CN111033532B - Training method and system for generating countermeasure network, electronic device and storage medium - Google Patents
Training method and system for generating countermeasure network, electronic device and storage medium Download PDFInfo
- Publication number
- CN111033532B CN111033532B CN201980002600.2A CN201980002600A CN111033532B CN 111033532 B CN111033532 B CN 111033532B CN 201980002600 A CN201980002600 A CN 201980002600A CN 111033532 B CN111033532 B CN 111033532B
- Authority
- CN
- China
- Prior art keywords
- network
- generating
- networks
- training
- picture
- 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.)
- Active
Links
- 238000012549 training Methods 0.000 title claims abstract description 108
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000003860 storage Methods 0.000 title claims description 7
- 238000012545 processing Methods 0.000 claims description 9
- 238000010586 diagram Methods 0.000 description 8
- 238000004519 manufacturing process Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 230000006870 function Effects 0.000 description 6
- 230000009471 action Effects 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 230000008485 antagonism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003042 antagnostic effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a training method and system for generating an countermeasure network. The generation countermeasure network includes a plurality of generation networks and a plurality of discrimination networks, each of the generation networks connecting the plurality of discrimination networks, the method comprising: generating a picture based on the generation network; judging the probability that the picture is a real picture or a generated picture by utilizing a plurality of judging networks connected to the generating network; selecting a designated discrimination network corresponding to the generation network according to the judged probability of the real picture or the generated picture; and updating parameters of the plurality of generating networks according to the specified judging network.
Description
Technical Field
The embodiment of the disclosure relates to the field of deep learning, in particular to a training method and system for generating an countermeasure network, electronic equipment and a storage medium.
Background
The generation of the antagonism network (GAN, generative Adversarial Networks) is a deep learning model, and is one of the most promising methods for unsupervised learning on complex distributions in recent years. The model is built up of (at least) two modules in a frame: the mutual game learning of the generating network and the discriminating network produces a fairly good output.
At present, generation of countermeasure networks has a wide range of applications including, for example, image generation, image conversion, watermarking, semantic segmentation, high resolution picture generation, and the like. However, at least two problems remain in current applications:
1, convergence is difficult when generating countermeasure network training; a large amount of training data is required to be used, for example, pictures are generated based on high-dimensional vectors and are trained;
2, generating a high-resolution picture is difficult; the precision of the production countermeasure network needs to be continuously improved, and when the precision of the model is high enough, the generated high-resolution picture can meet the requirement.
The above description of the discovery process of the problem is merely for aiding in understanding the technical solution of the present disclosure, and does not represent an admission that the above is prior art.
Disclosure of Invention
To solve at least one problem with the prior art, at least one embodiment of the present disclosure provides a training method of generating an countermeasure network including a plurality of generating networks and a plurality of discriminating networks, each of the generating networks connecting the plurality of discriminating networks, the method comprising:
generating a picture based on the generation network;
judging the probability that the picture is a real picture or a generated picture by utilizing a plurality of judging networks connected to the generating network;
selecting a designated discrimination network corresponding to the generation network according to the probability of the real picture or the generated picture judged by each discrimination network;
and updating parameters of the plurality of generating networks according to the specified judging network.
In an embodiment of the training method for generating the countermeasure network, the plurality of generating networks have the same structure and different initial parameters; the plurality of judgment networks have the same structure and different initial parameters.
In an embodiment of the training method for generating an countermeasure network, before the step of generating a picture based on the generation network, the method further comprises:
judging whether the generated countermeasure network meets a training target or not;
wherein the training target comprises: and for each generation network of the generation countermeasure network, any one of the connected judgment networks converges the probability that the judged picture is a real picture or the probability that the picture is generated.
In an embodiment of the training method for generating the countermeasure network, the specified discrimination network is determined based on the picture generated by the updated generation network.
In an embodiment of the training method for generating the countermeasure network, the step of selecting the designated discrimination network corresponding to the generation network according to the judged probability of the real picture or the generation picture includes:
and selecting a discrimination network with the lowest true picture probability or highest generated picture probability as a designated discrimination network.
In an embodiment of the training method for generating the countermeasure network, the number of the generating networks and the number of the discriminating networks in the generating countermeasure network are two or more.
In an embodiment of the training method for generating the countermeasure network, when one generating network meets a training target, a next generating network is trained for the plurality of generating networks.
In an embodiment of the training method for generating the countermeasure network, for the plurality of generating networks, updating parameters for each generating network are iterated in turn until the plurality of generating networks are all satisfied with the training target.
In an embodiment of the training method for generating an countermeasure network, the method further comprises:
and training the to-be-trained generating network by utilizing the generated countermeasure network after the training is completed.
In an embodiment of the training method for generating an countermeasure network, the step of training the generating network to be trained using the generated countermeasure network after completion of training includes:
calculating a Loss value between each generating network and the generating network to be trained to obtain a plurality of Loss values;
and training the to-be-trained generation network according to the plurality of Loss values, so that the sum of the plurality of Loss values converges to the minimum.
In an embodiment of the training method of generating countermeasure networks, each discrimination network is connected to only one generation network.
Another aspect of the present disclosure also proposes a computing processing device comprising: a processor and a memory; the processor is configured to execute the foregoing steps of generating the training method for the countermeasure network by calling the program or instructions stored in the memory.
Another aspect of the present disclosure also proposes a non-transitory computer-readable storage medium storing a program or instructions that cause a computer to perform the steps of the aforementioned training method of generating an countermeasure network.
It can be seen that at least one embodiment of the present disclosure provides a training scheme for generating an countermeasure network, so as to achieve the effects of improving the convergence speed of the generating countermeasure network and reducing the convergence difficulty.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings used in the embodiments or the description of the prior art will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present disclosure, and other drawings may be obtained according to these drawings to those of ordinary skill in the art.
Fig. 1 is a flowchart of a training method for generating an countermeasure network according to an embodiment of the present disclosure.
Fig. 2A and 2B are schematic block diagrams of two types of generation countermeasure networks provided in embodiments of the present disclosure.
Fig. 3 is a flowchart of a training method for generating an countermeasure network according to a second embodiment of the present disclosure.
Fig. 4 is a flowchart of the sub-steps of step S106 of the second embodiment of the present disclosure.
Fig. 5 is a block diagram of a system for generating an countermeasure network training provided by an embodiment of the present disclosure.
Fig. 6 is a block diagram of a computing processing device provided by an embodiment of the present disclosure.
Detailed Description
In order that the above-recited objects, features and advantages of the present disclosure may be more clearly understood, a more particular description of the disclosure will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings. It is to be understood that the described embodiments are some, but not all, of the embodiments of the present disclosure. The specific embodiments described herein are to be considered in an illustrative rather than a restrictive sense. All other embodiments derived by a person of ordinary skill in the art based on the described embodiments of the present disclosure fall within the scope of the present disclosure.
It should be noted that in this document, relational terms such as "first" and "second" and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions.
Aiming at the problems that the convergence of the generated countermeasure network is slow and the convergence difficulty is increased in the prior art, the embodiment of the disclosure provides a training scheme for generating the countermeasure network, and achieves the effects of improving the convergence speed of the generated countermeasure network and reducing the convergence difficulty, so that a foundation is provided for generating a better effect of the countermeasure network in application.
In the training method for generating an countermeasure network according to an embodiment of the present disclosure, the generating countermeasure network includes a plurality of generating networks and a plurality of discriminating networks, each of the generating networks is connected to the plurality of discriminating networks, fig. 1 is a flowchart illustrating a training method for generating an countermeasure network according to an embodiment of the present disclosure, and as shown in fig. 1, the method may include the following steps:
s101, generating a picture based on the generation network;
s102, judging the probability that the picture is a real picture or a generated picture by utilizing a plurality of judging networks connected to the generating network;
s103, selecting a designated discrimination network corresponding to the generation network according to the judged probability of the real picture or the generated picture;
s104, updating parameters of the plurality of generating networks according to the specified judging network;
these steps are described in detail below.
Step S101, generating a picture based on the generation network;
fig. 2A is a schematic diagram illustrating generation of an antagonizing network 100a according to an embodiment of the present disclosure. As described above, the generation countermeasure network 100a includes N generation networks and M discrimination networks. In fig. 2A, N production networks are represented by production networks 110, 120, 130, and M production networks are represented by three discrimination networks 210, 220, 230. In some embodiments, when the generating countermeasure network 100a is trained, the generating network 200 may be trained using the generating countermeasure network 100a, and the generating network 200 may be used for subsequent processing after the generating countermeasure network 100a is trained. In some embodiments, training of the generating network 200 may also employ the discrimination network 500 for auxiliary updating, and the training process of the generating network 200 may be more efficiently completed by the auxiliary updating of the discrimination network 500.
Each of the N generation networks is connected to M discrimination networks. That is, the generation network 110 as shown in fig. 2A is connected to the three discrimination networks 210, 220, and 230 shown, and the generation network 120 and the generation network 130 are also connected to the three discrimination networks 210, 220, and 230 shown, respectively.
In step S101, each of the generation networks 110, 120, 130 of the generation countermeasure network 100a generates pictures, respectively, which may be pictures generated based on random vectors, referred to as "generated pictures"; correspondingly, the picture actually photographed may become a "real picture". The generated pictures can be used as training data, input into the generated countermeasure network, and are used together with a plurality of real pictures for training the generated countermeasure network, and through continuous iteration, the judging network in the generated countermeasure network can judge that the received pictures are real pictures and generated pictures more accurately, so that feedback is carried out on the generated network, and through continuous countermeasure iteration, the generated network can generate pictures with spurious reality, and the training target is completed.
The training target can be a target which is preset to generate whether the picture generated by the countermeasure network meets the requirement. The sum of the probability that a picture is a real picture and the probability of generating a picture is 1, and the probability that a predicted picture is a real picture or the probability of generating a picture can be close to 0.5 due to the convergence property of the function. When it is determined that the training target is not reached, the foregoing steps S101 to S104 may be continuously performed until it is determined that the probability of being a real picture or generating a picture satisfies the requirement.
After performing step S101, step S102 may be performed as follows:
s102, judging the probability that the picture is a real picture or a generated picture by utilizing a plurality of judging networks connected to the generating network;
in an embodiment, the sum of the probability that the picture is a real picture and the probability of generating a picture is 1.
In this step, the discrimination network may discriminate whether the picture is a real picture or a generated picture, and output a probability of being a real picture or a probability of being a generated picture. Each discrimination network makes decisions independently, respectively probability values.
In some embodiments, the input content of the discrimination network is only derived from the generation network, the pictures received by the discrimination network are all generated pictures, and for a certain discrimination network, when the discrimination network judges that the picture is generated, the higher the probability that the discrimination network judges that the discrimination network is accurate, the parameters of the generation network can be modified by taking the discrimination network as a reference in a back propagation way, so that the generation network is optimized, and the pictures which cannot be accurately judged by the discrimination network are generated. The plurality of generating networks can independently generate pictures at the same time, and send the pictures to a plurality of judging networks connected with the pictures, and the pictures are respectively judged by the plurality of judging networks. As shown in fig. 2A, the multiple discrimination networks 210, 220, and 230 may, for example, first discriminate the picture generated by the generating network 110 to obtain multiple probabilities respectively, and then discriminate the picture generated by the generating network 120 to obtain multiple probabilities respectively; finally, the pictures generated by the generation network 130 are discriminated to obtain a plurality of probabilities respectively.
After performing step S102, step S103 may be performed as follows:
s103, selecting a designated discrimination network corresponding to the generation network according to the judged probability of the real picture or the generated picture;
in this step, when one discrimination network judges that the probability of the picture being the generated picture is higher or the probability value of the picture being the true picture is lower, the judgment result of the discrimination network is proved to be more accurate, the discrimination network is taken as the designated discrimination network corresponding to the generated network.
After performing step S103, step S104 may be performed as follows:
s104, updating parameters of the plurality of generating networks according to the specified judging network;
in this step, after the specified discrimination network is determined, the network parameters of the plurality of generation networks may be back-propagated updated according to the probability of the specified discrimination network, for example, after the specified discrimination network is determined, the parameters are modified for the other N-1 generation networks using the probability of the specified discrimination network, so that the pictures generated by the generation networks are further falsified.
For example, suppose the generation network is g (z), where z is a random noise, and the generation network converts this random noise into data type x. Using the picture generation scene example, the output of the generation network is a picture. For a discrimination network, the output of any input x, D (x) is a real number in the range of 0-1, which is used to determine how large the picture is a real picture or the probability of generating a picture is. Let Pr and Pg represent the distribution of the real image and the distribution of the generated image respectively, and the objective function of the discrimination network is obtained as follows:
the goal of a similar generating network is to make the discriminating network unable to distinguish between the real picture and the generated picture, then the overall optimization objective function is as follows:
the optimal mode of the maximum and minimum objective functions is various, and the most intuitive processing method can be to perform interactive iteration on the judging network parameters and the generating network parameters respectively, fixedly generate the network parameters, optimize the judging network parameters, and then fixedly optimize the judging network parameters to generate the network parameters until the process converges.
After executing step S104, when it is determined that the preset condition is not met, the operation of generating the picture in step S101 may be continuously executed by using the updated generating network, and the image may be sent to the plurality of discriminating networks to perform the determination, and a new discriminating network may be selected. After one iteration, the probability of judging that the picture is output by the judging network for generating the picture is reduced to 60%, but the critical value is not reached, and then a new iteration can be continued until the probability of judging the network output is continuously reduced.
Further, for a plurality of generating networks in the generating countermeasure network, the generating networks 110 to 130 train one by one in turn, after one generating network trains to be completed (i.e., the probability of the real picture or the generated picture output by each connected discriminating network converges to, for example, 0.5), the next generating network trains again, or the generating networks 110 to 130 train one by one until the probability of the output by the plurality of discriminating networks connected to the plurality of generating networks converges to 0.5.
The generation countermeasure network 100a as shown in fig. 2A trains a next generation network when one generation network satisfies a training target for a plurality of generation networks therein. For example, training may be performed on the generating network 110, multiple probabilities may be obtained by using multiple discriminating networks 210-230, respectively, a designated discriminating network may be selected, the parameters of the generating network 110 may be updated by back propagation, the designated discriminating network may be selected based on the picture output by the updated generating network 110, and the parameters of the generating network 110 may be updated. Until the training of the generating network 110 is completed, then training is performed on the generating network 120 by using the plurality of judging networks 210-230 until the training of the generating network 120 is completed, then training is performed on the generating network 130 until the training of the generating network 130 is completed, and then the training target of the generating countermeasure network 100a is completed.
Likewise, for multiple ones of the generation countermeasure networks 100a, the generation networks 110-130 may also be trained in turn before the training objectives are not met, iteratively updating parameters for each generation network in turn until the multiple generation networks are met that each meet the training objectives. For example, the generating network 110 may be trained first, multiple probabilities are obtained by using multiple discriminating networks 210-230, respectively, a designated discriminating network is selected, the parameters of the generating network 110 are updated in a back propagation manner, then the designated discriminating network is selected based on the picture output by the updated generating network 110, the generating network 120 is trained after the update, multiple probabilities are obtained by using multiple discriminating networks 210-230, a designated discriminating network is selected, the parameters of the generating network 120 are updated in a back propagation manner, and then the designated discriminating network is selected based on the picture output by the updated generating network 130, so the training is circulated until the generating countermeasure network 100b completes the training objective.
In other embodiments, the discrimination network to which each generation network is connected is different from the discrimination networks to which other generation networks are connected. Fig. 2B is a schematic diagram of another embodiment of the present invention for generating an countermeasure network 100B. As shown in fig. 2B, the number of discrimination networks may be N times that of the generation network, each discrimination network being connected to a unique generation network. As shown in fig. 2B, discrimination networks 210-230 are connected to generation network 110, discrimination networks 310-330 are connected to generation network 120, and discrimination networks 410-430 are connected to generation network 130. In training for generating the countermeasure network 100b, each of the generating networks 110-130 may be trained independently until the probability value output by each of the discriminating networks 210-430 in the generating countermeasure network 100b converges to 0.5.
In addition to the generation of the countermeasure network 100B, the generation network 200 and the discrimination network 500 to which the generation countermeasure network 100B is connected are also shown in fig. 2B. In some embodiments, when the generating countermeasure network 100b is trained, the generating network 200 may be trained using the generating countermeasure network 100b, and the generating network 200 may be used for subsequent processing after the generating countermeasure network 100b is trained. In some embodiments, training of the generating network 200 may also employ the discrimination network 500 for auxiliary updating, and the training process of the generating network 200 may be more efficiently completed by the auxiliary updating of the discrimination network 500.
As can be seen from the foregoing, the embodiment of the present disclosure provides a training method for generating an countermeasure network, which aims at the problems of slow convergence and high convergence difficulty of the countermeasure network generated in the prior art, and provides a training scheme for generating the countermeasure network.
FIG. 3 is a flowchart of generating a challenge model according to another embodiment of the present invention, as shown in FIG. 3, in an alternative embodiment, step S100 may be performed as follows before step S101 of the present invention:
s100, judging whether the generated countermeasure network meets the training target.
In this step, a determination may be made as to whether the generated countermeasure network meets the training objectives. As described above, the generated countermeasure network model is confirmed to have a convergence property, and when the condition that the discrimination value of each discrimination network converges to 0.5 is not satisfied, training of the generated countermeasure network may be continued until the training target of the generated countermeasure network is satisfied by a plurality of iterations, that is, the probability that the connected discrimination network judges that the picture is a true picture or that the generated picture converges to 0.5 for each picture output by the generated network, and after judging that the convergence to 0.5 is satisfied, the generated countermeasure network training is considered to be qualified, and the training is stopped.
Thus, in an alternative embodiment, the method may further comprise the steps of:
and S105, when judging that the probability of the judged picture being a real picture is close to the probability of the picture generation for each judgment network of the network generation countermeasure, ending the training of the network generation countermeasure.
In this step, when it is judged that the probability of the real picture or the generated picture output by each discrimination network is close to 0.5 and converges to 0.5 (for example, the probability of the generation network is not more than a specified range up and down around 0.5), the generation network is considered to have sufficiently converged, at which time the training for the generation of the antagonism network is ended.
In an alternative embodiment, the plurality of generating networks have the same structure and different initial parameters; the plurality of judgment networks have the same structure and different initial parameters.
The same structure here may include: the structure of the generated network is the same, and the structure of the discrimination network is the same.
For example, each discrimination network is a convolutional neural network model (CNN) having the same layer structure, each generation network is similar to but opposite to the convolutional neural network model (CNN), and these generation networks also have the same layer structure with each other.
In an alternative embodiment, the steps further comprise:
and S106, training the to-be-trained generating network by using the generated countermeasure network after training.
As shown in fig. 2A, after the generation of the reactive network training is completed, the generation network to be trained 200 is trained based on the trained generation network. For example, a Loss value between each generating network and the generating network to be trained 200 may be calculated, to obtain N Loss values. The Loss value refers to a distance between the generated result of the generated network and the generated result of the generated network 200.
The generating network 200 is trained based on the sum of the N Loss values such that the sum of the N Loss values converges to a minimum.
Thus, in an alternative embodiment, as shown in fig. 4, the step S106, that is, the step of training the generating network to be trained using the generated challenge network after completion of training, may include the following sub-steps:
s1061, calculating a Loss value between each generating network and the generating network to be trained to obtain a plurality of Loss values;
and S1062, training the to-be-trained generation network according to the plurality of Loss values, so that the sum of the plurality of Loss values is converged to the minimum.
In the above substep, the Loss value refers to, for example, the distance between the generation results of the generation networks 210 to 230 and the generation result of the generation network 200 in fig. 2A. When training the generating network 200, the comprehensive distance between the picture of the generating network 200 and the generated pictures of the trained N generating networks 210-230 can be minimized, and the trained generating network 200 can be obtained.
In some embodiments, as shown in FIG. 2A, a discrimination network 500 is connected to the generation network 200 for auxiliary updating of the generation network 200.
In an alternative embodiment, the number of the generating networks and the number of the discriminating networks are two or more. In some embodiments, the number of discriminating networks is n times the number of generating networks, wherein the number of generating networks is greater than 2, and n is a natural number.
Fig. 5 is a block diagram of a system 300 for generating an countermeasure network training provided by an embodiment of the present disclosure.
As shown in fig. 5, generating the countermeasure network training system 300 may include, but is not limited to, the following elements: a generation unit 301 for generating a picture based on a generation network; a judging unit 302 for judging a probability that the picture is a real picture or a generated picture using a plurality of judging networks connected to the generating network; a selecting unit 303, configured to select a specified discrimination network corresponding to the generation network according to the judged probability of the real picture or the generated picture; an updating unit 304, configured to update parameters of the plurality of generating networks according to the specified discrimination network.
The generation unit 301 is configured to generate a picture based on a generation network. In the schematic diagram of the generation of the countermeasure network 100a according to an embodiment of the present disclosure as shown in fig. 2A, the generation of the countermeasure network 100a includes N generation networks and M discrimination networks. In fig. 2A, N production networks are represented by production networks 110, 120, 130, and M production networks are represented by three discrimination networks 210, 220, 230.
Each of the N generation networks is connected to M discrimination networks. That is, the generation network 110 as shown in fig. 2A is connected to the three discrimination networks 210, 220, and 230 shown, and the generation network 120 and the generation network 130 are also connected to the three discrimination networks 210, 220, and 230 shown, respectively.
Each of the generation countermeasure networks 100a generates a respective picture, which may be a picture generated based on random vectors, referred to as a "generated picture" as training data to indicate that they are randomly generated pictures rather than real pictures. In contrast, a picture actually taken may be referred to as a "real picture". These generated pictures are input as training data to the generating countermeasure network 100a, and are used together with a plurality of real pictures to train the generating countermeasure network, so that the discrimination network in the generating countermeasure network can more accurately judge the real pictures and the generated pictures through continuous iteration, thereby feeding back the generating network, and enabling the generating network to generate the false and spurious pictures.
As mentioned above, the "training target" may be a target that is preset to generate a picture of the countermeasure network to meet the requirement. In one embodiment, due to the converging nature of the function, the training goal to generate the countermeasure network may be, for example, to discriminate that the probability predicted by the network meets a specified requirement, e.g., is approximately 0.5. When the training target is judged to be not reached, the cyclic training can be continued until the probability is judged to meet the requirement.
The judging unit 302 is configured to judge a probability that the picture is a real picture or a generated picture using a plurality of discrimination networks connected to the generating network. The discriminating network can discriminate whether the characteristic value of the picture is like or real. Each discrimination network can output a discrimination probability value with the characteristic value of the picture being like. As the characteristic values of the pictures are fakes, for a certain judging network, the higher the probability that the judging network judges that the picture is the fake, the more accurate the judging of the judging network is proved.
The selecting unit 303 is configured to select a specified discrimination network corresponding to the generation network according to the judged probability of the real picture or the generated picture.
When one discrimination network judges that the probability of the picture being the generated picture is higher or the probability value of the picture being the true picture is lower, the discrimination network is used as the appointed discrimination network corresponding to the generated network if the judgment result of the discrimination network is more accurate.
The updating unit 304 is configured to, after determining the specified discrimination network, perform back propagation updating on network parameters of the plurality of generating networks according to the probability of the specified discrimination network, for example, modify parameters for the other N-1 generating networks using the probability of the specified discrimination network after determining the specified discrimination network, so that pictures generated by the generating networks are further spurious.
In some embodiments, the division of each unit in the countermeasure network training system 300 is only one logic function division, and other division manners may be implemented in actual implementation, for example, the generating unit 301, the judging unit 302, the selecting unit 303 and the updating unit 304 may be implemented as one unit; the generating unit 301, the judging unit 302, the selecting unit 303, or the updating unit 304 may also be divided into a plurality of sub-units. It is understood that each unit or sub-unit can be implemented in electronic hardware, or in combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Those skilled in the art can implement the described functionality using different methods for each particular application.
Fig. 6 is a schematic structural diagram of a computing processing device according to an embodiment of the present disclosure. The computing processing device includes: at least one processor 401, at least one memory 402, and at least one communication interface 403. The various components in the in-vehicle device are coupled together by a bus system 404. A communication interface 403 for information transmission with an external device. It is appreciated that the bus system 404 serves to facilitate connected communications between these components. The bus system 404 includes a power bus, a control bus, and a status signal bus in addition to the data bus. But for clarity of illustration the various buses are labeled as bus system 404 in fig. 6.
It will be appreciated that the memory 402 in this embodiment can be either volatile memory or nonvolatile memory, or can include both volatile and nonvolatile memory.
In some implementations, the memory 402 stores the following elements, executable units or data structures, or a subset thereof, or an extended set thereof: an operating system and application programs.
The operating system includes various system programs, such as a framework layer, a core library layer, a driving layer, and the like, and is used for realizing various basic services and processing hardware-based tasks. Applications, including various applications such as Media Player (Media Player), browser (Browser), etc., are used to implement various application services. A program for implementing the training method for generating an countermeasure network provided by the embodiment of the present disclosure may be included in an application program.
In the embodiment of the present disclosure, the processor 401 is configured to execute the steps of each embodiment of the training method for generating the countermeasure network provided in the embodiment of the present disclosure by calling the program or the instruction stored in the memory 402, specifically, the program or the instruction stored in the application program.
The training method for generating an countermeasure network provided by the embodiments of the present disclosure may be applied to the processor 401 or implemented by the processor 401. The processor 401 may be an integrated circuit chip having signal processing capability. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in the processor 401 or by instructions in the form of software. The processor 401 described above may be a general purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), an off-the-shelf programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The steps of the training method for generating the countermeasure network provided in the embodiments of the present disclosure may be directly embodied and executed by a hardware decoding processor, or may be executed by a combination of hardware and software units in the decoding processor. The software elements may be located in a random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers, etc. as well known in the art. The storage medium is located in a memory 402 and the processor 401 reads the information in the memory 402 and in combination with its hardware performs the steps of the method.
It should be noted that, for simplicity of description, the foregoing method embodiments are all expressed as a series of combinations of actions, but those skilled in the art can appreciate that the disclosed embodiments are not limited by the order of actions described, as some steps may occur in other orders or concurrently in accordance with the disclosed embodiments. In addition, those skilled in the art will appreciate that the embodiments described in the specification are all alternatives.
Embodiments of the present disclosure also provide a non-transitory computer readable storage medium storing a program or instructions that cause a computer to perform steps such as generating embodiments of a training method for an countermeasure network, and are not described in detail herein to avoid repetition of the description.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.
Those skilled in the art will appreciate that while some embodiments described herein include some features but not others included in other embodiments, combinations of features of different embodiments are meant to be within the scope of the disclosure and form different embodiments.
Those skilled in the art will appreciate that the descriptions of the various embodiments are each focused on, and that portions of one embodiment that are not described in detail may be referred to as related descriptions of other embodiments.
Although embodiments of the present disclosure have been described with reference to the accompanying drawings, various modifications and variations may be made by those skilled in the art without departing from the spirit and scope of the disclosure, and such modifications and variations fall within the scope defined by the appended claims.
Industrial applicability
At least one embodiment of the present disclosure provides a training scheme for generating an countermeasure network, which achieves the effects of improving convergence speed of the generated countermeasure network and reducing convergence difficulty, and has industrial applicability.
Claims (13)
1. A training method of generating an countermeasure network, the generating countermeasure network including a plurality of generating networks and a plurality of discriminating networks, each of the generating networks being connected to a plurality of discriminating networks, each of the discriminating networks being connected to a plurality of the generating networks, the method comprising:
generating a picture based on the generation network;
judging the probability that the picture is a real picture or a generated picture by utilizing a plurality of judging networks connected to the generating network;
selecting a designated discrimination network corresponding to the generation network according to the probability of the real picture or the generated picture judged by each discrimination network;
and updating parameters of the plurality of generating networks according to the specified judging network.
2. The method of training for generating an countermeasure network of claim 1, wherein the plurality of generating networks have a same structure and different initial parameters; the plurality of discrimination networks have the same structure and different initial parameters.
3. The method of training for generating an countermeasure network of claim 1, wherein prior to the step of generating pictures based on the generation network, the method further comprises:
judging whether the generated countermeasure network meets a training target or not;
wherein the training target comprises: and for each generation network of the generation countermeasure network, any one of the connected judgment networks converges the probability that the judged picture is a real picture or the probability that the picture is generated.
4. The method of training for generating an countermeasure network of claim 1,
the specified discrimination network is determined based on the picture generated by the updated generation network.
5. The training method for generating an countermeasure network according to claim 1, wherein the step of selecting a specified discrimination network corresponding to the generation network in accordance with the judged probability of the true picture or the generation picture includes:
and selecting a discrimination network with the lowest true picture probability or highest generated picture probability as a designated discrimination network.
6. The training method for generating an countermeasure network according to claim 1, wherein the number of the generating networks and the number of the discriminating networks in the generating countermeasure network are two or more.
7. The method of training a generation of countermeasure networks of claim 1, wherein for the plurality of generation networks, when one generation network meets a training goal, a next generation network is trained.
8. The training method for generating a countermeasure network according to claim 1, wherein, for the plurality of generating networks, updating parameters for each generating network are iterated in turn until the plurality of generating networks are satisfied, all satisfying a training target.
9. Training method for generating an countermeasure network according to any of claims 1-8, characterized in that the method further comprises:
and training the to-be-trained generating network by utilizing the generated countermeasure network after the training is completed.
10. A method of training a generation network in accordance with claim 9, wherein the step of training the generation network to be trained with the training-completed generation network comprises:
calculating a Loss value between each generating network and the generating network to be trained to obtain a plurality of Loss values;
and training the to-be-trained generation network according to the plurality of Loss values, so that the sum of the plurality of Loss values converges to the minimum.
11. Training method for generating countermeasure networks according to claim 1, characterized in that in the generating countermeasure networks each discrimination network is connected to only one generating network.
12. A computing processing device, comprising: a processor and a memory;
the processor is adapted to perform the steps of the method according to any of claims 1 to 11 by invoking a program or instruction stored in the memory.
13. A non-transitory computer readable storage medium storing a program or instructions that cause a computer to perform the steps of the method of any one of claims 1 to 11.
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2019/121028 WO2021102697A1 (en) | 2019-11-26 | 2019-11-26 | Method and system for training generative adversarial network, and electronic device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111033532A CN111033532A (en) | 2020-04-17 |
| CN111033532B true CN111033532B (en) | 2024-04-02 |
Family
ID=70198326
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201980002600.2A Active CN111033532B (en) | 2019-11-26 | 2019-11-26 | Training method and system for generating countermeasure network, electronic device and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111033532B (en) |
| WO (1) | WO2021102697A1 (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563995A (en) * | 2017-08-14 | 2018-01-09 | 华南理工大学 | A kind of confrontation network method of more arbiter error-duration models |
| CN109785258A (en) * | 2019-01-10 | 2019-05-21 | 华南理工大学 | A kind of facial image restorative procedure generating confrontation network based on more arbiters |
| CN109801292A (en) * | 2018-12-11 | 2019-05-24 | 西南交通大学 | A kind of bituminous highway crack image partition method based on generation confrontation network |
| CN110059465A (en) * | 2019-04-24 | 2019-07-26 | 腾讯科技（深圳）有限公司 | Auth method, confrontation generate training method, device and the equipment of network |
| CN110222705A (en) * | 2019-04-23 | 2019-09-10 | 华为技术有限公司 | A kind of training method and relevant apparatus of network model |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170249339A1 (en) * | 2016-02-25 | 2017-08-31 | Shutterstock, Inc. | Selected image subset based search |
| CN107220600B (en) * | 2017-05-17 | 2019-09-10 | 清华大学深圳研究生院 | A kind of Picture Generation Method and generation confrontation network based on deep learning |
| CN107392973B (en) * | 2017-06-06 | 2020-01-10 | 中国科学院自动化研究所 | Pixel-level handwritten Chinese character automatic generation method, storage device and processing device |
| CN107945204B (en) * | 2017-10-27 | 2021-06-25 | 西安电子科技大学 | Pixel-level image matting method based on generation countermeasure network |
-
2019
- 2019-11-26 WO PCT/CN2019/121028 patent/WO2021102697A1/en active Application Filing
- 2019-11-26 CN CN201980002600.2A patent/CN111033532B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563995A (en) * | 2017-08-14 | 2018-01-09 | 华南理工大学 | A kind of confrontation network method of more arbiter error-duration models |
| CN109801292A (en) * | 2018-12-11 | 2019-05-24 | 西南交通大学 | A kind of bituminous highway crack image partition method based on generation confrontation network |
| CN109785258A (en) * | 2019-01-10 | 2019-05-21 | 华南理工大学 | A kind of facial image restorative procedure generating confrontation network based on more arbiters |
| CN110222705A (en) * | 2019-04-23 | 2019-09-10 | 华为技术有限公司 | A kind of training method and relevant apparatus of network model |
| CN110059465A (en) * | 2019-04-24 | 2019-07-26 | 腾讯科技（深圳）有限公司 | Auth method, confrontation generate training method, device and the equipment of network |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2021102697A1 (en) | 2021-06-03 |
| CN111033532A (en) | 2020-04-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113905391B (en) | Integrated learning network traffic prediction method, system, equipment, terminal and medium | |
| CN112418292B (en) | Image quality evaluation method, device, computer equipment and storage medium | |
| CN113011529B (en) | Training method, training device, training equipment and training equipment for text classification model and readable storage medium | |
| CN111625258B (en) | Mercker tree updating method, device, equipment and readable storage medium | |
| CN111784699B (en) | Method and device for carrying out target segmentation on three-dimensional point cloud data and terminal equipment | |
| CN111898750A (en) | Neural network model compression method and device based on evolutionary algorithm | |
| CN111680480A (en) | Template-based job approval method and device, computer equipment and storage medium | |
| CN116644804B (en) | Distributed training system, neural network model training method, device and medium | |
| CN112651436A (en) | Optimization method and device based on uncertain weight graph convolution neural network | |
| CN113467487A (en) | Path planning model training method, path planning device and electronic equipment | |
| CN114417739B (en) | Method and device for recommending process parameters under abnormal working conditions | |
| CN112329470B (en) | Intelligent address identification method and device based on end-to-end model training | |
| CN114297934A (en) | Model parameter parallel simulation optimization method and device based on proxy model | |
| CN111079930A (en) | Method and device for determining quality parameters of data set and electronic equipment | |
| CN110705622A (en) | Decision-making method and system and electronic equipment | |
| CN111033532B (en) | Training method and system for generating countermeasure network, electronic device and storage medium | |
| CN117829253A (en) | Offline reinforcement learning method, device, equipment and storage medium | |
| CN115412401B (en) | Method and device for training virtual network embedding model and virtual network embedding | |
| CN117010272A (en) | Decision generation method, device, computer equipment and medium based on reinforcement learning | |
| CN115601098A (en) | Sequence recommendation method and system based on coupling relation between article attributes and time sequence modes | |
| CN114329317B (en) | Output power scheduling method and device based on hybrid generator set and terminal | |
| CN114819096A (en) | Model training method and device, electronic equipment and storage medium | |
| CN115099875A (en) | Data classification method based on decision tree model and related equipment | |
| CN114707655A (en) | Quantum line conversion method, quantum line conversion system, storage medium and electronic equipment | |
| CN113239034A (en) | Big data resource integration method and system based on artificial intelligence and cloud platform |
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 |