Virtual scene generation methods, devices and electronic equipment

Abstract

This disclosure provides a method, apparatus, and electronic device for generating virtual scenes, relating to the field of computer technology, particularly to application areas such as machine learning, large language models, metaverse, digital world, game development, smart cities, outdoor scene design, traffic design and planning, traffic management and simulation, and architectural design. The specific implementation scheme is as follows: Obtaining a scene generation request; responding to the scene generation request, partitioning the original virtual scene to obtain a basic virtual scene; wherein the basic virtual scene includes multiple functional areas; using each of the multiple functional areas as a target area, using a target large model, filling the target area with multiple target instance objects to obtain a scene area corresponding to the target area; and generating the target virtual scene based on the scene area.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to application areas such as machine learning, large language models, metaverse, digital world, game development, smart cities, outdoor scene design, traffic design and planning, traffic management and simulation, and architectural design. Specifically, it relates to a virtual scene generation method, device and electronic equipment. Background Technology

[0002] Virtual scenes are simulation environments built using computer technology and digital simulation technology, allowing users to immerse themselves in the environment and interact with it. Summary of the Invention

[0003] This disclosure provides a method, apparatus, and electronic device for generating virtual scenes.

[0004] According to a first aspect of this disclosure, a method for generating a virtual scene is provided, comprising:

[0005] Obtain the scene generation request;

[0006] In response to a scene generation request, the original virtual scene is partitioned to obtain a basic virtual scene; the basic virtual scene includes multiple functional areas.

[0007] Each of the multiple functional areas is taken as the target area. Using the target large model, multiple target instance objects are filled in the target area to obtain the scene area corresponding to the target area.

[0008] Generate the target virtual scene based on the scene area.

[0009] According to a second aspect of this disclosure, a virtual scene generation apparatus is provided, comprising:

[0010] The request retrieval unit is used to retrieve the scene generation request;

[0011] The scene partitioning unit is used to partition the original virtual scene in response to the scene generation request to obtain the basic virtual scene; the basic virtual scene includes multiple functional areas.

[0012] The object filling unit is used to take each of the multiple functional areas as the target area, and use the target large model to fill multiple target instance objects in the target area to obtain the scene area corresponding to the target area.

[0013] The scene generation unit is used to generate a target virtual scene based on a scene region.

[0014] According to a third aspect of this disclosure, an electronic device is provided, comprising:

[0015] At least one processor;

[0016] The memory that is communicatively connected to the at least one processor;

[0017] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method provided in the first aspect of this disclosure.

[0018] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform the method provided according to a first aspect of this disclosure.

[0019] According to a fifth aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method provided according to a first aspect of this disclosure.

[0020] Using this disclosure can reduce the waste of human resources and costs, while enhancing the reusability and scalability of the target virtual scene, which is conducive to the effective utilization of scene resources.

[0021] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0022] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0023] Figure 1 A flowchart illustrating a virtual scene generation method provided in this embodiment of the disclosure;

[0024] Figure 2 A schematic diagram illustrating a basic virtual scene acquisition process provided in this embodiment of the disclosure;

[0025] Figure 3 This is a schematic diagram of a preliminary instance object selection process provided in an embodiment of the present disclosure;

[0026] Figure 4 A schematic diagram illustrating the representational meaning of a parameter expression provided in an embodiment of this disclosure;

[0027] Figure 5 A schematic diagram of a target virtual scene provided in an embodiment of this disclosure;

[0028] Figure 6 A diagram illustrating the complete flow of a virtual scene generation method provided in this disclosure.

[0029] Figure 7 This is a schematic diagram illustrating an application scenario of a virtual scene generation method provided in this embodiment of the disclosure;

[0030] Figure 8 This is a schematic structural block diagram of a virtual scene generation device provided in an embodiment of the present disclosure;

[0031] Figure 9 This is a schematic structural block diagram of an electronic device provided in an embodiment of the present disclosure. Detailed Implementation

[0032] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0033] As described in the background section, a virtual scene is a simulation environment constructed using computer technology and digital simulation technology, allowing users to immerse themselves in the experience and interact with it. Currently, the construction of virtual scenes mainly relies on meticulous manual modeling. This process not only consumes a large amount of human resources and costs, but also results in virtual scenes that are poor in terms of reusability and scalability, which is not conducive to the effective utilization of scene resources.

[0034] To address the above problems, this disclosure provides a virtual scene generation method that can be applied to electronic devices. The electronic device can be a server or a terminal device. Here, the terminal device can be a workbench, mainframe computer, conventional computer (e.g., desktop computer, laptop computer, in-vehicle computer, etc.), personal digital processing unit, or other similar computing device. The following will be combined with... Figure 1 The flowchart shown illustrates a virtual scene generation method provided in this disclosure. It should be noted that although a logical order is shown in the flowchart, in some cases, the steps shown or described in the flowchart may be executed in a different order.

[0035] Step S101: Obtain the scene generation request.

[0036] The scene generation request can include a target scene style, which can be used to further determine the scene type and scene characteristics. For example, a scene generation request could be "Generate a modern city scene". This request includes a target scene style of "modern city", and based on this target scene style, the scene type can be further determined to be "city" and the scene characteristic to be "modern".

[0037] Furthermore, in this embodiment of the disclosure, the scene generation request can be issued by the developer through a terminal device and received by an electronic device. In one example, the developer can input the scene building intent on the terminal device through text input or voice input, and the terminal device can generate a scene generation request based on the scene building intent, and then send the scene generation request to the electronic device.

[0038] Step S102: In response to the scene generation request, the original virtual scene is partitioned to obtain the basic virtual scene.

[0039] The original virtual scene can be a blank scene substrate, such as a two-dimensional (2D) blank scene substrate; the basic virtual scene can include multiple functional areas, such as residential areas, commercial areas, green areas, open areas, industrial areas, etc.

[0040] Step S103: Take each of the multiple functional areas as the target area, and use the target large model to fill the target area with multiple target instance objects to obtain the scene area corresponding to the target area.

[0041] The target large model can be a Large Language Model (LLM) or a trained LLM. Here, the LLM can be a pre-trained neural network model that possesses general language knowledge, world knowledge, and domain-specific expertise (e.g., expertise in the metaverse, architectural design, etc.), stored internally as parameters. In one example, the LLM can be an autoregressive generative model based on the Transformer architecture.

[0042] After designating each of the multiple functional areas as the target area, multiple target instance objects related to the target area can be populated within the target area based on the target large model to obtain the scene area corresponding to the target area. For example, if the target area is a residential area, the multiple target instance objects related to it can include residential buildings, trees, benches, convenience stores, medical stations, etc.; if the target area is a commercial area, the multiple target instance objects related to it can include shopping malls, parking lots, restaurants, banks, cinemas, etc.; and if the target area is a green area, the multiple target instance objects related to it can include trees, grass, viewing platforms, pavilions, etc.

[0043] Step S104: Generate the target virtual scene based on the scene area.

[0044] After obtaining multiple scene regions that correspond one-to-one with multiple functional areas, a target virtual scene can be generated based on these scene regions. For example, a target virtual scene that includes multiple scene regions can be generated. The target virtual scene can be a three-dimensional (3D) virtual scene.

[0045] The virtual scene generation method provided in this disclosure can, upon receiving a scene generation request, partition the original virtual scene to obtain a basic virtual scene including multiple functional areas. Each functional area is then used as a target area. Using a target large model, multiple target instance objects are filled into the target area to obtain a scene area corresponding to the target area. Finally, the target virtual scene is generated based on the scene area. Compared to existing virtual scene generation methods that rely on meticulous manual modeling, this disclosure, on the one hand, reduces waste of human resources and costs by replacing manual operation with an automated process based on a target large model; on the other hand, through a regionalized design concept, scene areas can be reused and expanded as templates, thereby enhancing the reusability and scalability of the target virtual scene and facilitating the effective utilization of scene resources.

[0046] In some optional implementations, step S102, namely, "in response to the scene generation request, partitioning the original virtual scene to obtain a basic virtual scene," may include:

[0047] Based on the scene generation request, determine the target scene style;

[0048] Obtain scene partitioning instructions;

[0049] Using a partitioning model, the original virtual scene is partitioned based on the target scene style and scene partitioning instructions to obtain the basic virtual scene.

[0050] The scene zoning instructions can be obtained from the area design guidance document. This document can be a guideline developed and published by design engineers specializing in fields such as metaverse and architectural design, explaining how to rationally plan, layout, and optimize various functional areas within the virtual scene. For example, the first scene zoning instruction can be obtained from the area design guidance document: the straight-line distance between residential areas and industrial areas is greater than or equal to 20 kilometers (km); the second scene zoning instruction can be obtained from the same document: the straight-line distance between residential areas and commercial areas is less than or equal to 10 km; and the third scene zoning instruction can be obtained from the same document: the straight-line distance between residential areas and green areas is less than 5 km.

[0051] After generating a scene request, determining the target scene style, and obtaining scene partitioning instructions, a partitioning model can be used to partition the original virtual scene based on the target scene style and scene partitioning instructions to obtain a basic virtual scene. The partitioning model can be an LLM or a trained LLM. Furthermore, as mentioned above, in this embodiment, the basic virtual scene can include multiple functional areas. Specifically, when partitioning the original virtual scene using the partitioning model based on the target scene style and scene partitioning instructions to obtain the basic virtual scene, object selection requirements for each functional area can also be obtained. These object selection requirements can include the categories of initial selected instance objects to be filled, the number of initial selected instance objects in each category, and the specific filling position of each initial selected instance object.

[0052] Please combine Figure 2 For example, after determining the target scene style as "modern city" based on the scene generation request "Generate a modern city scene", scene partitioning instructions can be obtained. Using a partitioning model, based on the target scene style and scene partitioning instructions, the original virtual scene can be partitioned to obtain a basic virtual scene comprising 12 functional areas, and object selection requirements for each of these 12 functional areas. These 12 functional areas include 4 residential areas, 2 commercial areas, 3 green areas, 2 open areas, and 1 industrial area. The object selection requirements include the categories of initial selected instance objects to be filled, the number of initial selected instance objects in each category, and the specific filling location for each initial selected instance object.

[0053] Suppose that for the first residential area out of four residential areas, the selection criteria for the corresponding target population include:

[0054] The initial selection of instance object categories to be filled are: residential buildings, trees, and benches.

[0055] The number of initial instance objects for the residential building is 1; the number of initial instance objects for the tree is 2 (including the first and second trees); the number of instance objects for the bench is 1.

[0056] The specific filling positions of the initial instance object of the residential building are (X1, Y1); the specific filling positions of the initial instance object of the first tree are (X2, Y2); the specific filling positions of the initial instance object of the second tree are (X3, Y3); and the specific filling positions of the initial instance object of the lounge chair are (X4, Y4).

[0057] Through the above methods, in this embodiment of the disclosure, the target scene style can be determined based on the scene generation request, and scene partitioning instructions can be obtained. Then, using a partitioning model, the original virtual scene can be partitioned based on the target scene style and the scene partitioning instructions to obtain a basic virtual scene. In other words, in this embodiment of the disclosure, on the one hand, by using a partitioning model to directly partition the original virtual scene to obtain a basic virtual scene, the waste of human resources and costs can be further reduced; on the other hand, when the partitioning model partitions the original virtual scene to obtain a basic virtual scene, it needs to use both the target scene style and the scene partitioning instructions as partitioning references, rather than relying solely on the target scene style. Therefore, it can ensure that multiple functional areas in the basic virtual scene have reasonable planning and layout.

[0058] Furthermore, as mentioned above, in this embodiment of the disclosure, the partitioning model can be a trained LLM. The process of training a first LLM to obtain the partitioning model will be briefly described below:

[0059] Generate request samples based on the scenario and determine the target scenario style samples;

[0060] Obtain scene partitioning instructions;

[0061] Using the first LLM, the original virtual scene sample is partitioned based on the target scene style sample and scene partitioning indicator to obtain the basic virtual scene sample;

[0062] The first LLM is trained using basic virtual scene samples.

[0063] The scene generation request sample has the same data structure and representational meaning as the scene generation request. For example, the scene generation request sample may include the target scene style sample, and the scene type sample and scene feature sample can be further determined based on the target scene style sample.

[0064] After generating request samples based on the scene, determining the target scene style sample, and obtaining the scene partitioning indication, the first LLM can be used to partition the original virtual scene sample based on the target scene style sample and the scene partitioning indication to obtain the basic virtual scene sample. The basic virtual scene sample can include multiple functional area samples. Specifically, when using the first LLM to partition the original virtual scene sample based on the target scene style sample and the scene partitioning indication to obtain the basic virtual scene sample, object selection requirement samples for each functional area sample can also be obtained. These object selection requirement samples can include the categories of initial selected instance object samples to be filled, the number of initial selected instance object samples in each category, and the specific filling position of each initial selected instance object sample.

[0065] After obtaining the basic virtual scene samples and object selection requirement samples, the scene loss between the basic virtual scene samples and the reference basic virtual scene, and the requirement loss between the object selection requirement samples and the reference object selection requirements, can be calculated. Based on the scene loss and requirement loss, the first LLM is trained. The reference basic virtual scene and the basic virtual scene samples have the same data structure and representational meaning; the reference object selection requirements and the object selection requirement samples have the same data structure and representational meaning. Moreover, the reference basic virtual scene and the reference object selection requirements can be obtained through manual annotation, which will not be elaborated here.

[0066] Furthermore, it should be noted that in this embodiment, the process of training the first LLM described above can be executed cyclically until the trained first LLM satisfies the first convergence condition, and the trained first LLM is used as the partitioning model. The first convergence condition can be that the scene loss meets a preset scene loss requirement and the required loss meets a preset requirement loss requirement. Here, the preset scene loss requirement and the preset requirement loss requirement can be determined based on the actual application scenario, and this embodiment will not elaborate on them.

[0067] In some optional implementations, the target large model includes a first target large model and a second target large model. Based on this, step S103, which includes "using the target large model to fill the target area with multiple target instance objects to obtain a scene area corresponding to the target area," may include:

[0068] Using the first target model, select multiple initial instance objects related to the target region from the instance object library;

[0069] Using the second target model, multiple initial instance objects are adjusted to obtain multiple target instance objects that correspond one-to-one with the multiple initial instance objects;

[0070] Multiple target instance objects are filled into the target area to obtain the scene area corresponding to the target area.

[0071] The first objective large model can be an LLM or a trained LLM; the second objective large model can also be an LLM or a trained LLM.

[0072] In one example, "using the first target large model, selecting multiple initial instance objects related to the target region from the instance object library" can include:

[0073] Obtain the object selection requirements related to the target area;

[0074] Using the first target model, a request is generated based on the scene to generate an overall object style related to the target area;

[0075] Using the first target model, based on the object selection requirements and the overall object style, select multiple initial instance objects related to the target area from the instance object library.

[0076] The overall object style can include the initial positioning attributes of each initially selected instance object to be filled in the target area. For example, for the initially selected instance object of a residential building, its initial positioning attributes can include building shape, floor height, facade color, facade texture, door and window design style, etc.; for the initially selected instance object of a tree, its initial positioning attributes can include tree species, tree height, tree tilt, foliage density, leaf color, etc.; and for the initially selected instance object of a lounge chair, its initial positioning attributes can include chair material, chair color, chair size, etc.

[0077] In addition, as mentioned above, in this embodiment of the disclosure, the object selection requirements may include the initial selection instance object category to be filled, the number of initial selection instance objects in each category, and the specific filling position of each initial selection instance object, etc.

[0078] After obtaining the overall object style and object selection requirements related to the target area, the first target large model can be used to select multiple preliminary instance objects related to the target area from the instance object library based on the overall object style and object selection requirements. The instance object library can include a massive number of candidate instance objects, each with attribute tags. For example, for the candidate instance object A1 (residential building), its attribute tags could include building shape tags, floor height tags, facade color tags, facade texture tags, and door and window design style tags; for the candidate instance object A2 (tree), its attribute tags could include tree species tags, tree height tags, tree tilt tags, foliage density tags, and leaf color tags; and for the candidate instance object A3 (leisure seating), its attribute tags could include seating material tags, seating color tags, and seating size tags.

[0079] Based on this, in a specific example, after obtaining the overall object style and object selection requirements related to the target area, the first target large model can be used to compare the overall object style and object selection requirements with the attribute labels of each candidate instance object in the instance object library, and select multiple preliminary instance objects related to the target area from the instance object library.

[0080] Please combine Figure 3 For example, the object selection requirements related to the target area include:

[0081] The initial selection of instance object categories to be filled are: residential buildings, trees, and benches.

[0082] The number of initial instance objects for the residential building is 1; the number of initial instance objects for the tree is 2 (including the first and second trees); the number of instance objects for the bench is 1.

[0083] The specific filling positions of the initial instance object of the residential building are (X1, Y1); the specific filling positions of the initial instance object of the first tree are (X2, Y2); the specific filling positions of the initial instance object of the second tree are (X3, Y3); and the specific filling positions of the initial instance object of the lounge chair are (X4, Y4).

[0084] Assuming that, using the first target large model, a request is generated based on the scene, and the overall object style generated is related to the target area, it includes:

[0085] The initial positioning attributes for the residential building as the initial selected instance are as follows: the building shape is a regular square (including square, rectangle, etc.), the floor height is low (e.g., the floor height is 1 to 7 floors), the exterior color is gray (including pure dark gray, pure light gray, patterned gray, etc.), the exterior texture is raised pattern (including raised decorative pattern, raised mural pattern, etc.), and the door and window design style is retro (e.g., Chinese retro style, American retro style, etc.).

[0086] Initial positioning attributes for the first tree as the initial selected instance object: tree species is pine (including black pine, Scots pine, oil pine, etc.), tree height is extra tall (e.g., 20 meters (M) to 30M), tree tilt is no tilt, foliage density is dense, and foliage color is green (including dark green, dark green, light green, etc.).

[0087] The initial positioning attributes for the second tree as a preliminary instance are: tree species is ginkgo (including yellow-leaf ginkgo, pyramidal ginkgo, split ginkgo, weeping ginkgo, variegated ginkgo, etc.), tree height is low (e.g., 5M~10M), tree tilt is not tilted, leaf density is dense, and leaf color is yellow (including dark yellow, light yellow, etc.).

[0088] ...

[0089] Therefore, using the first target model, based on the object selection requirements and the overall object style, the multiple preliminary instance objects related to the target area selected from the instance object library can include residential building B1, tree B2, and tree B3.

[0090] In the example above, object selection requirements related to the target region can be obtained. Using the first target large model, a scene generation request is generated to produce an overall object style related to the target region. Then, using the first target large model, multiple preliminary instance objects related to the target region are selected from the instance object library based on the object selection requirements and the overall object style. In this process, the first target large model is responsible for generating the overall object style related to the target region based on the scene generation request, and selecting multiple preliminary instance objects from the instance object library based on the object selection requirements and the overall object style. Since the first target large model can be an LLM or a trained LLM, the matching degree between the multiple preliminary instance objects and the scene generation request can be ensured.

[0091] In one example, "using the second target large model, adjusting multiple initial selected instance objects to obtain multiple target instance objects that correspond one-to-one with the multiple initial selected instance objects" can include:

[0092] Each of the multiple initial instance objects is taken as the object to be adjusted. The overall object style is refined using the second target large model to obtain the specific object style related to the object to be adjusted.

[0093] Get object adjustment instructions;

[0094] Using the second target model, based on the specific object style and object adjustment instructions, the object to be adjusted is adjusted to obtain the target instance object corresponding to the object to be adjusted.

[0095] The second target model can be an LLM or a trained LLM; the specific object style can include the target localization attributes of the object to be adjusted (here, the target localization attributes are more specific and clear than the initial localization attributes). For example, if the object to be adjusted is a residential building, its target localization attributes can include specific building shape, specific floor height, specific facade color, specific facade texture, specific door and window design style, etc.; as another example, if the object to be adjusted is a tree, its target localization attributes can include specific tree species, specific tree height, specific tree tilt, specific foliage density, specific leaf color, etc.; as yet another example, if the object to be adjusted is a bench, its target localization attributes can include specific bench material, specific bench color, specific bench size, etc.

[0096] Furthermore, in this embodiment of the disclosure, the object adjustment instructions can be obtained from a guiding object adjustment document. This guiding object adjustment document can be a document developed and published by design engineers in fields such as metaverse and architectural design, to explain how to reasonably adjust the object. For example, a first object adjustment instruction can be obtained from the guiding object adjustment document: the exterior color of a residential building cannot be pure dark gray; a second object adjustment instruction can be obtained from the guiding object adjustment document: the height of trees cannot exceed 25 meters; and a third object adjustment instruction can be obtained from the guiding object adjustment document: oleanders cannot be planted within residential areas.

[0097] After refining the overall object style using the second target model to obtain specific object styles related to the object to be adjusted, and acquiring object adjustment instructions, the second target model can then be used to adjust the object to be adjusted based on the specific object styles and object adjustment instructions to obtain the target instance object corresponding to the object to be adjusted. In a specific example, "using the second target model, based on the specific object styles and object adjustment instructions, to adjust the object to be adjusted to obtain the target instance object corresponding to the object to be adjusted" can include:

[0098] Obtain the parameter expression of the object to be adjusted; wherein, the parameter expression includes multiple initial parameters, and each of the multiple initial parameters corresponds to an object property of the object to be adjusted;

[0099] Using the second target model, based on specific object styles and object adjustment instructions, multiple initial parameters are adjusted to obtain multiple target parameters that correspond one-to-one with the multiple initial parameters;

[0100] Based on multiple target parameters, the object to be adjusted is adjusted to obtain the target instance object corresponding to the object to be adjusted.

[0101] For example, if the object to be adjusted is a residential building, its initial parameters may include initial parameters corresponding to the specific building shape, the specific floor height, the specific facade color, the specific facade texture, and the specific door and window design style, etc.; as another example, if the object to be adjusted is a tree, its initial parameters may include initial parameters corresponding to the specific tree species, the specific tree height, the tree tilt, the density of leaves, and the color of leaves, etc.; as yet another example, if the object to be adjusted is a bench, its initial parameters may include initial parameters corresponding to the specific bench material, the specific bench color, and the specific bench size, etc.

[0102] After obtaining the specific object style and object adjustment instructions, as well as the parameter expression of the object to be adjusted, the second target model can be used to adjust multiple initial parameters based on the specific object style and object adjustment instructions to obtain multiple target parameters that correspond one-to-one with the multiple initial parameters. Based on the multiple target parameters, the object to be adjusted is then adjusted to obtain the target instance object corresponding to the object to be adjusted.

[0103] Please combine Figure 4 Continuing with the previous example, the initial selected instance objects include residential building B1 and tree B2.

[0104] When taking residential building B1 as the object to be adjusted, the overall object style can be refined to obtain specific object styles related to the object to be adjusted, and object adjustment instructions can be obtained. Then, the parameter expressions of the object to be adjusted can be obtained (the parameter expressions include initial parameters corresponding to the specific building shape object attribute, initial parameters corresponding to the specific floor height object attribute, initial parameters corresponding to the specific facade color object attribute, initial parameters corresponding to the specific facade texture object attribute, initial parameters corresponding to the specific door and window design style object attribute, etc.). Using the second target large model, based on the specific object style and object adjustment instructions, multiple initial parameters are adjusted to obtain multiple target parameters that correspond one-to-one with the multiple initial parameters. Based on the multiple target parameters, the object to be adjusted is adjusted to obtain the target instance object corresponding to the object to be adjusted.

[0105] When tree B2 is taken as the object to be adjusted, the overall object style can be refined to obtain the specific object style related to the object to be adjusted, and the object adjustment instruction can be obtained. Then, the parameter expression of the object to be adjusted can be obtained (the parameter expression includes the initial parameter corresponding to the specific tree species object attribute, the initial parameter corresponding to the specific tree height object attribute, the initial parameter corresponding to the tree tilt object attribute, the initial parameter corresponding to the foliage density object attribute, the initial parameter corresponding to the leaf color object attribute, etc.). Using the second target large model, based on the specific object style and the object adjustment instruction, multiple initial parameters are adjusted to obtain multiple target parameters that correspond one-to-one with the multiple initial parameters. Based on the multiple target parameters, the object to be adjusted is adjusted to obtain the target instance object corresponding to the object to be adjusted.

[0106] In the example above, each of the multiple initial instance objects can be considered as the object to be adjusted. Using the second target model, the overall object style is refined to obtain a specific object style related to the object to be adjusted, and object adjustment instructions are acquired. Then, using the second target model, based on the specific object style and the object adjustment instructions, the object to be adjusted is modified to obtain the target instance object corresponding to the object to be adjusted. This improves the diversity and rationality of the target instance objects, thereby enhancing the visual effect of the target virtual scene.

[0107] In summary, through the above methods, in this embodiment of the disclosure, a first target large model can be used to select multiple preliminary instance objects related to the target region from the instance object library, and a second target large model can be used to adjust these preliminary instance objects to obtain multiple target instance objects that correspond one-to-one with the preliminary instance objects. These target instance objects are then filled into the target region to obtain the scene region corresponding to the target region. In other words, in this embodiment of the disclosure, on the one hand, the first target large model can be used to select multiple preliminary instance objects related to the target region from the instance object library to improve the generation efficiency of the target virtual scene; on the other hand, after obtaining multiple initial instance objects, the second target large model can be used to adjust these preliminary instance objects to obtain multiple target instance objects that correspond one-to-one with the preliminary instance objects, thereby improving the diversity and rationality of the target instance objects and thus enhancing the visual effect of the target virtual scene.

[0108] Furthermore, as mentioned above, in this embodiment of the disclosure, the first target model can be a trained LLM. The process of training a second LLM to obtain the first target model will be briefly described below:

[0109] Obtain a sample of objects related to the target region;

[0110] Using the second LLM, request samples are generated based on the scene, and overall object style samples related to the target area samples are generated.

[0111] Using the second LLM, multiple initial instance object samples are selected from the instance object library based on the object selection requirement sample and the overall object style sample;

[0112] The second LLM is trained using multiple initial sample instances.

[0113] Among them, the target region sample has the same data structure and representational meaning as the target region; the object selection requirement sample has the same data structure and representational meaning as the object selection requirement; the scene generation request sample has the same data structure and representational meaning as the scene generation request; the overall object style sample has the same data structure and representational meaning as the overall object style; and the initial instance object sample has the same data structure and representational meaning as the initial instance object. These will not be elaborated further here.

[0114] After obtaining multiple initial sample instances, a first object loss can be calculated between these initial sample instances and multiple reference initial sample instances. Based on this first object loss, a second LLM is trained. The reference initial sample instances have the same data structure and representational meaning as the initial sample instances. Furthermore, the reference initial sample instances can be obtained through manual annotation, which will not be elaborated upon here.

[0115] It should be noted that, in this embodiment of the disclosure, the process of training the second LLM described above can be executed cyclically until the trained second LLM satisfies the second convergence condition, and the trained second LLM is used as the first target large model. The second convergence condition can be that the first object loss meets the first object loss requirement. Here, the first object loss requirement can be determined based on the actual application scenario, and this embodiment of the disclosure will not elaborate on it.

[0116] Similarly, as mentioned above, in this embodiment of the disclosure, the second target model can be a trained LLM. The process of training a third LLM to obtain the second target model will be briefly described below:

[0117] Using the third LLM, the overall object style sample is refined to obtain specific object style samples related to the object sample to be adjusted;

[0118] Get object adjustment instructions;

[0119] Using the third LLM, based on specific object style samples and object adjustment instructions, the object sample to be adjusted is adjusted to obtain the target instance object sample corresponding to the object sample to be adjusted;

[0120] The third LLM is trained using target instance object samples.

[0121] Among them, the overall object style sample has the same data structure and representational meaning as the overall object style; the object sample to be adjusted has the same data structure and representational meaning as the object to be adjusted; the specific object style sample has the same data structure and representational meaning as the specific object style; and the target instance object sample has the same data structure and representational meaning as the target instance object. These will not be elaborated here.

[0122] After selecting the target instance object samples, a second object loss can be calculated between the target instance object samples and the reference target instance objects. Based on the second object loss, the third LLM is trained. The reference target instance objects have the same data structure and representational meaning as the target instance object samples. Moreover, the reference target instance objects can be obtained through manual annotation, which will not be elaborated upon here.

[0123] It should be noted that, in this embodiment of the disclosure, the process of training the third LLM described above can be executed cyclically until the trained third LLM satisfies the third convergence condition, and the trained third LLM is used as the second target large model. The third convergence condition can be that the second object loss meets the second object loss requirement. Here, the second object loss requirement can be determined based on the actual application scenario, and this embodiment of the disclosure will not elaborate on it.

[0124] In some alternative implementations, step S104, namely, "generating the target virtual scene based on the scene region," can be:

[0125] Based on the scene area, construct the initial virtual scene;

[0126] If the initial virtual scene meets the preset scene requirements, the target virtual scene is generated based on the initial virtual scene.

[0127] The preset scenario requirements can be set according to actual application needs, and this disclosure does not impose any restrictions on them.

[0128] In one example, "determining that the initial virtual scene meets the preset scene requirements" may include:

[0129] From each of the multiple evaluation dimensions, a single evaluation result for the initial virtual scene is obtained, so as to obtain multiple single evaluation results that correspond one-to-one with the multiple evaluation dimensions;

[0130] Based on the results of multiple individual evaluations, the overall evaluation result of the initial virtual scene is obtained;

[0131] If the overall evaluation results meet the preset evaluation requirements, the initial virtual scene is determined to meet the preset scene requirements.

[0132] The evaluation dimensions can include at least one of the following: functional requirements evaluation dimension, spatial relationship evaluation dimension, and architectural style evaluation dimension. Correspondingly, the multiple individual evaluation results can include a first individual evaluation result corresponding to the functional requirements evaluation dimension, a second individual evaluation result corresponding to the spatial relationship evaluation dimension, and a third individual evaluation result corresponding to the architectural style evaluation dimension. Here, the first individual evaluation result is used to characterize whether the initial virtual scene meets the scene generation request; the second individual evaluation result is used to characterize whether the spatial relationship of the initial virtual scene is reasonable; and the third individual evaluation result is used to characterize whether the architectural style of the initial virtual scene is harmonious.

[0133] After obtaining multiple individual evaluation results, an overall evaluation result for the initial virtual scene can be derived based on these results. If the overall evaluation result meets preset evaluation requirements, the initial virtual scene is deemed to meet those requirements. These preset evaluation requirements may include: a first individual evaluation result indicating that the initial virtual scene meets the scene generation request; a second individual evaluation result indicating that the spatial relationships within the initial virtual scene are reasonable; and a third individual evaluation result indicating that the architectural style of the initial virtual scene is harmonious.

[0134] Furthermore, it should be noted that in this embodiment of the disclosure, the LLM can also be used to determine whether the initial virtual scene meets the preset scene requirements.

[0135] In another example, "generating the target virtual scene based on the initial virtual scene" could include:

[0136] Obtain target atmosphere information;

[0137] Based on the target atmosphere information, the initial virtual scene is optimized to generate the target virtual scene.

[0138] The target atmosphere information includes at least one of 3D terrain information and weather information.

[0139] After obtaining the target atmosphere information, the parameter modification function corresponding to the target atmosphere information can be called to optimize the initial virtual scene, generate the target virtual scene, and send the target virtual scene to the terminal device for display. For example, the target virtual scene can be as follows: Figure 5 As shown.

[0140] Through the above methods, in this embodiment of the disclosure, an initial virtual scene can be constructed based on a scene area. If the initial virtual scene meets preset scene requirements, a target virtual scene can be generated based on the initial virtual scene. Furthermore, when the initial virtual scene meets the preset scene requirements, it is evaluated from multiple evaluation dimensions to ensure the reliability of the evaluation results. In addition, when generating the target virtual scene based on the initial virtual scene, target atmosphere information can be obtained, and the initial virtual scene can be optimized based on the target atmosphere information to generate the target virtual scene. In this way, not only can the rationality of the target virtual scene be ensured, but also its realism, thereby further improving the visual effect of the target virtual scene.

[0141] The following, combined with Figure 6 The complete process of the virtual scene generation method provided in the embodiments of this disclosure will be described.

[0142] (1) Obtain the scene generation request

[0143] The scene generation request can include a target scene style, which can be used to further determine the scene type and scene characteristics. For example, a scene generation request could be "Generate a modern city scene". This request includes a target scene style of "modern city", and based on this target scene style, the scene type can be further determined to be "city" and the scene characteristic to be "modern".

[0144] Furthermore, in this embodiment of the disclosure, the scene generation request can be issued by the developer through a terminal device and received by an electronic device. In one example, the developer can input the scene building intent on the terminal device through text input or voice input, and the terminal device can generate a scene generation request based on the scene building intent, and then send the scene generation request to the electronic device.

[0145] (2) Obtaining the basic virtual scene

[0146] In one example, the original virtual scene can be partitioned in response to a scene generation request to obtain a base virtual scene.

[0147] In a specific example, "in response to a scene generation request, partitioning the original virtual scene to obtain a basic virtual scene" may include:

[0148] Based on the scene generation request, determine the target scene style;

[0149] Obtain scene partitioning instructions;

[0150] Using a partitioning model, the original virtual scene is partitioned based on the target scene style and scene partitioning instructions to obtain the basic virtual scene.

[0151] The scene zoning instructions can be obtained from the area design guidance document. This document can be a guideline developed and published by design engineers specializing in fields such as metaverse and architectural design, explaining how to rationally plan, layout, and optimize various functional areas within the virtual scene. For example, the first scene zoning instruction from the guidance document might be: the straight-line distance between residential and industrial areas is greater than or equal to 20km; the second scene zoning instruction might be: the straight-line distance between residential and commercial areas is less than or equal to 10km; and the third scene zoning instruction might be: the straight-line distance between residential and green areas is less than 5km.

[0152] After generating a scene request, determining the target scene style, and obtaining scene partitioning instructions, a partitioning model can be used to partition the original virtual scene based on the target scene style and scene partitioning instructions to obtain a basic virtual scene. The partitioning model can be an LLM or a trained LLM. Furthermore, as mentioned above, in this embodiment, the basic virtual scene can include multiple functional areas. Specifically, when partitioning the original virtual scene using the partitioning model based on the target scene style and scene partitioning instructions to obtain the basic virtual scene, object selection requirements for each functional area can also be obtained. These object selection requirements can include the categories of initial selected instance objects to be filled, the number of initial selected instance objects in each category, and the specific filling position of each initial selected instance object.

[0153] (3) Select multiple initial instance objects

[0154] Specifically, each of the multiple functional areas can be taken as the target area, and multiple preliminary instance objects related to the target area can be selected from the instance object library using the first target large model.

[0155] In one example, "using the first target large model, selecting multiple initial instance objects related to the target region from the instance object library" can include:

[0156] Obtain the object selection requirements related to the target area;

[0157] Using the first target model, a request is generated based on the scene to generate an overall object style related to the target area;

[0158] Using the first target model, based on the object selection requirements and the overall object style, select multiple initial instance objects related to the target area from the instance object library.

[0159] The overall object style can include the initial positioning attributes of each initially selected instance object to be filled in the target area. For example, for the initially selected instance object of a residential building, its initial positioning attributes can include building shape, floor height, facade color, facade texture, door and window design style, etc.; for the initially selected instance object of a tree, its initial positioning attributes can include tree species, tree height, tree tilt, foliage density, leaf color, etc.; and for the initially selected instance object of a lounge chair, its initial positioning attributes can include chair material, chair color, chair size, etc.

[0160] In addition, as mentioned above, in this embodiment of the disclosure, the object selection requirements may include the initial selection instance object category to be filled, the number of initial selection instance objects in each category, and the specific filling position of each initial selection instance object, etc.

[0161] After obtaining the overall object style and object selection requirements related to the target area, the first target large model can be used to select multiple preliminary instance objects related to the target area from the instance object library based on the overall object style and object selection requirements. The instance object library can include a massive number of candidate instance objects, each with attribute tags. For example, for the candidate instance object A1 (residential building), its attribute tags could include building shape tags, floor height tags, facade color tags, facade texture tags, and door and window design style tags; for the candidate instance object A2 (tree), its attribute tags could include tree species tags, tree height tags, tree tilt tags, foliage density tags, and leaf color tags; and for the candidate instance object A3 (leisure seating), its attribute tags could include seating material tags, seating color tags, and seating size tags.

[0162] Based on this, in a specific example, after obtaining the overall object style and object selection requirements related to the target area, the first target large model can be used to compare the similarity between the overall object style and object selection requirements and the attribute labels of each candidate instance object in the instance object library, so as to select multiple preliminary instance objects related to the target area from the instance object library.

[0163] (4) Adjusting multiple initial instance objects

[0164] Specifically, the second target model can be used to adjust multiple initial instance objects to obtain multiple target instance objects that correspond one-to-one with the multiple initial instance objects.

[0165] In one example, "using the second target large model, adjusting multiple initial selected instance objects to obtain multiple target instance objects that correspond one-to-one with the multiple initial selected instance objects" can include:

[0166] Each of the multiple initial instance objects is taken as the object to be adjusted. The overall object style is refined using the second target large model to obtain the specific object style related to the object to be adjusted.

[0167] Get object adjustment instructions;

[0168] Using the second target model, based on the specific object style and object adjustment instructions, the object to be adjusted is adjusted to obtain the target instance object corresponding to the object to be adjusted.

[0169] The second target model can be an LLM or a trained LLM; the specific object style can include the target localization attributes of the object to be adjusted (here, the target localization attributes are more specific and clear than the initial localization attributes). For example, if the object to be adjusted is a residential building, its target localization attributes can include specific building shape, specific floor height, specific facade color, specific facade texture, specific door and window design style, etc.; as another example, if the object to be adjusted is a tree, its target localization attributes can include specific tree species, specific tree height, specific tree tilt, specific foliage density, specific leaf color, etc.; as yet another example, if the object to be adjusted is a bench, its target localization attributes can include specific bench material, specific bench color, specific bench size, etc.

[0170] Furthermore, in this embodiment of the disclosure, the object adjustment instructions can be obtained from a guiding object adjustment document. This guiding object adjustment document can be a document developed and published by design engineers in fields such as metaverse and architectural design, to explain how to reasonably adjust the object. For example, a first object adjustment instruction can be obtained from the guiding object adjustment document: the exterior color of a residential building cannot be pure dark gray; a second object adjustment instruction can be obtained from the guiding object adjustment document: the height of trees cannot exceed 25 meters; and a third object adjustment instruction can be obtained from the guiding object adjustment document: oleanders cannot be planted within residential areas.

[0171] After refining the overall object style using the second target model to obtain specific object styles related to the object to be adjusted, and acquiring object adjustment instructions, the second target model can then be used to adjust the object to be adjusted based on the specific object styles and object adjustment instructions to obtain the target instance object corresponding to the object to be adjusted. In a specific example, "using the second target model, based on the specific object styles and object adjustment instructions, to adjust the object to be adjusted to obtain the target instance object corresponding to the object to be adjusted" can include:

[0172] Obtain the parameter expression of the object to be adjusted; wherein, the parameter expression includes multiple initial parameters, and each of the multiple initial parameters corresponds to an object property of the object to be adjusted;

[0173] Using the second target model, based on specific object styles and object adjustment instructions, multiple initial parameters are adjusted to obtain multiple target parameters that correspond one-to-one with the multiple initial parameters;

[0174] Based on multiple target parameters, the object to be adjusted is adjusted to obtain the target instance object corresponding to the object to be adjusted.

[0175] For example, if the object to be adjusted is a residential building, its initial parameters may include initial parameters corresponding to the specific building shape, the specific floor height, the specific facade color, the specific facade texture, and the specific door and window design style, etc.; as another example, if the object to be adjusted is a tree, its initial parameters may include initial parameters corresponding to the specific tree species, the specific tree height, the tree tilt, the density of leaves, and the color of leaves, etc.; as yet another example, if the object to be adjusted is a bench, its initial parameters may include initial parameters corresponding to the specific bench material, the specific bench color, and the specific bench size, etc.

[0176] After obtaining the specific object style and object adjustment instructions, as well as the parameter expression of the object to be adjusted, the second target model can be used to adjust multiple initial parameters based on the specific object style and object adjustment instructions to obtain multiple target parameters that correspond one-to-one with the multiple initial parameters. Based on the multiple target parameters, the object to be adjusted is then adjusted to obtain the target instance object corresponding to the object to be adjusted.

[0177] Furthermore, it should be noted that in this embodiment of the disclosure, after obtaining multiple target instance objects that correspond one-to-one with multiple initially selected instance objects, the multiple target instance objects can be filled into the target area to obtain the scene area corresponding to the target area, and an initial virtual scene is constructed based on the scene area.

[0178] (5) Evaluate the initial virtual scene

[0179] The purpose of evaluating the initial virtual scene is to determine whether the initial virtual scene meets the preset scene requirements.

[0180] In one example, the LLM can determine whether the initial virtual scene meets the preset scene requirements, where "determining that the initial virtual scene meets the preset scene requirements" can include:

[0181] From each of the multiple evaluation dimensions, a single evaluation result for the initial virtual scene is obtained, so as to obtain multiple single evaluation results that correspond one-to-one with the multiple evaluation dimensions;

[0182] Based on the results of multiple individual evaluations, the overall evaluation result of the initial virtual scene is obtained;

[0183] If the overall evaluation results meet the preset evaluation requirements, the initial virtual scene is determined to meet the preset scene requirements.

[0184] The evaluation dimensions can include at least one of the following: functional requirements evaluation dimension, spatial relationship evaluation dimension, and architectural style evaluation dimension. Correspondingly, the multiple individual evaluation results can include a first individual evaluation result corresponding to the functional requirements evaluation dimension, a second individual evaluation result corresponding to the spatial relationship evaluation dimension, and a third individual evaluation result corresponding to the architectural style evaluation dimension. Here, the first individual evaluation result is used to characterize whether the initial virtual scene meets the scene generation request; the second individual evaluation result is used to characterize whether the spatial relationship of the initial virtual scene is reasonable; and the third individual evaluation result is used to characterize whether the architectural style of the initial virtual scene is harmonious.

[0185] After obtaining multiple individual evaluation results, an overall evaluation result for the initial virtual scene can be derived based on these results. If the overall evaluation result meets preset evaluation requirements, the initial virtual scene is deemed to meet those requirements. These preset evaluation requirements may include: a first individual evaluation result indicating that the initial virtual scene meets the scene generation request; a second individual evaluation result indicating that the spatial relationships within the initial virtual scene are reasonable; and a third individual evaluation result indicating that the architectural style of the initial virtual scene is harmonious.

[0186] If the initial virtual scene does not meet the preset scene requirements, LLM can be used to provide adjustment suggestions for the initial virtual scene and make adjustments accordingly.

[0187] (6) Generate the target virtual scene

[0188] If the initial virtual scene meets the preset scene requirements, the target atmosphere information can be obtained, and the initial virtual scene can be optimized based on the target atmosphere information to generate the target virtual scene.

[0189] The target atmosphere information includes at least one of 3D terrain information and weather information.

[0190] After obtaining the target atmosphere information, the parameter modification function corresponding to the target atmosphere information can be called to optimize the initial virtual scene and generate the target virtual scene.

[0191] (7) Send the target virtual scene to the terminal device

[0192] The terminal device is used to display the target virtual scene.

[0193] Furthermore, it should be noted that in this embodiment of the disclosure, if the developer is not satisfied with the target virtual scene, they can propose a scene modification requirement as a new scene generation requirement to modify the target virtual scene.

[0194] Please see Figure 7 This is a schematic diagram illustrating an application scenario of a virtual scene generation method provided in this embodiment of the disclosure.

[0195] The virtual scene generation method provided in this disclosure is applied to an electronic device. The electronic device can be a server or a terminal device. Here, the terminal device can be a workbench, a mainframe computer, a conventional computer (e.g., a desktop computer, a laptop computer, an in-vehicle computer, etc.), a personal digital processing unit, or other similar computing devices.

[0196] Here, electronic devices are used for:

[0197] Obtain the scene generation request;

[0198] In response to a scene generation request, the original virtual scene is partitioned to obtain a basic virtual scene; the basic virtual scene includes multiple functional areas.

[0199] Each of the multiple functional areas is taken as the target area. Using the target large model, multiple target instance objects are filled in the target area to obtain the scene area corresponding to the target area.

[0200] Generate the target virtual scene based on the scene area.

[0201] It should be noted that, in the embodiments disclosed herein, Figure 7 The application scenario diagrams shown are for illustrative purposes only and are not restrictive. Those skilled in the art can use them as a basis for their own interpretation. Figure 7 The examples may be modified in various obvious ways and / or substitutions, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of this disclosure.

[0202] To better implement the aforementioned virtual scene generation method, this disclosure also provides a virtual scene generation apparatus, which can be integrated into an electronic device. The electronic device can be a server or a terminal device. Here, the terminal device can be a workbench, a mainframe computer, a conventional computer (e.g., a desktop computer, a laptop computer, an in-vehicle computer, etc.), a personal digital processing unit, or other similar computing devices. The following will be combined with… Figure 8 The schematic structural block diagram shown illustrates a virtual scene generation device 800 provided in the disclosed embodiment.

[0203] Virtual scene generation device 800, including:

[0204] Request retrieval unit 801 is used to retrieve scene generation requests;

[0205] Scene partitioning unit 802 is used to partition the original virtual scene in response to a scene generation request to obtain a basic virtual scene; wherein, the basic virtual scene includes multiple functional areas;

[0206] The object filling unit 803 is used to take each of the multiple functional areas as the target area, and use the target large model to fill multiple target instance objects in the target area to obtain the scene area corresponding to the target area.

[0207] Scene generation unit 804 is used to generate target virtual scenes based on scene regions.

[0208] In some alternative implementations, the object filling unit 803 is used for:

[0209] Using the first target model, select multiple initial instance objects related to the target region from the instance object library;

[0210] Using the second target model, multiple initial instance objects are adjusted to obtain multiple target instance objects that correspond one-to-one with the multiple initial instance objects;

[0211] Multiple target instance objects are filled into the target area to obtain the scene area corresponding to the target area.

[0212] In some alternative implementations, the object filling unit 803 is used for:

[0213] Obtain the object selection requirements related to the target area;

[0214] Using the first target model, a request is generated based on the scene to generate an overall object style related to the target area;

[0215] Using the first target model, based on the object selection requirements and the overall object style, select multiple initial instance objects related to the target area from the instance object library.

[0216] In some alternative implementations, the object filling unit 803 is used for:

[0217] Each of the multiple initial instance objects is taken as the object to be adjusted. The overall object style is refined using the second target large model to obtain the specific object style related to the object to be adjusted.

[0218] Get object adjustment instructions;

[0219] Using the second target model, based on the specific object style and object adjustment instructions, the object to be adjusted is adjusted to obtain the target instance object corresponding to the object to be adjusted.

[0220] In some alternative implementations, the object filling unit 803 is used for:

[0221] Obtain the parameter expression of the object to be adjusted; wherein, the parameter expression includes multiple initial parameters, and each of the multiple initial parameters corresponds to an object property of the object to be adjusted;

[0222] Using the second target model, based on specific object styles and object adjustment instructions, multiple initial parameters are adjusted to obtain multiple target parameters that correspond one-to-one with the multiple initial parameters;

[0223] Based on multiple target parameters, the object to be adjusted is adjusted to obtain the target instance object corresponding to the object to be adjusted.

[0224] In some alternative implementations, the scene generation unit 804 is used for:

[0225] Based on the scene area, construct the initial virtual scene;

[0226] If the initial virtual scene meets the preset scene requirements, the target virtual scene is generated based on the initial virtual scene.

[0227] In some alternative implementations, the scene generation unit 804 is used for:

[0228] From each of the multiple evaluation dimensions, a single evaluation result for the initial virtual scene is obtained, so as to obtain multiple single evaluation results that correspond one-to-one with the multiple evaluation dimensions;

[0229] Based on the results of multiple individual evaluations, the overall evaluation result of the initial virtual scene is obtained;

[0230] If the overall evaluation results meet the preset evaluation requirements, the initial virtual scene is determined to meet the preset scene requirements.

[0231] In some alternative implementations, the scene generation unit 804 is used for:

[0232] Obtain target atmosphere information;

[0233] Based on the target atmosphere information, the initial virtual scene is optimized to generate the target virtual scene.

[0234] In some alternative implementations, the scene partitioning unit 802 is used for:

[0235] Based on the scene generation request, determine the target scene style;

[0236] Obtain scene partitioning instructions;

[0237] Using a partitioning model, the original virtual scene is partitioned based on the target scene style and scene partitioning instructions to obtain the basic virtual scene.

[0238] In this embodiment of the present disclosure, the specific functions and examples of each unit in the virtual scene generation device 800 can be found in the relevant descriptions of the corresponding steps in the aforementioned virtual scene generation method embodiments, and will not be repeated here.

[0239] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0240] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0241] Figure 9 A schematic structural block diagram of an example electronic device 900 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as in-vehicle computing devices, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0242] like Figure 9As shown, the electronic device 900 includes a computing unit 901, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 902 or a computer program loaded from a storage unit 908 into a random access memory (RAM) 903. The RAM 903 may also store various programs and data required for the operation of the electronic device 900. The computing unit 901, ROM 902, and RAM 903 are interconnected via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0243] Multiple components in electronic device 900 are connected to I / O interface 905, including: input unit 906, such as keyboard, mouse, etc.; output unit 907, such as various types of renderers, speakers, etc.; storage unit 908, such as disk, optical disk, etc.; and communication unit 909, such as network card, modem, wireless transceiver, etc. Communication unit 909 allows electronic device 900 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0244] The computing unit 901 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above, such as virtual scene generation methods. For example, in some embodiments, the virtual scene generation method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 908. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 900 via ROM 902 and / or communication unit 909. When the computer program is loaded into RAM 903 and executed by the computing unit 901, one or more steps of the virtual scene generation method described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured as a virtual scene generation method by any other suitable means (e.g., by means of firmware).

[0245] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0246] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data optimization device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0247] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM) or flash memory, optical fibers, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0248] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a rendering device (e.g., a cathode ray tube (CRT) renderer or a liquid crystal display (LCD)) for rendering information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices are also used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0249] The systems and technologies described herein can be implemented in computing systems that include back-end components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected via digital data communication (e.g., a communication network) of any form or medium. Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0250] A computer system can include client and server components. Clients and servers are generally located far apart and typically interact via a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, a server in a distributed system, or a server incorporating blockchain technology.

[0251] This disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute a virtual scene generation method.

[0252] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements a virtual scene generation method.

[0253] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein. Furthermore, in this disclosure, relational terms such as "first," "second," and "third" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Additionally, "multiple" in this disclosure can be understood as at least two.

[0254] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for generating a virtual scene, comprising: Obtain the scene generation request; In response to the scene generation request, the original virtual scene is partitioned to obtain a basic virtual scene; wherein, the basic virtual scene includes multiple functional areas; Each of the multiple functional areas is taken as the target area. Using the target large model, multiple target instance objects are filled in the target area to obtain the scene area corresponding to the target area. Based on the aforementioned scene area, a target virtual scene is generated; The step of using a large target model to fill the target region with multiple target instance objects to obtain a scene region corresponding to the target region includes: Obtain object selection requirements related to the target area; using the first target large model, generate an overall object style related to the target area based on the scene generation request, and select multiple preliminary instance objects related to the target area from the instance object library based on the object selection requirements and the overall object style; Each of the multiple initial instance objects is taken as the object to be adjusted. The overall object style is refined using the second target large model to obtain a specific object style related to the object to be adjusted. An object adjustment instruction is obtained. Based on the specific object style and the object adjustment instruction, the object to be adjusted using the second target large model to obtain a target instance object corresponding to the object to be adjusted. Multiple target instance objects, each corresponding to one of the multiple initially selected instance objects, are filled into the target area to obtain the scene area corresponding to the target area.

2. The method according to claim 1, wherein, The step of using the second target large model, adjusting the object to be adjusted based on the specific object style and the object adjustment instruction to obtain the target instance object corresponding to the object to be adjusted, includes: Obtain the parameter expression of the object to be adjusted; wherein the parameter expression includes multiple initial parameters, and each of the multiple initial parameters corresponds to an object attribute of the object to be adjusted; Using the second target large model, based on the specific object style and the object adjustment instruction, the multiple initial parameters are adjusted to obtain multiple target parameters that correspond one-to-one with the multiple initial parameters; Based on the multiple target parameters, the object to be adjusted is adjusted to obtain a target instance object corresponding to the object to be adjusted.

3. The method according to claim 1 or 2, wherein, The step of generating a target virtual scene based on the scene region includes: Based on the aforementioned scene area, an initial virtual scene is constructed; If the initial virtual scene meets the preset scene requirements, the target virtual scene is generated based on the initial virtual scene.

4. The method according to claim 3, wherein, Determining that the initial virtual scene meets the preset scene requirements includes: From each of the multiple evaluation dimensions, a single evaluation result for the initial virtual scene is obtained, so as to obtain multiple single evaluation results that correspond one-to-one with the multiple evaluation dimensions; Based on the multiple individual evaluation results, the overall evaluation result of the initial virtual scene is obtained; If the overall evaluation result meets the preset evaluation requirements, the initial virtual scene is determined to meet the preset scene requirements.

5. The method according to claim 3, wherein, The process of generating the target virtual scene based on the initial virtual scene includes: Obtain target atmosphere information; Based on the target atmosphere information, the initial virtual scene is optimized to generate the target virtual scene.

6. The method according to claim 1, wherein, In response to the scene generation request, the original virtual scene is partitioned to obtain a basic virtual scene, including: Based on the scenario, a request is generated to determine the target scenario style; Obtain scene partitioning instructions; Using a partitioning model, the original virtual scene is partitioned based on the target scene style and the scene partitioning indication to obtain a basic virtual scene.

7. A virtual scene generation device, comprising: The request retrieval unit is used to retrieve the scene generation request; A scene partitioning unit is used to partition the original virtual scene in response to the scene generation request to obtain a basic virtual scene; wherein the basic virtual scene includes multiple functional areas. An object filling unit is used to take each of the multiple functional areas as a target area, and fill multiple target instance objects in the target area using a target large model to obtain a scene area corresponding to the target area. A scene generation unit is used to generate a target virtual scene based on the scene region; The object filling unit is used for: Obtain object selection requirements related to the target area; using the first target large model, generate an overall object style related to the target area based on the scene generation request, and select multiple preliminary instance objects related to the target area from the instance object library based on the object selection requirements and the overall object style; Each of the multiple initial instance objects is taken as the object to be adjusted. The overall object style is refined using the second target large model to obtain a specific object style related to the object to be adjusted. An object adjustment instruction is obtained. Based on the specific object style and the object adjustment instruction, the object to be adjusted using the second target large model to obtain a target instance object corresponding to the object to be adjusted. Multiple target instance objects, each corresponding to one of the multiple initially selected instance objects, are filled into the target area to obtain the scene area corresponding to the target area.

8. The apparatus according to claim 7, wherein, The object filling unit is used for: Obtain the parameter expression of the object to be adjusted; wherein the parameter expression includes multiple initial parameters, and each of the multiple initial parameters corresponds to an object attribute of the object to be adjusted; Using the second target large model, based on the specific object style and the object adjustment instruction, the multiple initial parameters are adjusted to obtain multiple target parameters that correspond one-to-one with the multiple initial parameters; Based on the multiple target parameters, the object to be adjusted is adjusted to obtain a target instance object corresponding to the object to be adjusted.

9. The apparatus according to claim 7 or 8, wherein, The scene generation unit is used for: Based on the aforementioned scene area, an initial virtual scene is constructed; If the initial virtual scene meets the preset scene requirements, the target virtual scene is generated based on the initial virtual scene.

10. The apparatus according to claim 9, wherein, The scene generation unit is used for: From each of the multiple evaluation dimensions, a single evaluation result for the initial virtual scene is obtained, so as to obtain multiple single evaluation results that correspond one-to-one with the multiple evaluation dimensions; Based on the multiple individual evaluation results, the overall evaluation result of the initial virtual scene is obtained; If the overall evaluation result meets the preset evaluation requirements, the initial virtual scene is determined to meet the preset scene requirements.

11. The apparatus according to claim 9, wherein, The scene generation unit is used for: Obtain target atmosphere information; Based on the target atmosphere information, the initial virtual scene is optimized to generate the target virtual scene.

12. The apparatus according to claim 7, wherein, The scene partitioning unit is used for: Based on the scenario, a request is generated to determine the target scenario style; Obtain scene partitioning instructions; Using a partitioning model, the original virtual scene is partitioned based on the target scene style and the scene partitioning indication to obtain a basic virtual scene.

13. An electronic device, comprising: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 6.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 6.

15. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images