CN108230278A - A kind of image based on generation confrontation network goes raindrop method - Google Patents
A kind of image based on generation confrontation network goes raindrop method Download PDFInfo
- Publication number
- CN108230278A CN108230278A CN201810157009.1A CN201810157009A CN108230278A CN 108230278 A CN108230278 A CN 108230278A CN 201810157009 A CN201810157009 A CN 201810157009A CN 108230278 A CN108230278 A CN 108230278A
- Authority
- CN
- China
- Prior art keywords
- network
- error
- image
- production network
- clear
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000004519 manufacturing process Methods 0.000 claims abstract description 41
- 230000000694 effects Effects 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims description 28
- 238000012360 testing method Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000003672 processing method Methods 0.000 abstract description 4
- 230000000644 propagated effect Effects 0.000 abstract description 4
- 238000013527 convolutional neural network Methods 0.000 description 13
- 238000009826 distribution Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 2
- 230000009977 dual effect Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000003252 repetitive effect Effects 0.000 description 2
- 230000000630 rising effect Effects 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 238000002910 structure generation Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The embodiment of the invention discloses a kind of images based on generation confrontation network to go raindrop method.This method mainly generates confrontation network by building, utilize depth rain algorithm, a kind of more efficiently significant image rain removing method is provided, picture is only needed to input in production network in actual use, result figure piece can obtain by a propagated forward, more efficient effect is had compared to traditional image processing method is played, in addition, the perceived relevance of introduced feature spatially in a model, the details of rain effect can be gone with adjustment member, so that the image of generation is more clear intuitively, better effect can be provided in terms of image enhancement.
Description
Technical field
The present invention relates to image filtering technical field, machine learning fields, and in particular to one kind is based on generation confrontation network
Image go raindrop method.
Background technology
With the high speed development of smart mobile phone in recent years, more and more people carry out the shooting of outdoor scene using mobile phone.And outside
When scape is shot often due to the rainy day, captured picture can carry some raindrop or very light rain in scene often
Among.Therefore, the image that be more clear needs to carry out image certain processing.With the hair of computer in recent years
Exhibition and the continuous research of deep learning, are seemed more using more effective deep learning method to solve traditional studying a question
It is effective and feasible.
Convolutional neural networks (CNN) are the mutation of multi-layer perception (MLP) (MLP), and performances of the CNN on traditional sample is not
MLP is outstanding like that, but the effect on image pattern is more outstanding, and to compared with traditional one-dimensional data sample as defeated
The way entered, CNN can make image directly as the data of network, and avoiding some needs to carry out feature extraction and partial data
The operation of processing.CNN exports upper strata using convolution kernel and carries out convolution operation, can voluntarily extract the characteristic pattern of upper strata output
Piece, including some colors, texture, abstract structure in shape；Particularly asking in displacement, scaling and other distortion invariance
There is good robustness in topic.
Generation confrontation network (GAN) is a kind of method of the training pattern proposed in 2014, and this method passes through two moulds
Dual training between type (production network G and discriminate network D), with reference to the think of of the MinMax problems inside game theory
Road, finally so that the effect of two models can increase.The target of GAN gives the set of an authentic specimen distribution, root
It is final that generator G is generated as far as possible from noise signal according to the set continuous repetitive exercise generator G and arbiter D
Meet the sample of authentic specimen distribution, and arbiter D can differentiate whether the sample meets authentic specimen from the distribution of sample
Distribution.
Perceived relevance proposed that main contributions are to propose a kind of new measurement, help to generate GAN in 2016
Clear image.Its method is the loss metric replaced with the loss metric usually in original pixels space in feature space.
During training GAN, in addition to the dual training loss in original GAN, two loss items are additionally added, altogether three loss items, point
It is not：Feature space loss, confrontation loss, pixel space loss.
Image rain removing method mainly has：Image, which is carried out, using single convolutional neural networks removes rain, the specific steps are：1) it obtains
Clear image and artificial plus rain figure picture, build image library；2) convolutional neural networks are designed；3) using in image library it is clear-
Band rain figure picture pair, is trained designed convolutional neural networks；4) one fixed wheel number of training, obtains trained convolutional Neural
Network；5) band rain figure picture is input in trained convolutional neural networks, obtains corresponding clear image.
But the effect of the picture generated using above-mentioned technology has segmental defect, especially in the background portion similar with raindrop
Point, often it is distorted.
Invention content
The purpose of the present invention is overcome the shortcomings of existing method, it is proposed that a kind of image based on generation confrontation network removes rain
Drop method.The present invention generates confrontation network by building, and using depth rain algorithm, provides a kind of more efficient significant image
Rain removing method only needs picture to input in production network in actual use, can be tied by a propagated forward
Fruit picture has a more efficient effect compared to a traditional image processing method is played, in addition, introduced feature is spatially in a model
Perceived relevance can go the details of rain effect so that the image of generation is more clear intuitively, in image enhancement side with adjustment member
Face can provide better effect.
To solve the above-mentioned problems, the present invention proposes a kind of image based on generation confrontation network and goes to raindrop method, institute
The method of stating includes：
Outdoor scene pictures are obtained from database；
Image preprocessing adds in effect of raining for accessed outdoor scene pictures, builds training set and test set；
Production network is built, input is band rain scene image, is exported as clear scene image；
Production network is trained according to the error in pixel space；
The error added on feature space trains production network again；
Discriminate network is built, the sample that input is generated for authentic specimen or by generator is exported as true or false
Single mark；
Among discriminate network is added to model, using the error backpropagation algorithm training production network；
Band rain scene graph in test set is inputted in trained production network, is exported as corresponding clear scene figure
Picture.
Preferably, the error added on feature space trains production network again, specially：
Trained comparator network is introduced, the clear picture of generation and practical clear picture are separately input to
In comparator network, characteristic pattern of the two in comparator network is obtained respectively, calculates the Euclidean distance between two characteristic patterns,
As the error on feature space.With reference to the error in the error and feature space in pixel space as global error, use
The error backpropagation algorithm training production network.
In embodiments of the present invention, it is proposed that a kind of image based on generation confrontation network goes raindrop method.The present invention is logical
Structure generation confrontation network is crossed, using depth rain algorithm, a kind of more efficiently significant image rain removing method is provided, in reality
Picture is only needed to input in production network in use, result figure piece can obtain by a propagated forward, passed compared to rising
The image processing method of system has more efficient effect, can be in addition, the perceived relevance of introduced feature spatially in a model
Adjustment member goes the details of rain effect so that the image of generation is more clear intuitively, can be provided in terms of image enhancement more preferable
Effect.
Description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention, for those of ordinary skill in the art, without creative efforts, can be with
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is the overall flow figure of the embodiment of the present invention；
Fig. 2 is the structure chart of the production network G of the embodiment of the present invention；
Fig. 3 is the structure chart of the discriminate network D of the embodiment of the present invention；
Fig. 4 is the overall schematic of the embodiment of the present invention.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present invention, the technical solution in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other without creative efforts
Embodiment shall fall within the protection scope of the present invention.
Fig. 1 is the overall flow figure of the embodiment of the present invention, as shown in Figure 1, this method includes：
S1 obtains outdoor scene pictures from database；
S2, image preprocessing add in effect of raining for accessed outdoor scene pictures, build training set and test set；
S3 builds production network, and input is band rain scene image, is exported as clear scene image；
S4 trains production network according to the error in pixel space；
S5, the error added on feature space train production network again；
S6, builds discriminate network, and the sample that input is generated for authentic specimen or by generator is exported as true or false
Single mark；
S7, among discriminate network is added to model, using the error backpropagation algorithm training production net
Network；
Band rain scene graph in test set is inputted in trained production network, exported as corresponding clear field by S8
Scape image.
Step S1 is specially that 1000 outdoor scene pictures are obtained from SUN databases.
Step S2, it is specific as follows：
S21, using PS softwares, using the method for structure dynamic fuzzy filter, set respectively fuzzy angle for 75 degree, 80
It spends, 90 degree, 100 degree, 105 degree；Distance is 30 pixels, 50 pixels；10 kinds of filters for imitating different rainy effects are built altogether.
S22 adds the filter described in S21 respectively for each outdoor scene figure, and each pictures correspond to 10 band rain and imitate
The picture of fruit forms 10000 pairs clear-fuzzy pair altogether；8000 pairs of pictures are randomly selected, about 3,000,000 pairs of 32x32 sizes of generation
Band rain-and clearly to block, as training set；Remaining 2000 pictures evaluate network as forecast set.
Step S3, it is specific as follows：
As shown in Fig. 2, structure production network, is made of 24 residual block convolution units, input is band rain scene graph
Picture is exported as clear scene image；Whole network is made of altogether 24 blocks, wherein include inside each block convolutional layer,
Criticize standardization layer (Batch Normalization, BN) and Swish layers；Since the image of input is coloured image, so first
The convolution kernel size of layer is 7x7x3, and convolution step-length is 1, shares 16 convolution karyogenesis 16 and opens output；2nd~23 layer of convolution
Core size is 3x3x16, and convolution step-length is 1, shares 16 convolution karyogenesis 16 and opens output；And the convolution kernel size of last layer
For 3x3x3, convolution step-length is 1, shares 3 convolution karyogenesis 3 and opens output；Output of wherein N+1 layers of the output for N+1 blocks
In addition output (the N of N-1 blocks>=2).
Step S4, it is specific as follows：
Training production network using error backpropagation algorithm, is made using Euclidean distance of the image in pixel space
For error, the neural network described in S3 is trained；Learning rate is set as 0.1, and every batch of inputs 40 pictures, with 200,000 iteration
As a complete training, initial training production network.
Wherein, the error L in pixel space1, using mean square error (Mean Squared Error, MSE), specific formula
It is as follows：
Wherein, θ g are the parameter in constructed model in S3, and G is the mapping equation of the model, XiFor i-th band rain figure
Piece, YiFor corresponding clear picture, sums of the n for training sample, therefore L1Calculated is to go rain figure by what model generated
Piece Gθg(Xi) and true picture YiBetween the sum of Euclidean distance square, by back-propagation method minimize this error go it is excellent
Change model.
Wherein, backpropagation uses the stochastic gradient descent based on standard backpropagation, and weight matrix passes through following expression
Formula is updated：
Wherein, l is the number of plies, and i is iteration wheel number, and η is learning rate,For gradient.
Using above-mentioned scheme, Maker model can be made tentatively to generate a picture for removing rain.
Step S5, it is specific as follows：
S51, trained VGG19 networks are as comparator network for model introducing, the clear picture and reality of generation
The clear picture on border is separately input in VGG19 networks, and obtains the two respectively in VGG19 networks in the 2nd maximum pond
Change the 2nd characteristic pattern before layer, the Euclidean distance between two characteristic patterns is calculated, as the error on feature space.
Error on feature space, circular are as follows：
Comparator model is using trained VGG19 models, respectively the picture G generated by generatorθg(Xi) and
Corresponding clear picture YiIt is input in comparator model, obtains the characteristic pattern of middle layer output, calculate Euclidean between the two
Apart from mean square error, formula is as follows：
Wherein characteristic pattern of the function phi corresponding to picture, other parameter and the ginseng in the error in the pixel space in S4
Number is identical.
S52 is missed with reference to the error on the feature space in the error and S5 in the pixel space mentioned in S4 as overall
Difference, using the error backpropagation algorithm training convolutional neural networks generator, one fixed wheel number of iteration optimization convolutional Neural
Network generator；The weight of error is set as 0.1 on feature space, and production network is trained using the method for backpropagation, with
200000 iteration train production network again.
Specifically, the global error of model is：
L=L1+λ1·L2
Wherein λ1For the weight about error on feature space in model global error.According to above-mentioned global error, in S4
It has already passed through on trained model, reuses the stochastic gradient descent method based on standard backpropagation and carry out gradient decline,
Update the network parameter in Maker model.
Using above-mentioned preferred embodiment, the picture that Maker model is generated can be caused to be more nearly nature in detail
Image.
Step S6, it is specific as follows：
As shown in figure 3, structure discriminate network, inputs the sample generated for authentic specimen or by generator, exports and be
Single mark, value are true or false；It is made of before whole network four convolution blocks, wherein includes convolution inside each block
Layer, batch standardization layer (Batch Normalization, BN) and Swish layers；The image of input be 32x32 generation image or
Person's true picture, the convolution kernel size of first layer are 5x5x3, and convolution step-length is 2, share 128 convolution karyogenesis 128 open it is defeated
Go out；The convolution kernel size of the second layer is 3x3x128, and convolution step-length is 2, shares 512 convolution karyogenesis 512 and opens output；Third
The convolution kernel size of layer is 3x3x512, and convolution step-length is 2, shares 1024 convolution karyogenesis 1024 and opens output；Then one is connected
The full articulamentum of a 1024 dimension output, ReLU layers of Leaky；Reconnection one one-dimensional full articulamentum and sigmoid obtain defeated
Go out.
Step S7, it is specific as follows：
Among discriminate network is added to model, model is totally as shown in Figure 4.With production network and discriminate network
Between confrontation error as new error term, be attached among the global error in S5, instructed using error backpropagation algorithm
Practice the convolutional neural networks generator；
Production network parameter, training discriminate network are first fixed in S71, often wheel training.It is clear that every batch of inputs 20 generations
Picture, using error backpropagation algorithm, is made as negative sample, 20 true clear pictures as positive sample with 10,000 wheel iteration
For primary complete training, training production network.
S72, fixed discriminate network parameter, training production network.Add in confrontation error caused by discriminate network
To among the global error of production network, production network is trained using the method for backpropagation, is given birth to 10,000 repetitive exercises
Accepted way of doing sth network；
S73 repeats two steps of S71 and S72 totally 10 wheel.
Wherein, described confrontation error, circular are as follows：
For newly added discriminate network D, input as authentic specimen or generate sample, training objective is：Work as input
During for authentic specimen, network output is 1；When input is generation sample, network output is 0.Therefore the expression formula of error is fought
It is as follows：
Wherein θ d are the parameter of discriminate network D；Wish that the value of above formula is the smaller the better for production network G, then it represents that
The picture that production network G is generated is more nearly with true clear image；The value of above formula, which is got over, is wished for discriminate network D
It is big better, then it represents that discriminate network can more accurately differentiate the picture of generation and true picture.
According to the training process of generation confrontation network, since production network G has been subjected to certain instruction in step S4 and S5
Practice, therefore the first parameter of locking production network G, use the method training discriminate network D of stochastic gradient descent；Training is certain
After number, the parameter of locking discriminate network D with reference to the global error in S5, using the method for stochastic gradient descent, is instructed again
Practice production network G.
Step S7 can cause model to ultimately generate the picture more similar to natural picture.
In embodiments of the present invention, it is proposed that a kind of image based on generation confrontation network goes raindrop method.The present invention is logical
Structure generation confrontation network is crossed, using depth rain algorithm, a kind of more efficiently significant image rain removing method is provided, in reality
Picture is only needed to input in production network in use, result figure piece can obtain by a propagated forward, passed compared to rising
The image processing method of system has more efficient effect, can be in addition, the perceived relevance of introduced feature spatially in a model
Adjustment member goes the details of rain effect so that the image of generation is more clear intuitively, can be provided in terms of image enhancement more preferable
Effect.
One of ordinary skill in the art will appreciate that all or part of step in the various methods of above-described embodiment is can
It is completed with instructing relevant hardware by program, which can be stored in a computer readable storage medium, storage
Medium can include：Read-only memory (ROM, Read Only Memory), random access memory (RAM, Random
Access Memory), disk or CD etc..
In addition, above to the embodiment of the present invention provided it is a kind of based on generation confrontation network image go raindrop method into
It has gone and has been discussed in detail, specific case used herein is expounded the principle of the present invention and embodiment, implements above
The explanation of example is merely used to help understand the method and its core concept of the present invention；Meanwhile for the general technology people of this field
Member, thought according to the present invention, there will be changes in specific embodiments and applications, in conclusion this explanation
Book content should not be construed as limiting the invention.
Claims (2)
1. a kind of image based on generation confrontation network goes raindrop method, which is characterized in that the method includes：
Outdoor scene pictures are obtained from database；
Image preprocessing adds in effect of raining for accessed outdoor scene pictures, builds training set and test set；
Production network is built, input is band rain scene image, is exported as clear scene image；
Production network is trained according to the error in pixel space；
The error added on feature space trains production network again；
Discriminate network is built, the sample that input is generated for authentic specimen or by generator is exported as the single of true or false
Mark；
Among discriminate network is added to model, using the error backpropagation algorithm training production network；
Band rain scene graph in test set is inputted in trained production network, is exported as corresponding clear scene image.
2. a kind of image based on generation confrontation network as described in claim 1 goes raindrop method, which is characterized in that described to add
The error entered on feature space trains production network again, specially：
Trained comparator network is introduced, the clear picture of generation and practical clear picture are separately input to compare
In device network, characteristic pattern of the two in comparator network is obtained respectively, calculates the Euclidean distance between two characteristic patterns, as
Error on feature space.With reference to the error in the error and feature space in pixel space as global error, using error
The back-propagation algorithm training production network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810157009.1A CN108230278B (en) | 2018-02-24 | 2018-02-24 | Image raindrop removing method based on generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810157009.1A CN108230278B (en) | 2018-02-24 | 2018-02-24 | Image raindrop removing method based on generation countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108230278A true CN108230278A (en) | 2018-06-29 |
| CN108230278B CN108230278B (en) | 2021-08-06 |
Family
ID=62662141
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810157009.1A Active CN108230278B (en) | 2018-02-24 | 2018-02-24 | Image raindrop removing method based on generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108230278B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109344778A (en) * | 2018-10-10 | 2019-02-15 | 成都信息工程大学 | Based on the unmanned plane road extraction method for generating confrontation network |
| CN109360156A (en) * | 2018-08-17 | 2019-02-19 | 上海交通大学 | Single image rain removing method based on the image block for generating confrontation network |
| CN109447918A (en) * | 2018-11-02 | 2019-03-08 | 北京交通大学 | Removing rain based on single image method based on attention mechanism |
| CN109472757A (en) * | 2018-11-15 | 2019-03-15 | 央视国际网络无锡有限公司 | It is a kind of that logo method is gone based on the image for generating confrontation neural network |
| CN109760635A (en) * | 2019-01-08 | 2019-05-17 | 同济大学 | A kind of line traffic control windshield wiper control system based on GAN network |
| CN109886975A (en) * | 2019-02-19 | 2019-06-14 | 武汉大学 | It is a kind of that raindrop method and system is gone based on the image optimization processing for generating confrontation network |
| CN110163813A (en) * | 2019-04-16 | 2019-08-23 | 中国科学院深圳先进技术研究院 | A kind of image rain removing method, device, readable storage medium storing program for executing and terminal device |
| CN110189272A (en) * | 2019-05-24 | 2019-08-30 | 北京百度网讯科技有限公司 | For handling the method, apparatus, equipment and storage medium of image |
| CN110363793A (en) * | 2019-07-24 | 2019-10-22 | 北京华捷艾米科技有限公司 | A kind of tracking and device of object |
| CN110472467A (en) * | 2019-04-08 | 2019-11-19 | 江西理工大学 | The detection method for transport hub critical object based on YOLO v3 |
| CN110533581A (en) * | 2019-08-12 | 2019-12-03 | 广东工业大学 | A kind of raindrop removing method, system and storage medium based on raindrop probability graph |
| CN110570358A (en) * | 2018-09-04 | 2019-12-13 | 阿里巴巴集团控股有限公司 | vehicle loss image enhancement method and device based on GAN network |
| CN112581377A (en) * | 2019-09-30 | 2021-03-30 | 上海商汤临港智能科技有限公司 | Image processing method and device and electronic equipment |
| CN117422689A (en) * | 2023-10-31 | 2024-01-19 | 南京邮电大学 | Rainy day insulator defect detection method based on improved MS-PReNet and GAM-YOLOv7 |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106204499A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | Single image rain removing method based on convolutional neural networks |
| US20170300839A1 (en) * | 2007-09-28 | 2017-10-19 | Great-Circle Technologies, Inc. | Bundling of automated work flow |
| CN107730458A (en) * | 2017-09-05 | 2018-02-23 | 北京飞搜科技有限公司 | A kind of fuzzy facial reconstruction method and system based on production confrontation network |
-
2018
- 2018-02-24 CN CN201810157009.1A patent/CN108230278B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170300839A1 (en) * | 2007-09-28 | 2017-10-19 | Great-Circle Technologies, Inc. | Bundling of automated work flow |
| CN106204499A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | Single image rain removing method based on convolutional neural networks |
| CN107730458A (en) * | 2017-09-05 | 2018-02-23 | 北京飞搜科技有限公司 | A kind of fuzzy facial reconstruction method and system based on production confrontation network |
Non-Patent Citations (2)
| Title |
|---|
| HAOFENG LI等: "Context-Aware Semantic Inpainting", 《COMPUTER VISION AND PATTERN RECOGNITION》 * |
| RUI QIAN等: "Attentive Generative Adversarial Network for Raindrop Removal from A Single Image", 《COMPUTER VISION AND PATTERN RECOGNITION》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109360156A (en) * | 2018-08-17 | 2019-02-19 | 上海交通大学 | Single image rain removing method based on the image block for generating confrontation network |
| CN110570358A (en) * | 2018-09-04 | 2019-12-13 | 阿里巴巴集团控股有限公司 | vehicle loss image enhancement method and device based on GAN network |
| CN109344778A (en) * | 2018-10-10 | 2019-02-15 | 成都信息工程大学 | Based on the unmanned plane road extraction method for generating confrontation network |
| CN109447918A (en) * | 2018-11-02 | 2019-03-08 | 北京交通大学 | Removing rain based on single image method based on attention mechanism |
| CN109472757A (en) * | 2018-11-15 | 2019-03-15 | 央视国际网络无锡有限公司 | It is a kind of that logo method is gone based on the image for generating confrontation neural network |
| CN109760635A (en) * | 2019-01-08 | 2019-05-17 | 同济大学 | A kind of line traffic control windshield wiper control system based on GAN network |
| CN109886975A (en) * | 2019-02-19 | 2019-06-14 | 武汉大学 | It is a kind of that raindrop method and system is gone based on the image optimization processing for generating confrontation network |
| CN110472467A (en) * | 2019-04-08 | 2019-11-19 | 江西理工大学 | The detection method for transport hub critical object based on YOLO v3 |
| CN110163813A (en) * | 2019-04-16 | 2019-08-23 | 中国科学院深圳先进技术研究院 | A kind of image rain removing method, device, readable storage medium storing program for executing and terminal device |
| CN110163813B (en) * | 2019-04-16 | 2022-02-01 | 中国科学院深圳先进技术研究院 | Image rain removing method and device, readable storage medium and terminal equipment |
| CN110189272A (en) * | 2019-05-24 | 2019-08-30 | 北京百度网讯科技有限公司 | For handling the method, apparatus, equipment and storage medium of image |
| CN110363793A (en) * | 2019-07-24 | 2019-10-22 | 北京华捷艾米科技有限公司 | A kind of tracking and device of object |
| CN110363793B (en) * | 2019-07-24 | 2021-09-21 | 北京华捷艾米科技有限公司 | Object tracking method and device |
| CN110533581A (en) * | 2019-08-12 | 2019-12-03 | 广东工业大学 | A kind of raindrop removing method, system and storage medium based on raindrop probability graph |
| CN112581377A (en) * | 2019-09-30 | 2021-03-30 | 上海商汤临港智能科技有限公司 | Image processing method and device and electronic equipment |
| CN112581377B (en) * | 2019-09-30 | 2024-06-11 | 上海商汤临港智能科技有限公司 | Image processing method and device and electronic equipment |
| CN117422689A (en) * | 2023-10-31 | 2024-01-19 | 南京邮电大学 | Rainy day insulator defect detection method based on improved MS-PReNet and GAM-YOLOv7 |
| CN117422689B (en) * | 2023-10-31 | 2024-05-31 | 南京邮电大学 | Rainy day insulator defect detection method based on improved MS-PReNet and GAM-YOLOv7 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108230278B (en) | 2021-08-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108230278A (en) | A kind of image based on generation confrontation network goes raindrop method | |
| CN103971329B (en) | A kind of multisource image anastomosing method based on genetic optimization cell neural network | |
| CN110458844A (en) | A kind of semantic segmentation method of low illumination scene | |
| CN108416755A (en) | A kind of image de-noising method and system based on deep learning | |
| CN109508650A (en) | A kind of wood recognition method based on transfer learning | |
| CN108765344A (en) | A method of the single image rain line removal based on depth convolutional neural networks | |
| CN108510194A (en) | Air control model training method, Risk Identification Method, device, equipment and medium | |
| CN107016406A (en) | The pest and disease damage image generating method of network is resisted based on production | |
| CN111476249B (en) | Construction method of multi-scale large-receptive-field convolutional neural network | |
| CN111681178B (en) | Knowledge distillation-based image defogging method | |
| CN109635511A (en) | A kind of high-rise residential areas forced-ventilated schemes generation design method generating confrontation network based on condition | |
| CN107680077A (en) | A kind of non-reference picture quality appraisement method based on multistage Gradient Features | |
| CN109063724A (en) | A kind of enhanced production confrontation network and target sample recognition methods | |
| CN111242841A (en) | Image background style migration method based on semantic segmentation and deep learning | |
| CN110458060A (en) | A kind of vehicle image optimization method and system based on confrontation study | |
| CN104504007B (en) | The acquisition methods and system of a kind of image similarity | |
| CN109146061A (en) | The treating method and apparatus of neural network model | |
| CN107103285A (en) | Face depth prediction approach based on convolutional neural networks | |
| CN109087375A (en) | Image cavity fill method based on deep learning | |
| CN113807215B (en) | Tea tender shoot grading method combining improved attention mechanism and knowledge distillation | |
| CN108447048B (en) | Convolutional neural network image feature processing method based on attention layer | |
| CN109584162A (en) | A method of based on the image super-resolution reconstruct for generating network | |
| CN108235003A (en) | Three-dimensional video quality evaluation method based on 3D convolutional neural networks | |
| CN109284812A (en) | A kind of video-game analogy method based on improvement DQN | |
| Jiang et al. | Application of back propagation neural network in the classification of high resolution remote sensing image: take remote sensing image of Beijing for instance |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |