Image auditing method and related device

Abstract

The invention provides an image auditing method and a related device. The embodiment of the invention can be applied to the field of artificial intelligence. The method comprises the steps of extracting N candidate images to be audited from a database, inputting the N candidate images into a target screening model for screening, obtaining a target corresponding relation set composed of X preset tags and target images corresponding to preset tags in the X preset tags, and displaying the target corresponding relation set through an interactive interface, and checking the target images corresponding to the preset labels in the X preset labels obtained through screening by checking personnel, and inputting a checking result of the target images in the target corresponding relation set. According to the method, the target image meeting the requirement is screened out from the candidate images by using the target screening model, and the target image is audited by the auditor, so that the range of images needing to be audited by the auditor is reduced, and the auditing efficiency is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an image review method and related devices. Background Art

[0002] With the continuous development of computer vision technology, more and more users are showing their image creations to other users through the Internet. In order to ensure a civilized and friendly Internet environment, each Internet platform needs to review users' image creations to purify the image works on each Internet platform.

[0003] Since the forms and contents of image creation vary, in order to ensure the accuracy of the review results, the Internet platform will set up special staff to review the image creation uploaded on the Internet platform in real time. This review method can ensure that the Internet platform conducts a civilization and friendliness review for each uploaded image creation, achieving the goal of purifying image works on the Internet platform.

[0004] However, the occurrence of sudden events may cause a certain type of image works to be unsuitable for display to platform users at a certain time. For example, when a social event ferments excessively and causes the parties involved in the social event to be cyberbullied, the image works related to the social event or the party involved are no longer suitable for widespread dissemination at this stage. Internet platforms are in urgent need of finding the image works related to the social event and the party involved for secondary review. As the number of images to be reviewed for secondary review is growing exponentially, how to improve the review efficiency has become an urgent problem to be solved. Summary of the invention

[0005] The embodiments of the present application provide a method and related devices for image review, which are used to screen out target images that meet the requirements from multiple candidate images and provide these target images to reviewers for review, thereby reducing the number of images that reviewers need to review and improving review efficiency.

[0006] One aspect of the present application provides a method for image review, comprising:

[0007] Get N candidate images, where N is a positive integer;

[0008] Inputting N candidate images into a target screening model to obtain a target correspondence set, the target correspondence set including X preset labels and correspondences between the preset labels and the target image in the X preset labels, the target image being included in the N candidate images, where X is a natural number;

[0009] When the target correspondence relationship set is not an empty set, the target correspondence relationship set is displayed through an interactive interface;

[0010] In response to the input for the target correspondence set, a review result of the target image in the target correspondence set is obtained.

[0011] Another aspect of the present application provides an image review device, comprising:

[0012] An acquisition module, used for acquiring N candidate images, where N is a positive integer;

[0013] An input module, used for inputting N candidate images into a target screening model to obtain a target correspondence set, wherein the target correspondence set includes X preset labels and correspondences between the preset labels in the X preset labels and the target image, wherein the target image is included in the N candidate images, and X is a natural number;

[0014] An interactive module, used for displaying the target corresponding relationship set through an interactive interface when the target corresponding relationship set is not an empty set;

[0015] The interaction module is also used to obtain the review result of the target image in the target correspondence set in response to the input of the target correspondence set.

[0016] In another implementation of the embodiment of the present application, the acquisition module is specifically used to acquire N candidate videos;

[0017] An input module is specifically used to input N candidate videos into a target screening model to obtain a target correspondence set, wherein the target correspondence set includes X preset tags and a correspondence between the preset tags and the target video in the X preset tags, the target video is included in the N candidate videos, and the screen of the target video at a target time corresponds to the preset tag;

[0018] The interactive module is specifically used to obtain the review result of the target video in the target correspondence set in response to the input of the target correspondence set.

[0019] In another implementation of the embodiment of the present application, the acquisition module is further used to obtain a processing solution corresponding to the audit result of the target image in the target correspondence set;

[0020] The device also includes an execution module, which is used to execute a processing solution corresponding to the audit result on the target image in the target correspondence set.

[0021] In another implementation of the embodiment of the present application, the review result of the target image in the target correspondence set includes that the target image is a violation image, and the processing scheme corresponding to the review result of the target image in the target correspondence set is to delete the target image;

[0022] Or, the review result of the target image in the target correspondence set includes that the target image is a compliant image, and the processing scheme corresponding to the review result of the target image in the target correspondence set is to retain the target image.

[0023] In another implementation of the embodiment of the present application, the interaction module is further used to obtain a target tag in response to an input of an operation instruction, where the target tag is included in the X preset tags;

[0024] The device also includes a selection module, which is used to select a target image corresponding to the target label from the target correspondence set;

[0025] An interactive module, specifically used to display a target image corresponding to a target label through an interactive interface;

[0026] The interactive module is specifically used to obtain the review result of the target image corresponding to the target tag in response to the input of the target image corresponding to the target tag.

[0027] In another implementation of the embodiment of the present application, the interactive module is specifically used to display any target image among the target images corresponding to the target tag through an interactive interface;

[0028] The interactive module is specifically used to obtain the review result of any target image in the target object corresponding to the target tag in response to the input of any target image in the target object corresponding to the target tag.

[0029] In another implementation of the embodiment of the present application, when the target correspondence relationship set is an empty set, the device further includes a generating module for generating a preset identifier, the preset identifier indicating that no review is required;

[0030] The interactive module is also used to display preset logos through an interactive interface.

[0031] In another implementation of the embodiment of the present application, the acquisition module is further used to acquire X preset labels and training images corresponding to each preset label in the X preset labels;

[0032] The generation module is further used to generate a target screening model according to the X preset labels and the training images corresponding to each of the X preset labels.

[0033] In another implementation of the embodiment of the present application, the acquisition module is specifically used to:

[0034] Get X preset tags;

[0035] Retrieve each preset label in the X preset labels in the database, obtain positive samples and negative samples corresponding to each preset label in the X preset labels, and the training images include positive samples and negative samples.

[0036] In another implementation of the embodiment of the present application, the ratio of A to M is a preset ratio, and A and M are positive integers;

[0037] Get the module, specifically for:

[0038] Retrieve a preset label from among X preset labels in a database, and obtain C positive samples and B negative samples corresponding to the preset label from among the X preset labels, where B and C are positive integers, C is less than A, and the sum of A and B is M;

[0039] Perform at least one of cropping, scaling, and flipping operations on the C positive samples to obtain A positive samples.

[0040] In another implementation of the embodiment of the present application, the generating module is specifically used to:

[0041] Generate an intermediate model according to the X preset labels and the training images corresponding to the preset labels in the X preset labels;

[0042] Get m verification images corresponding to the preset labels among X preset labels, where m is a positive integer;

[0043] According to m verification images and the intermediate model, calculate the accuracy index of the intermediate model;

[0044] If the accuracy index of the intermediate model is greater than the threshold, the intermediate model is used as the target screening model.

[0045] In another implementation of the embodiment of the present application, the execution module is further configured to, if the accuracy index of the intermediate model is less than or equal to the threshold, add a training image corresponding to each preset label in the X preset labels to obtain an updated training image corresponding to each preset label;

[0046] The generation module is further used to generate an updated intermediate model according to the X preset labels and the updated training images corresponding to each preset label;

[0047] The device also includes a calculation module, which is used to calculate the accuracy index of the updated intermediate model based on the m verification images and the updated intermediate model;

[0048] The generation module is also used to use the accuracy index of the updated intermediate model as a target screening model if the accuracy index of the updated intermediate model is greater than a threshold.

[0049] Another aspect of the present application provides a computer device, comprising:

[0050] memories, transceivers, processors, and bus systems;

[0051] Wherein, the memory is used to store programs;

[0052] The processor is used to execute the program in the memory, including executing the above-mentioned methods;

[0053] The bus system is used to connect the memory and the processor so that the memory and the processor can communicate with each other.

[0054] Another aspect of the present application provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer is enabled to execute the above-mentioned methods.

[0055] Another aspect of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided by the above aspects.

[0056] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0057] The embodiment of the present application provides a method and related device for image review. In the embodiment of the present application, N candidate images to be reviewed are extracted from a database, and the N candidate images are input into a target screening model for screening, and a target correspondence relationship set consisting of X preset tags and target images corresponding to the preset tags in the X preset tags is obtained. The target correspondence relationship set is then displayed through an interactive interface, so that the reviewer can review the target images corresponding to the preset tags in the X preset tags obtained after screening, and input the review results of the target images in the target correspondence relationship set. The target screening model is used to screen out target images that meet the requirements from multiple candidate images, and the reviewer reviews the target images, which narrows the range of images that the reviewer needs to review, thereby improving the review efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic diagram of an architecture of an image review system provided in a certain embodiment of the present application;

[0059] Figure 2 A flow chart of a method for image review provided in a certain embodiment of the present application;

[0060] Figure 3a A schematic diagram of an interactive interface provided by a certain embodiment of the present application;

[0061] Figure 3b Another schematic diagram of an interactive interface provided in a certain embodiment of the present application;

[0062] Figure 3c Another schematic diagram of an interactive interface provided in a certain embodiment of the present application;

[0063] Figure 4a Another schematic diagram of an interactive interface provided in a certain embodiment of the present application;

[0064] Figure 4b Another schematic diagram of an interactive interface provided in a certain embodiment of the present application;

[0065] Figure 4c Another schematic diagram of an interactive interface provided in a certain embodiment of the present application;

[0066] Figure 5 Another flow chart of a method for image review provided in a certain embodiment of the present application;

[0067] Figure 6a Another schematic diagram of an interactive interface provided in a certain embodiment of the present application;

[0068] Figure 6b Another schematic diagram of an interactive interface provided in a certain embodiment of the present application;

[0069] Figure 7 A schematic diagram of a flow chart of a model application phase provided for a certain embodiment of the present application;

[0070] Figure 8 A schematic diagram of a process flow of a video review stage provided for a certain embodiment of the present application;

[0071] Fig. 9 Another flow chart of a method for image review provided in a certain embodiment of the present application;

[0072] Fig.10a A schematic diagram of the relationship between the classification loss function weight and the accuracy of model recognition provided in an embodiment of the present application;

[0073] Fig.10b A schematic diagram of a PR curve provided in a certain embodiment of the present application;

[0074] Fig.11 A schematic diagram of a flow chart of a model training phase provided in a certain embodiment of the present application;

[0075] Fig.12 Another flow chart of a method for image review provided in a certain embodiment of the present application;

[0076] Fig.13 A schematic diagram of the structure of an image review device provided in one embodiment of the present application;

[0077] Fig.14 A schematic diagram of the structure of an image review device provided in one embodiment of the present application;

[0078] Fig.15 A schematic diagram of a server structure provided for a certain embodiment of the present application. DETAILED DESCRIPTION

[0079] The embodiments of the present application provide a method and related devices for image review, which reduces the number of images that auditors need to review and improves review efficiency by screening out target images that meet the requirements from multiple candidate images and providing these target images to auditors for review.

[0080] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "corresponding to" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0081] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0082] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0083] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0084] Artificial intelligence can be used to classify things in different application scenarios. For example, artificial intelligence can help humans classify the content displayed in an image, or artificial intelligence can help humans extract features from images, so that humans can know what objects have appeared in the image without browsing the entire image content. In further needs, artificial intelligence can also help humans filter the content in images, and filter out images with a certain type of features or containing a certain type of objects from multiple images.

[0085] To facilitate understanding of the technical solutions provided in the embodiments of the present application, some key terms used in the embodiments of the present application are explained here:

[0086] The YOLO (you look only once) algorithm uses a separate CNN model to achieve end-to-end target detection. The core idea is to use the entire image as the input of the network and directly regress the position of the bounding box and its category in the output layer.

[0087] With the continuous development of computer vision technology, more and more Internet platforms for image sharing have appeared in our lives. Users can publish their own image creations through video platforms. In order to ensure a civilized and friendly Internet environment, each Internet platform needs to review users' image creations.

[0088] Since the forms and contents of image creations vary, in order to ensure the accuracy of the audit results, the Internet platform will set up special auditors to audit the latest uploaded image creations on the Internet platform. This audit method can ensure that the Internet platform conducts a civilization and friendliness audit on each uploaded image creation. For example, it will delete, warn and ban accounts for video works that contain insulting words, blood, violence and bad guidance, thus achieving the goal of purifying image works on the Internet platform.

[0089] However, sudden events may cause a certain type of image work to be unsuitable for display to platform users at a certain time. For example, when a social event ferments excessively and causes the parties involved in the social event to be cyberbullied, the image works related to the social event or the party involved are no longer suitable for widespread dissemination at this stage. The Internet platform urgently needs to find the image works related to the social event and the party involved for secondary review. Since the secondary review of image works requires auditors to quickly review the existing image works on the Internet, the number of works to be reviewed has increased exponentially, and how to improve the review efficiency has become an urgent problem to be solved.

[0090] In order to solve the problem of the huge number of works to be reviewed, this application proposes that a target screening model can be used to screen the collected works to be reviewed, obtain works to be reviewed that meet the requirements, and provide the works to be reviewed that meet the requirements to reviewers for screening, thereby reducing the number of works to be reviewed by reviewers to the number of works to be reviewed that meet the requirements, greatly reducing the number of works that reviewers need to review, and improving the review efficiency while ensuring the quality of the review.

[0091] For easier understanding, see Figure 1 , Figure 1 FIG. 1 is an application environment diagram of the method for image review in the embodiment of the present application, such as Figure 1 As shown, the method for image audit in the embodiment of the present application is applied to the system for image audit. The system for image audit includes: a server and a terminal device; wherein the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected by wired or wireless communication, and the embodiment of the present application is not limited here.

[0092] The server first obtains N candidate images, where N is a positive integer; secondly, the N candidate images are input into the target screening model to obtain a target correspondence set corresponding to the N candidate images, wherein the target correspondence set includes X preset tags and the correspondence between the preset tags and the target image in the X preset tags, and the target image is included in the N candidate images, where X is a natural number. Finally, when the target correspondence set is not an empty set, the target correspondence set is displayed through an interactive interface, and in response to the data for the target correspondence set, the auditor's audit result for the target image in the target correspondence set is obtained.

[0093] The following is an introduction to the image review method in this application from the server's perspective. Figure 2 The image review method provided in the embodiment of the present application includes: step S110 to step S190. Specifically:

[0094] S110, obtaining N candidate images;

[0095] Obtain N candidate images, where N is a positive integer. The N candidate images may be pictures or video images.

[0096] For example, the auditor can collect N images from the database as candidate images. For example, when a hot event occurs, the Internet platform analyzes and finds that the hot event begins to ferment, and the spread of image works related to the hot event has caused online violence against the parties involved in the hot event. The Internet platform decides to urgently remove the image works related to the hot event and the parties involved. The auditor of the Internet platform can obtain N images from the Internet platform as candidate images, where the N images can be composed of all videos or pictures on the Internet platform.

[0097] Optionally, still using the above example, in order to further improve the reviewer's review efficiency and speed, the N candidate images can be all image works uploaded after the hot event occurs, or, the N candidate images can be all image works in the subdivided field corresponding to the hot event.

[0098] It is understandable that the description of the method and quantity of obtaining N candidate images here is only an example. In actual application, it should be set in combination with specific application scenarios and is not limited here.

[0099] S120, inputting N candidate images into a target screening model to obtain a target correspondence set;

[0100] After obtaining N candidate images, the N candidate images are input into a target screening model to obtain a target correspondence set, wherein the target correspondence set includes X preset labels and correspondences between the preset labels in the X preset labels and the target image, the target image is included in the N candidate images, and X is a natural number.

[0101] After obtaining N candidate images, a pre-trained target screening model is used to screen out a target image from the N candidate images, and the label corresponding to the target image is any one of the X preset labels.

[0102] For example, taking the example described in S110 above, when the hot event is school violence and the name of the person involved is Xiao Ming, the preset labels may be "school violence" and "Xiao Ming", and the target image may be an image work labeled "school violence" and "Xiao Ming" among the N candidate images.

[0103] It is understandable that the description here of the preset labels in the target screening model and the target images corresponding to the preset labels is only an example. In actual application, they should be set in combination with specific application scenarios and are not limited here.

[0104] S130, determining whether the target correspondence relationship set is an empty set;

[0105] After obtaining the target correspondence relationship set, it is determined whether the target correspondence relationship set is an empty set.

[0106] Exemplarily, when X preset labels in the target screening model are 0 preset labels, the target correspondence set is an empty set, or when the N candidate images do not include a target image corresponding to the X preset labels, the target correspondence set is an empty set.

[0107] It can be understood that the explanation here of the reason why the target correspondence set is an empty set is only an example. In actual application, it should be set in combination with specific application scenarios and is not limited here.

[0108] If not, execute step S140;

[0109] If so, execute step S180.

[0110] S140, displaying the target correspondence relationship set through an interactive interface;

[0111] When the target correspondence set is not an empty set, the target correspondence set is displayed through the interactive interface. There are multiple ways to display the target correspondence set through the interactive interface, which will be introduced below in conjunction with the accompanying drawings:

[0112] Method 1: If Figure 3aAs shown, the target correspondence set can be displayed through an interactive interface. Specifically, a list of correspondences in the target correspondence set can be displayed through an interactive interface. The auditor can obtain the target image in the correspondence by clicking on each correspondence in the list of correspondences, and audit the target image.

[0113] Method 2: If Figure 3b As shown, the target correspondence set can be displayed through an interactive interface. Specifically, the target correspondence set can be displayed through an interactive interface, with each preset label and a thumbnail of the corresponding target image. The reviewer can further browse the target image by clicking on each thumbnail and review the target image.

[0114] Method 3: If Figure 3c As shown, displaying the target correspondence set through the interactive interface can specifically display the target images corresponding to the preset labels in the target correspondence set in sequence through the interactive interface, and the auditor can review the target images in sequence.

[0115] It is understandable that the description of displaying the target correspondence set through the interactive interface here is only an example. In actual application, it should be set in combination with specific application scenarios, and no limitation is made here.

[0116] S150, in response to an input for a target correspondence set, obtaining a review result of a target image in the target correspondence set;

[0117] After the target correspondence set is displayed through the interactive interface, in response to input for the target correspondence set, a review result of the target image in the target correspondence set is obtained.

[0118] Due to the different ways of displaying the target correspondence set through the interactive interface, the ways of obtaining the audit result of the target image in the target correspondence set in response to the input of the target correspondence set are slightly different. Figure 3a , Figure 3b and Figure 3c To explain:

[0119] When the target correspondence set is displayed through the interactive interface, Figure 3a As shown, corresponding to the input of the target correspondence set, the method for obtaining the audit result of the target image in the target correspondence set is as follows Figure 4a As shown, the auditor can obtain the target image in the corresponding relationship after clicking each corresponding relationship in the list of corresponding relationships, select the preset audit result for the target image in the corresponding relationship through the interactive interface, or type the audit result through the preset area of ​​the interactive interface. There is no restriction here.

[0120] In an embodiment of the present application, a variety of review options are provided for the reviewer. When the reviewer reviews each target image corresponding to the preset label in turn according to the correspondence list, it avoids the situation where the target image is missed, and improves the accuracy of the reviewer's image review.

[0121] Furthermore, after reviewing the target image in at least one corresponding relationship in the correspondence list, the auditor can also click the "one-click review" area in the interactive interface to perform "one-click review" on the target images that have not completed the review and input a unified result for these target images that have not completed the review, or, uniformly type the review results for these target images that have not completed the review through the preset area of ​​the interactive interface.

[0122] In the embodiment of the present application, depending on the specific circumstances of the audit, the auditor can also complete the audit of all target images with one click in order to efficiently complete the audit, thereby providing more possibilities for the implementation of the solution.

[0123] When the target correspondence set is displayed through the interactive interface, Figure 3b As shown, corresponding to the input of the target correspondence set, the method for obtaining the audit result of the target image in the target correspondence set is as follows Figure 4b As shown, the auditor can lock the approximate audit scope through preset tags, click on the thumbnail of any target image, and select the preset audit result for the target image through the interactive interface, or type the audit result through the preset area of ​​the interactive interface. There is no restriction here.

[0124] Furthermore, the auditor clicks on the thumbnail of any target image and completes the review of the target image. After completing the review of the target image, the auditor can also select a preset tag and click the "one-click review" area in the interactive interface to achieve a unified review of all target images with unfinished reviews corresponding to the preset tag. The specific review method can be to uniformly input a unified result for these target images with unfinished reviews, or to uniformly type the review results for these target images with unfinished reviews through the preset area of ​​the interactive interface.

[0125] When the target correspondence set is displayed through the interactive interface, Figure 3c As shown, corresponding to the input of the target correspondence set, the method for obtaining the audit result of the target image in the target correspondence set is as follows Figure 4c As shown, the auditor can review the target image displayed on the interactive interface in turn, select a preset review result for the target image through the interactive interface, or type the review result through the preset area of ​​the interactive interface. There is no limitation here.

[0126] Furthermore, the auditor can also directly click on the "one-click audit" area in the interactive interface to realize a unified audit of all target images that have not completed the audit. The specific audit method can be to uniformly input a unified result for these target images that have not completed the audit, or to uniformly type the audit results for these target images that have not completed the audit through the preset area of ​​the interactive interface.

[0127] It can be understood that the description here of obtaining the review result of the target image in the target correspondence set in response to the input of the target correspondence set is an example. In actual application, it should be set in combination with the specific application scenario and is not limited here.

[0128] S160, obtaining a processing solution corresponding to the audit result of the target image in the target correspondence set;

[0129] After obtaining the audit result of the target image in the target correspondence set, a processing solution corresponding to the audit result of the target image in the target correspondence set is obtained.

[0130] Exemplarily, when obtaining the target correspondence relationship and the audit results of the target image, when the audit is completed by selecting a preset audit result, the processing scheme corresponding to the audit result of the target image in the target correspondence relationship set can be a processing scheme corresponding to the preset audit result set in advance by the technician through the interactive interface.

[0131] When the audit results of the target correspondence and the target image are obtained, and the auditor completes the audit by typing the audit results in the preset area of ​​the interactive interface, the processing plan corresponding to the audit results of the target image in the target correspondence set can be configured by the auditor through the interactive interface.

[0132] It can be understood that the processing scheme for obtaining the audit results of the target image in the target correspondence set here is only an example. In actual application, it should be set in combination with the specific application scenario and is not limited here.

[0133] S170, executing a processing scheme corresponding to the audit result on the target image in the target correspondence relationship set;

[0134] After obtaining the processing scheme corresponding to the audit result of the target image in the target correspondence set, the processing scheme corresponding to the audit result is executed on the target image in the target correspondence set.

[0135] Exemplarily, when the audit result of the target image in the target correspondence set is "the target image is an illegal image", the processing scheme corresponding to the audit result of the target image in the target correspondence set is "delete the target image" or "hide the target image" or "hide the target image for a certain period of time". According to the processing scheme corresponding to the audit result of the target image in the target correspondence set, the target image is deleted, hidden, or hidden for a certain period of time, and there is no limitation here.

[0136] When the audit result of the target image in the target correspondence set is "the target image is a compliant image", the processing scheme corresponding to the audit result of the target image in the target correspondence set is "retain the target image". The target image is retained according to the processing scheme corresponding to the audit result of the target image in the target correspondence set, and there is no restriction here.

[0137] In the embodiments of the present application, a variety of processing methods are provided for the cases where the target image is an illegal image or the target image is a compliant image, thereby improving the flexibility of the solution.

[0138] It can be understood that the description of the processing scheme corresponding to the audit result of the target image in the target correspondence set here is only an example. In actual application, it should be set in combination with the specific application scenario and is not limited here.

[0139] In the embodiment of the present application, the target image in the target correspondence set is processed according to the processing scheme corresponding to the audit result of the target image in the target correspondence set, which improves the automation of the scheme and reduces the workload of the auditor.

[0140] S180, generating a preset identifier;

[0141] When the target correspondence set is an empty set, a preset identifier is generated, wherein the preset identifier indicates that no review is required.

[0142] For example, when the target correspondence set is an empty set, an indication is generated that there is no target image among the N candidate images, or an indication is generated that there is no target image corresponding to the X preset labels among the N candidate images.

[0143] It is understandable that the description of the specific content of the preset logo here is only an example. In actual application, the preset logo can be any pre-agreed logo that informs the reviewer that N candidate images do not need to be reviewed, and there is no limitation here.

[0144] S190: Displaying a preset logo through an interactive interface.

[0145] After the preset mark is generated, it is displayed to the auditor through an interactive interface.

[0146] In an embodiment of the present application, when the target correspondence set is an empty set, a preset identifier is generated, and the target identifier is displayed to the auditor through an interactive interface, so that the auditor can quickly know that there is no image that needs to be reviewed among the N candidate images, providing a specific implementation method for the rapid implementation of the plan.

[0147] In the embodiment of the present application, N candidate images to be reviewed are extracted from the database, and the N candidate images are input into the target screening model for screening, and a target correspondence relationship set consisting of X preset labels and target images corresponding to the preset labels in the X preset labels is obtained, and then the target correspondence relationship set is displayed through an interactive interface, so that the reviewer can review the target images corresponding to the preset labels in the X preset labels obtained after screening, and input the review results of the target images in the target correspondence relationship set. The target screening model is used to screen out target images that meet the requirements from multiple candidate images, and the reviewer reviews the target images, which narrows the range of images that the reviewer needs to review, thereby improving the review efficiency.

[0148] In this application Figure 2 In an optional embodiment of the method for image review provided in the corresponding embodiment, please refer to Figure 5 , Figure 5 When the candidate image is a video, the image review method provided in the embodiment of the present application is further described. The image review method provided in the embodiment of the present application includes: step S210 to step S280. Specifically:

[0149] S210, obtaining N candidate videos;

[0150] Get N candidate videos, where N is a positive integer.

[0151] For example, the auditors can collect N videos from the database as candidate videos. For example, when an international security incident occurs, in order to avoid excessive dissemination of the security incident, which causes the Internet platform's video works to be filled with violence and conflict, causing the Internet platform's users to fall into panic and violent emotions, the auditors of the Internet platform can use all videos of the Internet platform as N candidate images, or select all videos under certain categories of the Internet platform as N candidate videos, such as selecting all videos under the "Current Affairs" category of the Internet platform as N candidate videos.

[0152] Furthermore, the N candidate videos may also be all video works uploaded by the Internet platform after the international security incident occurs, or all videos uploaded by a specific partition under the Internet platform after the international security incident occurs.

[0153] It is understandable that the description of the method and quantity of obtaining N candidate videos here is only an example, and in actual application, it should be set in combination with specific application scenarios, and is not limited here.

[0154] S220, inputting N candidate videos into a target screening model to obtain a target correspondence set;

[0155] After obtaining N candidate videos, the N candidate videos are input into a target screening model to obtain a target correspondence set, wherein the target correspondence set includes X preset tags and a correspondence between the preset tags in the X preset tags and a target moment of the target video, the target video is included in the N candidate videos, and X is a natural number.

[0156] After obtaining N candidate videos, a pre-trained target screening model is used to screen out a target video from the N candidate videos, and the label corresponding to the target video is any one of the X preset labels.

[0157] For example, when the international security incident is a war, the preset labels may be "plane", "tank" and "war", and the target moment of the target video may be the moment when features corresponding to the preset labels appear in the video works labeled "plane", "tank" and "war" among the N candidate videos.

[0158] For example, when the international case source accident is a war, the target correspondence set may include "tagname: airplane, videoID: 001, timestamp: 01:24".

[0159] It is understandable that the description here of the preset labels in the target screening model and the target images corresponding to the preset labels is only an example. In actual application, they should be set in combination with specific application scenarios and are not limited here.

[0160] S230, determining whether the target correspondence relationship set is an empty set;

[0161] After obtaining the target correspondence relationship set, it is determined whether the target correspondence relationship set is an empty set.

[0162] Exemplarily, when the X preset tags in the target screening model are 0 preset tags, the target correspondence set is an empty set, or when the N candidate video images do not include videos corresponding to the X preset tags, the target correspondence set is an empty set.

[0163] It can be understood that the explanation here of the reason why the target correspondence set is an empty set is only an example. In actual application, it should be set in combination with specific application scenarios and is not limited here.

[0164] If not, execute step S240;

[0165] If so, execute step S270.

[0166] S240, in response to input of the operation instruction, obtaining a target tag;

[0167] When the target corresponding relationship set is not an empty set, in response to the input of the operation instruction, a target label is obtained, wherein the target label is included in X preset labels.

[0168] The auditor selects a target label from X preset labels in the target correspondence set displayed on the interactive interface, and the target label is obtained in response to the auditor's input for the X preset labels.

[0169] For example, the interactive interface displays preset labels such as "plane", "tank" and "war". Based on the severity of international security accidents, auditors can give priority to the preset label "war", that is, the target label is "war".

[0170] It is understandable that the description of the method of obtaining the target tag here is only an example. In actual application, it should be set in combination with the specific usage scenario and is not limited here.

[0171] S250, displaying the target video corresponding to the target tag through an interactive interface;

[0172] After obtaining the target tag, the target video corresponding to the target tag is displayed through the interactive interface. There are multiple ways to display the target video corresponding to the target tag through the interactive interface, which will be introduced below:

[0173] Exemplarily, the target videos corresponding to the target tags can be displayed in sequence through the interactive interface. For example, when the target tag is "war", the video contents corresponding to the corresponding relationships such as "tagname: war, videoID: 0019, timestamp: 00:30", "tagname: war, videoID: 1980, timestamp: 00:05" and "tagname: war, videoID: 0310, timestamp: 02:18" are displayed in sequence. After watching the target video, the reviewer can complete the review of the target video corresponding to the target tag by clicking on the preset review result to obtain the review result of the target video corresponding to the target tag, or input the review result for the target video corresponding to the target video to obtain the review result of the target video.

[0174] Optionally, the target video corresponding to the target tag can be displayed through the interactive interface. The thumbnail of the target moment of the target video corresponding to the target tag can also be displayed through the interactive interface. For example, when the target tag is "War", the thumbnails of "tagname: War, videoID: 0019, timestamp: 00:30", "tagname: War, videoID: 1980, timestamp: 00:05" and "tagname: War, videoID: 0310, timestamp: 02:18" are displayed. The reviewer can complete a rough review of the target video by roughly viewing the thumbnails of the target moments of the target video, complete the review of the target video corresponding to the target tag by clicking on the preset review result, and obtain the review result of the target video corresponding to the target tag, or input the review result of the target video corresponding to the target video to obtain the review result of the target video.

[0175] Furthermore, when the thumbnail cannot meet the reviewer's review requirements, the reviewer can also click on the thumbnail of the target video to further review the target video. There is no restriction here.

[0176] In the embodiment of the present application, when the auditor uses the thumbnail of the target video for auditing and finds a video of concern, he or she can click to watch the full version of the video, which improves the flexibility of the solution and avoids missing out on special plots in a certain frame.

[0177] It is understandable that the description here of displaying the target video corresponding to the target tag through the interactive interface is only an example. In actual application, it should be set in combination with the specific application scenario and is not limited here.

[0178] S260, in response to an input of a target video corresponding to a target tag, obtaining a review result of the target video corresponding to the target tag;

[0179] After the target video corresponding to the target tag is displayed through the interactive interface, in response to the input of the target video corresponding to the target tag, the review result of the target video corresponding to the target tag is obtained.

[0180] Due to the different ways of displaying the target video corresponding to the target tag through the interactive interface, the ways of obtaining the review result of the target image in the target correspondence set in response to the input of the target video corresponding to the target tag are slightly different, which are described below in conjunction with the accompanying drawings:

[0181] For example, Figure 6aWhen the target tag is "War", the video contents corresponding to the corresponding relationships such as "tagname: War, videoID: 0019, timestamp: 00:30", "tagname: War, videoID: 1980, timestamp: 00:05" and "tagname: War, videoID: 0310, timestamp: 02:18" are displayed in sequence. The auditor can set the speed review and watch the entire content of the target video at the speed after focusing on the content of the target moment of the target video. After the review is completed, the auditor can click the preset review result in the interactive interface to complete the review of the target video corresponding to the target tag, and can also type the review result in the preset area of ​​the interactive interface to complete the review of the target video corresponding to the target tag. There is no restriction here.

[0182] Furthermore, after reviewing at least one of the target videos of the target tag, the reviewer can also click the "one-click review" area in the interactive interface to quickly review the target videos that have not completed the review, and uniformly assign review results to the target videos that have not completed the review.

[0183] Optional, such as Figure 6b As shown, when the target tag is "War", thumbnails of "tagname: War, videoID: 0019, timestamp: 00:30", "tagname: War, videoID: 1980, timestamp: 00:05" and "tagname: War, videoID: 0310, timestamp: 02:18" are displayed. The auditor can review the target video by roughly viewing the thumbnails of the target moments of the target videos corresponding to the target tags, and batch processing of the target videos corresponding to the target tags can be achieved by clicking the "One-click Review" area in the interactive interface.

[0184] Furthermore, when the auditor has doubts about a target video in the target video corresponding to the target tag during the rough review process, the auditor can click on the target video in the target video corresponding to the target tag for further review and enter the review result for the target video in the target video corresponding to the target tag. The review of other videos is still completed in batches, and the specific review result input method is the same as the above Figure 6b Similar to , no restrictions are imposed here.

[0185] It can be understood that the description here of obtaining the review results of the target video corresponding to the target tag in response to the input of the target video corresponding to the target tag are all examples. In actual applications, they should be set in combination with specific application scenarios and are not limited here.

[0186] In the embodiment of the present application, the method in which the auditor sets the target label through the interactive interface and audits the target image corresponding to the target label further reduces the auditor's audit workload and improves the auditor's work efficiency and the mobility of the solution.

[0187] S270, generating a preset identifier;

[0188] When the target correspondence set is an empty set, a preset identifier is generated, wherein the preset identifier indicates that no review is required.

[0189] For example, when the target correspondence set is an empty set, an identification is generated that there is no target video among the N candidate videos, or, a target video corresponding to X preset tags is generated that there is no target video among the N candidate videos.

[0190] It is understandable that the description of the specific content of the preset identifier here is only an example. In actual application, the preset identifier can be any pre-agreed identifier that informs the reviewer that N candidate videos do not need to be reviewed, and there is no limitation here.

[0191] S280. Displaying a preset logo through an interactive interface.

[0192] After the preset mark is generated, it is displayed to the auditor through an interactive interface.

[0193] In the embodiment of the present application, by collecting videos from the database as videos to be reviewed, the videos to be reviewed are input into the target screening model for screening, and a target correspondence set is obtained, the target correspondence set includes the correspondence between X preset tags and the target moments of the target videos corresponding to the preset tags in the X preset tags, the pictures at the target moments of the target videos correspond to the preset tags, the auditor reviews the target videos corresponding to the target tags displayed on the interactive interface, and inputs the review results for the target images corresponding to the target tags, thereby completing the review results of the target images. The target screening model is used to screen out the target moments of the target videos that meet the preset tags, further narrowing the scope of the auditor's review when reviewing the videos, thereby further improving the auditor's review efficiency.

[0194] In order to facilitate the understanding of the above Figure 5 The image review method introduced below is combined with Figure 7 A brief introduction to the process of the model application phase:

[0195] First, the system reads all the video files in the video library of the Internet platform in the background. Secondly, the generated model is used to identify and analyze the video files in the video library. The content tags that the auditor plans to review and the time of the video corresponding to the content tags are identified in all the video files in the video library of the Internet platform, and a target correspondence set is generated through an algorithm. Thirdly, the generated target correspondence set is stored in the corresponding field tags in the video table of the background database for later use. Finally, it is determined whether there are any video files that have not been processed by the generated model. If not, the screening is completed. If so, the video files that have not been processed by the generated model are input into the generated model for identification and analysis.

[0196] In order to facilitate the understanding of the above Figure 5 The image review method introduced in the previous Figure 5 In one possible implementation of the method for image review introduced in Figure 8 A brief introduction to the technical process of the video review stage:

[0197] Step 1: The reviewer enters the target label;

[0198] Step 2: The system background searches the tags field in the video table and returns a list of videos that meet the requirements, that is, a collection of target moments of target videos corresponding to target tags;

[0199] Step 3: Determine whether the video list is empty;

[0200] If yes, the auditor is prompted that there is no video that needs to be reviewed;

[0201] If not, proceed to step 4.

[0202] Step 4: Show the videos in the video list to the reviewer;

[0203] Step 5 or Step 6 can be performed depending on the auditor's preference.

[0204] Step 5: Perform one-click review of the target video corresponding to the target tag based on the result input by the reviewer;

[0205] Step 6: The reviewer reviews the target videos one by one to determine whether the target videos are compliant. If they are compliant, the target videos are marked as passed. If they are not compliant, the target videos are marked as failed.

[0206] In this application Figure 2 In an optional embodiment of the method for image review provided in the corresponding embodiment, please refer to Fig. 9 , Fig. 9Before using the target screening model, the process of obtaining the target screening model by training the model using the X preset tags and the training pictures corresponding to the preset tags in the X preset tags is described. The image review method provided in the embodiment of the present application includes: step S310 to step S360. Specifically:

[0207] S310, obtaining X preset tags;

[0208] Get X preset tags.

[0209] Exemplarily, the X preset tags are X preset tags set by the auditor or technician through the interactive interface, wherein the X preset tags may be content tags that the auditor plans to review.

[0210] Optionally, the X preset labels may also be preset labels preset by a technician and inputted to meet certain conditions.

[0211] It is understandable that the description of the method for obtaining the X preset tags here is only an example, and in actual application, it should be set in combination with specific application scenarios.

[0212] S320, searching each preset label in the X preset labels in the database to obtain M training images corresponding to each preset label in the X preset labels;

[0213] After obtaining X preset labels, each of the X preset labels is retrieved from the database to obtain M training images corresponding to each of the X preset labels. The M training images include A positive samples and B negative samples corresponding to each of the X preset labels, where A and B are positive integers, the sum of A and B is M, M is a positive integer, and the ratio of M to A is a preset ratio.

[0214] Specifically, the target screening model is based on image review caused by international security incidents, and when the international security incident is a war, the X preset labels can be "aircraft", "tank" and "war". Among them, the M training images corresponding to each preset label in the X preset labels include M training images corresponding to any one of the labels "aircraft", "tank" and "war".

[0215] Here, taking the case where the number of training images corresponding to each preset label in the X preset labels is M, and the number of training images corresponding to “aircraft” is M, as an example, obtaining M training images is described:

[0216] Retrieve "airplane" from the database and obtain A positive samples and B negative samples.

[0217] Optionally, when the number of positive samples obtained by retrieving “airplane” from the database is greater than A, A positive samples are randomly selected from the positive samples obtained by retrieval as positive samples of “airplane”.

[0218] Optionally, “aircraft” is retrieved from the database to obtain C positive samples and B negative samples, where C is a positive integer and C is less than A. Since the number of positive samples obtained by retrieving “aircraft” from the database is less than expected, at least one of cropping, scaling and flipping is performed on the retrieved C positive samples to obtain A positive samples, where the preset ratio of A to M is 4:5.

[0219] In an embodiment of the present application, when the collected training samples are insufficient, the collected training samples are cropped, scaled, and flipped for at least one operation to obtain sufficient training samples, thereby improving the reliability of the target screening model obtained through training.

[0220] Furthermore, in order to improve the training efficiency of the model, technicians or reviewers can also manually mark the category of "aircraft" and the location information of "aircraft" in the picture, and there is no restriction here.

[0221] It can be understood that the description here of retrieving a preset label in X preset labels in the database and obtaining M training images corresponding to the preset label in the X preset labels is only an example. In actual application, it should be set in combination with specific application scenarios, and no limitation is made here.

[0222] In an embodiment of the present application, when obtaining training samples corresponding to preset labels from a database, positive samples and negative samples in the collected training samples are set according to a certain ratio to ensure that the model obtained by training has a certain robustness.

[0223] S330, generating an intermediate model according to the X preset labels and M training images corresponding to the preset labels among the X preset labels;

[0224] An intermediate model is generated according to the X preset labels and M training images corresponding to the preset labels in the X preset labels.

[0225] Specifically, X preset labels and M training images corresponding to the preset labels in the X preset labels are input into the YOLOv8 model, the batch size of the YOLOv8 model is set to 16, the weight of the classification loss function is set to 0.8, and the training model of YOLOve is used for training to obtain the intermediate model. The relationship between the weight of the classification loss function and the accuracy of model recognition is as follows: Fig.10a .

[0226] It can be understood that the description of the specific implementation method of generating an intermediate model according to X preset labels and M training images corresponding to the preset labels in the X preset labels is only an example. In actual application, it should be set in combination with specific application scenarios and is not limited here.

[0227] S340, calculating the accuracy index of the intermediate model according to the m verification images and the intermediate model;

[0228] After the intermediate model is generated, the accuracy index of the intermediate model is calculated based on the M verification images and the intermediate model.

[0229] Specifically, after training the YOLOv8 model to obtain the intermediate model, use the YOLOv8 Val mode to verify the intermediate model and draw the following Fig.10b The PR curve shown is Fig.10b The PR curve shown gives the precision index of the intermediate model.

[0230] It is understandable that the description of the accuracy index for calculating the intermediate model here is only an example. In actual application, it should be set according to the specific situation.

[0231] S350, determining whether the accuracy index of the intermediate model is greater than a threshold;

[0232] After calculating the precision index of the intermediate model, it is determined whether the precision index of the intermediate model is greater than a threshold.

[0233] If yes, execute step S360;

[0234] If not, add training pictures to obtain updated training pictures, and use the updated training pictures to execute step S330.

[0235] Optionally, if not, the weight of the classification loss function may be increased, and step S330 may be performed, wherein the amount of the increased weight of the classification loss function is not limited here and should be set in combination with the specific application scenario.

[0236] It is understandable that the description here of the operation performed when the value of the accuracy index of the intermediate model is less than or equal to the threshold is only an example. In actual applications, the number of training images can be increased, the weight of the classification loss function can be increased, and step S330 can be executed. There is no limitation here.

[0237] In an embodiment of the present application, when the accuracy index of the intermediate model is less than or equal to a threshold, training images are added to obtain updated training images, and step S330 is executed using the updated training images to further train the model, thereby improving the reliability of the intermediate model and obtaining a target screening model that meets the requirements, thereby ensuring the reliability of the target screening model.

[0238] S360. Use the intermediate model as the target screening model.

[0239] If the precision index of the intermediate model is greater than the threshold, the intermediate model is used as the target screening model to complete the training of the target screening model.

[0240] In the embodiment of the present application, whether the accuracy index of the intermediate model is greater than a threshold is used as a basis for judging whether the accuracy of the intermediate model reaches the target, thereby further improving the accuracy of the target screening model obtained through training.

[0241] In the embodiment of the present application, after the auditor sets the preset label for the planned audit, the preset label is retrieved from the database to obtain the training image corresponding to each preset label in the M preset labels. When the number of training images corresponding to the preset label in the database is insufficient, the number of training images corresponding to the preset label is obtained by performing at least one of cropping, scaling and flipping on the retrieved training images. The number of training images of the target screening model is guaranteed, and the robustness of the target screening model is improved.

[0242] In order to facilitate the understanding of the above Fig. 9 The image review method introduced in the previous Fig. 9 In one possible implementation of the method for image review introduced in Fig.11 A brief introduction to the technical process of the model training phase:

[0243] Step 1: Prepare the content;

[0244] The auditor sets the preset labels for the planned audit through the interactive interface. The system searches the Internet for preset labels based on the preset labels set by the auditor to obtain 1,000 training images that match the label content.

[0245] Step 2: Preprocess the dataset;

[0246] The system filters, crops, scales, and flips the training images, and auditors manually mark the categories and location information of objects; secondly, it ensures that the ratio of positive samples to negative samples in image classification is 8:2.

[0247] Step 3: Training model;

[0248] First, select YOLOv8 as the base model and set the training model to YOLOv8l; second, set the batch size to 16 and the weight of the classification loss function to 0.8; third, use the training mode of YOLOv8 to train the model.

[0249] Step 4: Validate the model;

[0250] Use the test data set to evaluate the performance of the model. Determine whether the model performance meets the requirements by judging whether the model's accuracy index reaches an available height. If so, save and deploy the model. If not, adjust the model parameters or adjust the training image collection to further train the model.

[0251] To facilitate understanding of the complete solution proposed in this application, Fig.11 A method for image auditing of an image auditing device is introduced. Fig.12 It can be seen that the image review method is divided into two stages: the initialization stage and the video review stage.

[0252] In the initialization phase, the auditor inputs preset labels into the system, and the system fetches images (training images) corresponding to the preset labels from the Internet. The auditor preprocesses the set of training images to obtain a data set, and the developer uses the data set and preset labels to train the target screening model. The target screening model is then used to generate a target correspondence set.

[0253] During the video review stage, the auditor selects the target tag and inputs it into the system. The system displays the video content corresponding to the target tag to the auditor through an interactive interface. The auditor can review the video content corresponding to the target tag in batches or one by one, and input the review results into the system.

[0254] The following is a detailed description of the image review device in this application. Fig.13 . Fig.13 This is a schematic diagram of an embodiment of an image review device 10 in an embodiment of the present application. The image review device 10 includes:

[0255] An acquisition module 110 is used to acquire N candidate images, where N is a positive integer;

[0256] An input module 120 is used to input N candidate images into a target screening model to obtain a target correspondence set, wherein the target correspondence set includes X preset labels and correspondences between the preset labels in the X preset labels and the target image, wherein the target image is included in the N candidate images, and X is a natural number;

[0257] An interactive module 130, configured to display the target correspondence set through an interactive interface when the target correspondence set is not an empty set;

[0258] The interaction module 130 is further configured to obtain the review result of the target image in the target correspondence set in response to the input of the target correspondence set.

[0259] Optionally, an acquisition module 110 is specifically configured to acquire N candidate videos;

[0260] The input module 120 is specifically configured to input N candidate videos into a target screening model to obtain a target correspondence set, wherein the target correspondence set includes X preset tags and a correspondence between the preset tags in the X preset tags and the target video, wherein the target video is included in the N candidate videos, and a frame at a target moment of the target video corresponds to the preset tag;

[0261] The interaction module 130 is specifically configured to obtain the review result of the target video in the target correspondence set in response to the input of the target correspondence set.

[0262] In this application Fig.13 In an optional embodiment of the apparatus for image review provided in the corresponding embodiment, please refer to Fig.14 , the acquisition module 110 is also used to obtain a processing solution corresponding to the audit result of the target image in the target correspondence set;

[0263] The device also includes an execution module 140, which is used to execute a processing solution corresponding to the audit result on the target image in the target correspondence set.

[0264] Optionally, the review result of the target image in the target correspondence set includes that the target image is a violation image, and the processing scheme corresponding to the review result of the target image in the target correspondence set is to delete the target image;

[0265] Or, the review result of the target image in the target correspondence set includes that the target image is a compliant image, and the processing scheme corresponding to the review result of the target image in the target correspondence set is to retain the target image.

[0266] Optionally, the interaction module 130 is further configured to obtain a target tag in response to an input of an operation instruction, where the target tag is included in the X preset tags;

[0267] The device also includes a selection module 150, which is used to select a target image corresponding to the target label from the target correspondence set;

[0268] An interactive module 130 is specifically used to display a target image corresponding to a target tag through an interactive interface;

[0269] The interaction module 130 is specifically configured to obtain a review result of the target image corresponding to the target tag in response to an input of the target image corresponding to the target tag.

[0270] Optionally, the interactive module 130 is specifically configured to display any target image among the target images corresponding to the target tag through an interactive interface;

[0271] The interaction module 130 is specifically configured to obtain a review result of any target image in the target object corresponding to the target tag in response to an input of any target image in the target object corresponding to the target tag.

[0272] Optionally, when the target correspondence set is an empty set, the device further comprises a generating module 160, which is used to generate a preset mark, and the preset mark indicates that no review is required;

[0273] The interactive module 130 is further used to display a preset logo through an interactive interface.

[0274] Optionally, the acquisition module 110 is further configured to acquire X preset labels and training images corresponding to each of the X preset labels;

[0275] The generating module 160 is further configured to generate a target screening model according to the X preset labels and the training images corresponding to each of the X preset labels.

[0276] Optionally, the acquisition module 110 is specifically configured to:

[0277] Get X preset tags;

[0278] Retrieve each preset label in the X preset labels in the database, obtain positive samples and negative samples corresponding to each preset label in the X preset labels, and the training images include positive samples and negative samples.

[0279] Optionally, the ratio of A to M is a preset ratio, and A and M are positive integers;

[0280] The acquisition module 110 is specifically used for:

[0281] Retrieve a preset label from among X preset labels in a database, and obtain C positive samples and B negative samples corresponding to the preset label from among the X preset labels, where B and C are positive integers, C is less than A, and the sum of A and B is M;

[0282] Perform at least one of cropping, scaling, and flipping operations on the C positive samples to obtain A positive samples.

[0283] Optionally, the generating module 160 is specifically configured to:

[0284] Generate an intermediate model according to the X preset labels and the training images corresponding to the preset labels in the X preset labels;

[0285] Get m verification images corresponding to the preset labels among X preset labels, where m is a positive integer;

[0286] According to m verification images and the intermediate model, calculate the accuracy index of the intermediate model;

[0287] If the accuracy index of the intermediate model is greater than the threshold, the intermediate model is used as the target screening model.

[0288] In this application Fig.13In an optional embodiment of the apparatus for image review provided in the corresponding embodiment, please refer to Fig.14 , the execution module 140 is also used to, if the accuracy index of the intermediate model is less than or equal to the threshold, add a training image corresponding to each preset label in the X preset labels to obtain an updated training image corresponding to each preset label;

[0289] The generating module 160 is further used to generate an updated intermediate model according to the X preset labels and the updated training images corresponding to each preset label;

[0290] The apparatus further includes a calculation module 170 for calculating a precision index of the updated intermediate model based on the m verification images and the updated intermediate model;

[0291] The generating module 160 is further configured to use the precision index of the updated intermediate model as a target screening model if the precision index of the updated intermediate model is greater than a threshold.

[0292] Fig.15 : is a schematic diagram of a server structure provided in an embodiment of the present application. The server 300 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 322 (for example, one or more processors) and memory 332, and one or more storage media 330 (for example, one or more mass storage devices) storing application programs 342 or data 344. Among them, the memory 332 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the server. Furthermore, the central processing unit 322 can be configured to communicate with the storage medium 330 to execute a series of instruction operations in the storage medium 330 on the server 300.

[0293] The server 300 may also include one or more power supplies 326, one or more wired or wireless network interfaces 350, one or more input and output interfaces 358, and / or one or more operating systems 341, such as Windows Server 2000. TM , Mac OS X TM , Unix TM ,Linux TM , FreeBSD TM etc.

[0294] The steps performed by the server in the above embodiment can be based on the Fig.15 The server structure shown.

[0295] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0296] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0297] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0298] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0299] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., and other media that can store program codes.

[0300] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for image review, characterized in that: include: Obtain N candidate images, where N is a positive integer; Inputting the N candidate images into a target screening model to obtain a target correspondence set, wherein the target correspondence set includes X preset labels and correspondences between preset labels in the X preset labels and target images, wherein the target image is included in the N candidate images, and X is a natural number; When the target correspondence relationship set is not an empty set, displaying the target correspondence relationship set through an interactive interface; In response to the input for the target correspondence set, a review result of the target image in the target correspondence set is obtained.

2. The method according to claim 1, characterized in that The obtaining of N candidate images comprises: Get N candidate videos; The step of inputting the N candidate images into a target screening model to obtain a target correspondence set includes: Input the N candidate videos into a target screening model to obtain a target correspondence set, wherein the target correspondence set includes X preset tags and a correspondence between a preset tag in the X preset tags and a target video, wherein the target video is included in the N candidate videos, and a screen at a target moment of the target video corresponds to the preset tag; The step of obtaining the review result of the target image in the target correspondence set in response to the input of the target correspondence set includes: In response to an input for a target correspondence set, a review result of a target video in the target correspondence set is obtained.

3. The method according to claim 1, characterized in that After obtaining the review result of the target image in the target correspondence set in response to the input of the target correspondence set, the method further includes: Obtaining a processing solution corresponding to the audit result of the target image in the target correspondence set; A processing scheme corresponding to the audit result is executed on the target image in the target correspondence set.

4. The method according to claim 3, characterized in that The audit result of the target image in the target correspondence set includes that the target image is a violation image, and the processing scheme corresponding to the audit result of the target image in the target correspondence set is to delete the target image; Alternatively, the audit result of the target image in the target correspondence set includes that the target image is a compliant image, and the processing scheme corresponding to the audit result of the target image in the target correspondence set is to retain the target image.

5. The method according to claim 1, characterized in that Before displaying the target correspondence relationship set through the interactive interface, the method further includes: In response to input of an operation instruction, acquiring a target tag, wherein the target tag is included in the X preset tags; Before displaying the target correspondence set through the interactive interface, the method further includes: Selecting a target image corresponding to the target label from the target correspondence set; The displaying of the target correspondence set through an interactive interface includes: Displaying the target image corresponding to the target label through an interactive interface; The step of obtaining the review result of the target image in the target correspondence set in response to the input of the target correspondence set includes: In response to an input of a target image corresponding to the target tag, a review result of the target image corresponding to the target tag is obtained.

6. The method according to claim 5, characterized in that The displaying of the target image corresponding to the target tag through the interactive interface includes: Displaying any target image among the target images corresponding to the target label through an interactive interface; The step of obtaining, in response to an input of a target image corresponding to the target tag, a review result of the target image corresponding to the target tag comprises: In response to an input of any target image in the target object corresponding to the target tag, a review result of any target image in the target object corresponding to the target tag is obtained.

7. The method according to claim 1 or 2, characterized in that: When the target correspondence relationship set is an empty set, the method further includes: generating a preset mark indicating that no review is required; The preset logo is displayed through the interactive interface.

8. The method according to claim 1, 2 or 7, characterized in that: Before inputting the N candidate images into the target screening model to obtain the target correspondence set, the method further includes: Obtain the X preset labels and training images corresponding to each of the X preset labels; The target screening model is generated according to the X preset tags and the training images corresponding to each preset tag in the X preset tags.

9. The method according to claim 8, characterized in that The obtaining of the X preset labels and training images corresponding to the preset labels among the X preset labels includes: Obtain the X preset tags; Retrieve each preset tag in the X preset tags in the database, obtain a positive sample and a negative sample corresponding to each preset tag in the X preset tags, and the training image includes the positive sample and the negative sample.

10. The method according to claim 9, characterized in that The ratio of A to M is a preset ratio, and A and M are positive integers; The step of retrieving a preset label from the X preset labels in a database to obtain a positive sample and a negative sample corresponding to the preset label from the X preset labels includes: Retrieving a preset label from the X preset labels in a database, obtaining C positive samples and B negative samples corresponding to the preset labels from the X preset labels, where B and C are positive integers, C is less than A, and the sum of A and B is M; Perform at least one of cropping, scaling, and flipping on the C positive samples to obtain the A positive samples.

11. The method according to claim 8, characterized in that The generating the target screening model according to the X preset tags and the training pictures corresponding to the preset tags in the X preset tags includes: Generate an intermediate model according to the X preset tags and training images corresponding to the preset tags in the X preset tags; Obtain m verification images corresponding to preset labels among the X preset labels, where m is a positive integer; Calculating the precision index of the intermediate model according to the m verification images and the intermediate model; If the accuracy index of the intermediate model is greater than a threshold, the intermediate model is used as a target screening model.

12. The method according to claim 11, characterized in that The method further comprises: If the accuracy index of the intermediate model is less than or equal to the threshold, adding a training image corresponding to each preset label in the X preset labels to obtain an updated training image corresponding to each preset label; Generate an updated intermediate model according to the X preset labels and the updated training images corresponding to each preset label; Calculating the accuracy index of the updated intermediate model according to the m verification images and the updated intermediate model; If the accuracy index of the updated intermediate model is greater than the threshold, the accuracy index of the updated intermediate model is used as the target screening model.

13. An image review device, characterized in that: include: An acquisition module, used to acquire N candidate images, where N is a positive integer; An input module, used for inputting the N candidate images into a target screening model to obtain a target correspondence set, wherein the target correspondence set includes X preset labels and correspondences between preset labels in the X preset labels and a target image, wherein the target image is included in the N candidate images, and X is a natural number; An interactive module, used for displaying the target correspondence set through an interactive interface when the target correspondence set is not an empty set; The interaction module is further configured to obtain a review result of a target image in the target correspondence set in response to an input to the target correspondence set.

14. A computer device, characterized in that: include: memories, transceivers, processors, and bus systems; Wherein, the memory is used to store programs; The processor is used to execute the program in the memory, including executing the image review method according to any one of claims 1 to 12; The bus system is used to connect the memory and the processor so that the memory and the processor can communicate with each other.

15. A computer-readable storage medium comprising instructions, which, when executed on a computer, causes the computer to perform the image review method according to any one of claims 1 to 12.

16. A computer program product comprising a computer program, characterized in that The computer program is used by a processor to execute the image review method according to any one of claims 1 to 12.

Images