional neural networks | |
CN110428476A (en) | A kind of image conversion method and device based on multi-cycle production confrontation network | |
CN110084193A (en) | Data processing method, equipment and medium for Facial image synthesis | |
CN113505855B (en) | Training method for challenge model | |
CN116342516B (en) | Model integration-based method and system for assessing bone age of X-ray images of hand bones of children | |
CN109978074A (en) | Image aesthetic feeling and emotion joint classification method and system based on depth multi-task learning | |
CN110533178A (en) | A kind of neural network model training method, apparatus and system | |
CN110210540A (en) | Across social media method for identifying ID and system based on attention mechanism | |
CN114299006A (en) | Self-adaptive multi-channel graph convolution network for joint graph comparison learning | |
CN113012811A (en) | Traditional Chinese medicine syndrome diagnosis and health evaluation method combining deep convolutional network and graph neural network | |
CN113988310A (en) | Deep learning model selection method and device, computer equipment and medium | |
CN114677263A (en) | Cross-mode conversion method and device for CT image and MRI image | |
CN113239825A (en) | High-precision tobacco beetle detection method in complex scene | |
CN105590341A (en) | Corn colony three-dimensional reconstruction method and apparatus | |
CN110175588A (en) | A kind of few sample face expression recognition method and system based on meta learning | |
Dong et al. | PDDD-PreTrain: a series of commonly used pre-trained models support image-based plant disease diagnosis | |
Qian et al. | COM: Contrastive Masked-attention model for incomplete multimodal learning | |
CN112101528A (en) | Terminal contribution measurement method based on back propagation | |
CN107908696A (en) | A kind of parallel efficiently multidimensional space data clustering algorithm GRIDEN based on grid and density | |
CN205665729U (en) | Cell convolutional neural network intelligent vision pays accelerator |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191108 |