Task verification method and model verification method

Abstract

Embodiments of the present specification provide a task verification method and a model verification method, wherein the task verification method comprises: obtaining an image pair, wherein the image pair comprises a source domain image and a target domain image, the source domain image and the target domain image are images containing the same image content and having different image attributes; performing a target task on the source domain image by using a source task policy to obtain a first predicted image, and performing the target task on the source domain image by using a task policy to be verified to obtain a second predicted image; adjusting the source domain image according to the first predicted image and the target domain image to obtain a first verification image, and adjusting the source domain image according to the second predicted image and the target domain image to obtain a second verification image, with the goal of balancing the difficulty of task execution; and determining a task verification result according to the first verification image and the second verification image. The performance of the task policy itself is completely retained, and lossless task verification is achieved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present specification relate to the technical field of computer technology, and particularly relate to a task verification method. BACKGROUND

[0002] With the development of computer technology, image processing is widely used in medical image analysis, computer vision, remote sensing image processing, robot vision, security monitoring and other fields, and has important application value and research significance. Since the cost of designing a task strategy for a specific application scenario is often huge, unscrupulous users will use malicious software infection or internal leakage methods to avoid the expensive strategy generation process. Therefore, protecting the copyright of the task strategy has gradually become a research focus.

[0003] Currently, the copyright of the task strategy can be verified by watermarking. However, watermarking will intrusively embed verification information into the task strategy, which will change the performance of the task strategy itself and thus affect the task processing performance. Therefore, there is an urgent need for a task verification scheme that does not affect the task processing performance. SUMMARY

[0004] Therefore, the embodiments of the present specification provide a task verification method. One or more embodiments of the present specification also relate to a model verification method, a model verification device, a task verification device, a computing device, a computer-readable storage medium, and a computer program to solve the technical defects in the prior art.

[0005] According to a first aspect of the embodiments of the present specification, a task verification method is provided, comprising:

[0006] Obtaining an image pair, wherein the image pair includes a source domain image and a target domain image, and the source domain image and the target domain image are images containing the same image content and having different image attributes;

[0007] Using a source task strategy to perform a target task on the source domain image to obtain a first predicted image, and using a task strategy to be verified to perform the target task on the source domain image to obtain a second predicted image;

[0008] Balancing the difficulty of task execution as the target, adjusting the source domain image according to the first predicted image and the target domain image to obtain a first verification image, and adjusting the source domain image according to the second predicted image and the target domain image to obtain a second verification image;

[0009] According to the first verification image and the second verification image, a task verification result is determined.

[0010] According to a second aspect of the embodiments of the present specification, a model verification method is provided, comprising:

[0011] obtaining an image pair, wherein the image pair comprises a source domain image and a target domain image, and the source domain image and the target domain image are images containing the same image content and having different image attributes;

[0012] performing the target task on the source domain image by using the source task model to obtain a first predicted image, and performing the target task on the source domain image by using the task model to be verified to obtain a second predicted image;

[0013] adjusting the source domain image according to the first predicted image and the target domain image to obtain a first verification image, and adjusting the source domain image according to the second predicted image and the target domain image to obtain a second verification image, with the goal of balancing the difficulty of task execution;

[0014] determining a model verification result according to the first verification image and the second verification image.

[0015] According to a third aspect of an embodiment of the present specification, a task verification method is provided, comprising:

[0016] receiving a task verification request sent by a target object, wherein the task verification request carries a task policy to be verified;

[0017] obtaining a first verification image, wherein the first verification image is obtained by adjusting the source domain image according to the first predicted image and the target domain image with the goal of balancing the difficulty of task execution, the first predicted image is obtained by performing the target task on the source domain image based on the source task policy, and the source domain image and the target domain image are images containing the same image content and having different image attributes;

[0018] performing the target task on the source domain image by using the task policy to be verified to obtain a second predicted image;

[0019] adjusting the source domain image according to the second predicted image and the target domain image to obtain a second verification image, with the goal of balancing the difficulty of task execution;

[0020] determining a task verification result according to the first verification image and the second verification image.

[0021] According to a fourth aspect of an embodiment of the present specification, a task verification device is provided, comprising:

[0022] a first obtaining module configured to obtain an image pair, wherein the image pair comprises a source domain image and a target domain image, and the source domain image and the target domain image are images containing the same image content and having different image attributes;

[0023] a first performing module configured to perform the target task on the source domain image by using the source task policy to obtain a first predicted image, and perform the target task on the source domain image by using the task model to be verified to obtain a second predicted image;

[0024] The first adjusting module is configured to adjust the original domain image according to the first predicted image and the target domain image to obtain a first verification image, and adjust the original domain image according to the second predicted image and the target domain image to obtain a second verification image, with the task execution difficulty balance as a target.

[0025] The first determining module is configured to determine a task verification result according to the first verification image and the second verification image.

[0026] According to a fifth aspect of an embodiment of the present specification, a model verification apparatus is provided, comprising:

[0027] The second obtaining module is configured to obtain an image pair, wherein the image pair comprises an original domain image and a target domain image, and the original domain image and the target domain image are images containing the same image content and having different image attributes.

[0028] The second executing module is configured to execute a target task on the original domain image by using a source task model to obtain a first predicted image, and execute the target task on the original domain image by using a task model to be verified to obtain a second predicted image.

[0029] The second adjusting module is configured to adjust the original domain image according to the first predicted image and the target domain image to obtain a first verification image, and adjust the original domain image according to the second predicted image and the target domain image to obtain a second verification image, with the task execution difficulty balance as a target.

[0030] The second determining module is configured to determine a model verification result according to the first verification image and the second verification image.

[0031] According to a sixth aspect of an embodiment of the present specification, a task verification apparatus is provided, comprising:

[0032] The receiving module is configured to receive a task verification request sent by a target object, wherein the task verification request carries a task policy to be verified.

[0033] The third obtaining module is configured to obtain a first verification image, wherein the first verification image is obtained by adjusting an original domain image according to a first predicted image and a target domain image, with the task execution difficulty balance as a target, the first predicted image is obtained by executing a target task on the original domain image based on a source task policy, and the original domain image and the target domain image are images containing the same image content and having different image attributes.

[0034] The third executing module is configured to execute a target task on the original domain image by using a task policy to be verified to obtain a second predicted image.

[0035] The third adjusting module is configured to adjust the original domain image according to the second predicted image and the target domain image to obtain a second verification image, with the task execution difficulty balance as a target.

[0036] The third determining module is configured to determine a task verification result according to the first verification image and the second verification image.

[0037] According to a seventh aspect of an embodiment of the present specification, a computing device is provided, comprising:

[0038] a memory and a processor;

[0039] The memory is configured to store computer-executable instructions, and the processor is configured to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method provided in the first aspect or the second aspect or the third aspect.

[0040] According to an eighth aspect of an embodiment of the present specification, a computer-readable storage medium is provided, which stores computer-executable instructions, which, when executed by a processor, implement the steps of the method provided in the first aspect or the second aspect or the third aspect.

[0041] According to a ninth aspect of an embodiment of the present specification, a computer program is provided, which, when executed in a computer, causes the computer to perform the steps of the method provided in the first aspect or the second aspect or the third aspect.

[0042] The task verification method provided by one embodiment of the present specification acquires an image pair, wherein the image pair comprises a source domain image and a target domain image, and the source domain image and the target domain image are images containing the same image content and having different image attributes; a target task is performed on the source domain image by using a source task policy to obtain a first predicted image, and the target task is performed on the source domain image by using a to-be-verified task policy to obtain a second predicted image; a first verification image is obtained by adjusting the source domain image according to the first predicted image and the target domain image, and a second verification image is obtained by adjusting the source domain image according to the second predicted image and the target domain image, with the goal of balancing the difficulty of task execution; and a task verification result is determined according to the first verification image and the second verification image. The characteristics of the source task policy can be determined through the first verification image, the characteristics of the to-be-verified task policy can be determined through the second verification image, and the task verification result can be further determined according to the respective characteristics of the source task policy and the to-be-verified task policy. In this process, no watermark information needs to be embedded in the source task policy and the to-be-verified task policy, the performance of the task policy itself is completely preserved, and lossless task verification is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 is an architecture diagram of a task verification system provided by one embodiment of the present specification;

[0044] Figure 2 is an architecture diagram of another task verification system provided by one embodiment of the present specification;

[0045] Figure 3 is a flow chart of a task verification method provided by one embodiment of the present specification;

[0046] Figure 4 is a schematic diagram of a performance boundary field in a task verification method provided by one embodiment of the present specification;

[0047] Figure 5 is a flow chart of another task verification method provided by one embodiment of the present specification;

[0048] Figure 6 is a flow chart of a model verification method provided by one embodiment of the present specification;

[0049] Figure 7 is a flow chart of a processing procedure of a task verification method provided by one embodiment of the present specification;

[0050] Figure 8 is a flow chart of a processing procedure of another task verification method provided by one embodiment of the present specification;

[0051] Figure 9 is an interface schematic diagram of a task verification interface provided by one embodiment of the present specification;

[0052] Figure 10 is a structural schematic diagram of a task verification device provided by one embodiment of the present specification;

[0053] Figure 11 is a structural schematic diagram of a model verification device provided by one embodiment of the present specification;

[0054] Figure 12 is a structural schematic diagram of another task verification device provided by one embodiment of the present specification;

[0055] Figure 13 is a structural block diagram of a computing device provided by one embodiment of the present specification. DETAILED DESCRIPTION

[0056] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present specification. However, the present specification can be practiced without the specific details, other than in the examples, set forth in this description. Those skilled in the art, in light of the description, can implement the present specification without limiting the scope of the present specification.

[0057] The terminology used in this disclosure, one or more embodiments of the present specification, is for the purpose of describing particular embodiments only and is not intended to be limiting of one or more embodiments of the present specification. As used in this disclosure and the appended claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be understood by those within the art that, in some instances, terms used herein have been adopted in order to provide explicit description for understanding by those of ordinary skill in the art. It will be apparent, however, to one of ordinary skill in the art that many of the most useful implementations of one or more embodiments of the present specification can not recite all of the features herein described. Therefore, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0058] It should be understood that although the terms first, second, etc. can be used herein to describe various information, these terms are not intended to denote a temporal or chronological order. These terms are used only to distinguish one from another. For example, without departing from the scope of one or more embodiments of the present specification, first can be termed second, and similarly, second can be termed first. The term "if' as used herein can be interpreted as meaning "when" or "in response to determining" depending on the context.

[0059] In addition, it should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in one or more embodiments of the present specification are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation portal for user to choose authorization or refusal.

[0060] First, the nomenclature involved in one or more embodiments of the present specification is explained.

[0061] Image processing: Image processing refers to a technology of analyzing, processing and changing digital images. Image processing includes operations such as restoration, enhancement, quality improvement and redundancy removal of images, in order to improve the understanding, analysis and utilization of images.

[0062] Image processing tasks include image denoising, super-resolution, deblurring, derain, low-light image enhancement, and other tasks, which are widely used in medical image analysis, computer vision, remote sensing image processing, robot vision, security monitoring, and other fields, and have important application value and research significance. The cost of designing and training image processing models, especially deep learning models, for specific application scenarios is often huge, such as the cost of hardware resources, data collection, labeling, manpower, etc. To share the results, the current approach is to share pre-trained models, and many companies and institutions also provide paid pre-trained model services. This provides an incentive for unscrupulous users to plagiarize or steal models, such as using malware infection or internal leaks to circumvent the expensive model training process. Therefore, it is very important to protect the copyright of the model.

[0063] Currently, there are few copyright protection schemes for image processing models, such as verifying the copyright of the model through model watermarking. However, model watermarking will intrusively embed watermark information into the model and change the model parameters, which will in turn change the performance of the model itself and affect the model performance. In addition, model watermarking includes white-box models and black-box models, where the white-box scheme is quite sensitive to model structure attacks, and the black-box scheme will embed an insecure backdoor mapping that can be exploited by malicious attackers.

[0064] To solve the above problems, the embodiments of the present specification explore the direction of non-intrusive copyright protection of image processing models and propose a zero-watermark protection scheme. Taking model zero-watermarking as an example, the model zero-watermarking protection scheme only extracts the intrinsic information of the model to verify the copyright of the model, and does not embed any information into the model or change any performance of the model. On the basis of ensuring the performance of the model, the copyright of the model is verified.

[0065] Specifically, the embodiment of the present specification provides a task verification method, an image pair is obtained, wherein the image pair includes a source domain image and a target domain image, the source domain image and the target domain image are images containing the same image content and having different image attributes; a target task is performed on the source domain image by using a source task policy to obtain a first predicted image, and the target task is performed on the source domain image by using a task policy to be verified to obtain a second predicted image; taking task execution difficulty balance as a target, the source domain image is adjusted according to the first predicted image and the target domain image to obtain a first verification image, and the source domain image is adjusted according to the second predicted image and the target domain image to obtain a second verification image; and a task verification result is determined according to the first verification image and the second verification image. The characteristics of the source task policy can be determined through the first verification image, the characteristics of the task policy to be verified can be determined through the second verification image, and the task verification result can be further determined according to the characteristics of the source task policy and the task policy to be verified. In this process, no watermark information needs to be embedded into the source task policy and the task policy to be verified, the performance of the task policy itself is completely preserved, and lossless task verification is realized.

[0066] In the present specification, a task verification method is provided, and the present specification also relates to a model verification method, a model verification device, a task verification device, a computing device, and a computer readable storage medium, which are described in detail one by one in the following embodiments.

[0067] Referring to Figure 1 , Figure 1 An architecture diagram of a task verification system provided by an embodiment of the present specification is shown, and the task verification system can include a client 100 and a server 200;

[0068] The client 100 is configured to send a source task policy and a task policy to be verified to the server 200.

[0069] The server 200 is configured to obtain an image pair, wherein the image pair includes a source domain image and a target domain image, the source domain image and the target domain image are images containing the same image content and having different image attributes; a target task is performed on the source domain image by using a source task policy to obtain a first predicted image, and the target task is performed on the source domain image by using a task policy to be verified to obtain a second predicted image; taking task execution difficulty balance as a target, the source domain image is adjusted according to the first predicted image and the target domain image to obtain a first verification image, and the source domain image is adjusted according to the second predicted image and the target domain image to obtain a second verification image; and a task verification result is determined according to the first verification image and the second verification image; and the task verification result is sent to the client 100.

[0070] The client 100 is further configured to receive the task verification result sent by the server 200.

[0071] By applying the scheme of the embodiments of the present specification, the characteristics of the source task policy can be determined through the first verification image, the characteristics of the task policy to be verified can be determined through the second verification image, and the task verification result can be further determined according to the characteristics of the source task policy and the task policy to be verified. In this process, there is no need to embed watermark information into the source task policy and the task policy to be verified, the performance of the task policy itself is completely preserved, and lossless task verification is achieved.

[0072] Referring to Figure 2 , Figure 2 An architecture diagram of another task verification system provided by an embodiment of the present specification is shown. The task verification system can include a server 200 and a plurality of clients 100. The client 100 can be an end-side device, and the server 200 can be a cloud-side device. The plurality of clients 100 can establish a communication connection through the server 200. In a task verification scenario, the server 200 is used to provide task verification services between the plurality of clients 100. The plurality of clients 100 can respectively act as a sending end or a receiving end, and implement communication through the server 200.

[0073] A user can interact with the server 200 through the client 100 to receive data sent by other clients 100, or send data to other clients 100, and the like. In a task verification scenario, the user can publish a data stream to the server 200 through the client 100. The server 200 generates a task verification result according to the data stream, and pushes the task verification result to other clients that establish a communication connection.

[0074] The client 100 and the server 200 establish a connection through a network. The network provides a medium for a communication link between the client 100 and the server 200. The network can include various connection types, such as wired, wireless communication links, or optical fiber cables, and the like. The data transmitted by the client 100 can need to be processed through encoding, transcoding, compression, and the like before being published to the server 200.

[0075] The client 100 can be a browser, an APP (Application), or a web application such as an H5 (HyperText Markup Language 5) application, or a light application (also known as a small program, a lightweight application), or a cloud application, etc. The client 100 can be developed based on a software development kit (SDK) of a corresponding service provided by the server 200, such as a real-time communication (RTC) SDK, etc. The client 100 can be deployed in an electronic device, and needs to rely on the device or some APP in the device, etc. The electronic device can have a display screen and support information browsing, etc., such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, etc. Various other types of applications can also be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0076] The server 200 can include servers that provide various services, such as servers that provide communication services for multiple clients, servers that provide support for models used on clients for background training, servers that process data sent by clients, etc. It should be noted that the server 200 can be implemented as a distributed server cluster composed of multiple servers, or as a single server. The server can also be a server of a distributed system, or a server combined with a blockchain. The server can also be a cloud server of cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, etc. Basic cloud computing services, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0077] It should be noted that the task verification method provided in the embodiments of the present specification is generally executed by the server, but in other embodiments of the present specification, the client can also have similar functions as the server, so as to execute the task verification method provided in the embodiments of the present specification. In other embodiments, the task verification method provided in the embodiments of the present specification can also be executed by the client and the server together.

[0078] Referring to Figure 3 , Figure 3 A flowchart of a task verification method provided in an embodiment of the present specification is shown, which specifically includes the following steps:

[0079] Step 302: obtaining an image pair, wherein the image pair comprises a source domain image and a target domain image, and the source domain image and the target domain image are images containing the same image content and different image attributes.

[0080] In one or more embodiments of the present specification, in order to perform task verification, at least one image pair can be obtained, and the source domain image and the target domain image in the image pair are used to verify the task policy to be verified, thereby obtaining a task verification result.

[0081] Specifically, the image pair is composed of a source domain image and a target domain image. The number of image pairs can be one or more. The target domain image can be obtained based on the source domain image, and the source domain image can be obtained based on the target domain image. The image attributes include but are not limited to image scene attributes and image quality attributes. The image scene attribute is used to represent the scene information of the image, such as rain or no rain, etc. The image quality attribute is used to represent the quality information of the image, such as image clarity, image brightness, etc. The source domain image and the target domain image contain the same image content, but the image content of the two may not be exactly the same. For example, the source domain image is a park in a rainy scene, and the target domain image is an image in a non-rainy scene, and the source domain image and the target domain image include the same park.

[0082] Exemplarily, taking the processing of mapping the source domain image to another target domain image as an example, the image denoising task maps the noisy image to the clear image, and the noisy image is the source domain image and the clear image is the target domain image; the low-light enhancement task maps the low-light image to the normal light image, and the low-light image is the source domain image and the normal light image is the target domain image; the rain removal task maps the rainy image to the non-rainy image, and the rainy image is the source domain image and the non-rainy image is the target domain image.

[0083] In actual application, there are various ways to obtain the image pair, which can be selected according to actual conditions, and the present specification does not make any limitation on this. In a possible implementation manner of the present specification, the image pair sent by the user can be received.

[0084] In another possible implementation manner of the present specification, the target domain image can be obtained, and the source domain image can be obtained by degrading the target domain image. That is, the above-mentioned obtaining the image pair can comprise the following steps:

[0085] Obtaining a target processing strategy corresponding to the target domain image and the target domain image.

[0086] Degradation processing the target domain image according to the target processing strategy to obtain the source domain image.

[0087] Specifically, the target processing strategy can be referred to as a target image degradation strategy, a target image inverse processing strategy. The target processing strategy is used to perform degradation processing on the target image to obtain the source domain image. The degradation processing is used to reduce the image quality of the target domain image to obtain the source domain image with lower quality. The target processing strategy includes at least one of an image denoising strategy, an image super-resolution strategy, an image deblurring strategy, and an image light enhancement strategy.

[0088] Exemplarily, it is assumed that the target domain image is X, the degradation processing process is D t , and the source domain image is D t (X). Specifically, the image denoising strategy is as shown in the following formula (1), and the source domain image can be obtained by adding additive noise to the target domain image; the image super-resolution strategy is as shown in the following formula (2), and the source domain image can be obtained by processing the target domain image based on a super-resolution downsampling rate; the image deblurring strategy is as shown in the following formula (3), and the source domain image can be obtained by processing the target domain image according to a blur kernel and additive noise; the image light enhancement strategy is as shown in the following formula (4), and the source domain image can be obtained by processing the target domain image according to an index of low-light conversion and performing normalization processing on the processed image; and the image rain removal strategy is as shown in the following formula (5), and the source domain image can be obtained by adding a rain layer to the target domain image:

[0089] D t (X)=X+N (1)

[0090] D t (X)=X ↓d (2)

[0091]

[0092] D t (X)=Norm(X γ ) (4)

[0093] D t (X)=X+R (5)

[0094] Wherein, N is additive noise, which can be simply defined as zero-mean Gaussian white noise in the embodiments of the present specification; is a super-resolution downsampling rate; K is a blur kernel; is a tensor product; γ is an index of low-light conversion, Norm is an image normalization process, and a simple normalization process such as MinMax can be used in the embodiments of the present specification; and R is a simply synthesized rain layer.

[0095] It should be noted that, since the range of image pixels is usually [0, 255], in the embodiments of the present specification, the range of image pixels can be normalized to [0, 1] or [-1, 1] to facilitate the convergence of the task verification process.

[0096] By applying the scheme of the present specification, the target domain image and the target processing strategy corresponding to the target domain image are obtained; and the target domain image is degraded according to the target processing strategy to obtain the original domain image. By degrading the target domain image using the target processing strategy to obtain the original domain image, the accuracy of the original domain image is ensured.

[0097] In practical applications, there are multiple ways to obtain the target domain image and the target processing strategy corresponding to the target domain image, which are selected according to actual conditions, and the embodiments of the present specification do not make any limitation thereon. In a possible implementation manner of the present specification, the target domain image sent by the user can be received, and any image processing strategy can be randomly selected from the pre-set multiple image processing strategies as the target processing strategy.

[0098] In another possible implementation manner of the present specification, since the source task strategy and the task to be verified strategy are used to execute the target task on the original domain image in the task verification process, the original domain image and the target domain image should meet the task type of the target task. For example, if the target task is an image rain removal task, the original image should be an image including a rain layer, and the target domain image should be an image not including a rain layer. That is, the above-mentioned obtaining of the target domain image and the target processing strategy corresponding to the target domain image can include the following steps:

[0099] identifying the task type of the target task;

[0100] obtaining the target domain image according to the task type;

[0101] filtering the target processing strategy from the pre-set multiple image processing strategies according to the task type, wherein the target processing strategy includes at least one of an image denoising strategy, an image super-resolution strategy, an image deblurring strategy, and an image light enhancement strategy.

[0102] Specifically, the target task is an image processing task, and the target task includes image denoising, super-resolution, deblurring, rain removal, low-light image enhancement, etc. The task type includes but is not limited to rain removal type, super-resolution type, and deblurring type. Since it is difficult to accurately model the real degradation process from the target domain image to the original domain image, such as real noise and motion blur data, in the embodiments of the present specification, the real degradation process is simulated by data modeling, multiple image processing strategies are pre-set, and the image processing strategy can also be customized based on a specific model when setting the image processing strategy, to ensure the flexibility of the image processing strategy.

[0103] It should be noted that there are various ways to identify the task type of the target task, which are selected according to actual conditions, and the embodiments of the present specification do not make any limitation in this regard. In a possible implementation manner of the present specification, the target task can be identified by using a type identification model to determine the task type of the target task. In another possible implementation manner of the present specification, the target task can be matched with sample tasks corresponding to various types, and the type corresponding to the sample task matched by the target task is taken as the task type of the target task.

[0104] Further, after identifying the task type of the target task, the processing types corresponding to the plurality of image processing strategies can be obtained, a target processing type same as the task type is filtered out from the plurality of processing types, and the image processing strategy corresponding to the target processing type is taken as the target processing strategy.

[0105] By applying the scheme of the embodiments of the present specification, the task type of the target task is identified, the target domain image is obtained according to the task type, and the target processing strategy is filtered out from the plurality of image processing strategies pre-set according to the task type, which ensures that the target domain image and the target processing strategy are related to the target task, and further ensures that the original domain image is related to the target task.

[0106] Step 304: performing the target task on the original domain image by using the source task strategy to obtain a first predicted image, and performing the target task on the original domain image by using the to-be-verified task strategy to obtain a second predicted image.

[0107] In one or more embodiments of the present specification, after obtaining the image pair, further, the target task can be performed on the original domain image by using the source task strategy to obtain a first predicted image, and the target task can be performed on the original domain image by using the to-be-verified task strategy to obtain a second predicted image.

[0108] Specifically, the to-be-verified task strategy refers to a suspicious task strategy relative to the source task strategy. When verifying the to-be-verified task strategy, it can be considered that the source task strategy has copyright relative to the to-be-verified task strategy, and the to-be-verified task strategy is a task strategy suspected of stealing the source task strategy. The source task strategy and the to-be-verified task strategy can both perform image processing tasks on images. The task strategy can be a task model or a task execution code, which is selected according to actual conditions, and the embodiments of the present specification do not make any limitation in this regard. The first predicted image is the result predicted by the source task strategy, and the second predicted image is the result predicted by the to-be-verified task strategy.

[0109] It should be noted that in order to ensure the accuracy of task verification, the source task strategy and the to-be-verified task strategy can be caused to perform the same target task when processing the original domain image.

[0110] In an optional embodiment of the present specification, the source task strategy comprises a source task model, and the to-be-verified task strategy comprises a to-be-verified task model; the above-mentioned performing the target task on the original domain image by using the source task strategy to obtain the first predicted image, and performing the target task on the original domain image by using the to-be-verified task strategy to obtain the second predicted image can comprise the following steps:

[0111] performing the target task on the original domain image by using the source task model to obtain the first predicted image, and performing the target task on the original domain image by using the to-be-verified task model to obtain the second predicted image.

[0112] Specifically, the source task model refers to a copyright image processing model, and the to-be-verified task model refers to a suspicious task model relative to the source task model.

[0113] In actual application, the implementation manner of "performing the target task on the original domain image by using the source task model to obtain the first predicted image" is the same as that of "performing the target task on the original domain image by using the to-be-verified task model to obtain the second predicted image". Taking performing the target task on the original domain image by using the source task model to obtain the first predicted image as an example, the original domain image can be input into the source task model to obtain the first predicted image. For example, the original domain image comprising a rain layer is input into the source task model to obtain the first predicted image.

[0114] By applying the scheme of the embodiments of the present specification, the target task is performed on the original domain image by using the source task model to obtain the first predicted image, and the target task is performed on the original domain image by using the to-be-verified task model to obtain the second predicted image, thereby realizing copyright verification on the model.

[0115] Step 306: adjusting the original domain image according to the first predicted image and the target domain image to obtain the first verification image, and adjusting the original domain image according to the second predicted image and the target domain image to obtain the second verification image, with the target of balancing the difficulty of task execution.

[0116] In one or more embodiments of the present specification, an image pair is acquired; after performing the target task on the original domain image by using the source task strategy to obtain the first predicted image, and performing the target task on the original domain image by using the to-be-verified task strategy to obtain the second predicted image, further, the original domain image can be adjusted according to the first predicted image and the target domain image to obtain the first verification image, and the original domain image can be adjusted according to the second predicted image and the target domain image to obtain the second verification image, with the target of balancing the difficulty of task execution.

[0117] Specifically, when the task execution difficulty is balanced, it is verified that the original domain image corresponding to the image is just on the performance boundary domain of the task strategy "easy to predict" and "difficult to predict", that is, the original domain image is adjusted, and when the adjusted original domain image is on the performance boundary domain, the target domain image corresponding to the adjusted original domain image is the verification image. The verification image is located on the target domain, and the verification image can be called a critical image, a zero watermark image, and an intrinsic information of the task strategy. At the same time, the verification image can be used as a fingerprint of the task strategy.

[0118] Referring to Figure 4 , Figure 4 A schematic diagram of a performance boundary domain in a task verification method provided by one embodiment of the present specification is shown, as shown in Figure 4 The shadow area inside is a small error prediction area, the shadow area outside is a large error prediction area, the shadow area is a performance boundary domain, and the black circle inside the shadow area is an original domain image corresponding to a verification image. Figure 4 It can be seen that the original domain image corresponding to the first verification image and the original domain image corresponding to the second verification image are both located on the performance boundary domain.

[0119] It should be noted that the implementation of "adjusting the original domain image according to the first prediction image and the target domain image to obtain the first verification image" is the same as that of "adjusting the original domain image according to the second prediction image and the target domain image to obtain the second verification image".

[0120] In an optional embodiment of the present specification, the above-mentioned adjusting the original domain image according to the first prediction image and the target domain image to obtain the first verification image can include the following steps:

[0121] determining a first prediction loss according to the first prediction image and the target domain image;

[0122] determining a first total variation loss according to the target domain image;

[0123] adjusting the original domain image according to the first prediction loss and the first total variation loss to obtain the first verification image.

[0124] It should be noted that in general, the smoother the image is, the easier it is to process, and the smoothness of the image can be measured by a total variation (TV) loss, which is defined as the L1 norm of the image gradient, that is, The total variation loss is used to promote the spatial smoothness of the image. The smaller the total variation loss, the smoother the image, the easier to process, and the closer to the small error prediction area. On the contrary, the larger the total variation loss, the less smooth the image, the more difficult to process, and the closer to the large error prediction area. Therefore, the first total variation loss can reflect the difficulty of the source task strategy in processing the original domain image. The prediction loss is defined as the L2 norm of the image gradient. The smaller the prediction loss, the closer to the small error prediction area. The larger the prediction loss, the closer to the large error prediction area. Therefore, the first prediction loss can reflect the difficulty of the source task strategy in predicting the first predicted image.

[0125] In practical applications, the first prediction loss can be determined by the following formula (6), and the first total variation loss can be determined by the following formula (7):

[0126]

[0127]

[0128] Wherein, L is the first prediction loss; X is the target domain image; M is the source task strategy; D t (X) is the original domain image; M(D t (X)) is the first predicted image. is the first total variation loss; is the image gradient, which is obtained by subtracting the horizontal and vertical coordinates of the target domain image respectively; i is the i-th element.

[0129] Further, when adjusting the original domain image according to the first prediction loss and the first total variation loss to obtain the first verification image, the source task strategy can be fixed unchanged, and the original domain image or the target domain image can be gradient descended according to the first prediction loss and the first total variation loss to optimize the image, and the first verification image can be obtained by iteration.

[0130] In practical applications, an image X0 with a size of m×n can be initialized, and the initialization method can be random initialization, so that X0 is random noise and is located in the large error prediction area. Further optimization of X0 can be performed to determine the first verification image by the following formula (8). Alternatively, an image X0 with a size of m×n can be initialized, and the initialization method can be constant or all-0 initialization, so that X0 is located in the small error prediction area, and the first verification image can be determined by the following formula (9):

[0131]

[0132]

[0133] Wherein, S is the first verification image; argmin(A) is the value of the variable when A reaches the minimum value; λ is the weight.

[0134] It is worth mentioning that the process of obtaining the first verification image by adjusting the source domain image according to the first prediction loss and the first total variation loss can be regarded as a confrontation between the first prediction loss and the first total variation loss. For example, in formula (8), the first prediction loss makes the target domain image move to the small error prediction area, and the first total variation loss makes the target domain image stay in the large error prediction area, so that the verification image located at the junction of the large error prediction area and the small error prediction area (i.e. the performance boundary domain) will be obtained after optimization.

[0135] In an optional embodiment of the present specification, when determining the first verification image, customization can be made for a specific task strategy, such as optimization of learning rate initialization, optimizer, learning rate scheduling, etc., so as to increase the uniqueness and robustness of the zero watermark.

[0136] According to the scheme of the embodiment of the present specification, the first prediction loss is determined according to the first prediction image and the target domain image, the first total variation loss is determined according to the target domain image, and the first verification image is obtained by adjusting the source domain image according to the first prediction loss and the first total variation loss. The adjustment process of the source domain image is jointly constrained by the first total variation loss and the first prediction loss, so as to ensure the accuracy of the first verification image.

[0137] In actual application, there are various ways to obtain the first verification image by adjusting the source domain image according to the first prediction loss and the first total variation loss, which are selected according to actual conditions, and the embodiments of the present specification do not make any limitation on this.

[0138] In a possible implementation manner of the present specification, the source domain image can be adjusted according to the first prediction loss and the first total variation loss to obtain an adjusted source domain image, and the step of performing the target task on the source domain image by using the source task strategy to obtain the first prediction image is returned to be executed until a preset stopping condition is reached to obtain the first verification image.

[0139] In another possible implementation manner of the present specification, the adjustment of the source domain image can be realized by adjusting the target domain image, that is, the above-mentioned step of adjusting the source domain image according to the first prediction loss and the first total variation loss to obtain the first verification image can include the following steps:

[0140] adjusting the target domain image according to the first prediction loss and the first total variation loss to obtain an adjusted target domain image;

[0141] determining an adjusted source domain image according to the adjusted target domain image, and returning to execute the step of performing the target task on the source domain image by using the source task strategy to obtain the first prediction image until a preset stopping condition is reached to obtain the first verification image.

[0142] It should be noted that, according to the adjusted target domain image, the adjusted source domain image is determined, and the adjusted target domain image can be degraded by using a target processing strategy to obtain the adjusted source domain image. The preset stop condition is that the adjusted source domain image is on the performance boundary domain. When the adjusted source domain image is on the performance boundary domain, the adjusted target domain image corresponding to the adjusted source domain image can be taken as the first verification image.

[0143] According to the scheme of the embodiment of the present specification, the target domain image is adjusted according to the first prediction loss and the first total variation loss to obtain an adjusted target domain image; the adjusted source domain image is determined according to the adjusted target domain image, and the step of executing the target task on the source domain image by using the source task strategy to obtain the first prediction image is returned until the preset stop condition is reached to obtain the first verification image, which guarantees the accuracy of the first verification image.

[0144] Step 308: determining a task verification result according to the first verification image and the second verification image.

[0145] In one or more embodiments of the present specification, an image pair is obtained; a first prediction image is obtained by executing the target task on the source domain image by using the source task strategy, and a second prediction image is obtained by executing the target task on the source domain image by using the to-be-verified task strategy; with the goal of balancing the difficulty of task execution, the source domain image is adjusted according to the first prediction image and the target domain image to obtain the first verification image, and the source domain image is adjusted according to the second prediction image and the target domain image to obtain the second verification image. Further, the task verification result can be determined according to the first verification image and the second verification image.

[0146] Specifically, the task verification result can be a verification result of the to-be-verified task strategy, such as whether the to-be-verified task strategy is a task strategy stolen from the source task strategy. The task verification result can also be a verification result of the source task strategy, such as whether the source task strategy is stolen by the to-be-verified task strategy.

[0147] According to the scheme of the embodiment of the present specification, the characteristics of the source task strategy can be determined through the first verification image, the characteristics of the to-be-verified task strategy can be determined through the second verification image, and the task verification result can be determined according to the characteristics of the source task strategy and the to-be-verified task strategy. In this process, there is no need to embed watermark information into the source task strategy and the to-be-verified task strategy, the performance of the task strategy itself is completely preserved, and lossless task verification is achieved.

[0148] In actual application, there are many ways to determine the task verification result according to the first verification image and the second verification image, which are selected according to actual conditions, and the present specification does not make any limitation in this regard.

[0149] In a possible implementation of the present specification, the determining of the task verification result according to the first verification image and the second verification image can include the following steps.

[0150] performing feature extraction on the first verification image to obtain first feature information, and performing feature extraction on the second verification image to obtain second feature information;

[0151] determining the task verification result according to the first feature information and the second feature information.

[0152] In actual application, the similarity between the first verification image and the second verification image can be compared to determine whether the first verification image and the second verification image come from the same task strategy, and the task verification result is obtained. If the similarity is greater than a preset threshold, it can be considered that the task strategy to be verified is the same as the source task strategy. Specifically, hypothesis test can be performed by using the first feature information and the second feature information to determine the task verification result. The hypothesis test includes but is not limited to t-test, multinomial-test.

[0153] It should be noted that the manner of "performing feature extraction on the first verification image to obtain first feature information" and the manner of "performing feature extraction on the second verification image to obtain second feature information" are the same. The manner of feature extraction includes but is not limited to Local Binary Patterns (LBP), Opposite Color Local Binary Patterns (OC-LBP), Local Vector Pattern (LVP), Monogenic Binary Patterns (MBP), Local Gradient Pattern (LGP) + image color histogram. Preferably, the manner of Local Gradient Pattern + image color histogram can be used to perform feature extraction on the first verification image to obtain the first feature information.

[0154] By applying the scheme of the present specification, the first feature information is obtained by performing feature extraction on the first verification image, and the second feature information is obtained by performing feature extraction on the second verification image. The task verification result is determined according to the first feature information and the second feature information, which realizes automatic verification of the first verification image and the second verification image, and improves the efficiency of obtaining the task verification result.

[0155] In another possible implementation of the present specification, the determining of the task verification result according to the first verification image and the second verification image can include the following steps.

[0156] sending the first verification image and the second verification image to an image verification party, and receiving image verification information sent by the image verification party;

[0157] generating a task verification result based on the image verification information.

[0158] Specifically, the image verification party can be a staff of a third-party authority. The image verification information can represent whether the first verification image and the second verification image are the same, and can also represent the similarity of the first verification image and the second verification image.

[0159] It should be noted that, since the verification image includes rich image information, the first verification image and the second verification image can be directly sent to the image verification party, and the image verification party can visually verify the first verification image and the second verification image manually to obtain the image verification information.

[0160] Further, the image verification information sent by the image verification party can be received. If the image verification information indicates that the first verification image and the second verification image are the same, it is determined that the task verification result is that the to-be-verified task strategy and the source task strategy are the same, and the to-be-verified task strategy is a task strategy that steals the source task strategy. If the image verification information indicates that the first verification image and the second verification image are different, it is determined that the task verification result is that the to-be-verified task strategy and the source task strategy are different, and the to-be-verified task strategy is not a task strategy that steals the source task strategy.

[0161] In actual application, the task verification result can be generated by combining the automatic verification and the manual visual verification.

[0162] By applying the scheme of the embodiments of the present specification, the first verification image and the second verification image are sent to the image verification party, and the image verification information sent by the image verification party is received. The task verification result is generated based on the image verification information, thereby ensuring the accuracy of the task verification result.

[0163] In an optional embodiment of the present specification, after the task verification result is determined according to the first verification image and the second verification image, the following step can be further included:

[0164] In the case where the task verification result is that the to-be-verified task strategy and the source task strategy are the same, verification alarm information is generated.

[0165] It should be noted that, if the to-be-verified task strategy and the source task strategy are the same, it indicates that the to-be-verified task strategy is a task strategy that steals the source task strategy. At this time, the verification alarm information can be generated. The verification alarm information includes but is not limited to alarm voice and alarm pop-up window. The verification alarm information can notify the source task strategy owner that the source task strategy is stolen, or can notify the to-be-verified task strategy owner that the to-be-verified task strategy is a stolen task strategy.

[0166] According to the scheme of the embodiment of the present specification, in the case where the task verification result is that the to-be-verified task policy is the same as the source task policy, verification alarm information is generated, so that the source task policy owner and the to-be-verified task policy owner can know the task verification result in time.

[0167] In an optional embodiment of the present specification, after the task verification result is determined according to the first verification image and the second verification image, the following steps can be further included:

[0168] In the case where the task verification result is that the to-be-verified task policy is the same as the source task policy, first verification information corresponding to the source task policy is obtained, and second verification information corresponding to the to-be-verified task policy is obtained.

[0169] The task verification result is verified according to the first verification information and the second verification information, and a verification result is obtained.

[0170] Specifically, the verification information includes but is not limited to task policy sharing time stamp, policy generation record, etc. The verification information can be understood as evidence information. There are various ways to obtain the verification information, which can receive the verification information sent by the user, or read the verification information from the database.

[0171] It should be noted that, in the above task verification process, it is considered that the source task policy has copyright relative to the to-be-verified task policy, but the source task policy may be a task policy stolen from the to-be-verified task policy. Therefore, in order to ensure the accuracy of the task verification result, after the task verification result is generated, the first verification information corresponding to the source task policy and the second verification information corresponding to the to-be-verified task policy are obtained, so as to determine the task policy that truly has copyright by using the first verification information and the second verification information.

[0172] In actual application, in the case where the task verification result is that the to-be-verified task policy is the same as the source task policy, it is determined that the to-be-verified task policy is a task policy stolen from the source task policy. At this time, if the policy sharing time stamp corresponding to the source task policy is later than the sharing time stamp of the to-be-verified task policy, it can be determined that the to-be-verified task policy is the task policy that has copyright, and the verification result is generated as inaccurate, and the task verification result can be further adjusted as "the source task policy is a task policy stolen from the to-be-verified task policy".

[0173] According to the scheme of the embodiment of the present specification, in the case where the task verification result is that the to-be-verified task policy is the same as the source task policy, first verification information corresponding to the source task policy is obtained, and second verification information corresponding to the to-be-verified task policy is obtained; the task verification result is verified according to the first verification information and the second verification information, and a verification result is obtained. By introducing the verification information to verify the task verification result, the interests of the task policy copyright owner are maintained.

[0174] In one optional embodiment of this specification, after determining the task verification result based on the first verification image and the second verification image, the following steps may be further included:

[0175] Based on the display requirements, the task verification result is sent to the client so that the client can display the task verification result to the user.

[0176] It should be noted that the display requirement information represents the user's need to view the task verification results. Display requirement information includes, but is not limited to, displaying only the task verification results, displaying the task verification results, the first verification image, and the second verification image. The specific display requirement information is set according to the user's actual needs, and this specification does not impose any limitations on it in the embodiments.

[0177] By applying the solution in the embodiments of this specification, the task verification result is sent to the client according to the display requirement information, so that the client can display the task verification result to the user, thereby increasing interaction with the user and improving user satisfaction.

[0178] See Figure 5 , Figure 5 This specification shows a flowchart of another task verification method provided in one embodiment, which specifically includes the following steps:

[0179] Step 502: Receive a task verification request sent by the target object, wherein the task verification request carries the strategy of the task to be verified.

[0180] Step 504: Obtain the first verification image, wherein the first verification image is obtained by adjusting the original domain image based on the first predicted image and the target domain image with the goal of balancing the difficulty of task execution, the first predicted image is obtained by performing the target task on the original domain image based on the source task strategy, and the original domain image and the target domain image are images containing the same image content but with different image attributes.

[0181] Step 506: Perform the target task on the original domain image using the strategy to be verified to obtain the second predicted image.

[0182] Step 508: With the goal of balancing the difficulty of task execution, adjust the original domain image based on the second predicted image and the target domain image to obtain the second verification image.

[0183] Step 510: Determine the task verification result based on the first verification image and the second verification image.

[0184] Specifically, the target object refers to the object for which the task verification requirement is met, such as the owner of the source task strategy.

[0185] It should be noted that the source task policy owner can send the source task policy to the server in advance, and the server generates and stores the first verification image. When the source task policy owner has a task verification requirement later, the task policy to be verified can be sent to the server, and thus the server only needs to generate the second verification image each time the task verification is performed, and compare the second verification image with the first verification image stored in advance to determine the task verification result.

[0186] In practical application, the implementation manners of steps 504 to 510 are the same as those of steps 302 to 308, and thus the description of the embodiments of the present specification will not be repeated.

[0187] According to the scheme of the embodiments of the present specification, only the second verification image needs to be generated each time the task verification is performed, and the second verification image is compared with the first verification image stored in advance to determine the task verification result, thereby improving the task verification efficiency.

[0188] Referring to Figure 6 , Figure 6 A flowchart of a model verification method provided by one embodiment of the present specification is shown, which specifically includes the following steps:

[0189] Step 602: Obtain an image pair, wherein the image pair includes a source domain image and a target domain image, and the source domain image and the target domain image are images containing the same image content and having different image attributes.

[0190] Step 604: Perform the target task on the source domain image by using the source task model to obtain a first predicted image, and perform the target task on the source domain image by using the task model to be verified to obtain a second predicted image.

[0191] Step 606: Balance the task execution difficulty, adjust the source domain image according to the first predicted image and the target domain image to obtain a first verification image, and adjust the source domain image according to the second predicted image and the target domain image to obtain a second verification image.

[0192] Step 608: Determine the model verification result according to the first verification image and the second verification image.

[0193] In practical application, the implementation manners of steps 604 to 608 are the same as those of steps 302 to 308, and thus the description of the embodiments of the present specification will not be repeated.

[0194] By applying the scheme of the embodiments of this specification, the features of the source task model can be determined by the first verification image, and the features of the task model to be verified can be determined by the second verification image. Furthermore, the model verification result can be determined based on the respective features of the source task model and the task model to be verified. In this process, there is no need to embed watermark information into the source task model and the task model to be verified, and the performance of the task model itself is fully preserved, thus realizing lossless model copyright verification.

[0195] In one optional embodiment of this specification, before performing the target task on the original domain image using the source task model to obtain the first predicted image, the following steps may be included:

[0196] Receive the source model information of the source task model and the verification model information of the task model to be verified, sent by the user through the client;

[0197] Construct the source task model based on the source model information, and construct the task model to be verified based on the model information to be verified.

[0198] It should be noted that the source model information can be the code information of the source task model, or it can be the code storage address information, code reading key, etc. The model information to be verified can be the code information of the task model to be verified, or it can be the code storage address information, code reading key, etc.

[0199] The scheme implemented in this specification receives source model information of the source task model and verification model information of the task model to be verified, sent by the user through the client; constructs the source task model based on the source model information and constructs the task model to be verified based on the verification model information, further improving the interaction with the user and making model verification more flexible.

[0200] See Figure 7 , Figure 7 This document illustrates a flowchart of a task verification method provided in one embodiment. The overall technical approach of this task verification method is an ownership (copyright) verification method. It can extract zero-watermarks from both the source task strategy and the task strategy to be verified, obtaining their respective verification images, which are then compared and verified. Based on the similarity of the two verification images and a statistical hypothesis testing scheme, it is determined whether a task strategy theft problem exists. Specifically, it includes:

[0201] Extract the first verification image corresponding to the source task strategy, and extract the second verification image corresponding to the task strategy to be verified;

[0202] The similarity of the first verification image and the second verification image is verified, and specifically, feature extraction can be performed on the first verification image and the second verification image respectively, and whether the first verification image and the second verification image are the same is determined according to the extracted features. If the first verification image and the second verification image are the same, it is determined that the to-be-verified task policy is a stealing task policy, and an alarm is performed; if the first verification image and the second verification image are different, it is determined that the verification is passed, and the to-be-verified task policy is an irrelevant task policy.

[0203] By using the zero watermark scheme, the scheme of the embodiment of the present specification does not change any parameter of the task policy, completely retains the performance of the task policy itself, and realizes lossless task verification. Moreover, through experimental verification on different image processing tasks, the task verification scheme proposed in the embodiment of the present specification can obtain a distinctive verification image as a zero watermark, and is robust to common attacks such as lightweight policy compression, fine-tuning, and quantization. Therefore, the task verification scheme has high distinctiveness and robustness.

[0204] Referring to Figure 8 , Figure 8 A processing process flowchart of another task verification method provided by an embodiment of the present specification is shown. Considering that a dispute between only two parties may still exist, a fair third-party authority can be introduced in the embodiment of the present specification, which specifically includes:

[0205] The task policy copyright owner registers the source task policy in the third-party authority, and uploads the policy information and related optimization process parameters (such as target processing policy, iteration number, etc.) to the third-party authority;

[0206] The third-party authority calculates and stores the verification image of the source task policy;

[0207] If the task policy copyright owner finds a controversial to-be-verified task policy, the to-be-verified task policy can be uploaded to the third-party authority;

[0208] The third-party authority calculates the verification image of the to-be-verified task policy, and compares and verifies the verification image with the stored verification image of the source task policy, and determines whether the to-be-verified task policy has a plagiarism behavior according to the similarity.

[0209] It should be noted that the task verification scheme proposed in the embodiments of this specification aims to obtain a verification image. The degraded version of the original image of the verification image lies precisely on the performance boundary between "easy to predict" and "difficult to predict," meaning that the processing error of the task strategy is precisely at a balance point. For example, given an image Q, if the processing result of model P has a small error compared to the true result, then this image Q can be considered to be located in the "easy to predict" region of model P. Conversely, if the processing result of model P has a large error compared to the true result, then this image Q can be considered to be located in the "difficult to predict" region of model P.

[0210] See Figure 9 , Figure 9 This diagram illustrates a task verification interface according to an embodiment of this specification. The task verification interface is divided into a request input interface and a result display interface. The request input interface includes a request input box, an "OK" control, and a "Cancel" control. The result display interface includes a result display box.

[0211] The user inputs a task verification request through a request input box displayed on the client. This request carries the strategy for the task to be verified. The user clicks the "OK" button. The server receives the task strategy from the client, obtains a first verification image, which is adjusted from the original domain image based on a first predicted image and a target domain image, with the goal of balancing task execution difficulty. The first predicted image is obtained by performing the target task on the original domain image based on the source task strategy. The original and target domain images contain the same image content but have different image attributes. The server then performs the target task on the original domain image using the task strategy to obtain a second predicted image. The server adjusts the original domain image based on the second predicted image and the target domain image, with the goal of balancing task execution difficulty, to obtain a second verification image. Based on the first and second verification images, the server determines the task verification result and sends it to the client. The client displays the task verification result in a result display box.

[0212] In practical applications, users can interact with controls in various ways, including clicking, double-clicking, touching, hovering, swiping, long-pressing, voice control, or shaking. The specific method chosen depends on the actual situation, and this specification does not impose any limitations on this.

[0213] Corresponding to the above method embodiments, this specification also provides embodiments of a task verification device. Figure 10 A schematic diagram of a task verification device according to one embodiment of this specification is shown. Figure 10 As shown, the device includes:

[0214] The first obtaining module 1002 is configured to obtain an image pair, wherein the image pair comprises a source domain image and a target domain image, and the source domain image and the target domain image are images containing the same image content and having different image attributes;

[0215] The first executing module 1004 is configured to execute the target task on the source domain image by using a source task policy to obtain a first predicted image, and execute the target task on the source domain image by using a to-be-verified task policy to obtain a second predicted image.

[0216] The first adjusting module 1006 is configured to balance the difficulty of task execution, adjust the source domain image according to the first predicted image and the target domain image to obtain a first verification image, and adjust the source domain image according to the second predicted image and the target domain image to obtain a second verification image.

[0217] The first determining module 1008 is configured to determine a task verification result according to the first verification image and the second verification image.

[0218] Optionally, the first adjusting module 1006 is further configured to determine a first predicted loss according to the first predicted image and the target domain image, determine a first total variation loss according to the target domain image, and adjust the source domain image according to the first predicted loss and the first total variation loss to obtain the first verification image.

[0219] Optionally, the first adjusting module 1006 is further configured to adjust the target domain image according to the first predicted loss and the first total variation loss to obtain an adjusted target domain image, determine an adjusted source domain image according to the adjusted target domain image, and return to execute the step of executing the target task on the source domain image by using the source task policy to obtain the first predicted image until a preset stop condition is reached to obtain the first verification image.

[0220] Optionally, the first obtaining module 1002 is further configured to obtain a target processing policy corresponding to the target domain image and the target domain image, and perform degradation processing on the target domain image according to the target processing policy to obtain the source domain image.

[0221] Optionally, the first obtaining module 1002 is further configured to identify a task type of the target task, obtain the target domain image according to the task type, and select a target processing policy from a plurality of image processing policies pre-set according to the task type, wherein the target processing policy comprises at least one of an image denoising strategy, an image super-resolution strategy, an image deblurring strategy, and an image light enhancement strategy.

[0222] Optionally, the source task policy comprises a source task model, and the to-be-verified task policy comprises a to-be-verified task model; the first execution module 1004 is configured to execute the target task on the original domain image by using the source task model to obtain a first predicted image, and execute the target task on the original domain image by using the to-be-verified task model to obtain a second predicted image.

[0223] Optionally, the first determination module 1008 is further configured to perform feature extraction on the first verification image to obtain first feature information, and perform feature extraction on the second verification image to obtain second feature information; and determine the task verification result according to the first feature information and the second feature information.

[0224] Optionally, the first determination module 1008 is further configured to send the first verification image and the second verification image to an image verification party, and receive image verification information sent by the image verification party; and generate the task verification result based on the image verification information.

[0225] Optionally, the apparatus further comprises a generation module configured to generate verification alarm information in a case where the task verification result is that the to-be-verified task policy is the same as the source task policy.

[0226] Optionally, the apparatus further comprises a verification module configured to, in a case where the task verification result is that the to-be-verified task policy is the same as the source task policy, acquire first verification information corresponding to the source task policy, and acquire second verification information corresponding to the to-be-verified task policy; and verify the task verification result according to the first verification information and the second verification information to obtain a verification result.

[0227] Optionally, the apparatus further comprises a sending module configured to send the task verification result to the client according to the display requirement information, so that the client displays the task verification result to the user.

[0228] By applying the scheme of the embodiments of the present specification, the features of the source task policy can be determined through the first verification image, the features of the to-be-verified task policy can be determined through the second verification image, and the task verification result can be further determined according to the respective features of the source task policy and the to-be-verified task policy. In this process, no watermark information needs to be embedded in the source task policy and the to-be-verified task policy, the performance of the task policy itself is completely preserved, and lossless task verification is achieved.

[0229] The above is a schematic scheme of a task verification apparatus of the present embodiment. It should be noted that the technical scheme of the task verification apparatus belongs to the same concept as the technical scheme of the task verification method described above, and the details of the technical scheme of the task verification apparatus that are not described in detail can be referred to the description of the technical scheme of the task verification method.

[0230] Corresponding to the method embodiments described above, the specification also provides model verification device embodiments, Figure 11 A structural schematic diagram of a model verification device provided by one embodiment of the specification is shown. As shown in the figure, Figure 11 The device comprises:

[0231] The second acquisition module 1102 is configured to acquire an image pair, wherein the image pair comprises a source domain image and a target domain image, and the source domain image and the target domain image are images containing the same image content and having different image attributes;

[0232] The second execution module 1104 is configured to execute a target task on the source domain image by using a source task model to obtain a first predicted image, and execute the target task on the source domain image by using a task model to be verified to obtain a second predicted image;

[0233] The second adjustment module 1106 is configured to balance the difficulty of task execution, adjust the source domain image according to the first predicted image and the target domain image to obtain a first verification image, and adjust the source domain image according to the second predicted image and the target domain image to obtain a second verification image;

[0234] The second determination module 1108 is configured to determine a model verification result according to the first verification image and the second verification image.

[0235] Optionally, the device further comprises a construction module configured to receive source model information of the source task model and to-be-verified model information of the task model to be verified sent by a user through a client; construct the source task model according to the source model information, and construct the task model to be verified according to the to-be-verified model information

[0236] By applying the scheme of the embodiments of the specification, the characteristics of the source task model can be determined through the first verification image, the characteristics of the task model to be verified can be determined through the second verification image, and the model verification result can be further determined according to the respective characteristics of the source task model and the task model to be verified. In this process, no watermark information needs to be embedded into the source task model and the task model to be verified, the performance of the task model itself is completely preserved, and lossless model copyright verification is achieved.

[0237] The above is a schematic scheme of a model verification device of the embodiment. It should be noted that the technical scheme of the model verification device belongs to the same concept as the technical scheme of the model verification method described above. The details of the technical scheme of the model verification device that are not described in detail can be referred to the description of the technical scheme of the model verification method.

[0238] Corresponding to the method embodiments described above, the specification also provides task verification device embodiments, Figure 12A structural diagram of another task verification device provided by an embodiment of the present specification is shown. As shown in Figure 12 The device includes:

[0239] The receiving module 1202 is configured to receive a task verification request sent by a target object, wherein the task verification request carries a task policy to be verified;

[0240] The third obtaining module 1204 is configured to obtain a first verification image, wherein the first verification image is obtained by adjusting an original domain image according to a first predicted image and a target domain image, the first predicted image is obtained by performing a target task on the original domain image based on a source task policy, and the original domain image and the target domain image are images containing the same image content and having different image attributes;

[0241] The third performing module 1206 is configured to perform the target task on the original domain image by using the task policy to be verified to obtain a second predicted image;

[0242] The third adjusting module 1208 is configured to adjust the original domain image according to the second predicted image and the target domain image to obtain a second verification image, with the balance of task execution difficulty as a target;

[0243] The third determining module 1210 is configured to determine a task verification result according to the first verification image and the second verification image.

[0244] By applying the scheme of the embodiment of the present specification, the owner of the source task policy can send the source task policy to the server in advance, and the server generates and stores the first verification image. When the owner of the source task policy has a task verification demand later, the owner can send the task policy to be verified to the server at any time. Thus, the server only needs to generate the second verification image each time the task verification is performed, compare the second verification image with the first verification image stored in advance to determine the task verification result, and the task verification efficiency is improved.

[0245] The above is a schematic scheme of a task verification device of the present embodiment. It should be noted that the technical scheme of the task verification device belongs to the same concept as the technical scheme of the task verification method described above, and the details of the technical scheme of the task verification device that are not described in detail can be referred to the description of the technical scheme of the task verification method.

[0246] Figure 13 A structural block diagram of a computing device provided by an embodiment of the present specification is shown. The components of the computing device 1300 include but are not limited to a memory 1310 and a processor 1320. The processor 1320 is connected with the memory 1310 through a bus 1330, and a database 1350 is used to save data.

[0247] The computing device 1300 also includes an access device 1340 that enables the computing device 1300 to communicate via one or more networks 1360. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of networks such as the Internet. The access device 1340 can include one or more of any type of network interface (for example, a network interface card (NIC)) such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, or the like.

[0248] In one embodiment of the present specification, the above-mentioned components of the computing device 1300 and other components not shown in the Figure 13 may be connected to each other, for example, through a bus. It should be understood that Figure 13 The computing device structure diagram shown is merely for the purpose of example, and is not a limitation on the scope of the present specification. Those skilled in the art can add or replace other components as needed.

[0249] The computing device 1300 can be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (for example, a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, and the like), a mobile phone (for example, a smartphone), a wearable computing device (for example, a smart watch, smart glasses, and the like), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 1300 can also be a mobile or stationary server.

[0250] The processor 1320 is configured to execute computer-executable instructions, which, when executed by the processor, implement the steps of the task verification method or the model verification method described above.

[0251] The above is a schematic scheme of the computing device of the embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the task verification method and the model verification method belong to the same concept, and the details of the technical scheme of the computing device that are not described in detail can be referred to the description of the technical scheme of the task verification method or the model verification method.

[0252] An embodiment of the present specification further provides a computer readable storage medium storing computer executable instructions, and the computer executable instructions are executed by a processor to implement the steps of the task verification method or the model verification method.

[0253] The above is a schematic scheme of the computer readable storage medium of the embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the task verification method and the model verification method belong to the same concept, and the details of the technical scheme of the storage medium that are not described in detail can be referred to the description of the technical scheme of the task verification method or the model verification method.

[0254] An embodiment of the present specification further provides a computer program, and the computer program causes a computer to execute the steps of the task verification method or the model verification method when the computer program is executed in the computer.

[0255] The above is a schematic scheme of the computer program of the embodiment. It should be noted that the technical scheme of the computer program and the technical scheme of the task verification method and the model verification method belong to the same concept, and the details of the technical scheme of the computer program that are not described in detail can be referred to the description of the technical scheme of the task verification method or the model verification method.

[0256] The above describes specific embodiments of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the acts or steps in a claim can be performed in an order different than the order in which the acts or steps are recited in the embodiments, and still accomplish the desired result. Also, the process depicted in the accompanying figures does not necessarily require the particular order shown, or sequential order to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0257] The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, software distribution medium, etc. It should be noted that the computer readable medium can include appropriate additions or deletions according to the requirements of patent practice, for example, according to the patent practice in some regions, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0258] It should be noted that for the foregoing method embodiments, in order to facilitate description, they are all expressed as a combination of a series of actions, but those skilled in the art should know that the embodiments of the present specification are not limited by the order of the described actions, because according to the embodiments of the present specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments of the present specification.

[0259] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0260] The preferred embodiments of the present specification disclosed above are only used to help explain the present specification. The alternative embodiments do not describe all the details and limit the invention to the specific embodiments described. Obviously, according to the content of the embodiments of the present specification, many modifications and changes can be made. The present specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of the present specification, so that those skilled in the art can well understand and use the present specification. The present specification is limited only by the claims and their entire scope and equivalents.

Claims

1. A task verification method, comprising: obtaining an image pair, wherein the image pair comprises a source domain image and a target domain image, the source domain image and the target domain image being images containing the same image content and having different image attributes; performing a target task on the source domain image by using a source task policy to obtain a first predicted image, and performing the target task on the source domain image by using a task policy to be verified to obtain a second predicted image; adjusting the source domain image according to the first predicted image and the target domain image to obtain a first verification image, and adjusting the source domain image according to the second predicted image and the target domain image to obtain a second verification image, wherein the first verification image is a target domain image corresponding to the adjusted source domain image, the adjusted source domain image is obtained by adjusting the source domain image based on a target function, the target function is obtained according to a first prediction loss and a first total variation loss, the first prediction loss is obtained according to the first predicted image and the target domain image, the first total variation loss is obtained according to the target domain image, the implementation of adjusting the source domain image according to the first predicted image and the target domain image to obtain the first verification image is the same as the implementation of adjusting the source domain image according to the second predicted image and the target domain image to obtain the second verification image, and the adjusted source domain image is located on a performance boundary domain of easy and difficult prediction of the task policy when the task execution difficulty balance is targeted; determining a task verification result according to the first verification image and the second verification image.

2. The method of claim 1, wherein adjusting the source domain image according to the first predicted image and the target domain image to obtain the first verification image comprises: determining a first prediction loss according to the first predicted image and the target domain image; determining a first total variation loss according to the target domain image; adjusting the source domain image according to the first prediction loss and the first total variation loss to obtain the first verification image.

3. The method of claim 2, wherein adjusting the source domain image according to the first prediction loss and the first total variation loss to obtain the first verification image comprises: adjusting the target domain image according to the first prediction loss and the first total variation loss to obtain an adjusted target domain image; determining an adjusted source domain image according to the adjusted target domain image, and returning to perform the step of performing the target task on the source domain image by using the source task policy to obtain the first predicted image until a preset stopping condition is reached to obtain the first verification image.

4. The method of claim 1, wherein obtaining the image pair comprises: obtaining a target domain image and a target processing policy corresponding to the target domain image; performing degradation processing on the target domain image according to the target processing policy to obtain a source domain image.

5. The method of claim 4, wherein obtaining the target domain image and the target processing policy corresponding to the target domain image comprises: identifying a task type of a target task; obtaining a target domain image according to the task type; screening a target processing strategy from a plurality of image processing strategies pre-set according to the task type, wherein the target processing strategy comprises at least one of an image denoising strategy, an image super-resolution strategy, an image deblurring strategy, and an image light enhancement strategy.

6. The method of claim 1, wherein the source task strategy comprises a source task model, and the task strategy to be verified comprises a task model to be verified. The target task is performed on the source domain image by using the source task strategy to obtain a first predicted image, and the target task is performed on the source domain image by using the task strategy to be verified to obtain a second predicted image, comprising: The target task is performed on the source domain image by using the source task model to obtain the first predicted image, and the target task is performed on the source domain image by using the task model to be verified to obtain the second predicted image.

7. The method of claim 1, wherein the task verification result is determined according to the first verification image and the second verification image, comprising: feature information of the first verification image is extracted, and feature information of the second verification image is extracted; the task verification result is determined according to the first feature information and the second feature information.

8. The method of claim 1, wherein the task verification result is determined according to the first verification image and the second verification image, comprising: the first verification image and the second verification image are sent to an image verification party, and image verification information sent by the image verification party is received; the task verification result is generated based on the image verification information.

9. The method of claim 1, after the task verification result is determined according to the first verification image and the second verification image, further comprising: in a case where the task verification result is that the task strategy to be verified is the same as the source task strategy, first verification information corresponding to the source task strategy is obtained, and second verification information corresponding to the task strategy to be verified is obtained; the task verification result is verified according to the first verification information and the second verification information to obtain a verification result.

10. A model verification method, comprising: obtaining an image pair, wherein the image pair comprises a source domain image and a target domain image, and the source domain image and the target domain image are images containing the same image content and having different image attributes; a target task is performed on the source domain image by using a source task model to obtain a first predicted image, and the target task is performed on the source domain image by using a task model to be verified to obtain a second predicted image; According to the first predicted image and the target domain image, the original domain image is adjusted to obtain a first verification image, and according to the second predicted image and the target domain image, the original domain image is adjusted to obtain a second verification image, wherein the first verification image is a target domain image corresponding to the adjusted original domain image, the adjusted original domain image is obtained by adjusting the original domain image based on a target function, the target function is obtained according to a first prediction loss and a first total variation loss with the task execution difficulty balance as the target, the first prediction loss is obtained according to the first predicted image and the target domain image, and the first total variation loss is obtained according to the target domain image. The implementation of adjusting the original domain image according to the first predicted image and the target domain image to obtain the first verification image is the same as the implementation of adjusting the original domain image according to the second predicted image and the target domain image to obtain the second verification image, and when the task execution difficulty balance is taken as the target, the adjusted original domain image is located on the performance boundary domain of easy prediction and difficult prediction of the task strategy. According to the first verification image and the second verification image, a model verification result is determined.

11. The method of claim 10, before the step of performing the target task on the original domain image by using the source task model to obtain the first predicted image, further comprising: receiving source model information of a source task model and to-be-verified model information of a to-be-verified task model sent by a user through a client; constructing a source task model according to the source model information and constructing a to-be-verified task model according to the to-be-verified model information.

12. A task verification method, comprising: receiving a task verification request sent by a target object, wherein the task verification request carries a to-be-verified task strategy; obtaining a first verification image, wherein the first verification image is obtained by adjusting an original domain image according to a first predicted image and a target domain image with a task execution difficulty balance as the target, the first verification image is a target domain image corresponding to the adjusted original domain image, the adjusted original domain image is obtained by adjusting the original domain image based on a target function, the target function is obtained according to a first prediction loss and a first total variation loss with the task execution difficulty balance as the target, the first prediction loss is obtained according to the first predicted image and the target domain image, and the first total variation loss is obtained according to the target domain image, when the task execution difficulty balance is taken as the target, the adjusted original domain image is located on a performance boundary domain of easy prediction and difficult prediction of the task strategy, the first predicted image is obtained by performing a target task on the original domain image based on a source task strategy, and the original domain image and the target domain image are images containing the same image content and different image attributes; performing a target task on the original domain image by using the to-be-verified task strategy to obtain a second predicted image; According to the second predicted image and the target domain image, the original domain image is adjusted to obtain a second verification image, wherein the implementation of adjusting the original domain image according to the first predicted image and the target domain image to obtain the first verification image with the task execution difficulty balance as the target is the same as the implementation of adjusting the original domain image according to the second predicted image and the target domain image to obtain the second verification image; According to the first verification image and the second verification image, a task verification result is determined.

13. A computing device, comprising: a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions, and the computer executable instructions, when executed by the processor, implement the steps of the method in any one of claims 1 to 9 or any one of claims 10 to 11 or claim 12.

14. A computer readable storage medium storing computer executable instructions, and the computer executable instructions, when executed by a processor, implement the steps of the method in any one of claims 1 to 9 or any one of claims 10 to 11 or claim 12.

Images