CN108537776A - A kind of image Style Transfer model generating method and mobile terminal - Google Patents
A kind of image Style Transfer model generating method and mobile terminal Download PDFInfo
- Publication number
- CN108537776A CN108537776A CN201810202121.2A CN201810202121A CN108537776A CN 108537776 A CN108537776 A CN 108537776A CN 201810202121 A CN201810202121 A CN 201810202121A CN 108537776 A CN108537776 A CN 108537776A
- Authority
- CN
- China
- Prior art keywords
- layer
- style
- image
- neural network
- style transfer
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- 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
- G06T2207/30168—Image quality inspection
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
An embodiment of the present invention provides a kind of image Style Transfer model generating method and mobile terminals, wherein the method includes：Build style training data pair；Building structure sequence is followed successively by the neural network of input layer, prefilter layer, at least one residual error feature construction layer, back layer and output layer；The mean square deviation of mean square deviation loss and depth perception feature based on training artwork and style figure is lost, and loss function is built；According to the style training data pair and the loss function, the neural network is trained, generates image Style Transfer model.The image Style Transfer model that the image Style Transfer model generating method provided through the embodiment of the present invention is generated, the high fidelity visual after Style Transfer can will be quickly obtained in pending image input picture Style Transfer model, without executing cumbersome iterative operation, it is light to take short processor live load, and gained Style Transfer image space distortion can be avoided.
Description
Technical field
The present embodiments relate to technical field of image processing more particularly to a kind of image Style Transfer model generating methods
And mobile terminal.
Background technology
The research of image Style Transfer is long-standing, will be transformed into another image with reference to the style of style image.It is existing
Have in technology and mainly image style is migrated by the Style Transfer algorithm based on neural network, although the algorithm can be learned
The overall color transformation of image and structural element are practised, but it is poor for the learning ability of pixel details, cause style to be moved
The Style Transfer image generated after shifting is fuzzy, and the Style Transfer image generated can have straight-line bending, texture distortion
The problem of equal spatial warpings.
To solve the problems, such as that above-mentioned image space distortion, those skilled in the art provide a kind of removal image space distortion
Method, this method only acts on map function on color space.Although this kind of method can inhibit image space to distort
Problem, but this kind of method needs to execute cumbersome interative computation, and time-consuming and processor live load weight.
Invention content
The embodiment of the present invention provides a kind of image Style Transfer model generating method, to solve carrying out figure in the prior art
Time-consuming when as Style Transfer and the problem of processor live load weight.
In order to solve the above-mentioned technical problem, the invention is realized in this way：
In a first aspect, an embodiment of the present invention provides a kind of image Style Transfer model generating methods, wherein the method
Including：Build style training data pair, wherein each data are to including a trained artwork and a style figure；Building structure
Sequence is followed successively by the neural network of input layer, prefilter layer, at least two residual error feature construction layers, back layer and output layer；Base
It is lost in training artwork and the mean square deviation of the loss of the mean square deviation of style figure and depth perception feature, builds loss function；According to institute
Style training data pair and the structure loss function are stated, the neural network is trained, generates image Style Transfer mould
Type.
Second aspect, an embodiment of the present invention provides a kind of mobile terminals, wherein the mobile terminal includes：Build mould
Block, for building style training data pair, wherein each data are to including a trained artwork and a style figure；Build mould
Block is followed successively by input layer, prefilter layer, at least two residual error feature construction layers, back layer and output for building structure sequence
The neural network of layer；Loss function builds module, for based on the mean square deviation loss and depth perception for training artwork and style figure
The mean square deviation of feature is lost, and loss function is built；Training module, for being damaged according to the style training data pair and the structure
Function is lost, the neural network is trained, generates image Style Transfer model.
The third aspect an embodiment of the present invention provides a kind of mobile terminal, including processor, memory and is stored in described
It is real when the computer program is executed by the processor on memory and the computer program that can run on the processor
The step of showing any image Style Transfer model generating method described in the embodiment of the present invention.
Fourth aspect, an embodiment of the present invention provides a kind of computer readable storage medium, the computer-readable storage
Computer program is stored on medium, and any described in the embodiment of the present invention is realized when the computer program is executed by processor
The step of kind image Style Transfer model generating method.
In embodiments of the present invention, by the way that neural network is built into input layer, prefilter layer, at least one residual error feature structure
The structure of build-up layers, back layer and output layer, the mean square deviation loss based on training artwork figure and style figure and depth perception feature
Mean square deviation loss structure loss function, will training artwork and style figure constitute style training data pair, the style based on structure
Training data is trained neural network, loss function, generates image Style Transfer model, and the image style generated is moved
Shifting formwork type to pending image when carrying out Style Transfer processing, it is only necessary to will be in pending image input picture Style Transfer model
The high fidelity visual after Style Transfer can be quickly obtained, without executing cumbersome iterative operation, it is negative to take short processor work
Lotus is light, and can avoid gained Style Transfer image space distortion.
Above description is only the general introduction of technical solution of the present invention, in order to better understand the technical means of the present invention,
And can be implemented in accordance with the contents of the specification, and in order to allow above and other objects of the present invention, feature and advantage can
It is clearer and more comprehensible, below the special specific implementation mode for lifting the present invention.
Description of the drawings
Fig. 1 is one of the flow chart of the image Style Transfer model generating method provided according to embodiments of the present invention；
Fig. 2 is the two of the flow chart of the image Style Transfer model generating method provided according to embodiments of the present invention；
Fig. 3 is one of structure diagram of mobile terminal of the embodiment of the present invention；
Fig. 4 is the two of the structure diagram of the mobile terminal of the embodiment of the present invention；
Fig. 5 is the structure diagram of the training module of the mobile terminal of the embodiment of the present invention；
Fig. 6 is the hardware architecture diagram of the mobile terminal provided according to embodiments of the present invention.
Specific implementation mode
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 describes, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, the every other implementation that those of ordinary skill in the art are obtained without creative efforts
Example, shall fall within the protection scope of the present invention.
Referring to Fig.1, one of the flow chart of image Style Transfer model generating method provided in an embodiment of the present invention is shown.
The image Style Transfer model generating method of the embodiment of the present invention includes the following steps：
Step 101：Build style training data pair.
Wherein, each data are to including a trained artwork and a style figure.
The depth photography style conversion method that Cornell University Fujun doctors Luan proposition specifically may be used carrys out structure
Training data is built, to a certain true style figure, obtains the image pair of preset quantity using this method, each image is to being then one
A data pair, including a trained artwork and a style figure.Using the data pair for including training artwork and style figure, to being taken
The neural network built is trained, the obtained image Style Transfer model of training can be sufficiently reserved style transfer effect and and
The detail accuracy of artwork always.
Step 102：Building structure sequence is followed successively by input layer, prefilter layer, at least one residual error feature construction layer, back layer
And the neural network of output layer.
Wherein, prefilter layer, back layer and each residual error feature construction layer are made of multilayer convolutional layer, the embodiment of the present invention
In include for each layer the specific number of plies of convolutional layer be not particularly limited.
Such as：It includes five layers of convolutional layer that prefilter layer, which is arranged,；Back layer includes five layers of convolutional layer；Each residual error feature construction layer is equal
Including three-layer coil lamination.
By input layer, the first layer packet of last layer of prefilter layer, last layer of each residual error feature construction layer, back layer
The characteristic pattern number of plies contained is identical, picture size can be kept to be remained unchanged in neural network structure, avoid image fault.
Step 103：The mean square deviation of mean square deviation loss and depth perception feature based on training artwork and style figure is lost, structure
Build loss function.
It is based only upon the mean square deviation loss structure loss function of trained artwork and style figure, constructed loss in the prior art
For the image Style Transfer model that function is trained when handling image, the Style Transfer image generated is fuzzy.The present invention is real
It applies in example, the mean square deviation of mean square deviation loss and depth perception feature based on training artwork and style figure is lost, structure loss letter
Number trains the image Style Transfer model generated when carrying out Style Transfer to image, energy to train built neural network
Enough promote the fineness of generated Style Transfer image.
Step 104：According to style training data pair and constructed loss function, neural network is trained, is generated
Image Style Transfer model.
For the concrete mode being trained to neural network based on training data, loss function, with reference to existing related skill
Art is not particularly limited this in the embodiment of the present invention.Such as：Each style training data creates input respectively
In neural network and advance trained image recognition model, loss function is adjusted based on parameters obtained, based on the damage after adjustment
It loses function and adjusts neural network parameter, until after neural network convergence reaches preset standard, complete the training to neural network
Generate image Style Transfer model.
Image Style Transfer model generating method provided in an embodiment of the present invention, by the way that neural network is built into input
The structure of layer, prefilter layer, at least one residual error feature construction layer, back layer and output layer, based on training artwork and style figure
Mean square deviation loss and depth perception feature mean square deviation loss structure loss function, will training artwork and style figure constitute style
Training data pair, the style training data based on structure are trained neural network, loss function, generate image style and move
Shifting formwork type, the image Style Transfer model generated to pending image when carrying out Style Transfer processing, it is only necessary to will be pending
The high fidelity visual after Style Transfer can be quickly obtained in image input picture Style Transfer model, without executing cumbersome change
Generation operation, time-consuming short processor live load is light, and can avoid gained Style Transfer image space distortion.
With reference to Fig. 2, the two of the flow chart of image Style Transfer model generating method provided in an embodiment of the present invention are shown.
Image Style Transfer model generating method in the embodiment of the present invention specifically includes following steps：
Step 201：Build style training data pair.
Wherein, each data are to including a trained artwork and a style figure.
In structure style training data clock synchronization, can be achieved by the steps of：
First, the method based on the full convolutional network of structure feedforward proposed based on doctor Johnson is generated as style figure
Method.
Secondly, select the image data collection of Microsoft as training artwork collection, the picture scene which includes is rich
Richness, content various kinds and image are of moderate size.Using the method pair for the full convolutional network that feedovered based on structure that doctor Johnson proposes
Each trained artwork is handled, and the corresponding style figure of each trained artwork is obtained, and the style figure that this kind of method is trained can be kept
The gradient information consistent with training artwork can ensure the validity of the style training data pair generated.
Step 202：Building structure sequence is followed successively by input layer, prefilter layer, at least one residual error feature construction layer, back layer
And the neural network of output layer.
Input layer is the training artwork for waiting for style processing, and size is height x width x channel, wherein
Width, height are respectively the width and height of image, and channel is the Color Channel number of image, usually RGB and red, green, blue
Three Color Channels.Output layer is the style figure after stylization, and size consistent with input layer is height x width x
channel。
Prefilter layer, back layer and intermediate residual error feature construction layer, by convolutional layer, the example normalizing being at least repeated once
Change layer and nonlinear activation layer composition, example normalizes layer has better effect, nonlinear activation for image stylization
ReLU i.e. Rectified Linear Units activation primitives may be used in layer.
Specifically, in the convolutional neural networks in the embodiment of the present invention, prefilter layer can be by the convolutional layer structure of 5 layers of repetition
At adjacent convolution interlayer is provided with example normalization layer and nonlinear activation layer.Residual error feature construction layer is 5 layers, and back layer is by 5
The convolutional layer that layer repeats is constituted, and adjacent convolution interlayer is provided with example normalization layer and nonlinear activation layer.Wherein, Group tables
Show that the quantity of grouping convolution, Dilation indicate the size of empty convolution.
Specifically, 5 layers of design parameter of prefilter layer are：
It is 3x3, Group=1, Dilation=that the convolutional layer of preposition first layer, which uses 24 characteristic patterns, convolution kernel size,
1；
It is 3x3, Group=24, Dilation=that the convolutional layer of the preposition second layer, which uses 24 characteristic patterns, convolution kernel size,
2；
It is 1x1, Group=1, Dilation=that the convolutional layer of preposition third layer, which uses 48 characteristic patterns, convolution kernel size,
1；
It is 3x3, Group=48, Dilation=that preposition 4th layer of convolutional layer, which uses 48 characteristic patterns, convolution kernel size,
4；
It is 1x1, Group=1, Dilation=that the convolutional layer of preposition layer 5, which uses 96 characteristic patterns, convolution kernel size,
1。
Specifically, 5 layers of design parameter of back layer are：
It is 3x3, Group=96, Dilation=that the convolutional layer of postposition first layer, which uses 96 characteristic patterns, convolution kernel size,
4；
It is 1x1, Group=1, Dilation=that the convolutional layer of the postposition second layer, which uses 48 characteristic patterns, convolution kernel size,
1；
It is 3x3, Group=48, Dilation=that the convolutional layer of postposition third layer, which uses 48 characteristic patterns, convolution kernel size,
2；
It is 1x1, Group=1, Dilation=that the convolutional layer that the 4th layer of postposition, which uses 24 characteristic patterns, convolution kernel size,
1；
It is 3x3, Group=1, Dilation=1 that the convolutional layer of preposition layer 5, which uses 3 characteristic patterns, convolution kernel size,.
Specifically, residual error feature construction layer number is that 5,5 residual error feature construction layers are identical.Every layer of residual error feature construction layer
Including 3 layers of convolutional layer, example normalization layer and nonlinear activation layer, every layer of residual error feature structure are provided between adjacent convolutional layer
The convolutional layer design parameter that build-up layers are included is as follows：
It is 1x1, Group=1, Dilation=1 that first layer convolutional layer, which uses 48 characteristic patterns, convolution kernel size,；
It is 3x3, Group=48, Dilation=4 that second layer convolutional layer, which uses 48 characteristic patterns, convolution kernel size,；
It is 1x1, Group=1, Dilation=1 that third layer convolutional layer, which uses 96 characteristic patterns, convolution kernel size,.
The neural network of this kind of structure can gradually be put while not zoomed image Assurance of Size generates image detail
The pore size of big convolution experiences the visual field to amplify, to ensure the transformed picture quality of style.
Preferably, carrying out depth to prefilter layer, back layer and each residual error feature construction layer in neural network respectively can
Convolution sum point process of convolution is detached, parameter adjustment is carried out, the image Style Transfer generated by neural metwork training can be improved
The performance of model.
Step 203：The mean square deviation of mean square deviation loss and depth perception feature based on training artwork and style figure is lost, structure
Build loss function.
Constructed loss function is as follows：
Wherein, θ is all parameter sets of convolutional neural networks, ysFor the style figure of style training data centering, y is wind
The result figure that the training artwork of lattice training data centering obtains after Processing with Neural Network, CHW distinguish width, the height of correspondence image
And Color Channel number, φjFor the characteristic pattern of the jth layer of deep layer Perception Features network, CjHjWjIt is special that deep layer perception is corresponded to respectively
Width, height and the feature port number of the characteristic pattern of network jth layer are levied, λ is the weight coefficient of two losses, and λ is hyper parameter numerical value
It could be provided as 1.
Step 204：According to style training data pair and constructed loss function, neural network is trained, is generated
Image Style Transfer model.
A kind of mode being preferably trained to neural network is as follows：
First, one by one by each style training data to being input in neural network；
Secondly, for each style training data pair of input, by neural network to the instruction of style training data centering
Practice artwork to be handled, obtains result figure；
Again, by advance trained deep layer Perception Features network, to the wind of result figure and style training data centering
Trrellis diagram carries out Perception Features extraction, obtains Perception Features figure；
Wherein it is possible to by the relu1_2 of deep layer Perception Features network, relu2_2, relu3_3, tetra- layers of features of relu4_3
Perceptually feature determines that Perception Features figure, gained Perception Features figure are the jth of deep layer Perception Features network by Perception Features
The characteristic pattern of layer.
Finally, Perception Features figure, result figure are substituted into loss function according to style figure, generates target loss function；Foundation
Target loss function adjusts the parameter of neural network, to promote neural network convergence.
Include multiple parameters in the loss function created in step 203, when being trained to neural network, whenever input
One data is to then needing to obtain target loss function to adaptively adjusting loss function for the data.
The mean square deviation of mean square deviation loss and depth perception feature based on training artwork and style figure in the embodiment of the present invention
Loss, structure loss function train built neural network, image Style Transfer model that training generates to image into
When row Style Transfer, the fineness of generated Style Transfer image can be promoted.
When the neural network to structure is trained, the training artwork of input x, that is, data centering passes through full convolutional network
Obtain result figure y and the style figure y of data centeringsThe loss of content mean square deviation is calculated separately together and depth perception feature is equal
Variance is lost, and updates full convolutional network, training process mid-deep strata Perception Features to two losses with backpropagation after super ginseng fusion
The weight parameter of network does not update, and only is used for participating in the extraction of perception characteristic pattern.After extracting obtained Perception Features figure, according to sense
Know that characteristic pattern adjusts loss function, neural network is trained.Specifically, stochastic gradient descent method counting loss may be used
Gradients of the function L (θ) for the parameter θ in convolutional neural networksThe gradientIt is used to constantly update convolution
Parameter in neural network, furthermore it is also possible to according to the gradientThe value of undated parameter θIts
In, η is learning rate, is used for the newer amplitudes of control parameter θ.
Step 205：Pending image is input to image Style Transfer model.
After being trained to neural network by step 204, by neural network converge to preset standard after, complete to god
Training through network generates image Style Transfer model.
After the generation of image Style Transfer model, you can carried out at Style Transfer to pending image by the model of generation
Reason.Specifically, after pending image being input to image Style Transfer model as input data, the output of the model is then wind
Image after lattice migration.
Step 206：Style Transfer is carried out to pending image by image Style Transfer model, exports pending image pair
The Style Transfer image answered.
In the embodiment of the present invention, it is only necessary to deployment structure image Style Transfer model in advance, by building the image completed
When Style Transfer model carries out Style Transfer processing to pending image, it is only necessary to execute primary feedforward propagation and can be obtained by place
Reason takes image that is short and being applicable to any style as a result, without executing cumbersome iterative operation.
Image Style Transfer model generating method provided in an embodiment of the present invention, by the way that neural network is built into input
The structure of layer, prefilter layer, at least one residual error feature construction layer, back layer and output layer, based on training artwork and style figure
Mean square deviation loss and depth perception feature mean square deviation loss structure loss function, will training artwork and style figure constitute style
Training data pair, the style training data based on structure are trained neural network, loss function, generate image style and move
Shifting formwork type, the image Style Transfer model generated to pending image when carrying out Style Transfer processing, it is only necessary to will be pending
The high fidelity visual after Style Transfer can be quickly obtained in image input picture Style Transfer model, without executing cumbersome change
Generation operation, time-consuming short processor live load is light, and can avoid gained Style Transfer image space distortion.In addition, of the invention
In embodiment by depth separate convolution method and point convolution method, respectively in neural network prefilter layer, back layer with
And each residual error feature construction layer carries out depth and separates convolution sum point process of convolution, carries out parameter adjustment, can improve and pass through god
The performance of the image Style Transfer model generated through network training.
With reference to Fig. 3, one of structure diagram of mobile terminal of the embodiment of the present invention is shown.Before mobile terminal 30 can be realized
The details of the image Style Transfer model generating method in embodiment is stated, and reaches identical effect.
The mobile terminal of the embodiment of the present invention includes：Module 301 is built, for building style training data pair, wherein every
A data are to including a trained artwork and a style figure；Module 302 is built, input is followed successively by for building structure sequence
The neural network of layer, prefilter layer, at least one residual error feature construction layer, back layer and output layer；Loss function builds module
303, the mean square deviation for mean square deviation loss and depth perception feature based on training artwork and style figure is lost, structure loss letter
Number；Training module 304, for according to the style training data pair and the structure loss function, to the neural network into
Row training, generates image Style Transfer model.
With reference to Fig. 4, the two of the structure diagram of the mobile terminal of the embodiment of the present invention are shown, it is mobile on the basis of Fig. 3
Terminal 30 further includes following module：
Optimization module 305, for being followed successively by input layer, prefilter layer, at least in the module 302 building structure sequence of building
After the neural network of two residual error feature construction layers, back layer and output layer, respectively to described in the neural network
Prefilter layer, the back layer and each residual error feature construction layer carry out depth and separate convolution sum point process of convolution, carry out parameter
Adjustment；
Input module 306 will wait locating after generating image Style Transfer model described in the training module 304
Reason image is input to described image Style Transfer model；The pending image is carried out by described image Style Transfer model
Style Transfer exports the corresponding Style Transfer image of the pending image.
Preferably, the prefilter layer includes five layers of convolutional layer；The back layer includes five layers of convolutional layer；Each residual error feature
Structure layer includes three-layer coil lamination.
Preferably, the structure diagram of the training module 304 is as shown in figure 5, training module 304 includes following submodule：
Input submodule 3041, for one by one by each style training data to being input in the neural network；
Submodule 3042 is called, for each style training data pair for input, the style is instructed by the neural network
The training artwork for practicing data centering is handled, and result figure is obtained；Submodule 3043 is analyzed, for passing through advance trained depth
Layer Perception Features network carries out Perception Features extraction to the style figure of the result figure and the data centering, it is special to obtain perception
Sign figure；Submodule 3044 is generated, for the Perception Features figure, the result figure to be substituted into the loss according to the style figure
Function generates target loss function；Submodule 3045 is adjusted, for according to target loss function, adjusting the neural network
Parameter, to promote the neural network convergence
Mobile terminal provided in an embodiment of the present invention can realize that mobile terminal is realized in the embodiment of the method for Fig. 1 to Fig. 2
Each process, to avoid repeating, which is not described herein again.
Mobile terminal provided in an embodiment of the present invention, by the way that neural network is built into input layer, prefilter layer, at least two
The structure of residual error feature construction layer, back layer and output layer, the mean square deviation loss based on training artwork and style figure and depth
The mean square deviation loss structure loss function of Perception Features, constitutes style training data pair by training artwork and style figure, is based on structure
The style training data built is trained neural network, loss function, generates image Style Transfer model, the figure generated
As Style Transfer model to pending image when carrying out Style Transfer processing, it is only necessary to move pending image input picture style
The high fidelity visual after Style Transfer can be quickly obtained in shifting formwork type, without executing cumbersome iterative operation, take weakness reason
Device live load is light, and can avoid gained Style Transfer image space distortion.
With reference to Fig. 6, a kind of hardware architecture diagram of mobile terminal of the embodiment of the present invention is shown.
A kind of hardware architecture diagram of Fig. 6 mobile terminals of each embodiment to realize the present invention, the mobile terminal 600
Including but not limited to：Radio frequency unit 601, audio output unit 603, input unit 604, sensor 605, is shown network module 602
Show the components such as unit 6406, user input unit 607, interface unit 608, memory 609, processor 610 and power supply 611.
It will be understood by those skilled in the art that mobile terminal structure shown in Fig. 6 does not constitute the restriction to mobile terminal, it is mobile whole
End may include either combining certain components or different components arrangement than illustrating more or fewer components.In the present invention
In embodiment, mobile terminal includes but not limited to mobile phone, tablet computer, laptop, palm PC, car-mounted terminal, can wear
Wear equipment and pedometer etc..
Processor 610, for building style training data pair, wherein each data are to including a trained artwork and one
A style figure；Building structure sequence is followed successively by input layer, prefilter layer, at least one residual error feature construction layer, back layer and defeated
Go out the neural network of layer；The mean square deviation of mean square deviation loss and depth perception feature based on training artwork and style figure is lost, structure
Build loss function；According to the style training data pair and the structure loss function, the neural network is trained, it is raw
At image Style Transfer model.
Mobile terminal provided in an embodiment of the present invention, by the way that neural network is built into input layer, prefilter layer, at least one
The structure of residual error feature construction layer, back layer and output layer, the mean square deviation loss based on training artwork and style figure and depth
The mean square deviation loss structure loss function of Perception Features, constitutes style training data pair by training artwork and style figure, is based on structure
The style training data built is trained neural network, loss function, generates image Style Transfer model, the figure generated
As Style Transfer model to pending image when carrying out Style Transfer processing, it is only necessary to move pending image input picture style
The high fidelity visual after Style Transfer can be quickly obtained in shifting formwork type, without executing cumbersome iterative operation, take weakness reason
Device live load is light, and can avoid gained Style Transfer image space distortion.
It should be understood that the embodiment of the present invention in, radio frequency unit 601 can be used for receiving and sending messages or communication process in, signal
Send and receive, specifically, by from base station downlink data receive after, to processor 610 handle；In addition, by uplink
Data are sent to base station.In general, radio frequency unit 601 includes but not limited to antenna, at least one amplifier, transceiver, coupling
Device, low-noise amplifier, duplexer etc..In addition, radio frequency unit 601 can also by radio communication system and network and other set
Standby communication.
Mobile terminal has provided wireless broadband internet to the user by network module 602 and has accessed, and such as user is helped to receive
Send e-mails, browse webpage and access streaming video etc..
It is that audio output unit 603 can receive radio frequency unit 601 or network module 602 or in memory 609
The audio data of storage is converted into audio signal and exports to be sound.Moreover, audio output unit 603 can also be provided and be moved
The relevant audio output of specific function that dynamic terminal 600 executes is (for example, call signal receives sound, message sink sound etc.
Deng).Audio output unit 603 includes loud speaker, buzzer and receiver etc..
Input unit 604 is for receiving audio or video signal.Input unit 604 may include graphics processor
(Graphics Processing Unit, GPU) 6041 and microphone 6042, graphics processor 6041 is in video acquisition mode
Or the image data of the static images or video obtained by image capture apparatus (such as camera) in image capture mode carries out
Reason.Treated, and picture frame may be displayed on display unit 606.Through graphics processor 6041, treated that picture frame can be deposited
Storage is sent in memory 609 (or other storage mediums) or via radio frequency unit 601 or network module 602.Mike
Wind 6042 can receive sound, and can be audio data by such acoustic processing.Treated audio data can be
The format output of mobile communication base station can be sent to via radio frequency unit 601 by being converted in the case of telephone calling model.
Mobile terminal 600 further includes at least one sensor 6605, such as optical sensor, motion sensor and other biographies
Sensor.Specifically, optical sensor includes ambient light sensor and proximity sensor, wherein ambient light sensor can be according to environment
The light and shade of light adjusts the brightness of display panel 6061, and proximity sensor can close when mobile terminal 600 is moved in one's ear
Display panel 6061 and/or backlight.As a kind of motion sensor, accelerometer sensor can detect in all directions (general
For three axis) size of acceleration, size and the direction of gravity are can detect that when static, can be used to identify mobile terminal posture (ratio
Such as horizontal/vertical screen switching, dependent game, magnetometer pose calibrating), Vibration identification correlation function (such as pedometer, tap)；It passes
Sensor 605 can also include fingerprint sensor, pressure sensor, iris sensor, molecule sensor, gyroscope, barometer, wet
Meter, thermometer, infrared sensor etc. are spent, details are not described herein.
Display unit 606 is for showing information input by user or being supplied to the information of user.Display unit 606 can wrap
Display panel 6061 is included, liquid crystal display (Liquid Crystal Display, LCD), Organic Light Emitting Diode may be used
Forms such as (Organic Light-Emitting Diode, OLED) configure display panel 6061.
User input unit 607 can be used for receiving the number or character information of input, and generate the use with mobile terminal
Family is arranged and the related key signals input of function control.Specifically, user input unit 607 include touch panel 6071 and
Other input equipments 6072.Touch panel 6071, also referred to as touch screen collect user on it or neighbouring touch operation
(for example user uses any suitable objects or attachment such as finger, stylus on touch panel 6071 or in touch panel 6071
Neighbouring operation).Touch panel 6071 may include both touch detecting apparatus and touch controller.Wherein, touch detection
Device detects the touch orientation of user, and detects the signal that touch operation is brought, and transmits a signal to touch controller；Touch control
Device processed receives touch information from touch detecting apparatus, and is converted into contact coordinate, then gives processor 610, receiving area
It manages the order that device 610 is sent and is executed.Furthermore, it is possible to more using resistance-type, condenser type, infrared ray and surface acoustic wave etc.
Type realizes touch panel 6071.In addition to touch panel 6071, user input unit 607 can also include other input equipments
6072.Specifically, other input equipments 6072 can include but is not limited to physical keyboard, function key (such as volume control button,
Switch key etc.), trace ball, mouse, operating lever, details are not described herein.
Further, touch panel 6071 can be covered on display panel 6061, when touch panel 6071 is detected at it
On or near touch operation after, send processor 610 to determine the type of touch event, be followed by subsequent processing device 610 according to touch
The type for touching event provides corresponding visual output on display panel 6061.Although in figure 6, touch panel 6071 and display
Panel 6061 is to realize the function that outputs and inputs of mobile terminal as two independent components, but in some embodiments
In, can be integrated by touch panel 6071 and display panel 6061 and realize the function that outputs and inputs of mobile terminal, it is specific this
Place does not limit.
Interface unit 608 is the interface that external device (ED) is connect with mobile terminal 600.For example, external device (ED) may include having
Line or wireless head-band earphone port, external power supply (or battery charger) port, wired or wireless data port, storage card end
Mouth, port, the port audio input/output (I/O), video i/o port, earphone end for connecting the device with identification module
Mouthful etc..Interface unit 608 can be used for receiving the input (for example, data information, electric power etc.) from external device (ED) and
By one or more elements that the input received is transferred in mobile terminal 600 or can be used in 600 He of mobile terminal
Transmission data between external device (ED).
Memory 609 can be used for storing software program and various data.Memory 609 can include mainly storing program area
And storage data field, wherein storing program area can storage program area, application program (such as the sound needed at least one function
Sound playing function, image player function etc.) etc.；Storage data field can store according to mobile phone use created data (such as
Audio data, phone directory etc.) etc..In addition, memory 609 may include high-speed random access memory, can also include non-easy
The property lost memory, a for example, at least disk memory, flush memory device or other volatile solid-state parts.
Processor 610 is the control centre of mobile terminal, utilizes each of various interfaces and the entire mobile terminal of connection
A part by running or execute the software program and/or module that are stored in memory 609, and calls and is stored in storage
Data in device 609 execute the various functions and processing data of mobile terminal, to carry out integral monitoring to mobile terminal.Place
Reason device 610 may include one or more processing units；Preferably, processor 610 can integrate application processor and modulatedemodulate is mediated
Manage device, wherein the main processing operation system of application processor, user interface and application program etc., modem processor is main
Processing wireless communication.It is understood that above-mentioned modem processor can not also be integrated into processor 610.
Mobile terminal 600 can also include the power supply 611 (such as battery) powered to all parts, it is preferred that power supply 611
Can be logically contiguous by power-supply management system and processor 610, to realize management charging by power-supply management system, put
The functions such as electricity and power managed.
In addition, mobile terminal 600 includes some unshowned function modules, details are not described herein.
Preferably, the embodiment of the present invention also provides a kind of mobile terminal, including processor 610, and memory 609 is stored in
On memory 609 and the computer program that can be run on the processor 610, the computer program are executed by processor 610
Each process of the above-mentioned image Style Transfer model generating method embodiments of Shi Shixian, and identical technique effect can be reached, it is
It avoids repeating, which is not described herein again.
The embodiment of the present invention also provides a kind of computer readable storage medium, and meter is stored on computer readable storage medium
Calculation machine program, the computer program realize each of above-mentioned image Style Transfer model generating method embodiment when being executed by processor
A process, and identical technique effect can be reached, to avoid repeating, which is not described herein again.Wherein, described computer-readable to deposit
Storage media, such as read-only memory (Read-Only Memory, abbreviation ROM), random access memory (Random Access
Memory, abbreviation RAM), magnetic disc or CD etc..
It should be noted that herein, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that process, method, article or device including a series of elements include not only those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or device institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including this
There is also other identical elements in the process of element, method, article or device.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can add the mode of required general hardware platform to realize by software, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on this understanding, technical scheme of the present invention substantially in other words does the prior art
Going out the part of contribution can be expressed in the form of software products, which is stored in a storage medium
In (such as ROM/RAM, magnetic disc, CD), including some instructions are used so that a station terminal (can be mobile phone, computer, service
Device, air conditioner or network equipment etc.) execute method described in each embodiment of the present invention.
The embodiment of the present invention is described with above attached drawing, but the invention is not limited in above-mentioned specific
Embodiment, the above mentioned embodiment is only schematical, rather than restrictive, those skilled in the art
Under the inspiration of the present invention, without breaking away from the scope protected by the purposes and claims of the present invention, it can also make very much
Form belongs within the protection of the present invention.
Claims (12)
1. a kind of image Style Transfer model generating method, which is characterized in that the method includes：
Build style training data pair, wherein each data are to including a trained artwork and a style figure；
Building structure sequence is followed successively by input layer, prefilter layer, at least one residual error feature construction layer, back layer and output layer
Neural network；
The mean square deviation of mean square deviation loss and depth perception feature based on training artwork and style figure is lost, and loss function is built；
According to the style training data pair and the loss function, the neural network is trained, generates image style
Migration models.
2. according to the method described in claim 1, it is characterized in that, being followed successively by input layer, preposition in building structure sequence
After the neural network of layer, at least one residual error feature construction layer, back layer and output layer, the method further includes：
Carrying out depth to the prefilter layer, the back layer and each residual error feature construction layer in the neural network respectively can
Convolution sum point process of convolution is detached, parameter adjustment is carried out.
3. according to the method described in claim 1, it is characterized in that：
The prefilter layer includes five layers of convolutional layer；
The back layer includes five layers of convolutional layer；
Each residual error feature construction layer includes three-layer coil lamination.
4. according to the method described in claim 1, it is characterized in that, described according to the style training data pair and the loss
Function is trained the neural network, generates image Style Transfer model, including：
One by one by each style training data to being input in the neural network；
For each style training data pair of input, the training by the neural network to the style training data centering
Artwork is handled, and result figure is obtained；
By advance trained deep layer Perception Features network, to the style of the result figure and the style training data centering
Figure carries out Perception Features extraction, obtains Perception Features figure；
The Perception Features figure, the result figure are substituted into the loss function according to the style figure, generate target loss letter
Number；
According to target loss function, the parameter of the neural network is adjusted.
5. described according to the method described in claim 1, it is characterized in that, after the generation image Style Transfer model
Method further includes：
Pending image is input to described image Style Transfer model；
By described image Style Transfer model, Style Transfer is carried out to the pending image, exports the pending image
Corresponding Style Transfer image.
6. a kind of mobile terminal, which is characterized in that the mobile terminal includes：
Module is built, for building style training data pair, wherein each data are to including a trained artwork and a style
Figure；
Module is built, input layer, prefilter layer, at least one residual error feature construction layer, postposition are followed successively by for building structure sequence
The neural network of layer and output layer；
Loss function builds module, is lost for the mean square deviation based on training artwork and style figure square with depth perception feature
Differential loss loses, and builds loss function；
Training module, for according to the style training data pair and the loss function, being trained to the neural network,
Generate image Style Transfer model.
7. mobile terminal according to claim 6, which is characterized in that the mobile terminal further includes：
Optimization module, for being followed successively by input layer, prefilter layer, at least two residual errors spy in the module building structure sequence of building
After levying structure layer, the neural network of back layer and output layer, respectively to the prefilter layer, described in the neural network
Back layer and each residual error feature construction layer carry out depth and separate convolution sum point process of convolution, carry out parameter adjustment.
8. mobile terminal according to claim 6, it is characterised in that：
The prefilter layer includes five layers of convolutional layer；
The back layer includes five layers of convolutional layer；
Each residual error feature construction layer includes three-layer coil lamination.
9. mobile terminal according to claim 6, which is characterized in that the training module includes：
Input submodule, for one by one by each style training data to being input in the neural network；
Submodule is called, for each style training data pair for input, the style is instructed by the neural network
The training artwork for practicing data centering is handled, and result figure is obtained；
Submodule is analyzed, for by advance trained deep layer Perception Features network, being instructed to the result figure and the style
The style figure for practicing data centering carries out Perception Features extraction, obtains Perception Features figure；
Submodule is generated, for the Perception Features figure, the result figure to be substituted into the loss function according to the style figure,
Generate target loss function；
Submodule is adjusted, for according to target loss function, adjusting the parameter of the neural network.
10. mobile terminal according to claim 6, which is characterized in that further include in the mobile terminal：
Input module inputs pending image after generating image Style Transfer model described in the training module
To described image Style Transfer model；By described image Style Transfer model, Style Transfer is carried out to the pending image,
Export the corresponding Style Transfer image of the pending image.
11. a kind of mobile terminal, which is characterized in that including processor, memory and be stored on the memory and can be in institute
The computer program run on processor is stated, such as claim 1 to 5 is realized when the computer program is executed by the processor
Any one of described in image Style Transfer model generating method the step of.
12. a kind of computer readable storage medium, which is characterized in that store computer journey on the computer readable storage medium
Sequence realizes the image Style Transfer mould as described in any one of claim 1 to 5 when the computer program is executed by processor
The step of type generation method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810202121.2A CN108537776A (en) | 2018-03-12 | 2018-03-12 | A kind of image Style Transfer model generating method and mobile terminal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810202121.2A CN108537776A (en) | 2018-03-12 | 2018-03-12 | A kind of image Style Transfer model generating method and mobile terminal |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108537776A true CN108537776A (en) | 2018-09-14 |
Family
ID=63484350
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810202121.2A Pending CN108537776A (en) | 2018-03-12 | 2018-03-12 | A kind of image Style Transfer model generating method and mobile terminal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108537776A (en) |
Cited By (40)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109062811A (en) * | 2018-09-27 | 2018-12-21 | 华东师范大学 | A kind of method for generating test case based on neural Style Transfer |
| CN109166087A (en) * | 2018-09-29 | 2019-01-08 | 上海联影医疗科技有限公司 | Style conversion method, device, medical supply, image system and the storage medium of medical image |
| CN109345446A (en) * | 2018-09-18 | 2019-02-15 | 西华大学 | Image style transfer algorithm based on dual learning |
| CN109376830A (en) * | 2018-10-17 | 2019-02-22 | 京东方科技集团股份有限公司 | Two-dimensional code generation method and device |
| CN109472315A (en) * | 2018-11-15 | 2019-03-15 | 江苏木盟智能科技有限公司 | A kind of object detection method and system separating convolution based on depth |
| CN109472270A (en) * | 2018-10-31 | 2019-03-15 | 京东方科技集团股份有限公司 | Image style conversion method, device and equipment |
| CN109492735A (en) * | 2018-11-23 | 2019-03-19 | 清华大学 | Two-dimensional code generation method and computer readable storage medium |
| CN109635711A (en) * | 2018-12-07 | 2019-04-16 | 上海衡道医学病理诊断中心有限公司 | A kind of pathological image dividing method based on deep learning network |
| CN109670529A (en) * | 2018-11-14 | 2019-04-23 | 天津大学 | A kind of separable decomposition residual error modularity for quick semantic segmentation |
| CN109859096A (en) * | 2018-12-28 | 2019-06-07 | 北京达佳互联信息技术有限公司 | Image Style Transfer method, apparatus, electronic equipment and storage medium |
| CN109919829A (en) * | 2019-01-17 | 2019-06-21 | 北京达佳互联信息技术有限公司 | Image Style Transfer method, apparatus and computer readable storage medium |
| CN109949214A (en) * | 2019-03-26 | 2019-06-28 | 湖北工业大学 | A kind of image Style Transfer method and system |
| CN110033034A (en) * | 2019-04-01 | 2019-07-19 | 深圳大学 | A kind of image processing method, device and the computer equipment of non-homogeneous texture |
| CN110059586A (en) * | 2019-03-29 | 2019-07-26 | 电子科技大学 | A kind of Iris Location segmenting system based on empty residual error attention structure |
| CN110298230A (en) * | 2019-05-06 | 2019-10-01 | 深圳市华付信息技术有限公司 | Silent biopsy method, device, computer equipment and storage medium |
| CN110335206A (en) * | 2019-05-31 | 2019-10-15 | 平安科技（深圳）有限公司 | Smart filter method, apparatus and computer readable storage medium |
| CN110427948A (en) * | 2019-07-29 | 2019-11-08 | 杭州云深弘视智能科技有限公司 | The generation method and its system of character sample |
| CN110598781A (en) * | 2019-09-05 | 2019-12-20 | Oppo广东移动通信有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| TWI688276B (en) * | 2019-03-06 | 2020-03-11 | 明新科技大學 | A method of characterizing a color image sensor by probabilistic neural network |
| CN110909790A (en) * | 2019-11-20 | 2020-03-24 | Oppo广东移动通信有限公司 | Image style migration method, device, terminal and storage medium |
| CN110910351A (en) * | 2019-10-31 | 2020-03-24 | 上海交通大学 | Ultrasound image modality migration and classification method and terminal based on generation countermeasure network |
| CN110930295A (en) * | 2019-10-25 | 2020-03-27 | 广东开放大学（广东理工职业学院） | Image style migration method, system, device and storage medium |
| CN110956654A (en) * | 2019-12-02 | 2020-04-03 | Oppo广东移动通信有限公司 | Image processing method, device, equipment and storage medium |
| CN111127309A (en) * | 2019-12-12 | 2020-05-08 | 杭州格像科技有限公司 | Portrait style transfer model training method, portrait style transfer method and device |
| CN111124398A (en) * | 2018-10-31 | 2020-05-08 | 中国移动通信集团重庆有限公司 | User interface generation method, device, equipment and storage medium |
| CN111476708A (en) * | 2020-04-03 | 2020-07-31 | 广州市百果园信息技术有限公司 | Model generation method, model acquisition method, model generation device, model acquisition device, model generation equipment and storage medium |
| CN111667401A (en) * | 2020-06-08 | 2020-09-15 | 武汉理工大学 | Multi-level gradient image style migration method and system |
| CN111667399A (en) * | 2020-05-14 | 2020-09-15 | 华为技术有限公司 | Method for training style migration model, method and device for video style migration |
| CN111753908A (en) * | 2020-06-24 | 2020-10-09 | 北京百度网讯科技有限公司 | Image classification method and device and style migration model training method and device |
| CN111784566A (en) * | 2020-07-01 | 2020-10-16 | 北京字节跳动网络技术有限公司 | Image processing method, migration model training method, device, medium and equipment |
| WO2020228406A1 (en) * | 2019-05-15 | 2020-11-19 | 北京字节跳动网络技术有限公司 | Image stylization generation method and apparatus, and electronic device |
| CN111985544A (en) * | 2020-08-07 | 2020-11-24 | 中移（杭州）信息技术有限公司 | Fractal graph style fusion method and device, electronic equipment and storage medium |
| CN112439200A (en) * | 2020-12-03 | 2021-03-05 | 腾讯科技（深圳）有限公司 | Data processing method, data processing device, storage medium and electronic equipment |
| CN112740645A (en) * | 2018-10-12 | 2021-04-30 | 三星电子株式会社 | Display device, server, and control method thereof |
| CN112862669A (en) * | 2021-02-02 | 2021-05-28 | 百果园技术(新加坡)有限公司 | Training method, generating method, device and equipment of image generation model |
| CN113012036A (en) * | 2021-03-05 | 2021-06-22 | 清华大学 | Human motion style migration method and system based on generative flow model |
| CN113763230A (en) * | 2020-06-04 | 2021-12-07 | 北京达佳互联信息技术有限公司 | Image style migration model training method, style migration method and device |
| CN113841179A (en) * | 2019-04-29 | 2021-12-24 | 商汤集团有限公司 | Image generation method and device, electronic device and storage medium |
| CN114298895A (en) * | 2021-12-24 | 2022-04-08 | 成都索贝数码科技股份有限公司 | Image realistic style migration method, device, equipment and storage medium |
| CN117932043A (en) * | 2024-03-22 | 2024-04-26 | 杭州食方科技有限公司 | Dialogue style migration reply information display method, device, equipment and readable medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106651766A (en) * | 2016-12-30 | 2017-05-10 | 深圳市唯特视科技有限公司 | Image style migration method based on deep convolutional neural network |
| CN106952224A (en) * | 2017-03-30 | 2017-07-14 | 电子科技大学 | A kind of image style transfer method based on convolutional neural networks |
| CN107392842A (en) * | 2017-06-30 | 2017-11-24 | 北京奇虎科技有限公司 | Image stylization processing method, device, computing device and computer-readable storage medium |
| CN107392316A (en) * | 2017-06-30 | 2017-11-24 | 北京奇虎科技有限公司 | Network training method, device, computing device and computer-readable storage medium |
| CN107705242A (en) * | 2017-07-20 | 2018-02-16 | 广东工业大学 | A kind of image stylization moving method of combination deep learning and depth perception |
-
2018
- 2018-03-12 CN CN201810202121.2A patent/CN108537776A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106651766A (en) * | 2016-12-30 | 2017-05-10 | 深圳市唯特视科技有限公司 | Image style migration method based on deep convolutional neural network |
| CN106952224A (en) * | 2017-03-30 | 2017-07-14 | 电子科技大学 | A kind of image style transfer method based on convolutional neural networks |
| CN107392842A (en) * | 2017-06-30 | 2017-11-24 | 北京奇虎科技有限公司 | Image stylization processing method, device, computing device and computer-readable storage medium |
| CN107392316A (en) * | 2017-06-30 | 2017-11-24 | 北京奇虎科技有限公司 | Network training method, device, computing device and computer-readable storage medium |
| CN107705242A (en) * | 2017-07-20 | 2018-02-16 | 广东工业大学 | A kind of image stylization moving method of combination deep learning and depth perception |
Non-Patent Citations (2)
| Title |
|---|
| FRANÇOIS CHOLLET: "Xception: Deep Learning with Depthwise Separable Convolutions", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| JUSTIN JOHNSON，ALEXANDRE ALAHI，LI FEI-FEI: "Perceptual Losses for Real-Time Style Transfer and Super-Resolution", 《LECTURE NOTES IN COMPUTER SCIENCE》 * |
Cited By (63)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109345446A (en) * | 2018-09-18 | 2019-02-15 | 西华大学 | Image style transfer algorithm based on dual learning |
| CN109345446B (en) * | 2018-09-18 | 2022-12-02 | 西华大学 | Image style transfer algorithm based on dual learning |
| CN109062811A (en) * | 2018-09-27 | 2018-12-21 | 华东师范大学 | A kind of method for generating test case based on neural Style Transfer |
| CN109166087A (en) * | 2018-09-29 | 2019-01-08 | 上海联影医疗科技有限公司 | Style conversion method, device, medical supply, image system and the storage medium of medical image |
| CN112740645A (en) * | 2018-10-12 | 2021-04-30 | 三星电子株式会社 | Display device, server, and control method thereof |
| US11366581B2 (en) | 2018-10-12 | 2022-06-21 | Samsung Electronics Co., Ltd. | Display apparatus, server and control method for applying transfer style to images using artificial intelligence |
| CN109376830A (en) * | 2018-10-17 | 2019-02-22 | 京东方科技集团股份有限公司 | Two-dimensional code generation method and device |
| US11164059B2 (en) | 2018-10-17 | 2021-11-02 | Boe Technology Group Co., Ltd. | Two-dimensional code image generation method and apparatus, storage medium and electronic device |
| CN109376830B (en) * | 2018-10-17 | 2022-01-11 | 京东方科技集团股份有限公司 | Two-dimensional code generation method and device |
| CN111124398A (en) * | 2018-10-31 | 2020-05-08 | 中国移动通信集团重庆有限公司 | User interface generation method, device, equipment and storage medium |
| US10970830B2 (en) | 2018-10-31 | 2021-04-06 | Boe Technology Group Co., Ltd. | Image style conversion method, apparatus and device |
| CN109472270B (en) * | 2018-10-31 | 2021-09-24 | 京东方科技集团股份有限公司 | Image style conversion method, device and equipment |
| CN109472270A (en) * | 2018-10-31 | 2019-03-15 | 京东方科技集团股份有限公司 | Image style conversion method, device and equipment |
| CN109670529A (en) * | 2018-11-14 | 2019-04-23 | 天津大学 | A kind of separable decomposition residual error modularity for quick semantic segmentation |
| CN109472315B (en) * | 2018-11-15 | 2021-09-24 | 江苏木盟智能科技有限公司 | Target detection method and system based on depth separable convolution |
| CN109472315A (en) * | 2018-11-15 | 2019-03-15 | 江苏木盟智能科技有限公司 | A kind of object detection method and system separating convolution based on depth |
| CN109492735A (en) * | 2018-11-23 | 2019-03-19 | 清华大学 | Two-dimensional code generation method and computer readable storage medium |
| CN109635711A (en) * | 2018-12-07 | 2019-04-16 | 上海衡道医学病理诊断中心有限公司 | A kind of pathological image dividing method based on deep learning network |
| CN109859096A (en) * | 2018-12-28 | 2019-06-07 | 北京达佳互联信息技术有限公司 | Image Style Transfer method, apparatus, electronic equipment and storage medium |
| WO2020134556A1 (en) * | 2018-12-28 | 2020-07-02 | 北京达佳互联信息技术有限公司 | Image style transfer method, device, electronic apparatus, and storage medium |
| CN109919829A (en) * | 2019-01-17 | 2019-06-21 | 北京达佳互联信息技术有限公司 | Image Style Transfer method, apparatus and computer readable storage medium |
| CN109919829B (en) * | 2019-01-17 | 2023-12-26 | 北京达佳互联信息技术有限公司 | Image style migration method, device and computer readable storage medium |
| TWI688276B (en) * | 2019-03-06 | 2020-03-11 | 明新科技大學 | A method of characterizing a color image sensor by probabilistic neural network |
| CN109949214A (en) * | 2019-03-26 | 2019-06-28 | 湖北工业大学 | A kind of image Style Transfer method and system |
| CN110059586A (en) * | 2019-03-29 | 2019-07-26 | 电子科技大学 | A kind of Iris Location segmenting system based on empty residual error attention structure |
| CN110059586B (en) * | 2019-03-29 | 2022-03-15 | 电子科技大学 | Iris positioning and segmenting system based on cavity residual error attention structure |
| CN110033034B (en) * | 2019-04-01 | 2023-09-12 | 深圳大学 | Picture processing method and device for non-uniform texture and computer equipment |
| CN110033034A (en) * | 2019-04-01 | 2019-07-19 | 深圳大学 | A kind of image processing method, device and the computer equipment of non-homogeneous texture |
| CN113841179A (en) * | 2019-04-29 | 2021-12-24 | 商汤集团有限公司 | Image generation method and device, electronic device and storage medium |
| CN110298230A (en) * | 2019-05-06 | 2019-10-01 | 深圳市华付信息技术有限公司 | Silent biopsy method, device, computer equipment and storage medium |
| WO2020228406A1 (en) * | 2019-05-15 | 2020-11-19 | 北京字节跳动网络技术有限公司 | Image stylization generation method and apparatus, and electronic device |
| CN110335206B (en) * | 2019-05-31 | 2023-06-09 | 平安科技（深圳）有限公司 | Intelligent filter method, device and computer readable storage medium |
| CN110335206A (en) * | 2019-05-31 | 2019-10-15 | 平安科技（深圳）有限公司 | Smart filter method, apparatus and computer readable storage medium |
| CN110427948A (en) * | 2019-07-29 | 2019-11-08 | 杭州云深弘视智能科技有限公司 | The generation method and its system of character sample |
| CN110598781A (en) * | 2019-09-05 | 2019-12-20 | Oppo广东移动通信有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN110930295A (en) * | 2019-10-25 | 2020-03-27 | 广东开放大学（广东理工职业学院） | Image style migration method, system, device and storage medium |
| CN110910351B (en) * | 2019-10-31 | 2022-07-29 | 上海交通大学 | Ultrasound image modality migration and classification method and terminal based on generation countermeasure network |
| CN110910351A (en) * | 2019-10-31 | 2020-03-24 | 上海交通大学 | Ultrasound image modality migration and classification method and terminal based on generation countermeasure network |
| CN110909790A (en) * | 2019-11-20 | 2020-03-24 | Oppo广东移动通信有限公司 | Image style migration method, device, terminal and storage medium |
| CN110956654A (en) * | 2019-12-02 | 2020-04-03 | Oppo广东移动通信有限公司 | Image processing method, device, equipment and storage medium |
| CN110956654B (en) * | 2019-12-02 | 2023-09-19 | Oppo广东移动通信有限公司 | Image processing method, device, equipment and storage medium |
| CN111127309B (en) * | 2019-12-12 | 2023-08-11 | 杭州格像科技有限公司 | Portrait style migration model training method, portrait style migration method and device |
| CN111127309A (en) * | 2019-12-12 | 2020-05-08 | 杭州格像科技有限公司 | Portrait style transfer model training method, portrait style transfer method and device |
| CN111476708A (en) * | 2020-04-03 | 2020-07-31 | 广州市百果园信息技术有限公司 | Model generation method, model acquisition method, model generation device, model acquisition device, model generation equipment and storage medium |
| CN111667399B (en) * | 2020-05-14 | 2023-08-25 | 华为技术有限公司 | Training method of style migration model, video style migration method and device |
| CN111667399A (en) * | 2020-05-14 | 2020-09-15 | 华为技术有限公司 | Method for training style migration model, method and device for video style migration |
| CN113763230B (en) * | 2020-06-04 | 2024-05-17 | 北京达佳互联信息技术有限公司 | Image style migration model training method, style migration method and device |
| CN113763230A (en) * | 2020-06-04 | 2021-12-07 | 北京达佳互联信息技术有限公司 | Image style migration model training method, style migration method and device |
| CN111667401A (en) * | 2020-06-08 | 2020-09-15 | 武汉理工大学 | Multi-level gradient image style migration method and system |
| CN111667401B (en) * | 2020-06-08 | 2022-11-29 | 武汉理工大学 | Multi-level gradient image style migration method and system |
| CN111753908A (en) * | 2020-06-24 | 2020-10-09 | 北京百度网讯科技有限公司 | Image classification method and device and style migration model training method and device |
| CN111784566B (en) * | 2020-07-01 | 2022-02-08 | 北京字节跳动网络技术有限公司 | Image processing method, migration model training method, device, medium and equipment |
| CN111784566A (en) * | 2020-07-01 | 2020-10-16 | 北京字节跳动网络技术有限公司 | Image processing method, migration model training method, device, medium and equipment |
| CN111985544A (en) * | 2020-08-07 | 2020-11-24 | 中移（杭州）信息技术有限公司 | Fractal graph style fusion method and device, electronic equipment and storage medium |
| CN112439200B (en) * | 2020-12-03 | 2022-07-01 | 腾讯科技（深圳）有限公司 | Data processing method, data processing device, storage medium and electronic equipment |
| CN112439200A (en) * | 2020-12-03 | 2021-03-05 | 腾讯科技（深圳）有限公司 | Data processing method, data processing device, storage medium and electronic equipment |
| CN112862669A (en) * | 2021-02-02 | 2021-05-28 | 百果园技术(新加坡)有限公司 | Training method, generating method, device and equipment of image generation model |
| CN112862669B (en) * | 2021-02-02 | 2024-02-09 | 百果园技术(新加坡)有限公司 | Training method, generating method, device and equipment for image generating model |
| CN113012036B (en) * | 2021-03-05 | 2022-07-15 | 清华大学 | Human motion style migration method and system based on generative flow model |
| CN113012036A (en) * | 2021-03-05 | 2021-06-22 | 清华大学 | Human motion style migration method and system based on generative flow model |
| CN114298895A (en) * | 2021-12-24 | 2022-04-08 | 成都索贝数码科技股份有限公司 | Image realistic style migration method, device, equipment and storage medium |
| CN114298895B (en) * | 2021-12-24 | 2023-09-26 | 成都索贝数码科技股份有限公司 | Image realism style migration method, device, equipment and storage medium |
| CN117932043A (en) * | 2024-03-22 | 2024-04-26 | 杭州食方科技有限公司 | Dialogue style migration reply information display method, device, equipment and readable medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108537776A (en) | A kind of image Style Transfer model generating method and mobile terminal | |
| CN107766939A (en) | A kind of data processing method, device and mobile terminal | |
| CN107833178A (en) | A kind of image processing method, device and mobile terminal | |
| CN107948499A (en) | A kind of image capturing method and mobile terminal | |
| CN107734251A (en) | A kind of photographic method and mobile terminal | |
| CN107872639A (en) | Transmission method, device and the mobile terminal of communication video | |
| CN108495029A (en) | A kind of photographic method and mobile terminal | |
| CN107943390A (en) | A kind of word clone method and mobile terminal | |
| CN107566730A (en) | A kind of panoramic picture image pickup method and mobile terminal | |
| CN108989678A (en) | A kind of image processing method, mobile terminal | |
| CN107909583A (en) | A kind of image processing method, device and terminal | |
| CN107734260A (en) | A kind of image processing method and mobile terminal | |
| CN107886321A (en) | A kind of method of payment and mobile terminal | |
| CN107749046A (en) | A kind of image processing method and mobile terminal | |
| CN107958161A (en) | A kind of multitask display methods and mobile terminal | |
| CN107682637A (en) | A kind of image pickup method, mobile terminal and computer-readable recording medium | |
| CN107679514A (en) | A kind of face identification method and electronic equipment | |
| CN110213485A (en) | A kind of image processing method and terminal | |
| CN108008858A (en) | A kind of terminal control method and mobile terminal | |
| CN107895143A (en) | A kind of finger print information processing method, mobile terminal and computer-readable recording medium | |
| CN107888392A (en) | A kind of information processing method, server and mobile terminal | |
| CN107911812A (en) | It is preferred that public land mobile network PPLMN lists method of adjustment and mobile terminal | |
| CN108573307A (en) | A kind of method and terminal of processing neural network model file | |
| CN107845074A (en) | A kind of image rendering method and mobile terminal | |
| CN107786811A (en) | A kind of photographic method and mobile terminal |
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 |