Optimization method and system for automatically generating immersive 3D scenes based on AIGC

Abstract

This application discloses an optimization method and system for automatically generating immersive 3D scenes based on AIGC. For a single pending item in a set of pending items, the second historical scene description information of the single pending item is analyzed based on a number of reference historical scene description information. Then, a target intelligent feature map is built based on the first historical scene description information and the second historical scene description information. The target intelligent feature map is the intelligent feature map corresponding to the pending item, and the intelligent feature map includes individual nodes with connection relationships. Finally, the target intelligent feature map is used to determine the target item that meets the relationship requirements with the pending items in the set of pending items, and the 3D scene is optimized through the target item. The target item belongs to the set of pending items or several reference items. In this way, the implementation scene can be completely restored, making the user experience more real and reliable.

Description

Technical Field

[0001] The present disclosure relates to the technical field of image optimization, and in particular to an optimization method and system for automatically generating an immersive 3D scene based on AIGC. Background Art

[0002] AIGC is a milestone for human society and artificial intelligence. In the short term, AIGC changes the basic productivity tools, in the medium term, it will change the social production relations, and in the long term, it will promote a qualitative breakthrough in the productivity of the entire society. In such productivity tools, production relations, and productivity changes, the value of production factors - data is greatly magnified.

[0003] In traditional technology, 3D scene optimization generally involves manual analysis of each image, feature extraction, and reconstruction of the 3D model. This not only wastes manpower costs but also time. In addition, the restored scene reality has certain defects, resulting in a poor user experience. Therefore, there is an urgent need for an artificial intelligence 3D scene optimization method to improve the above technical problems. Summary of the invention

[0004] In order to improve the technical problems existing in the related technologies, the present disclosure provides an optimization method and system for automatically generating immersive 3D scenes based on AIGC.

[0005] In a first aspect, an optimization method for automatically generating an immersive 3D scene based on AIGC is provided, the method comprising:

[0006] Obtaining reference historical scenario description information of a set of pending items and a plurality of reference items, wherein the pending items in the set of pending items include independent pending items and non-independent pending items, the independent pending items are not included in the plurality of reference items, and the non-independent pending items are included in the plurality of reference items, and the reference historical scenario description information is used to indicate a relationship between at least two reference items in the plurality of reference items;

[0007] For a non-individual pending item, first historical scene description information of the non-individual pending item is determined according to the reference historical scene description information of the non-individual pending item, wherein the first historical scene description information is used to represent the relationship between the non-individual pending item and one or more reference items among the plurality of reference items except the non-individual pending item;

[0008] For a separate pending item, analyzing the second historical scene description information of the separate pending item according to the plurality of reference historical scene description information, wherein the second historical scene description information is used to indicate the relationship between the separate pending item and one or more of the plurality of reference items;

[0009] Building a target intelligent feature map based on the first historical scene description information and the second historical scene description information;

[0010] The target intelligent feature map is used to determine a target item that meets the relationship requirements with the items to be processed in the set of items to be processed, and the 3D scene is optimized through the target item, and the target item belongs to the set of items to be processed or the several reference items.

[0011] In the present application, the step of analyzing the second historical scene description information of the separate to-be-processed item according to the plurality of reference historical scene description information includes:

[0012] Determining common items of the individual items to be processed from the several reference items;

[0013] Obtaining third historical scene description information covering the common matters from the plurality of reference historical scene description information;

[0014] The second historical scene description information of the separate item to be processed is determined based on the third historical scene description information.

[0015] In the present application, the determining the second historical scene description information of the separate to-be-processed item based on the third historical scene description information includes:

[0016] Obtaining a debugging coefficient corresponding to the third historical scene description information;

[0017] The third historical scene description information whose debugging coefficient is higher than the first debugging target value is determined as the second historical scene description information of the separate matter to be processed.

[0018] In this application, the common items of the separate items to be processed are determined from the several reference items, including:

[0019] Obtaining descriptive factor data of the individual item to be processed and descriptive factor data of each reference item, wherein the descriptive factor data includes at least one of brightness information, scene type information, and color information;

[0020] From each of the reference items, common items whose descriptive factor data meet the commonality coefficient requirements with the individual items to be processed are determined.

[0021] In the present application, each reference historical scene description information corresponds to a debugging coefficient, and the determining of the first historical scene description information of the non-individual to-be-processed item according to the reference historical scene description information of the non-individual to-be-processed item includes:

[0022] Obtaining the debugging coefficient corresponding to the reference historical scenario description information of each non-individual pending item;

[0023] The reference historical scene description information of the non-individual to-be-processed item whose debugging coefficient is higher than the second debugging target value is determined as the first historical scene description information of the non-individual to-be-processed item.

[0024] In the present application, the method of using the target intelligent feature map to determine a target item that meets the relationship requirements with the items to be processed in the set of items to be processed, and optimizing the 3D scene through the target item, includes:

[0025] Obtaining the number of historical scene description information covered by the target intelligent feature map;

[0026] When the number of the historical scene description information exceeds the target number value, determining the associated items of the set of items to be processed from the target intelligent feature map, wherein the associated items are items that have a matching relationship with the set of items to be optimized;

[0027] From the associated items, a target item that meets the relationship requirement with the items to be processed in the set of items to be processed is determined.

[0028] In the present application, the target intelligent feature map includes the debugging coefficient of the historical scene description information, and determining the associated items of the set of items to be processed from the target intelligent feature map includes:

[0029] Determine, from the target intelligent feature map, pending items that meet the association requirements with the set of pending items;

[0030] Determining pending historical scenario description information covering the pending matters;

[0031] Obtaining a debugging coefficient for each pending historical scene description information from the target intelligent feature map;

[0032] Determine the pending historical scene description information whose debugging coefficient is higher than the third debugging target value as the selected historical scene description information;

[0033] The matters covered by the selected historical scene description information are determined as associated matters of the set of matters to be processed.

[0034] In the present application, the target intelligent feature map includes an item representation vector of each item to be processed and a relationship representation vector of the relationship requirement, and the target item that meets the relationship requirement with the items to be processed in the set of items to be processed is determined by using the target intelligent feature map, and the 3D scene is optimized by the target item, and further includes:

[0035] When the number of the historical scene description information is less than or equal to the target number, obtaining the first item representation vector of the to-be-processed item, the second item representation vectors of the remaining to-be-processed items, and the relationship representation vector of the relationship requirement, wherein the remaining to-be-processed items are the to-be-processed items in the to-be-processed item set other than the to-be-processed item;

[0036] Performing function processing on the first item representation vector and the relationship representation vector of the relationship requirement to obtain a function vector;

[0037] determining a vector difference between the function vector and each second item representation vector;

[0038] The items corresponding to the vector differences that are smaller than the difference target value are determined as target items that meet the relationship requirements with the items to be processed, and the 3D scene is optimized through the target items.

[0039] In the present application, the method further comprises:

[0040] Obtaining an original item representation vector of each reference item and an original relationship representation vector of each reference historical scene description information;

[0041] The original event representation vector and the original relationship representation vector are trained and optimized to obtain an optimized event representation vector for each reference event and an optimized relationship representation vector for each reference historical scene description information.

[0042] In the present application, the original event representation vector and the original relationship representation vector are trained and optimized to obtain an optimized event representation vector for each reference event and an optimized relationship representation vector for each reference historical scene description information, including:

[0043] Obtain a sample set of existential connections;

[0044] determining a non-existence of a connection example set based on the existence of the connection example set;

[0045] Building a quantitative evaluation strategy based on a first difference of each existing connection example in the existing connection example set and a second difference of each non-existing connection example in the non-existing connection example set;

[0046] Based on the quantitative evaluation strategy, each original event representation vector and each original relationship representation vector are trained and optimized to obtain an optimized event representation vector for each reference event and an optimized relationship representation vector for each reference historical scene description information.

[0047] In the present application, each of the existence connection examples includes a starting matter, a relationship, and a final matter, and determining the non-existence connection example set based on the existence connection example set includes:

[0048] Obtaining an a-th existence-connection example, where a=1, 2, ..., X, and X is the total number of existence-connection examples;

[0049] Replace the ath starting item in the ath existing connection example with the first arbitrary item to obtain the ath non-existing connection example, wherein the first arbitrary item is a different item from the ath starting item; or replace the ath final item in the ath existing connection example with the second arbitrary item to obtain the ath non-existing connection example, wherein the second arbitrary item is a different item from the ath final item.

[0050] In a second aspect, an optimization system for automatically generating an immersive 3D scene based on AIGC is provided, comprising a processor and a memory that communicate with each other, wherein the processor is used to read a computer program from the memory and execute it to implement the above method.

[0051] The technical solution provided by the embodiments of the present disclosure may include the following beneficial effects.

[0052] For non-individual pending items in the set of pending items, the first historical scene description information of the non-individual pending items is determined based on the reference historical scene description information of the non-individual pending items, wherein the non-individual pending items are included in several reference items; for individual pending items in the set of pending items, the second historical scene description information of the individual pending items is analyzed based on several reference historical scene description information, wherein the individual pending items are not included in several reference items; then, a target intelligent feature map is constructed based on the first historical scene description information and the second historical scene description information, and the target intelligent feature map is the intelligent feature map corresponding to the pending items, and the intelligent feature map includes individual nodes with connection relationships; finally, the target intelligent feature map is used to determine the target items that meet the relationship requirements with the pending items in the set of pending items, and the 3D scene is optimized through the target items, wherein the target items belong to the set of pending items or several reference items. In this way, the implementation scene can be completely restored, making the user experience more real and reliable.

[0053] It is to be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings herein are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the description, are used to explain the principles of the present application.

[0055] Figure 1 A flowchart of an optimization method for automatically generating an immersive 3D scene based on AIGC provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0057] Based on the above, please refer to Figure 1 , which is a flow chart of the optimization method for automatically generating an immersive 3D scene based on AIGC provided in an embodiment of the present application. Furthermore, the optimization method for automatically generating an immersive 3D scene based on AIGC may specifically include the contents described in the following steps S101 to S105.

[0058] In step S101, a set of pending items and reference historical scene description information of several reference items are obtained.

[0059] Among them, matters can be understood as three-dimensional scene objects.

[0060] In this step, the reference historical scene description information is used to indicate the relationship between at least two reference items among the plurality of reference items. For example, the reference historical scene description information may indicate the relationship between the reference items and the reference items. The reference items may be physical items or virtual items.

[0061] Through step S101, a set of items to be processed and reference historical scene description information of several reference items can be obtained to prepare for subsequent image data processing.

[0062] In step S102, for a non-individual item to be processed, first historical scene description information of the non-individual item to be processed is determined according to the reference historical scene description information of the non-individual item to be processed.

[0063] Among them, single pending items refer to images with little information, which are relatively simple and easy to process, such as images with only people. Non-single pending items refer to images with a lot of information, which may interfere with each other, such as images with people, trees, animals, etc.

[0064] In this step, since the non-individual pending items are pending items covered by several reference items, the reference historical scene description information of the non-individual pending items can be obtained from the reference historical scene description information.

[0065] In one possible implementation, each reference historical scene description information can be traversed to determine whether the reference historical scene description information includes non-separate pending matters. If it includes non-separate pending matters, the reference information is determined as the reference historical scene description information of non-separate pending matters.

[0066] In one possible implementation, each reference historical scene description information corresponds to a debugging coefficient. To improve accuracy, when determining the first historical scene description information, it is further determined whether the debugging coefficient meets the requirements. Based on this, the "determining the first historical scene description information of non-separate pending matters based on the reference historical scene description information of non-separate pending matters" in step S102 can be implemented in the following way.

[0067] Obtain the debugging coefficient corresponding to the reference historical scene description information of each non-individual pending matter, and then determine the reference historical scene description information of the non-individual pending matter whose debugging coefficient is higher than the second debugging target value as the first historical scene description information of the non-individual pending matter.

[0068] In the embodiment of the present application, the debugging coefficient is used to reflect the difference description of the actual relationship between matters in the historical scene description information, wherein the higher the debugging coefficient, the greater the difference description; conversely, the lower the debugging coefficient, the smaller the difference description.

[0069] In a possible implementation, a debugging coefficient of reference historical scene description information of each non-individual to-be-processed item may be obtained.

[0070] In an embodiment of the present application, the second debugging target value may be a value set in advance. For example, the second debugging target value may be 10, 15, 20, etc.

[0071] In a possible implementation embodiment, historical scene description information having a debugging coefficient higher than a second debugging target value in reference historical scene description information of non-individual items to be processed is determined as first historical scene description information.

[0072] In a possible implementation, the reference historical scene description information of non-individual pending items may be sorted in descending order of the debugging coefficient to obtain sorted reference historical scene description information; and then the first preset number of historical scene description information in the sorted reference historical scene description information is determined as the first historical scene description information. The first preset number is less than the number of the second historical scene description information, and illustratively, the first preset number may be 50, 60, 100, etc.

[0073] Through step S102, for non-single pending items, based on the reference historical scene description information of the non-single pending items, the historical scene description information covering any non-single pending item and with a debugging coefficient greater than the second debugging target value is determined as the first historical scene description information of the non-single pending item, thereby determining historical scene description information with high accuracy for the non-single pending items in the pending items set.

[0074] The description will continue with the above step S102.

[0075] In step S103, for a single item to be processed, second historical scene description information of the single item to be processed is analyzed based on a plurality of reference historical scene description information.

[0076] In this step, the second historical scene description information is used to represent the relationship between the separate pending item and one or more reference items among the multiple reference items. Since the separate pending item is not included in the multiple reference items, that is, it is impossible to determine the historical scene description information covering the separate pending item from the reference historical scene description information, the second historical scene description information of the separate pending item is determined based on the reference historical scene description information to establish the relationship between the separate pending item and the remaining pending items, so that in the subsequent training and optimization of the intelligent feature map, the separate node corresponding to the separate pending item can also be optimized, so that the separate node can be applied to the prediction and analysis of the relationship, thereby improving the accuracy of the prediction and analysis.

[0077] In a possible implementation example, "analyzing second historical scene description information of a separate item to be processed based on a plurality of reference historical scene description information" in step S103 can be implemented by following steps S1031 to S1033, which are described in detail below.

[0078] In step S1031, common items of the individual items to be processed are determined from a number of reference items.

[0079] In one possible implementation, the descriptive factor data of the individual item to be processed and the descriptive factor data of each reference item can be first obtained, wherein the descriptive factor data includes at least one of brightness information, scene type information and color information, and then common items that meet the commonality coefficient requirements with the descriptive factor data of the individual item to be processed are determined from each reference item.

[0080] In a possible implementation, reference items that meet common requirements with the brightness information, scene type information, and color information of the individual items to be processed can be determined as common items. In implementation, first determine a first reference item with the same brightness information as the individual items to be processed from each reference item; then determine a second reference item with the same scene type information as the individual items to be processed from each first reference item; and finally determine a reference item with the same color information as the individual items to be processed in each second reference item as a common item.

[0081] In one possible implementation, any one of brightness information, scene type information or color information may be used as a commonality coefficient requirement. For example, a reference item determined in each reference item to be the same as the brightness information (scene type information or color information) of a separate item to be processed may be determined as a commonality item, that is, a commonality item may be determined as long as the brightness information is the same. In other embodiments, any two of brightness information, scene type information or color information may be used as a commonality coefficient requirement. For example, a reference item determined in each reference item to be the same as the brightness information and scene type information of a separate item to be processed may be determined as a commonality item.

[0082] In step S1032, third historical scene description information covering common matters is obtained from a plurality of reference historical scene description information.

[0083] In this step, each reference historical scene description information can be traversed, and it can be determined whether each reference historical scene description information covers common matters. If covered, the reference historical scene description information is determined as the third historical scene description information; if not covered, the reference historical scene description information is deleted until the traversal and judgment of all reference historical scene description information are completed.

[0084] In step S1033, second historical scene description information of the separate item to be processed is determined based on the third historical scene description information.

[0085] In this step, the debugging coefficient corresponding to the third historical scene description information is first obtained, and then the third historical scene description information with a debugging coefficient higher than the first debugging target value is determined as the second historical scene description information of a separate item to be processed.

[0086] In the embodiment of the present application, the first debugging target value may be a value set in advance, and the first debugging target value may be the same as the second debugging target value or different from the second debugging target value. For example, the second debugging target value may be 10, 16, 25, etc.

[0087] In a possible implementation example, historical scene description information having a debugging coefficient higher than a first debugging target value in the third historical scene description information is determined as the second historical scene description information.

[0088] In a possible implementation, the third historical scene description information may be sorted in descending order of the debugging coefficient to obtain the sorted third historical scene description information; and then the first second preset number of historical scene description information in the sorted third historical scene description information is determined as the second historical scene description information. The second preset number is less than the number of the first historical scene description information, and the second preset number may be the same as the first preset number or different from the first preset number. For example, the second preset number may be 50, 65, 100, etc.

[0089] Through step S103, for separate pending matters, the common matters of the separate pending matters can be first determined from the reference matters, and then the third historical scene description information covering the common matters can be obtained from several reference historical scene description information, and the debugging coefficient of the third historical scene description information can be obtained at the same time; finally, the third historical scene description information with a debugging coefficient higher than the first debugging target value is determined as the second historical scene description information of the separate pending matters. In order to build the relationship between the separate pending matters and the remaining pending matters, a complete intelligent feature map can be built based on this. In addition, when the intelligent feature map is subsequently trained and optimized, the individual nodes corresponding to the separate pending matters can also be optimized, so that the individual nodes can be applied to the prediction and analysis of the relationship, thereby improving the accuracy of the prediction and analysis.

[0090] In step S104, a target intelligent feature map is constructed based on the first historical scene description information and the second historical scene description information.

[0091] The target intelligent feature map can be understood as the database in AIGC.

[0092] In step S105, the target intelligent feature map is used to determine the target items that meet the relationship requirements with the items to be processed in the set of items to be processed, and the 3D scene is optimized through the target items.

[0093] In this step, a pending item can be determined from the pending item set according to actual needs. The pending item can be any item in the item set.

[0094] In an embodiment of the present application, the target item belongs to a set of items to be processed or several reference items.

[0095] In one possible implementation embodiment, "using the target intelligent feature map to determine the target items that meet the relationship requirements with the items to be processed in the set of items to be processed, and optimizing the 3D scene through the target items" in step S105 can be achieved through the following steps S1051 to S1058, which are described in detail below.

[0096] In step S1051, the number of historical scene description information covered by the target intelligent feature map is obtained.

[0097] In this step, the total number of all historical scene description information covered in the target intelligent feature map is obtained, which can be recorded as the number of historical scene description information.

[0098] In step S1052, it is determined whether the number of historical scene description information is greater than the target value.

[0099] In this step, the target value of the number may be a value set in advance.

[0100] In the embodiment of the present application, the relationship between the number of historical scene description information and the number target value can be determined by comparing the size. When the number of historical scene description information is less than or equal to the number target value, the target intelligent feature map is not very large, and the analysis time required for relationship analysis using the target intelligent feature map is acceptable. The computing resources of the computer device can also support the analysis based on the target intelligent feature map, and then enter step S1053. If the number of historical scene description information is greater than the number target value, the target intelligent feature map is very large, and the analysis time required for relationship analysis using the target intelligent feature map may be several days or even weeks. The computing resources of the computer device are difficult to support the analysis based on the target intelligent feature map, and then enter step S1057.

[0101] In step S1053, the first item representation vector of the items to be processed, the second item representation vectors of the remaining items to be processed, and the relationship representation vectors of the relationship requirements are obtained.

[0102] In step S1054, function processing is performed on the first item representation vector and the relationship representation vector to obtain a function vector.

[0103] In this step, the function vector can be determined by a vector summation method.

[0104] In step S1055, the vector difference between the function vector and each second event representation vector is determined.

[0105] In this step, the vector difference between the function vector and each second item representation vector may be determined by using methods such as Euclidean difference, Harmanton difference, and Chebyshev difference.

[0106] In step S1056, the items corresponding to the vector differences that are smaller than the difference target value are determined as target items that meet the relationship requirements with the items to be processed, and the 3D scene is optimized through the target items. The process ends.

[0107] In this step, the difference target value is a difference set in advance. For example, the difference target value may be 0.1, 0.2, etc.

[0108] In a possible implementation, the size relationship between each vector difference and the difference target value can be determined by a size comparison method. If the vector difference is less than the difference target value, the item corresponding to the vector difference is determined as a target item that meets the relationship requirements with the item to be processed, and the 3D scene is optimized by the target item. The number of the target items can be several.

[0109] In step S1057, related items of the set of items to be processed are determined from the target intelligent feature map.

[0110] At this time, when the number of historical scene description information is greater than the target value, the target intelligent feature map is very large, and the analysis time required for relationship analysis using the target intelligent feature map may be several days or even weeks. The computing resources of computer equipment are difficult to support analysis based on the target intelligent feature map. Instead of directly using the target intelligent feature map for analysis, the associated items are determined, where the associated items are items that have a matching relationship with the set of items to be optimized.

[0111] In one possible implementation, the target intelligent feature map includes a debugging coefficient of historical scene description information, and step S1057 "determining the associated items of the set of pending items from the target intelligent feature map" can be implemented through the following steps S571 to S575, which are described in detail below.

[0112] In step S571, pending items that meet the association requirements with the set of pending items are determined from the target intelligent feature map.

[0113] In one possible implementation, matters belonging to the same region as the set of pending matters can be determined from matters in the target intelligent feature map, and the matters belonging to the same region can be determined as the first pending matters. Matters belonging to the same industry as the set of pending matters can also be determined from matters in the target intelligent feature map, and the matters belonging to the same industry can be determined as the second pending matters. Matters belonging to the same community as the set of pending matters can also be determined from matters in the target intelligent feature map, and the matters belonging to the same community can be determined as the third pending matters. Matters having a controlling interest in the set of pending matters can also be determined from matters in the target intelligent feature map, and the matters having a controlling interest can be determined as the fourth pending matters. Finally, the first pending matter, the second pending matter, the third pending matter, and the fourth pending matter are determined as pending matters.

[0114] In step S572, pending historical scene description information covering pending matters is determined.

[0115] In this step, each historical scene description information in the target intelligent feature map is obtained one by one, and it is determined whether each historical scene description information covers pending matters. If it covers pending matters, the historical scene description information covering the pending matters is determined as the pending historical scene description information; if it does not cover pending matters, the historical scene description information that does not cover pending matters is deleted.

[0116] In step S573, the debugging coefficient of each pending historical scene description information is obtained from the target intelligent feature map.

[0117] In this step, since the target intelligent feature map includes the debugging coefficients of each piece of historical scene description information, the debugging coefficients of each piece of pending historical scene description information are obtained from the target intelligent feature map.

[0118] In step S574, the pending historical scene description information having a debugging coefficient higher than the third debugging target value is determined as the selected historical scene description information.

[0119] In this step, the third debugging target value may be a value set in advance. The third debugging target value may be the same as the second debugging target value or may be different from the second debugging target value. Exemplarily, the third debugging target value may be 15, 20, 30, etc.

[0120] In an embodiment of the present application, historical scene description information whose debugging coefficient is higher than a third debugging target value in the pending historical scene description information is determined as the selected historical scene description information.

[0121] In step S575, the items covered by the selected historical scene description information are determined as associated items of the set of items to be processed.

[0122] In this step, the items covered in the selected historical scene description information are first obtained, and then the items covered in the selected historical scene description information are determined as related items, wherein the number of the related items may be several.

[0123] In step S1058, a target item that meets the relationship requirements with the items to be processed in the set of items to be processed is determined from the associated items, and the 3D scene is optimized through the target item.

[0124] In this step, the related items are used as the analysis scope to determine the target items that meet the relationship requirements with the items to be processed, and the 3D scene is optimized through the target items.

[0125] In a possible implementation embodiment, the method for determining the target item can refer to the above steps S1053 to S10546, but it is necessary to replace the "second item representation vector of each item in the target intelligent feature map" in step S1053 with the "third item representation vector of the associated item".

[0126] Through step S105, the pending items are first determined from the pending items set; then, if the number of historical scene description information covered by the target intelligent feature map does not exceed the target value, the target items that meet the relationship requirements with the pending items are determined based on all the items in the target intelligent feature map, and the 3D scene is optimized through the target items; and if the number of historical scene description information covered by the target intelligent feature map exceeds the target value, the related items are determined, and the target items that meet the relationship requirements with the pending items are determined from the related items, and the 3D scene is optimized through the target items, wherein the related items are very likely to be the target items. In this way, there is no need for full analysis, which can reduce the complexity of the analysis and improve the efficiency of relationship analysis.

[0127] In a possible implementation embodiment, before executing step S101, the original representation vector of each reference item and the original relationship representation vector of each reference historical scene description information are obtained, and then the original representation vector and the original relationship representation vector are trained and optimized to obtain an optimized item representation vector and an optimized relationship representation vector. Based on this, when determining the first historical scene description information and the second historical scene description information based on the reference historical scene description information, the optimized relationship representation vector corresponding to the first historical scene description information is determined as the relationship representation of the first historical scene description information, and the optimized relationship representation vector corresponding to the second historical scene description information is determined as the relationship representation of the second historical scene description information. Based on this, before executing step S101, the original item representation vector of each reference item and the original relationship representation vector of each reference historical scene description information are obtained; and the original item representation vector and the original relationship representation vector are trained and optimized to obtain an optimized item representation vector for each reference item and an optimized relationship representation vector for each reference historical scene description information.

[0128] In a possible implementation embodiment, the above-mentioned "training and optimizing the original event representation vector and the original relationship representation vector to obtain an optimized event representation vector for each reference event and an optimized relationship representation vector for each reference historical scene description information" can be achieved through the following steps S001 to S004, which are described in detail below.

[0129] In step S001, a set of existence connection examples is obtained.

[0130] In this step, an existence connection example set may be obtained from a public example library, where the existence connection example set is real examples.

[0131] In step S002, it is determined whether a connection example set does not exist based on the existence of the connection example set.

[0132] In a possible implementation example, each existing connection example includes a starting matter, a relationship and an ending matter, and step S002 "determining a non-existent connection example set based on an existing connection example set" can be implemented by the following method 1 and / or method 2, which are described in detail below.

[0133] Method 1: obtaining the ath existence connection example; replacing the ath starting item in the ath existence connection example with the first arbitrary item to obtain the ath non-existence connection example.

[0134] In the first method, a=1, 2, ..., X, where X is the total number of existence connection examples. That is, each existence connection example in the existence connection example set is obtained one by one.

[0135] In the embodiment of the present application, the first arbitrary matter and the ath starting matter are different matters.

[0136] Method 2: obtaining the ath existence connection example; replacing the ath final item in the ath existence connection example with the second arbitrary item to obtain the ath non-existence connection example.

[0137] In the second method, the method of obtaining the ath positive example is the same as the above method 1. Then, the ath final item in the ath existing relationship example can be replaced by the second arbitrary item in the reference item that is different from the ath final item, so as to obtain the ath non-existent relationship example, that is, the ath non-existent relationship example is composed of the ath starting item, the ath relationship and the second arbitrary item.

[0138] In the embodiment of the present application, the second arbitrary matter and the ath final matter are different matters.

[0139] In a possible implementation, each non-existent connection example can be obtained by the above-mentioned method 1 to obtain a non-existent connection example set; each non-existent connection example can also be obtained by the above-mentioned method 2 to obtain a non-existent connection example set; and a part of the non-existent connection examples can be obtained by the above-mentioned method 1, and another part of the non-existent connection examples can be obtained by the above-mentioned method 2 to obtain a non-existent connection example set, that is, the non-existent connection example set can be obtained by both methods.

[0140] In step S003, a quantitative evaluation strategy is established based on the first difference of each existing connection example in the existing connection example set and the second difference of each non-existing connection example in the non-existing connection example set.

[0141] In this step, it is necessary to first determine the first difference of each existing connection example in the existing connection example set, and also to determine the second difference of each non-existing connection example in the non-existing connection example set, wherein the determination method of the first difference is the same as the second difference.

[0142] Here, the first difference of the a-th existence connection example is taken as an example. First, the representation vector of the a-th starting item, the representation vector of the a-th relationship, and the representation vector of the a-th final item are obtained, wherein the a-th relationship is the relationship between the a-th starting item and the a-th final item; then, the representation vector of the a-th starting item and the representation vector of the a-th relationship are processed by function to obtain the a-th function vector; finally, the a-th vector difference between the a-th function vector and the representation vector of the a-th final item is determined, and the a-th vector difference is determined as the first difference of the a-th existence connection example.

[0143] In a possible implementation example, the above-mentioned example method can be used to determine the first difference of each example with a connection and the second difference of each example without a connection.

[0144] In step S004, each original event representation vector and each original relationship representation vector are trained and optimized based on a quantitative evaluation strategy to obtain a target intelligent feature map.

[0145] In one possible implementation, a preset number of training times and training optimizations may be performed to obtain a target intelligent feature map.

[0146] Through the above steps S001 to S004, the original event representation vector and the original relationship representation vector are obtained; then, a quantitative evaluation strategy is established based on the example set of existing connections and the example set of non-existing connections, and the original event representation vector and the original relationship representation vector are trained and optimized based on the quantitative evaluation strategy, so as to obtain the optimized event representation vector for each reference event and the optimized relationship representation vector for each reference historical scene description information.

[0147] In summary, when applying the above scheme, the optimization method and system for automatically generating immersive 3D scenes based on AIGC, for non-individual pending items in the pending items set, is to determine the first historical scene description information of the non-individual pending items based on the reference historical scene description information of the non-individual pending items, wherein the non-individual pending items are covered in several reference items; for individual pending items in the pending items set, is to analyze the second historical scene description information of the individual pending items based on several reference historical scene description information, wherein the individual pending items are not covered in several reference items; then, based on the first historical scene description information and the second historical scene description information, a target intelligent feature map is constructed, and the target intelligent feature map is the intelligent feature map corresponding to the pending items, and the intelligent feature map includes individual nodes with connection relationships; finally, the target intelligent feature map is used to determine the target items that meet the relationship requirements with the pending items in the pending items set, and the 3D scene is optimized through the target items, wherein the target items belong to the pending items set or several reference items. In this way, the implementation scene can be completely restored, and the user experience is more real and reliable.

[0148] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be performed without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. An optimization method for automatically generating an immersive 3D scene based on AIGC, characterized in that: The method comprises: Obtaining reference historical scenario description information of a set of pending items and a plurality of reference items, wherein the pending items in the set of pending items include independent pending items and non-independent pending items, the independent pending items are not included in the plurality of reference items, and the non-independent pending items are included in the plurality of reference items, and the reference historical scenario description information is used to indicate a relationship between at least two reference items in the plurality of reference items; For a non-individual pending item, first historical scene description information of the non-individual pending item is determined according to the reference historical scene description information of the non-individual pending item, wherein the first historical scene description information is used to represent the relationship between the non-individual pending item and one or more reference items among the plurality of reference items except the non-individual pending item; For a separate pending item, analyzing second historical scene description information of the separate pending item according to a plurality of reference historical scene description information, wherein the second historical scene description information is used to indicate a relationship between the separate pending item and one or more of the plurality of reference items; Building a target intelligent feature map based on the first historical scene description information and the second historical scene description information; Determine a target item that meets the relationship requirements with the items to be processed in the set of items to be processed by using the target intelligent feature map, optimize the 3D scene by using the target item, and the target item belongs to the set of items to be processed or the plurality of reference items; The step of using the target intelligent feature map to determine a target item that meets the relationship requirements with the items to be processed in the set of items to be processed, and optimizing the 3D scene through the target item includes: Obtaining the number of historical scene description information covered by the target intelligent feature map; When the number of the historical scene description information exceeds the target number value, determining the associated items of the set of items to be processed from the target intelligent feature map, wherein the associated items are items that have a matching relationship with the set of items to be optimized; Determine, from the associated items, a target item that meets the relationship requirements with the items to be processed in the set of items to be processed, The target intelligent feature map includes an item representation vector of each item to be processed and a relationship representation vector of a relationship requirement, and the target item that meets the relationship requirement with the items to be processed in the set of items to be processed is determined by using the target intelligent feature map, and the 3D scene is optimized by using the target item, and further includes: When the number of the historical scene description information is less than or equal to the target number, obtaining the first item representation vector of the to-be-processed item, the second item representation vectors of the remaining to-be-processed items, and the relationship representation vector of the relationship requirement, wherein the remaining to-be-processed items are the to-be-processed items in the to-be-processed item set other than the to-be-processed item; Performing function processing on the first item representation vector and the relationship representation vector of the relationship requirement to obtain a function vector; determining a vector difference between the function vector and each second item representation vector; The items corresponding to the vector differences that are smaller than the difference target value are determined as target items that meet the relationship requirements with the items to be processed, and the 3D scene is optimized through the target items.

2. The method according to claim 1, characterized in that The step of analyzing the second historical scene description information of the separate to-be-processed item according to the plurality of reference historical scene description information includes: Determining common items of the individual items to be processed from the several reference items; Obtaining third historical scene description information covering the common matters from the plurality of reference historical scene description information; The second historical scene description information of the separate item to be processed is determined based on the third historical scene description information.

3. The method according to claim 2, characterized in that The determining the second historical scene description information of the separate to-be-processed item based on the third historical scene description information includes: Obtaining a debugging coefficient corresponding to the third historical scene description information; The third historical scene description information whose debugging coefficient is higher than the first debugging target value is determined as the second historical scene description information of the separate matter to be processed.

4. The method according to claim 2, characterized in that Determining the common items of the individual items to be processed from the several reference items includes: Obtaining descriptive factor data of the individual item to be processed and descriptive factor data of each reference item, wherein the descriptive factor data includes at least one of brightness information, scene type information, and color information; From each of the reference items, common items whose descriptive factor data meet the commonality coefficient requirements with the individual items to be processed are determined.

5. The method according to claim 1, characterized in that Each reference historical scene description information corresponds to a debugging coefficient, and determining the first historical scene description information of the non-individual to-be-processed item according to the reference historical scene description information of the non-individual to-be-processed item includes: Obtaining the debugging coefficient corresponding to the reference historical scenario description information of each non-individual pending item; The reference historical scene description information of the non-individual to-be-processed item whose debugging coefficient is higher than the second debugging target value is determined as the first historical scene description information of the non-individual to-be-processed item.

6. The method according to claim 1, characterized in that The target intelligent feature map includes a debugging coefficient of historical scene description information, and determining related items of the set of items to be processed from the target intelligent feature map includes: Determine, from the target intelligent feature map, pending items that meet the association requirements with the set of pending items; Determining pending historical scenario description information covering the pending matters; Obtaining a debugging coefficient for each pending historical scene description information from the target intelligent feature map; Determine the pending historical scene description information whose debugging coefficient is higher than the third debugging target value as the selected historical scene description information; The matters covered by the selected historical scene description information are determined as associated matters of the set of matters to be processed.

7. The method according to claim 1, characterized in that The method further comprises: Obtaining an original item representation vector of each reference item and an original relationship representation vector of each reference historical scene description information; Performing training optimization on the original event representation vector and the original relationship representation vector to obtain an optimized event representation vector for each reference event and an optimized relationship representation vector for each reference historical scene description information; The original event representation vector and the original relationship representation vector are trained and optimized to obtain an optimized event representation vector for each reference event and an optimized relationship representation vector for each reference historical scene description information, including: Obtain a sample set of existential connections; determining a non-existence of a connection example set based on the existence of the connection example set; Building a quantitative evaluation strategy based on a first difference of each existing connection example in the existing connection example set and a second difference of each non-existing connection example in the non-existing connection example set; Based on the quantitative evaluation strategy, each original event representation vector and each original relationship representation vector are trained and optimized to obtain an optimized event representation vector for each reference event and an optimized relationship representation vector for each reference historical scene description information; Each of the existence connection examples includes a starting matter, a relationship, and a final matter, and determining the non-existence connection example set based on the existence connection example set includes: Obtaining an a-th existence-connection example, where a=1, 2, ..., X, and X is the total number of existence-connection examples; Replace the ath starting item in the ath existing connection example with the first arbitrary item to obtain the ath non-existing connection example, wherein the first arbitrary item is a different item from the ath starting item; or replace the ath final item in the ath existing connection example with the second arbitrary item to obtain the ath non-existing connection example, wherein the second arbitrary item is a different item from the ath final item.

8. An optimization system for automatically generating immersive 3D scenes based on AIGC, characterized in that: The invention comprises a processor and a memory communicating with each other, wherein the processor is used to read a computer program from the memory and execute the computer program to implement the method according to any one of claims 1 to 7.

Images