CN108470208A - It is a kind of based on be originally generated confrontation network model grouping convolution method - Google Patents
It is a kind of based on be originally generated confrontation network model grouping convolution method Download PDFInfo
- Publication number
- CN108470208A CN108470208A CN201810101710.1A CN201810101710A CN108470208A CN 108470208 A CN108470208 A CN 108470208A CN 201810101710 A CN201810101710 A CN 201810101710A CN 108470208 A CN108470208 A CN 108470208A
- Authority
- CN
- China
- Prior art keywords
- convolution
- network model
- originally generated
- generated confrontation
- confrontation network
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Neurology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of based on the grouping convolution method for being originally generated confrontation network model, belongs to deep learning field of neural networks, includes the following steps:S1, construction are originally generated confrontation network model;S2, constructing neural network serve as the function of generator and arbiter;S3, initialization random noise, input in generator;S4, setting number of packet N, make the convolution of neural network be carried out on N number of GPU;S5, N number of characteristic pattern is merged, updates loss function, is subsequently trained.The depth convolution based on grouping convolution of this method structure fights network model, convolution is grouped, it is made to be carried out at the same time in multiple GPU, is finally merged the result of convolution, to greatly reduce parameter amount, the efficiency of whole network training is improved.
Description
Technical field
The present invention relates to deep learning nerual network technique fields, and in particular to one kind is based on being originally generated confrontation network mould
The grouping convolution method of type.
Background technology
It is by Goodfellow that production, which fights network (Generative Adversarial Network, abbreviation GAN),
The deep learning frame proposed in 2014, it is based on the thought of " game theory ", construction generator (generator) and arbiter
(discriminator) two kinds of models, the former generates image by the Uniform noise or gaussian random noise of input (0,1), after
Person differentiates the image of input, and determination is image from data set or the image generated by generator.
In traditional confrontation network model, all convolution operations all only carry out in single GPU, in such case
Under, the parameter amount of whole network training is very huge, is required for being updated huge parameter amount every time after training, training
Efficiency it is more low.
Invention content
The purpose of the present invention is to solve drawbacks described above in the prior art, provide one kind and are based on being originally generated confrontation net
The grouping convolution method of network model.
The purpose of the present invention can be reached by adopting the following technical scheme that:
It is a kind of based on be originally generated confrontation network model grouping convolution method, the grouping convolution method includes following
Step:
S1, construction are originally generated confrontation network model, generate image by generator and are input to arbiter progress network
Training;
The neural network of S2, construction comprising N number of convolution kernel serves as the function of generator and arbiter;
S3, initialization random noise, input in generator;
S4, setting number of packet N, make the convolution of neural network be carried out on N number of GPU;
S5, N number of characteristic pattern is merged, updates loss function, is subsequently trained.
Further, the quantity N of the convolution kernel and the GPU are the complexities according to data images feature
It is configured.
Further, the step S4 processes are as follows:
S41, setting number of packet N;
S42, it convolution is assigned on N number of GPU is carried out at the same time.
Further, the step S5 processes are as follows:
S51, each characteristic pattern being grouped after GPU convolution is collected;
S52, N number of characteristic pattern is merged, updates loss function, is subsequently trained.
Further, the expression formula of the loss function is:
Wherein, D (x) indicates that differentiation of the arbiter to image, pr indicate that the distribution of data images, pg indicate to generate image
Distribution, λ is hyper parameter,For gradient.
The present invention has the following advantages and effects with respect to the prior art:
It is disclosed by the invention that convolution is grouped based on the grouping convolution method for being originally generated confrontation network model,
So that it is carried out at the same time in multiple GPU, finally merge the result of convolution, to greatly reduce parameter amount, improves
The efficiency of whole network training has high efficiency compared to traditional confrontation network model.
Description of the drawings
Fig. 1 is the overall flow figure for being originally generated confrontation network in the present invention and being trained by empty convolution;
Fig. 2 is that convolutional layer is grouped to the schematic diagram for carrying out convolution.
Specific implementation mode
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
The every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Embodiment
As shown in Figure 1 and Figure 2, present embodiment discloses a kind of based on the grouping volume for being originally generated confrontation network model
Product method, specifically includes the following steps:
Step S1, construction is originally generated confrontation network model, generates image by generator and is input to arbiter progress
Network training.
Step S2, the neural network for constructing multiple convolution kernels serves as the function of generator and arbiter;
Different convolution kernels is embodied in the difference of matrix numerical value, the difference of ranks number.
The neural network for constructing multiple convolution kernels, during handling image, different convolution kernels is meant to
Different characteristic of the study to generation image during network training.
In traditional confrontation network model, the operation of arbiter and generator to image progress convolution, is all same
It is carried out on a GPU.In this case, it is huge, each undated parameter that whole network, which trains required parameter scale,
It needs to take a substantial amount of time, trained efficiency is more low, and this method is by the way of being grouped convolution, by the operation of convolution
It is carried out at the same time on multiple GPU, finally merges the result of each grouping convolution, to improve whole network training
Efficiency.
In practical applications, it should which, according to the complexity of data images feature, the number of convolution kernel is set.
Step S3, random noise is initialized, is inputted in generator.
Step S4, number of packet N is set, the convolution of neural network is made to be carried out on N number of GPU.
The specific method is as follows:
S41, setting number of packet N;
S42, it convolution is assigned on N number of GPU is carried out at the same time.
In specific training process, it should determine the quantity of grouping convolution according to the complexity of image in data set
N。
Step S5, N number of characteristic pattern is merged, updates loss function, is subsequently trained.Detailed process is as follows:
S51, by the characteristic pattern after convolution in step S4, input arbiter is differentiated;
S52, N number of characteristic pattern is merged, updates loss function, is subsequently trained.
The effect of loss function is the ability weighed arbiter and judged generating image.The value of loss function is smaller, explanation
In current iteration, arbiter can have the generation image of preferable performance discrimination generator;Property that is on the contrary then illustrating arbiter
It can be poor.
The expression formula of loss function is:
Wherein, D (x) indicates that differentiation of the arbiter to image, pr indicate that the distribution of data images, pg indicate to generate image
Distribution, λ is hyper parameter,For gradient.
In conclusion present embodiment discloses a kind of based on the grouping convolution method for being originally generated confrontation network model, phase
Than in traditional original confrontation network model, the present invention, which changes arbiter and receives, to be learnt characteristics of image after picture
Mode.In traditional confrontation network model, the operation of arbiter and generator to image progress convolution, is all same
It is carried out on GPU.In this case, it is huge that whole network, which trains required parameter scale, and each undated parameter needs
It takes a substantial amount of time, trained efficiency is more low, and this method is existed the operation of convolution by the way of being grouped convolution
It is carried out at the same time, finally merges the result of each grouping convolution, to improve the effect of whole network training on multiple GPU
Rate.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, it is other it is any without departing from the spirit and principles of the present invention made by changes, modifications, substitutions, combinations, simplifications,
Equivalent substitute mode is should be, is included within the scope of the present invention.
Claims (5)
1. a kind of based on the grouping convolution method for being originally generated confrontation network model, which is characterized in that the grouping convolution side
Method includes the following steps:
S1, construction are originally generated confrontation network model, generate image by generator and are input to arbiter progress network training;
The neural network of S2, construction comprising N number of convolution kernel serves as the function of generator and arbiter;
S3, initialization random noise, input in generator;
S4, setting number of packet N, make the convolution of neural network be carried out on N number of GPU;
S5, N number of characteristic pattern is merged, updates loss function, is subsequently trained.
2. according to claim 1 a kind of based on the grouping convolution method for being originally generated confrontation network model, feature exists
In the quantity N of the convolution kernel and the GPU are configured according to the complexity of data images feature.
3. according to claim 1 a kind of based on the grouping convolution method for being originally generated confrontation network model, feature exists
In the step S4 processes are as follows:
S41, setting number of packet N;
S42, it convolution is assigned on N number of GPU is carried out at the same time.
4. according to claim 1 a kind of based on the grouping convolution method for being originally generated confrontation network model, feature exists
In the step S5 processes are as follows:
S51, each characteristic pattern being grouped after GPU convolution is collected;
S52, N number of characteristic pattern is merged, updates loss function, is subsequently trained.
5. according to claim 1 a kind of based on the grouping convolution method for being originally generated confrontation network model, feature exists
In the expression formula of the loss function is:
Wherein, D (x) indicates that differentiation of the arbiter to image, pr indicate that the distribution of data images, pg indicate to generate point of image
Cloth, λ are hyper parameter,For gradient.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810101710.1A CN108470208A (en) | 2018-02-01 | 2018-02-01 | It is a kind of based on be originally generated confrontation network model grouping convolution method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810101710.1A CN108470208A (en) | 2018-02-01 | 2018-02-01 | It is a kind of based on be originally generated confrontation network model grouping convolution method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108470208A true CN108470208A (en) | 2018-08-31 |
Family
ID=63265997
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810101710.1A Pending CN108470208A (en) | 2018-02-01 | 2018-02-01 | It is a kind of based on be originally generated confrontation network model grouping convolution method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108470208A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109344879A (en) * | 2018-09-07 | 2019-02-15 | 华南理工大学 | A kind of decomposition convolution method fighting network model based on text-image |
CN109461458A (en) * | 2018-10-26 | 2019-03-12 | 合肥工业大学 | A kind of audio method for detecting abnormality based on generation confrontation network |
CN112016639B (en) * | 2020-11-02 | 2021-01-26 | 四川大学 | Flexible separable convolution framework and feature extraction method and application thereof in VGG and ResNet |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106997380A (en) * | 2017-03-21 | 2017-08-01 | 北京工业大学 | Imaging spectrum safe retrieving method based on DCGAN depth networks |
CN107563493A (en) * | 2017-07-17 | 2018-01-09 | 华南理工大学 | A kind of confrontation network algorithm of more maker convolution composographs |
CN107563510A (en) * | 2017-08-14 | 2018-01-09 | 华南理工大学 | A kind of WGAN model methods based on depth convolutional neural networks |
CN107862377A (en) * | 2017-11-14 | 2018-03-30 | 华南理工大学 | A kind of packet convolution method that confrontation network model is generated based on text image |
-
2018
- 2018-02-01 CN CN201810101710.1A patent/CN108470208A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106997380A (en) * | 2017-03-21 | 2017-08-01 | 北京工业大学 | Imaging spectrum safe retrieving method based on DCGAN depth networks |
CN107563493A (en) * | 2017-07-17 | 2018-01-09 | 华南理工大学 | A kind of confrontation network algorithm of more maker convolution composographs |
CN107563510A (en) * | 2017-08-14 | 2018-01-09 | 华南理工大学 | A kind of WGAN model methods based on depth convolutional neural networks |
CN107862377A (en) * | 2017-11-14 | 2018-03-30 | 华南理工大学 | A kind of packet convolution method that confrontation network model is generated based on text image |
Non-Patent Citations (4)
Title |
---|
ALEC RADFORD ET AL.: "UNSUPERVISED REPRESENTATION LEARNING WITH DEEP CONVOLUTIONAL GENERATIVE ADVERSARIAL NETWORKS", 《MARCHINE LEARNING》 * |
ANDREW BROCK ET AL.: "NEURAL PHOTO EDITING WITH INTROSPECTIVE ADVERSARIAL NETWORKS", 《MARCHINE LEARNING》 * |
ISHAAN GULRAJANI ET AL.: "Improved Training of Wasserstein GANs", 《MARCHINE LEARNING》 * |
王裕民: "多GPU环境下的卷积神经网络并行算法", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109344879A (en) * | 2018-09-07 | 2019-02-15 | 华南理工大学 | A kind of decomposition convolution method fighting network model based on text-image |
CN109461458A (en) * | 2018-10-26 | 2019-03-12 | 合肥工业大学 | A kind of audio method for detecting abnormality based on generation confrontation network |
CN109461458B (en) * | 2018-10-26 | 2022-09-13 | 合肥工业大学 | Audio anomaly detection method based on generation countermeasure network |
CN112016639B (en) * | 2020-11-02 | 2021-01-26 | 四川大学 | Flexible separable convolution framework and feature extraction method and application thereof in VGG and ResNet |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107862377A (en) | A kind of packet convolution method that confrontation network model is generated based on text image | |
CN110097178A (en) | It is a kind of paid attention to based on entropy neural network model compression and accelerated method | |
CN105023240B (en) | The dictionary class image super-resolution system and method rebuild based on iterative projection | |
CN107563493A (en) | A kind of confrontation network algorithm of more maker convolution composographs | |
CN107292319A (en) | The method and device that a kind of characteristic image based on deformable convolutional layer is extracted | |
CN108765344A (en) | A method of the single image rain line removal based on depth convolutional neural networks | |
CN107871142A (en) | A kind of empty convolution method based on depth convolution confrontation network model | |
CN105488776B (en) | Super-resolution image reconstruction method and device | |
CN109271933A (en) | The method for carrying out 3 D human body Attitude estimation based on video flowing | |
CN108460720A (en) | A method of changing image style based on confrontation network model is generated | |
CN107968962A (en) | A kind of video generation method of the non-conterminous image of two frames based on deep learning | |
CN108470208A (en) | It is a kind of based on be originally generated confrontation network model grouping convolution method | |
CN108470196A (en) | A method of handwritten numeral is generated based on depth convolution confrontation network model | |
CN109360159A (en) | A kind of image completion method based on generation confrontation network model | |
CN107590518A (en) | A kind of confrontation network training method of multiple features study | |
CN107886169A (en) | A kind of multiple dimensioned convolution kernel method that confrontation network model is generated based on text image | |
CN107944546A (en) | It is a kind of based on be originally generated confrontation network model residual error network method | |
CN108734169A (en) | One kind being based on the improved scene text extracting method of full convolutional network | |
CN104143186B (en) | A kind of SLIC super-pixel piecemeal optimization method | |
CN108021979A (en) | It is a kind of based on be originally generated confrontation network model feature recalibration convolution method | |
CN112508190A (en) | Method, device and equipment for processing structured sparse parameters and storage medium | |
CN109816615A (en) | Image processing method, device, equipment and storage medium | |
CN110533594A (en) | Model training method, image rebuilding method, storage medium and relevant device | |
CN103942571A (en) | Graphic image sorting method based on genetic programming algorithm | |
CN109344879A (en) | A kind of decomposition convolution method fighting network model based on text-image |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180831 |
|
RJ01 | Rejection of invention patent application after publication |