Evaluation methods, devices, and media for generative models

Abstract

The embodiments of the present application provide a method, device, and medium for evaluating a generative model and generating a design drawing. The method includes: obtaining a set of images generated by a generative model; performing a diversity analysis on the images in the image set to determine image pairs with diversity associations; determining a subset of images that meet association conditions based on the image pairs with diversity associations, and configuring identification information for the image subsets; determining a diversity parameter for the generative model based on the amount of identification information; and providing the diversity parameter to facilitate selection of a desired generative model. This decouples diversity evaluation from image quality, more intuitively reflecting generative model diversity. The diversity parameter can also be provided to facilitate selection of a desired generative model to generate a design drawing.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method for evaluating a generation model, a method for evaluating a video generation model, a terminal device, and a machine-readable medium. Background Art

[0002] Currently, various product designs are manually designed. For example, in the fashion industry, the global fashion market accounts for 2% of global GDP and is projected to grow at an annual rate of 7.31%. Despite this enormous market potential, fashion design is still performed manually by designers, resulting in relatively low design efficiency.

[0003] Currently, image generation models are used to generate images to assist in design. While generative models have advantages in image quality, there is no way to evaluate the diversity of the generated images. Summary of the Invention

[0004] An embodiment of the present application provides a method for evaluating a generative model to evaluate the diversity of the generative model.

[0005] Accordingly, an embodiment of the present application also provides a method for evaluating a video generation model, an electronic device, and a machine-readable medium to ensure the implementation and application of the above method.

[0006] In order to solve the above problems, the present invention discloses a method for evaluating a generative model, which includes:

[0007] Get the image set generated by the generative model;

[0008] performing diversity analysis on the images in the image set to determine image pairs with diversity associations;

[0009] Determining an image subset that meets association conditions based on the image pairs with diverse associations, and configuring identification information for the image subset;

[0010] Determining a diversity parameter of the generation model according to the amount of identification information;

[0011] The diversity parameter is provided so as to select a desired generative model.

[0012] Optionally, performing diversity analysis on the images in the image set to determine image pairs with diversity associations includes:

[0013] forming image pairs from two images in the image set, and determining diversity features of the image pairs;

[0014] When the diversity feature satisfies a diversity threshold, the image pair is determined to be an image pair with diversity association.

[0015] Optionally, determining the diversity feature of the image pair comprises at least one step:

[0016] performing similarity analysis on the image pairs and determining a global similarity as a diversity feature;

[0017] performing similarity analysis on designated regions of the image pair and determining local similarity as a diversity feature;

[0018] Performing a difference analysis on the image pairs, and determining difference features as diversity features.

[0019] Optionally, determining the image subset that meets the association condition based on the image pairs with diverse associations includes:

[0020] An undirected graph is created based on the image set and the image pairs with diverse associations, and an image subset meeting an association condition is determined based on the undirected graph.

[0021] Optionally, creating an undirected graph based on the image set and the image pairs with diverse associations includes:

[0022] Images in the image set are used as nodes, edges are created between corresponding nodes of image pairs with diverse associations, and a corresponding undirected graph is generated.

[0023] Optionally, determining the image subset that meets the association condition based on the undirected graph includes:

[0024] Determining node paths with diverse associations based on edges between nodes in the undirected graph;

[0025] It is determined that the images corresponding to the nodes on the node path meet the association conditions and are added to the corresponding image subset.

[0026] Optionally, also include:

[0027] A generation model is selected based on the diversity parameter, and a plurality of design drawings of the target object are generated using the generation model.

[0028] Optionally, also include:

[0029] Displaying the design drawing on the design page;

[0030] In response to triggering the edit control, a selected design element in the design drawing is adjusted.

[0031] The present application also discloses a method for evaluating a video generation model, the method comprising:

[0032] Get the video collection generated by the video generation model;

[0033] Performing diversity analysis on the videos in the video collection to determine video pairs with diversity associations;

[0034] Determining a video subset that meets association conditions based on the video pairs with diverse associations, and configuring identification information for the video subset;

[0035] A diversity parameter of the video generation model is determined according to the amount of identification information, so as to select a video generation model based on the diversity parameter.

[0036] Optionally, performing diversity analysis on the videos in the video set to determine video pairs with diversity associations includes:

[0037] Forming video pairs from the video set in pairs, and determining diversity features of the video pairs;

[0038] When the diversity feature meets a diversity threshold, the video pair is determined to be a video pair with diversity association.

[0039] Optionally, determining a subset of videos meeting an association condition based on the video pairs having diverse associations includes:

[0040] An undirected graph is created based on the video set and video pairs with diverse associations, and a video subset meeting an association condition is determined based on the undirected graph.

[0041] An embodiment of the present application further discloses an electronic device, comprising: a processor; and a memory on which executable code is stored. When the executable code is executed, the processor executes the method described in the embodiment of the present application.

[0042] The embodiments of the present application further disclose one or more machine-readable media on which executable codes are stored. When the executable codes are executed, the processor executes the method described in the embodiments of the present application.

[0043] Compared with the prior art, the embodiments of the present application have the following advantages:

[0044] In an embodiment of the present application, a set of images generated by a generative model can be obtained, and then a diversity analysis can be performed on the images in the image set to determine image pairs with diversity associations. Then, analysis can be performed based on the image pairs with diversity associations to determine a subset of images that meet the association conditions, and identification information can be configured for the image subset. Based on the amount of identification information, the diversity parameters of the generative model can be determined, thereby decoupling the diversity evaluation from the image quality, more intuitively reflecting the diversity of the generative model, and providing parameters for the selection of generative model generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a schematic diagram of an example of an evaluation method for a generative model according to an embodiment of the present application;

[0046] Figure 2 This is a flowchart of the steps of an evaluation method for generating a model according to an embodiment of the present application;

[0047] Figure 3 is a schematic diagram of an example of an undirected graph in an embodiment of the present application;

[0048] Figure 4 is a flowchart of the steps of another generation model evaluation method according to an embodiment of the present application;

[0049] Figure 5 This is a flowchart of the steps of a design drawing generation method according to an embodiment of the present application;

[0050] Figure 6 This is a flowchart of another method for generating a design drawing according to an embodiment of the present application;

[0051] Figure 7 This is a flowchart of the steps of a method for evaluating a video generation model according to an embodiment of the present application;

[0052] Figure 8 It is a structural diagram of a device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0054] The embodiments of the present application can be applied to the evaluation and use scenarios of generative models, including various image generation models, video generation models, etc. Taking the generative model designed for the target object as an example, the target object is the object to be designed, which can be a physical object or a virtual object. The target object can be a commodity object or other objects. For example, in an educational scenario, the target object can be an object such as a teaching aid. Taking the commodity object as an example, it can include various types, such as fashion, electrical appliances, daily necessities, etc. For example, the teaching aid object can be a teaching aid such as a set square or a globe, or a teaching aid such as a PPT. The design drawing of the target object can be generated by the generative model. The model for generating the design drawing can also be called a design model. The design model can be generated based on various machine learning and network learning models. In an optional embodiment, the generative model can be generated using an adversarial network model, such as StyleGAN, StyleGAN2 model, etc. The adversarial network model can include a mapping network and a synthesis network. The style vector can be obtained based on the mapping network mapping, and then combined with the synthesis network to generate an image. Taking the video generation model as an example, this video generation model can be used to generate various long and short videos.

[0055] For generative models, one of their evaluation indicators is diversity, which can be understood as the number of different objects (such as images, videos, etc.) generated by the generative model. Taking the image generation model as an example, the existing technology often measures the diversity of the generative model based on image quality when evaluating it. This method is also related to the number of samples, the number of generated images, etc., and has low accuracy. The embodiment of the present application proposes a method for evaluating the diversity of the generative model, which is decoupled from the quality of the image and focuses more on the differences in the image content itself.

[0056] Reference Figure 1 , which shows a schematic diagram of an example of a generative model evaluation method according to an embodiment of the present application.

[0057] Reference Figure 2 , shows a step flow chart of a generation model evaluation method according to an embodiment of the present application.

[0058] Step 202: Obtain an image set generated by the generative model.

[0059] When evaluating each generative model, images may be generated using the generative model. Based on the input description information, each generative model may generate multiple images, with the generated images forming an image set. After obtaining the image set, the diversity of the generative model may be evaluated based on the images in the image set. The images may include various types of images, such as static images and dynamic images.

[0060] Step 202: Perform diversity analysis on the images in the image set to determine image pairs with diversity associations.

[0061] Diversity analysis is performed on the images in the image collection. Diversity analysis can be performed in various ways. For example, similarity analysis can be used to identify similar images, and then the image diversity can be determined based on the similar images. Analysis can also be performed based on image differences, and diversity can be assessed based on the differences. Diversity analysis can also include global diversity and local diversity. For example, similarity can be analyzed between every two images to determine the similarity of image pairs, and similar image pairs can be identified to analyze diversity.

[0062] In an optional embodiment, the diversity analysis of the images in the image set is performed to determine image pairs with diversity associations, including: combining the images in the image set into image pairs in pairs, determining the diversity features of the image pairs; and determining that the image pairs are image pairs with diversity associations when the diversity features meet a diversity threshold. The images in the image set are combined in pairs to form image pairs, and then a diversity analysis is performed on each image pair. The diversity analysis can be based on similarity or difference. A diversity threshold is also set for the diversity feature. If the diversity feature is a similarity feature, the diversity threshold is a similarity threshold. If the diversity feature is a difference feature, the diversity threshold is a difference threshold, and so on. It can be determined whether the diversity feature meets the diversity threshold. If the diversity feature meets the diversity threshold, the image pair is determined to be an image pair with diversity associations.

[0063] Determining the diversity feature of the image pair includes at least one of the following steps: performing a similarity analysis on the image pair to determine a global similarity as a diversity feature; performing a similarity analysis on a specified region of the image pair to determine a local similarity as a diversity feature; and performing a difference analysis on the image pair to determine a difference feature as a diversity feature. Diversity can be determined by analyzing global similarity. Thus, the image pair can be analyzed for similarity as a whole to obtain a corresponding global similarity as a diversity feature. Diversity can also be determined by analyzing local similarity. Specified regions can be determined in each of the two images of the image pair. These specified regions can be determined as needed, such as regions with identical coordinates or regions that are correlated with the image pair. Similarity analysis is then performed on these specified regions to determine local similarity as a diversity feature. Diversity can also be determined by analyzing differences, such as by analyzing differences based on pixel values, pixel distribution, or other methods. These differences can also include global and local differences, thereby obtaining a difference feature as a diversity feature.

[0064] Taking diversity based on similarity analysis as an example, the images in the image set are combined into image pairs in pairs, and the similarity information of the image pairs is determined; when the similarity information meets the similarity threshold, the image pairs are determined to be similar image pairs. The images in the image set are combined into image pairs in pairs, and then similarity analysis is performed on each image pair to determine the similarity of the image pairs. In the embodiment of the present application, a similarity threshold is also pre-set, and the similarity threshold is used to evaluate the similarity of image pairs. When the similarity information of a certain image pair meets the similarity threshold, the image pair is determined to be a similar image pair. Optionally, the determination of the similarity information of the image pair includes: performing feature extraction on the images in the image pair respectively to determine the image feature pair; and calculating the cosine similarity of the image feature pair as the similarity information of the image pair. Image similarity can be measured using features within the images. Therefore, feature extraction can be performed for each image in each image pair. The extracted features can be determined based on requirements, image content, and other factors. For example, in one example, a self-supervised feature vector can be extracted. This self-supervised feature vector can be learned from the image's attributes and extracted using a corresponding machine learning model. The self-supervised feature vector pair for each image pair is then used as the image feature pair. For each image pair, similarity information can be determined based on the image feature pair. For example, cosine similarity can be calculated, which evaluates similarity by calculating the cosine of the angle between the two image feature vectors in the image feature pair. In other examples, the vector distance between the two image feature vectors in the image feature pair, as well as other linear or nonlinear similarity measures, can be used. The resulting similarity is used as the similarity information for the image pair. A determination can then be made as to whether this similarity information exceeds a similarity threshold. If so, the image pair is determined to be similar. If not, the image pair is determined to be non-similar.

[0065] Step 206 : determining an image subset that meets the association condition based on the image pairs with diverse associations, and configuring identification information for the image subset.

[0066] In the embodiment of the present application, image pairs with diverse associations are used to propagate diverse associations, thereby determining an image subset that meets the association conditions and configuring identification information for the image subset. In one example, the images in the image subset are images with diverse associations and can share the same identification information.

[0067] Wherein, the method of determining the subset of images that meet the association conditions based on the image pairs with diversity associations includes: creating an undirected graph based on the image set and the image pairs with diversity associations, and determining the subset of images that meet the association conditions based on the undirected graph. Diversity can be propagated in the form of a graph, wherein the diversity characteristics of the associations between images have no directional relationship, so an undirected graph can be created to propagate diversity, thereby determining the subset of images that meet the association conditions based on the undirected graph. Specifically, the method of creating an undirected graph based on the image set and the image pairs with diversity associations includes: using images in the image set as nodes, creating edges between corresponding nodes of the image pairs with diversity associations, and generating a corresponding undirected graph. A node can be created for each image in the image set, and then two corresponding nodes can be found based on the image pairs with diversity associations, and edges can be created between the two nodes, thereby connecting the two corresponding nodes based on the nutritive relationship of the image pairs, and using the nodes to form a corresponding undirected graph, such as Figure 3 shown.

[0068] On the basis of constructing an undirected graph, the image subset that meets the association condition is determined based on the undirected graph, including: determining a node path with diverse associations based on the edges between nodes in the undirected graph; the image corresponding to each node on the node path meets the association condition, and a corresponding image subset is established. In the undirected graph, nodes with diverse associations can be determined based on the edges between nodes, thereby obtaining a node path with diverse associations. The nodes on the node path have diverse associations, that is, the diverse associations can be propagated through the node path, thereby treating all nodes on the node path as nodes with diverse associations. The images corresponding to these nodes with diverse associations meet the association condition and can be added to an image subset. Among them, some nodes have edges connecting one or two nodes, such as Figure 3 Node g has connecting edges with nodes h and f; some nodes have connecting edges with more than two nodes, such as Figure 3 If node c has connecting edges with nodes b, d, and f, all nodes that can be connected through the edges are added to the corresponding node path, such as Figure 3 One path is The images corresponding to the corresponding nodes a-i all belong to the same image, can be put into an image subset, and share the same identification information.

[0069] Taking diversity analysis based on similarity as an example, in the embodiment of the present application, similar images can be determined to belong to the same image, and the same identification information can be assigned to these images, so that the number of different images generated by the generative model can be determined, providing a data basis for diversity analysis. Therefore, multiple images belonging to the same image can be determined based on similar image pairs to form an image subset, thereby configuring identification information for the image subset. In the above steps, two-by-two image pairs are used to determine similar images. In order to improve the accuracy of diversity assessment, similarity can also be propagated based on similar image pairs. For example, if image A and image B are similar image pairs, and image B and image D are similar image pairs, then through the propagation of similarity, image A and image D are also similar images, and image A, image B, and image C are in the image subset belonging to the same image. Therefore, similarity propagation can be performed based on similar image pairs to determine multiple similar images, determine that the multiple similar images belong to the same image, and add them to the corresponding image subset.

[0070] In an optional embodiment, the method of determining the subset of images belonging to the same image based on the similar image pairs includes: creating an undirected graph based on the image set and similar image pairs, and determining the subset of images belonging to the same image based on the undirected graph. The propagation of similarity can be performed in the form of a graph, wherein the similarity between images has an undirected relationship, and thus an undirected graph can be created to propagate similarity, thereby determining the subset of images belonging to the same image based on the undirected graph. Specifically, the method of creating an undirected graph based on the image set and similar image pairs includes: using images in the image set as nodes, creating edges between corresponding nodes of similar image pairs, and generating a corresponding undirected graph. A node can be created for each image in the image set, and then two corresponding nodes can be found based on similar image pairs, and edges can be created between the two nodes, thereby connecting the two corresponding nodes based on the similarity relationship of the image pairs, and using the nodes to form a corresponding undirected graph, such as Figure 3 shown.

[0071] On the basis of constructing an undirected graph, the image subsets belonging to the same image are determined based on the undirected graph, including: determining a node path with similar relationships based on the edges between nodes in the undirected graph; the images corresponding to the nodes on the node path belong to the same image, and establishing a corresponding image subset. In the undirected graph, nodes with similar relationships can be determined based on the edges between nodes, thereby obtaining a node path with similar relationships. The nodes on the node path have similar relationships, that is, the similar relationships can be propagated through the node path, thereby treating all nodes on the node path as nodes with similar relationships, and the images corresponding to these nodes with similar relationships belong to the same image and can be added to an image subset. Among them, some nodes have edges connecting one or two nodes, such as Figure 3Node g has connecting edges with nodes h and f; some nodes have connecting edges with more than two nodes, such as Figure 3 If node c has connecting edges with nodes b, d, and f, all nodes that can be connected through the edges are added to the corresponding node path, such as Figure 3 One path is The images corresponding to the corresponding nodes a-i all belong to the same image, can be put into an image subset, and share the same identification information.

[0072] Thus, through the propagation of diversity, multiple nodes with diverse associations can be identified, and the images corresponding to these nodes can be placed into an image subset. If a node has no diversity association with other nodes, its corresponding image can also be treated as a separate image subset. Each image subset can be assigned an identification information, for example, numbered starting from 0, or other identification methods can be used, etc., which are not limited in the embodiments of the present application.

[0073] The above example uses an undirected graph to illustrate the propagation of diversity, identifying multiple images that meet the association criteria so that identification information can be uniformly determined. Alternatively, a matrix approach can be used, where elements in the matrix correspond to images. Each image-corresponding element is initially 0 and then numbered based on the diversity association. For example, the first two elements with a diversity association are numbered 1. Subsequently, all elements with a diversity association to the element numbered 1 are numbered 1. Subsequent elements that have no similarity to number 1 but have a diversity association with other elements are collectively numbered 2, and so on. This allows elements with a diversity association to share the same value, thereby determining that multiple elements with a diversity association are the same image, and the corresponding numbers can also serve as identification information for the image.

[0074] Step 208: Determine the diversity parameter of the generation model based on the amount of identification information.

[0075] In some examples, an image can be uniquely identified by identification information, or a diversity feature can be identified by identification information. Then, multiple images that meet the association conditions have the same identification information. Therefore, based on the number of identification information, the number of different images in the image set, the number of different diversity features, etc. can be determined, and the diversity parameter can be determined based on this number and the total number of images in the image set.

[0076] In another optional embodiment, the diversity parameter can be understood as the number of unique IDs UID 1M , UID 1MRepresents the number of images with unique IDs per 1 million generated images. Using a sampling scale of one million allows the diverse parameters to adapt to a broad and unbiased generation space. On this basis, counting the number of unique IDs can intuitively reflect the diversity of the generative model.

[0077] The embodiment of the present application uses propagation and clustering to delete duplicate images and calculate the number of unique IDs. Multiple experiments have verified that the same experimental results can be obtained, indicating that the embodiment is deterministic.

[0078] Step 210: providing the diversity parameter so as to select a desired generative model to generate an image.

[0079] After determining the diversity parameter of the generative model, the diversity parameter can be provided to users of the generative model, allowing them to select a generative model based on the diversity parameter. Of course, generative models also have other parameters, such as quality parameters and efficiency parameters, and users can comprehensively select the desired generative model based on these parameters. For example, if a user wants the generative model to produce a wider variety of design drawings and videos, they can select a generative model with a higher diversity parameter, thereby obtaining a wider variety of design drawings to assist in design or video production.

[0080] In summary, it is possible to obtain an image set generated by a generative model, then perform diversity analysis on the images in the image set to determine image pairs with diversity associations, and then perform analysis based on the image pairs with diversity associations to determine a subset of images that meet the association conditions, and configure identification information for the image subset. Based on the amount of identification information, the diversity parameters of the generative model are determined, thereby decoupling the diversity evaluation from the image quality, more intuitively reflecting the diversity of the generative model, and providing parameters for the selection of generative model generation.

[0081] Based on the above embodiments, the embodiments of the present application also provide a method for evaluating a generative model. Taking similarity analysis diversity as an example, it can combine undirected graphs to cluster similar images and determine the diversity parameters of the generative model.

[0082] Reference Figure 4 , shows a step flow chart of another generation model evaluation method according to an embodiment of the present application.

[0083] Step 402: Obtain an image set generated by the generative model.

[0084] Step 404: The images in the image set are grouped into image pairs, and similarity information of the image pairs is determined.

[0085] Herein, feature extraction may be performed on each of the images in the image pair to determine an image feature pair; and the cosine similarity of the image feature pair is determined as the similarity information of the image pair.

[0086] Step 406: Determine whether the similarity information is greater than a similarity threshold.

[0087] If yes, execute step 408; if no, return to step 404 to continue determining similarity information of other image pairs.

[0088] Step 408: Determine whether the image pair is a similar image pair.

[0089] When the similarity information meets a similarity threshold, the image pair is determined to be a similar image pair.

[0090] Step 410 : Using the images in the image set as nodes, creating edges between corresponding nodes of similar image pairs, and generating a corresponding undirected graph.

[0091] Step 410: Determine node paths with similar relationships based on the edges between nodes in the undirected graph.

[0092] Step 412: Determine that the images corresponding to the nodes on the node path belong to the same image, and add them to the corresponding image subset.

[0093] Step 414: configure identification information for the image subset.

[0094] Step 416: Determine the diversity parameter of the generation model based on the amount of identification information.

[0095] Step 418: providing the diversity parameter so as to select a desired generation model to generate a design diagram.

[0096] On this basis, after the user selects a generative model based on the diversity parameter, the generative model can be used to generate a design drawing. In an optional embodiment, a generative model is selected based on the diversity parameter and used to generate multiple design drawings of the target object. The input information required by the generative model can be input into the generative model, and the corresponding design drawings can be obtained through processing by the generative model.

[0097] On the basis of generating the design drawing, the design drawing may be displayed on the design page; and in response to triggering the editing control, the design elements selected in the design drawing may be adjusted.

[0098] After obtaining the design drawing of the target object, the design drawing can be output. The design drawing can be provided to the target user, for example, by sending it to the target user via a network, or by displaying it in the design page of the target user. In an embodiment of the present application, the target user can enter corresponding descriptive information on the design page, which can then trigger the generation of the design drawing, and the corresponding terminal device can generate a design drawing that meets the description in the above manner. It can then be displayed in the design area of ​​the design page for easy viewing by the user. Subsequent users can also perform required processing based on the design drawing, such as providing editing controls in the design page, and the user can edit and adjust the target object in the design drawing. For example, for the designer of the target object, the target object can be produced based on the design drawing.

[0099] On the basis of the above-mentioned embodiment, the embodiment of the present application can also predict the popularity of the design drawing and obtain the corresponding popularity information. Among them, the popularity information is used to characterize the popularity of the target object. After obtaining the design drawing, the popularity of the target object corresponding to the design drawing can also be predicted, so as to analyze whether it is used for production, etc. At least one product drawing similar to the design drawing can be searched; the popularity information of the at least one product drawing can be determined, and the popularity information of the design drawing can be predicted. The popularity of the target object in the design drawing can be predicted in combination with the current target object situation. Therefore, at least one product drawing similar to the design drawing can be searched, wherein at least one product drawing similar to the design drawing can be searched on e-commerce websites, social networking websites, etc., and the product evaluation information of the corresponding target object can be obtained. The product evaluation information may include information such as the number of likes, the number of collections, and the click-through rate. Therefore, the popularity information corresponding to each product drawing can be determined based on the product splicing information, and then the popularity information of the design drawing can be predicted based on the popularity information of each product drawing.

[0100] On the basis of the above embodiments, the embodiments of the present application further provide a method for generating a design drawing, which can combine the diversity parameter selection involved models to generate diverse design drawings.

[0101] Reference Figure 5 , showing a step flow chart of a method for generating a design drawing in an embodiment of the present application.

[0102] Step 502 : selecting a generative model based on a diversity parameter, wherein the diversity parameter represents the number of different images generated by the generative model, the different images having different identification information, and analyzing whether the images generated by the generative model belong to the same image.

[0103] The diversity parameters of the generation model can be determined through the above-mentioned embodiments, and the generation model can be selected based on the diversity parameters. For example, if the user hopes that the generation model can design more diverse design drawings, the generation model with a higher diversity parameter can be selected, so that more diverse design drawings can be obtained to assist in the design.

[0104] Step 504 : Generate multiple design drawings of the target object using the selected generation model.

[0105] After selecting a generative model, the required input information can be input into the generative model, and the corresponding design drawing can be obtained through processing by the generative model. As in the above embodiment, the required descriptive information can be obtained and input into the generative model, which can then process the information to obtain the corresponding design drawing. Each generative model can generate multiple design drawings.

[0106] Step 506: Provide the design drawing to the target user, so that the target user can perform required processing based on the design drawing.

[0107] The generated multiple design drawings can be provided to target users, allowing them to view the generated design drawings and manufacture corresponding products based on the design drawings. Alternatively, they can edit, modify, redesign, etc. based on the design drawings. Popularity prediction can also be performed based on the design drawings to assist in actual design.

[0108] The design of the embodiment of the present application can also be applied to teaching scenarios, such as teaching that requires the use of various teaching aids. For example, mathematics, geography, chemistry, physics, etc. all have their own teaching aids. English and Chinese also require some teaching aids such as pictures. Teaching aid objects can include physical objects and virtual objects. Taking physics teaching as an example, physical teaching aids can be ramps, carts, etc., while virtual teaching aids can be mechanics education videos of ramps and carts. The embodiment of the present application can automatically generate design drawings based on the description of the teaching aid objects.

[0109] Based on the above embodiments, the embodiments of the present application also provide a processing method based on commodity data, which can automatically generate a design drawing in combination with the description of the target object.

[0110] Reference Figure 6 , shows a step flow chart of another design drawing generation method according to an embodiment of the present application.

[0111] Step 602 : selecting a generative model based on a diversity parameter, wherein the diversity parameter represents the number of different images generated by the generative model, the different images having different identification information, and analyzing whether the images generated by the generative model belong to the same image.

[0112] The diversity parameters of the generation model can be determined through the above-mentioned embodiments, and the generation model can be selected based on the diversity parameters. For example, if the user hopes that the generation model can design more diverse design drawings, the generation model with a higher diversity parameter can be selected, so that more diverse design drawings can be obtained to assist in the design.

[0113] Step 604 : Obtain description information of the product object, and generate multiple design drawings of the product object based on the description information and the selected generation model.

[0114] After selecting a generative model, the required input information can be fed into the model, which then processes the information to generate the corresponding design. This input information is related to the specific product being designed, such as shoes, clothing, or figurines. The obtained product description is then fed into the generative model, which then processes the information to generate the corresponding design. Each generative model can generate multiple design drawings for a product.

[0115] Step 606: Provide the multiple design drawings to a target user, so that the target user can perform processing related to the product object based on the design drawings.

[0116] Multiple designs of product objects can be provided to target users, allowing them to view the generated designs and manufacture the corresponding products based on them. They can also edit, modify, and redesign the designs based on them. Popularity predictions can also be performed based on the designs to assist in actual design.

[0117] On the basis of the above embodiments, an embodiment of the present application further provides a method for evaluating a video generation model, which can evaluate the diversity of the video generation model based on the generated video data.

[0118] Reference Figure 7 , showing a step flow chart of a method for evaluating a video generation model in an embodiment of the present application.

[0119] Step 702: Obtain a video set generated by the video generation model.

[0120] Based on the output information, a video can be generated using a video generation model. For example, scripts, outlines, music, etc. can be used as input information to generate corresponding video data. The video data generated by the video generation model can be combined to form a video. Among them, the videos generated by the video generation model can include various types, such as short videos within a specified time range (such as within 5 minutes), and long videos exceeding the specified time range (such as 5 minutes).

[0121] Step 704: Perform diversity analysis on the videos in the video set to determine video pairs with diversity associations.

[0122] Diversity analysis is performed on the videos in the video collection. Diversity analysis can be performed in various ways. For example, similar videos can be determined through similarity analysis, and then the diversity of the videos can be determined based on the similar videos. Analysis can also be performed based on video differences, and diversity can be assessed based on the differences. In addition, diversity analysis can also include global diversity, local diversity, etc. For example, the similarity between each two videos can be analyzed to determine the similarity of video pairs, and similar video pairs can be identified to analyze diversity. Diversity analysis for videos can be based on images in the videos, for example, extracting all or part of the image frames from the video for diversity analysis.

[0123] The method of performing diversity analysis on the videos in the video collection to determine video pairs with diversity associations includes: combining the videos in the video collection into video pairs, determining the diversity characteristics of the video pairs; and determining the video pairs as video pairs with diversity associations when the diversity characteristics meet a diversity threshold. The videos in the video collection are combined into video pairs, and then performing diversity analysis on each video pair. This diversity analysis can be based on similarity or difference. A diversity threshold is also set for the diversity characteristics. If the diversity characteristics are similarity characteristics, the diversity threshold is a similarity threshold; if the diversity characteristics are difference characteristics, the diversity threshold is a difference threshold, etc. Whether the diversity characteristics meet the diversity threshold can be determined. If the diversity characteristics meet the diversity threshold, the video pairs are determined as video pairs with diversity associations. For example, all or part of the image frames can be extracted from the video for diversity analysis. Similar to the above-mentioned image analysis, the image pairs can be analyzed for similarity, difference, etc., to determine the diversity association of the corresponding video pairs based on the number or percentage of image pairs with diversity associations.

[0124] Step 706: Determine a video subset that meets the association condition based on the video pairs with diverse associations, and configure identification information for the video subset.

[0125] In the embodiment of the present application, the video pairs with diverse associations are used to propagate diverse associations, thereby determining a video subset that meets the association conditions and configuring identification information for the video subset. In one example, the videos in the video subset are videos with diverse associations and can share the same identification information.

[0126] The method of determining a subset of videos that meet the association conditions based on the video pairs with diversity associations includes: creating an undirected graph based on the video set and the video pairs with diversity associations, and determining a subset of videos that meet the association conditions based on the undirected graph. Diversity can be propagated in a graph manner, wherein the diversity characteristics of the associations between videos have no directional relationship, so an undirected graph can be created to propagate diversity, thereby determining a subset of videos that meet the association conditions based on the undirected graph. Specifically, the method of creating an undirected graph based on the video set and the video pairs with diversity associations includes: using the videos in the video set as nodes, creating edges between the corresponding nodes of the video pairs with diversity associations, and generating a corresponding undirected graph. A node can be created for each video in the video set, and then two corresponding nodes can be found based on the video pairs with diversity associations, and edges can be created between the two nodes, thereby connecting the two corresponding nodes based on the nurturing relationship of the video pairs, and using the nodes to form a corresponding undirected graph.

[0127] Based on constructing an undirected graph, determining a video subset that meets the association condition based on the undirected graph includes: determining a node path with diverse associations based on the edges between nodes in the undirected graph; the videos corresponding to each node on the node path meet the association condition, and establishing a corresponding video subset. In the undirected graph, nodes with diverse associations can be determined based on the edges between nodes, thereby obtaining a node path with diverse associations. The nodes on this node path have diverse associations, that is, the diverse associations can be propagated through the node path, so that all nodes on the node path are regarded as nodes with diverse associations. The videos corresponding to these nodes with diverse associations meet the association condition and can be added to a video subset.

[0128] Step 708: Determine a diversity parameter of the video generation model according to the amount of identification information, so as to select a video generation model based on the diversity parameter.

[0129] In some examples, a video can be uniquely identified by identification information, or a diversity feature can be identified by identification information. Then, multiple videos that meet the association conditions have the same identification information. Therefore, based on the number of identification information, the number of different videos in the video set, the number of different diversity features, etc. can be determined. The diversity parameter can be determined based on this number and the total number of videos in the video set.

[0130] After the diversity parameter of the generation model is determined, the diversity parameter can be provided to a user who uses the generation model, so that the user can select the generation model according to the diversity parameter.

[0131] This embodiment addresses the problem of being unable to independently measure the diversity of generative models. By decoupling diversity from image quality, this embodiment detects diversity based on the content characteristics of the image itself, achieving independent measurement of diversity. To address the inability to accurately distinguish different generative models when the number of generative patterns is large, this embodiment uses a sampling scale of millions of images, making it adaptable to a wide range of generative spaces; the results are more intuitive and better interpretable.

[0132] It should be noted that for the method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present application are not limited by the order of the actions described, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

[0133] On the basis of the above embodiments, this embodiment further provides an evaluation device for generating a model, which is applied to electronic devices such as terminal devices and server devices.

[0134] The acquisition module is used to obtain the image set generated by the generation model.

[0135] The diversity analysis module is used to perform diversity analysis on the images in the image set to determine image pairs with diversity associations.

[0136] The identification configuration module is used to determine an image subset that meets the association condition based on the image pairs with diverse associations, and configure identification information for the image subset.

[0137] The diversity analysis module is used to determine the diversity parameters of the generation model according to the amount of identification information.

[0138] The parameter providing module is used to provide the diversity parameters so as to select a desired generation model.

[0139] In summary, it is possible to obtain an image set generated by a generative model, then perform diversity analysis on the images in the image set to determine image pairs with diversity associations, and then perform analysis based on the image pairs with diversity associations to determine a subset of images that meet the association conditions, and configure identification information for the image subset. Based on the amount of identification information, the diversity parameters of the generative model are determined, thereby decoupling the diversity evaluation from the image quality, more intuitively reflecting the diversity of the generative model, and providing parameters for the selection of generative model generation.

[0140] The diversity analysis module is configured to form image pairs from each of the images in the image set, determine diversity features of the image pairs, and determine that the image pairs are image pairs with diversity association when the diversity features meet a diversity threshold.

[0141] The diversity analysis module is used to perform similarity analysis on the image pair and determine global similarity as a diversity feature; and / or perform similarity analysis on a specified area of ​​the image pair and determine local similarity as a diversity feature; and / or perform difference analysis on the image pair and determine difference features as a diversity feature.

[0142] The identification configuration module includes: a graph analysis submodule and an identification allocation submodule, wherein:

[0143] The graph analysis submodule is used to create an undirected graph based on the image set and the image pairs with diverse associations, and determine an image subset that meets the association conditions based on the undirected graph.

[0144] The identification allocation submodule is used to configure identification information for the image subset.

[0145] The graph analysis submodule is configured to use images in the image collection as nodes, create edges between corresponding nodes of pairs of images with diverse associations, and generate a corresponding undirected graph. Within the undirected graph, nodes with diverse associations are identified based on the edges between the nodes. Images corresponding to nodes on the node paths that meet association criteria are identified and added to the corresponding image subset.

[0146] It also includes: a model processing module, which is used to select a generation model based on the diversity parameters and use the generation model to generate multiple design drawings of the target object.

[0147] The adjustment module is used to display the design drawing on the design page; in response to triggering the editing control, adjust the design elements selected in the design drawing.

[0148] The prediction module is used to search for at least one product image similar to the design image; determine the popularity information of the at least one product image, and predict the popularity information of the design image.

[0149] Based on the above embodiment, the evaluation device of the generation model can also be applied to the evaluation of the video generation model.

[0150] The acquisition module is used to obtain the video collection generated by the video generation model.

[0151] The diversity analysis module is used to perform diversity analysis on the videos in the video set to determine video pairs with diversity associations.

[0152] The identification configuration module is used to determine a video subset that meets the association condition based on the video pairs with diverse associations, and configure identification information for the video subset.

[0153] The diversity analysis module is used to determine the diversity parameter of the video generation model according to the amount of identification information, so as to select the video generation model based on the diversity parameter.

[0154] The diversity analysis module is configured to form video pairs from the video set in pairs, determine diversity features of the video pairs, and determine that the video pairs are video pairs with diversity association when the diversity features meet a diversity threshold.

[0155] The identification configuration module is used to create an undirected graph based on the video set and video pairs with diverse associations, and determine a video subset that meets the association conditions based on the undirected graph.

[0156] The embodiment of the present application uses similarity propagation and clustering to delete duplicate images and calculate the number of unique IDs. Multiple experiments have verified that the same experimental results can be obtained, indicating that the embodiment is deterministic.

[0157] On the basis of the above embodiments, this embodiment further provides a design drawing generating device, which is applied to electronic devices such as terminal devices and server devices.

[0158] A selection module is used to select a generation model based on a diversity parameter, wherein the diversity parameter represents the number of different images generated by the generation model, the different images have different identification information, and the images generated by the generation model are analyzed to determine whether they belong to the same image.

[0159] The generation module is used to generate multiple design drawings of the target object by adopting the selected generation model.

[0160] A providing module is used to provide the design drawing to a target user, so that the target user performs required processing based on the design drawing.

[0161] Based on the above embodiments, a design scenario applied to a commodity object is taken as an example.

[0162] The selection module is configured to select a generative model based on a diversity parameter, wherein the diversity parameter represents the number of different images generated by the generative model, the different images having different identification information, and the images generated by the generative model are analyzed to determine whether they belong to the same image;

[0163] The generation module is configured to obtain description information of a product object and generate multiple design drawings of the product object based on the description information and a selected generation model;

[0164] The providing module is configured to provide the plurality of design drawings to a target user, so that the target user performs processing related to the commodity object based on the design drawings.

[0165] This embodiment addresses the problem of being unable to independently measure the diversity of generative models. By decoupling diversity from image quality, this embodiment detects diversity based on the content characteristics of the image itself, achieving independent measurement of diversity. To address the inability to accurately distinguish different generative models when the number of generative patterns is large, this embodiment uses a sampling scale of millions of images, making it adaptable to a wide range of generative spaces; the results are more intuitive and better interpretable.

[0166] An embodiment of the present application further provides a non-volatile readable storage medium, which stores one or more modules (programs). When the one or more modules are applied to a device, the device can execute instructions (instructions) of each method step in the embodiment of the present application.

[0167] The present application provides one or more machine-readable media having instructions stored thereon, which, when executed by one or more processors, cause an electronic device to perform one or more of the methods described in the above embodiments. In the present application, the electronic device includes various types of devices such as terminal devices and servers (clusters).

[0168] The embodiments of the present disclosure may be implemented as a device configured as desired using any appropriate hardware, firmware, software, or any combination thereof, which may include terminal devices, servers (clusters), and other electronic devices within a data center. Figure 8 An exemplary apparatus 800 that can be used to implement various embodiments described in this application is schematically illustrated.

[0169] For one embodiment, Figure 8 An exemplary apparatus 800 is shown having one or more processors 802, a control module (chip set) 804 coupled to at least one of the processor(s) 802, a memory 806 coupled to the control module 804, a non-volatile memory (NVM) / storage device 808 coupled to the control module 804, one or more input / output devices 810 coupled to the control module 804, and a network interface 812 coupled to the control module 804.

[0170] The processor 802 may include one or more single-core or multi-core processors, and the processor 802 may include any combination of general-purpose processors or dedicated processors (e.g., graphics processors, application processors, baseband processors, etc.). In some embodiments, the apparatus 800 can serve as a terminal device, server (cluster), or other device described in the embodiments of the present application.

[0171] In some embodiments, the apparatus 800 may include one or more computer-readable media (e.g., memory 806 or NVM / storage 808) having instructions 814 and one or more processors 802 configured in conjunction with the one or more computer-readable media to execute the instructions 814 to implement a module to perform the actions described in the present disclosure.

[0172] For one embodiment, the control module 804 may include any suitable interface controller to provide any suitable interface to at least one of the processor(s) 802 and / or any suitable device or component in communication with the control module 804 .

[0173] The control module 804 may include a memory controller module to provide an interface to the memory 806. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0174] The memory 806 can be used, for example, to load and store data and / or instructions 814 for the device 800. For one embodiment, the memory 806 can include any suitable volatile memory, such as a suitable DRAM. In some embodiments, the memory 806 can include double data rate type four synchronous dynamic random access memory (DDR4 SDRAM).

[0175] For one embodiment, the control module 804 may include one or more input / output controllers to provide interfaces to the NVM / storage device 808 and the input / output device(s) 810 .

[0176] For example, NVM / storage 808 may be used to store data and / or instructions 814. NVM / storage 808 may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable non-volatile storage device(s) (e.g., one or more hard disk drives (HDDs), one or more compact disk (CD) drives, and / or one or more digital versatile disk (DVD) drives).

[0177] The NVM / storage device 808 may include storage resources that are physically part of the device on which the apparatus 800 is installed, or it may be accessible to the device without being part of the device. For example, the NVM / storage device 808 may be accessible over a network via the input / output device(s) 810.

[0178] (One or more) input / output devices 810 may provide an interface for the apparatus 800 to communicate with any other appropriate device. The input / output device 810 may include a communication component, an audio component, a sensor component, etc. The network interface 812 may provide an interface for the apparatus 800 to communicate via one or more networks. The apparatus 800 may wirelessly communicate with one or more components of a wireless network according to any of one or more wireless network standards and / or protocols, for example, accessing a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof for wireless communication.

[0179] For one embodiment, at least one of the processor(s) 802 may be packaged together with the logic of one or more controllers of the control module 804 (e.g., a memory controller module). For one embodiment, at least one of the processor(s) 802 may be packaged together with the logic of one or more controllers of the control module 804 to form a system-in-package (SiP). For one embodiment, at least one of the processor(s) 802 may be integrated on the same die with the logic of one or more controllers of the control module 804. For one embodiment, at least one of the processor(s) 802 may be integrated on the same die with the logic of one or more controllers of the control module 804 to form a system-on-chip (SoC).

[0180] In various embodiments, the apparatus 800 may be, but is not limited to, a terminal device such as a server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). In various embodiments, the apparatus 800 may have more or fewer components and / or a different architecture. For example, in some embodiments, the apparatus 800 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including a touchscreen display), a non-volatile memory port, multiple antennas, a graphics chip, an application-specific integrated circuit (ASIC), and a speaker.

[0181] Among them, the main control chip can be used as a processor or control module in the detection device, sensor data, location information, etc. are stored in the memory or NVM / storage device, the sensor group can be used as an input / output device, and the communication interface may include a network interface.

[0182] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0183] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0184] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0185] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0186] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0187] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0188] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0189] The above is a detailed introduction to a generation model evaluation method and device, a design drawing generation method and device, a terminal device and a machine-readable medium provided by this application. Specific examples are used in this article to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method of this application and its core idea; at the same time, for general technical personnel in this field, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on this application.

Claims

1. A method for evaluating a generative model, characterized in that: The method comprises: Get the image set generated by the generative model; Performing diversity analysis on the images in the image set to determine image pairs with diversity associations, wherein the diversity analysis includes similarity analysis and / or difference analysis; Propagating diversity associations based on the image pairs with diversity associations, determining a subset of images that meet association conditions, and configuring identification information for the subset of images, wherein the propagation of the diversity associations includes: a graph method and / or a similarity analysis method, and the multiple images that meet the association conditions have the same identification information; Determining a diversity parameter of the generation model based on the amount of identification information, where the diversity parameter is the number of images corresponding to the identification information; The diversity parameter is provided so as to select a desired generative model, wherein the diversity parameter is positively correlated with the diversity of images generated by the generative model.

2. The method according to claim 1, characterized in that The performing diversity analysis on the images in the image set to determine image pairs with diversity associations includes: forming image pairs from two images in the image set, and determining diversity features of the image pairs; When the diversity feature satisfies a diversity threshold, the image pair is determined to be an image pair with diversity association.

3. The method according to claim 2, characterized in that Determining the diversity characteristics of the image pair comprises at least one step: performing similarity analysis on the image pairs and determining a global similarity as a diversity feature; performing similarity analysis on designated regions of the image pair and determining local similarity as a diversity feature; Performing a difference analysis on the image pairs, and determining difference features as diversity features.

4. The method according to claim 1, wherein The propagating of diversity associations based on the image pairs having diversity associations to determine a subset of images meeting association conditions includes: An undirected graph is created based on the image set and the image pairs with diverse associations, and an image subset meeting an association condition is determined based on the undirected graph.

5. The method according to claim 4, characterized in that The step of creating an undirected graph based on the image set and the image pairs with diverse associations includes: Images in the image set are used as nodes, edges are created between corresponding nodes of image pairs with diverse associations, and a corresponding undirected graph is generated.

6. The method according to claim 4, characterized in that The determining, based on the undirected graph, a subset of images meeting the association condition comprises: Determining node paths with diverse associations based on edges between nodes in the undirected graph; It is determined that the images corresponding to the nodes on the node path meet the association conditions and are added to the corresponding image subset.

7. The method according to claim 1, characterized in that Also includes: A generation model is selected based on the diversity parameter, and a plurality of design drawings of the target object are generated using the generation model.

8. The method according to claim 7, characterized in that Also includes: Displaying the design drawing on the design page; In response to triggering the edit control, a selected design element in the design drawing is adjusted.

9. A method for evaluating a video generation model, characterized in that: The method comprises: Get the video collection generated by the video generation model; Performing diversity analysis on the videos in the video collection to determine video pairs with diversity associations, wherein the diversity analysis includes similarity analysis and / or difference analysis; Propagating diversity associations based on the video pairs with diversity associations, determining a video subset that meets association conditions, and configuring identification information for the video subset, wherein the propagation of the diversity associations includes: a graph method and / or a similarity analysis method, and the multiple images that meet the association conditions have the same identification information; Based on the amount of identification information, a diversity parameter of the video generation model is determined so that a video generation model is selected based on the diversity parameter, wherein the diversity parameter is the number of videos corresponding to the identification information, wherein the diversity parameter is positively correlated with the diversity of videos generated by the video generation model.

10. The method according to claim 9, characterized in that The performing diversity analysis on the videos in the video set to determine video pairs with diversity associations includes: Forming video pairs from the video set in pairs, and determining diversity features of the video pairs; When the diversity feature meets a diversity threshold, the video pair is determined to be a video pair with diversity association.

11. The method according to claim 9, characterized in that The propagating of the diversity association based on the video pairs with diversity association to determine a subset of videos meeting association conditions includes: An undirected graph is created based on the video set and video pairs with diverse associations, and a video subset meeting an association condition is determined based on the undirected graph.

12. An electronic device, characterized in that: include: processor; and A memory having executable codes stored thereon, which, when executed, causes the processor to perform the method according to any one of claims 1 to 11.

13. One or more machine-readable media having executable codes stored thereon, which, when executed, cause a processor to perform the method according to any one of claims 1 to 11.

Images