Virtual scene construction method and virtual scene construction apparatus

Abstract

Embodiments of the present specification provide a virtual scene construction method and a virtual scene construction device, wherein the virtual scene construction method comprises: obtaining creation requirement information of a virtual scene, and generating a first perspective image of the virtual scene based on the creation requirement information; generating a second perspective image of the virtual scene based on the first perspective image and the creation requirement information; analyzing the second perspective image to determine spatial structured attribute information of the virtual scene; generating scene description data corresponding to the virtual scene based on the first perspective image, the second perspective image, and the spatial structured attribute information, and rendering the virtual scene based on the scene description data. According to the virtual scene construction method provided by the present specification, the accuracy of virtual scene construction is improved.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of artificial intelligence technology, and in particular to a virtual scene construction method and a virtual scene construction device. Background Technology

[0002] With the continuous development of artificial intelligence technology, people have increasingly higher requirements for the accuracy of virtual scene generation.

[0003] In scene generation practice, due to limitations in data acquisition costs and collection conditions, a single main scene image or panoramic image is typically used as the visual basis for scene modeling. Based on this single view, a pre-trained scene generation model is used for inference to generate a complete virtual 3D scene or a 2D multi-view layout. However, using a single view lacks detailed references to the virtual scene, making it difficult to accurately generate the virtual scene. Furthermore, due to the performance limitations of scene generation models, using the same scene generation model makes it difficult to adapt to different generation requirements, thus hindering the accurate generation of virtual scenes.

[0004] In summary, how to accurately generate virtual scenes has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, embodiments of this specification provide a method for constructing a virtual scene. One or more embodiments of this specification also relate to a game scene construction method, a virtual scene construction apparatus, a computing device, an electronic device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0006] According to a first aspect of the embodiments of this specification, a method for constructing a virtual scene is provided, comprising: Obtain the creation requirement information of the virtual scene, and generate a first-view image of the virtual scene based on the creation requirement information; Based on the first perspective image and the creation requirement information, a second perspective image for the virtual scene is generated; Analyze the second-view image to determine the spatial structured attribute information of the virtual scene; Based on the first perspective image, the second perspective image, and the spatial structured attribute information, scene description data corresponding to the virtual scene is generated, and the virtual scene is rendered based on the scene description data.

[0007] According to a second aspect of the embodiments of this specification, a game scene construction method is provided, comprising: Obtain the creation requirement text of the game virtual scene, and generate a first-person view image of the game virtual scene based on the creation requirement text; Based on the main view image and the creation requirement text, a top-down view image for the game virtual scene is generated; Analyze the top-down view image to identify at least one object to be constructed in the game virtual scene, which has semantic annotations and a movement structure. Based on the main view image, the top view image, and each object to be constructed in the game virtual scene with semantic annotations and movement structure, scene description data corresponding to the game virtual scene is generated, and the game virtual scene is rendered based on the scene description data.

[0008] According to a third aspect of the embodiments of this specification, a virtual scene construction apparatus is provided, comprising: The acquisition unit is configured to acquire the creation requirement information of the virtual scene and generate a first-view image of the virtual scene based on the creation requirement information; The generation unit is configured to generate a second-view image for the virtual scene based on the first-view image and the creation requirement information. The parsing unit is configured to parse the second-view image and determine the spatial structured attribute information of the virtual scene; The rendering unit is configured to generate scene description data corresponding to the virtual scene based on the first view image, the second view image, and the spatial structured attribute information, and to render the virtual scene based on the scene description data.

[0009] According to a fourth aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, they implement the steps of the virtual scene construction method described above.

[0010] According to a fifth aspect of the embodiments of this specification, an electronic device is provided, comprising: A memory and a processor, the memory and the processor being connected via a bus; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, they implement the steps of the virtual scene construction method described above.

[0011] According to a sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the virtual scene construction method described above.

[0012] According to a seventh aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the virtual scene construction method described above.

[0013] According to the virtual scene construction method provided in this manual, by acquiring the creation requirements information of the virtual scene and generating a first-view image accordingly, a visually anchoring scene can be generated based on user needs, providing a visual benchmark for subsequent multi-view generation and scene construction. Based on this, a second-view image is constructed from the first-view image and the virtual scene with creation requirements, allowing the spatial layout information of the scene to be presented from a different viewing angle than the first-view image, compensating for the shortcomings of a single viewpoint in expressing spatial structure. By parsing the second-view image to determine spatial structured attribute information, structured data such as spatial topological relationships, object positional relationships, and movement information of the scene can be automatically extracted from the image, avoiding the high-cost method of relying on manual annotation or manual modeling to obtain spatial information. Furthermore, the first-view image, second-view image, and spatial structured attribute information are fused to generate scene description data corresponding to the virtual scene, and the virtual scene is rendered accordingly. This ensures that the final generated virtual scene has both visual appearance reference and spatial structural constraints, effectively reducing spatial consistency problems such as mirror misalignment and object offset caused by the lack of spatial reference during cross-view scene generation, improving the accuracy of virtual scene generation and the reusability of scene assets. Attached Figure Description

[0014] Figure 1 This specification illustrates an architecture diagram of a virtual scene construction system provided in one embodiment. Figure 2 A flowchart of a virtual scene construction method according to an embodiment of this specification is shown; Figure 3 A flowchart illustrating the processing steps of a virtual scene construction method provided in one embodiment of this specification is shown. Figure 4 This specification shows a flowchart of a method for constructing a virtual game scene according to one embodiment. Figure 5 This specification shows a schematic diagram of the structure of a virtual scene construction apparatus according to one embodiment; Figure 6 A structural block diagram of a computing device according to one embodiment of this specification is shown; Figure 7 A structural block diagram of an electronic device provided according to one embodiment of this specification is shown. Detailed Implementation

[0015] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0016] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0017] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0018] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in one or more embodiments of this specification are obtained through open-source datasets or public datasets that comply with their license agreements, or are obtained with full authorization from the relevant parties. Moreover, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0019] In one or more embodiments of this specification, a large model refers to a deep learning model with a large number of model parameters, typically containing hundreds of millions, tens of billions, hundreds of billions, trillions, or even tens of trillions of model parameters. A large model can also be called a foundation model. It is pre-trained using large-scale unlabeled corpora to produce a pre-trained model with hundreds of millions of parameters. Such models can adapt to a wide range of downstream tasks and have good generalization ability. Examples include Large Language Models (LLMs) and multi-modal pre-training models.

[0020] In practical applications, large models only require a small number of samples to fine-tune the pre-trained model before they can be applied to different tasks. Large models can be widely used in fields such as Natural Language Processing (NLP) and Computer Vision. Specifically, they can be applied to computer vision tasks such as Visual Question Answering (VQA), Image Captioning (IC), and Image Generation, as well as natural language processing tasks such as text-based sentiment classification, text summarization, and machine translation. The main application scenarios of large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design.

[0021] The virtual scene construction method provided in this manual is applied to virtual scene generation.

[0022] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0023] Image generation models are models based on deep learning technology that can construct corresponding image content from text descriptions, reference images, or other conditional virtual scenes. Image generation models include diffusion-based models, generative adversarial network (GAN)-based models, and autoregressive models. Different image generation models have different strengths in terms of visual aesthetics, spatial structure reasoning, instruction compliance, and image quality.

[0024] Rendering refers to the process of generating a two-dimensional image from a corresponding viewpoint using computer graphics algorithms based on data such as the scene's geometric information, material properties, and lighting conditions. In virtual scene generation, rendering can generate scene images from any viewpoint based on data such as the three-dimensional structure, object positions, and material properties described in the scene description data.

[0025] With the development of computer and internet technologies, virtual scene generation technology has been widely applied in film and television production, animation creation, game development, and virtual reality. In the construction of virtual scenes, scene reference assets are crucial prior information for ensuring scene quality and cross-view consistency. Typically, a single master scene image or panoramic view is used as the visual basis for scene modeling, and a pre-trained scene generation model is used to infer from this single view to generate a complete virtual scene.

[0026] However, using a single view lacks reference for the spatial structure and multi-angle appearance of the virtual scene. When generating scene images from different perspectives, such as reverse shots, side shots, or moving shots, the model needs to infer the spatial layout relationships of the scene itself, which can easily lead to spatial inconsistencies such as mirror misalignment, object position displacement, and incorrect entrances and exits. In addition, different scene generation models have different focuses in terms of aesthetic expression, spatial structure reasoning ability, and image cleanliness. A single generation model can hardly simultaneously take into account visual aesthetics, structural accuracy, and image quality. Especially in complex scenes, it is prone to problems such as structural collapse or image dirtiness, affecting the accuracy of virtual scene generation.

[0027] In view of this, in order to solve the above-mentioned technical problems, this specification provides a virtual scene construction method. This specification also relates to a game scene construction method, a virtual scene construction device, a computing device, an electronic device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0028] See Figure 1 , Figure 1 This specification illustrates an architecture diagram of a virtual scene construction system according to an embodiment of the present specification. The virtual scene construction system may include a client 100 and a server 200. Client 100 is used to send creation request information to server 200; Server 200 is used to obtain the creation requirement information of the virtual scene, and generate a first-view image of the virtual scene based on the creation requirement information; generate a second-view image of the virtual scene based on the first-view image and the creation requirement information; parse the second-view image to determine the spatial structured attribute information of the virtual scene; generate scene description data corresponding to the virtual scene based on the first-view image, the second-view image and the spatial structured attribute information, and render the virtual scene based on the scene description data; and send the virtual scene to client 100. Client 100 is also used to receive virtual scenes sent by server 200.

[0029] The virtual scene construction system may include multiple clients 100 and a server 200. Clients 100 can be referred to as edge devices, and server 200 can be referred to as cloud devices. Multiple clients 100 can establish communication connections through server 200. In the virtual scene construction scenario, server 200 is used to provide virtual scene construction services between multiple clients 100. Each client 100 can act as a sender or receiver, communicating through server 200.

[0030] Users can interact with server 200 through client 100 to receive data sent by other clients 100, or send data to other clients 100, etc. In the virtual scene construction scenario, users can publish data streams to server 200 through client 100, server 200 can generate virtual scenes based on the data streams, and push the virtual scenes to other clients that have established communication.

[0031] In this system, client 100 and server 200 establish a connection via a network. The network provides the medium for communication between client 100 and server 200. The network can include various connection types, such as wired or wireless communication links or fiber optic cables. Data transmitted by client 100 may need to undergo encoding, transcoding, compression, or other processing before being published to server 200.

[0032] Client 100 can be a browser, an app (application), a web application such as an H5 (HyperText Markup Language 5) application, a lightweight application (also known as a mini-program), or a cloud application. Client 100 can be developed based on the software development kit (SDK) of the corresponding service provided by server 200, such as a real-time communication (RTC) SDK. Client 100 can be deployed on a computing device and depends on the device or certain apps on the device to run. The computing device may have a display screen and support information browsing, such as a personal mobile terminal like a mobile phone, tablet, or personal computer. Various other types of applications can also be configured on the computing device, such as human-computer interaction applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, and social media platform software.

[0033] Server 200 may include servers providing various services, such as servers providing communication services to multiple clients, servers supporting backend training of models used on clients, and servers processing data sent by clients. It should be noted that server 200 can be implemented as a distributed server cluster composed of multiple servers, or as a single server. The server can also be a server in a distributed system, or a server integrated with blockchain. The server can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0034] It is worth noting that the virtual scene construction method provided in the embodiments of this specification is generally executed by the server. However, in other embodiments of this specification, the client may also have similar functions to the server, thereby executing the virtual scene construction method provided in the embodiments of this specification. In other embodiments, the virtual scene construction method provided in the embodiments of this specification may also be executed jointly by the client and the server.

[0035] See Figure 2 , Figure 2 A flowchart of a virtual scene construction method according to an embodiment of this specification is shown, which specifically includes the following steps.

[0036] Step 202: Obtain the creation requirement information of the virtual scene, and generate a first-view image of the virtual scene based on the creation requirement information.

[0037] The creation requirement information can be understood as a set of feature data describing the virtual scene that the user expects to build, used to guide the generation direction and quality requirements of the virtual scene. Creation requirement information can cover multiple descriptive dimensions, including scene type information (e.g., indoor, outdoor, mixed scenes), style preference information (e.g., realistic style, cartoon style, cyberpunk style), theme description information (e.g., "ancient palace," "future city"), and spatial scale information (e.g., large, medium, small scenes). The creation requirement information can be obtained in various ways: users can directly describe the desired scene features through natural language text, select a basic template from a pre-set scene template library and customize it, or upload reference images to convey visual style preferences.

[0038] In one specific implementation, the creation requirement information may include structured parameter configurations, such as parameterized descriptions of scene size, primary color tone, and lighting type (natural light, artificial light, mixed light). In another specific implementation, the creation requirement information may include unstructured natural language descriptions, which are parsed by a natural language understanding module to extract key scene elements and convert them into an input format that the model can understand.

[0039] The first-person perspective image can be understood as the visual presentation of the virtual scene from the primary viewing direction, serving as a visual anchor. The first-person perspective image determines the overall visual tone of the scene, including color style, lighting atmosphere, material texture, and spatial depth. The first-person perspective image can be a front view. It is important to clarify that in this specification, the first-person perspective image is used to provide a basic perspective template for subsequent virtual scene generation. It is also important to clarify that subsequent second-person perspective images and multi-angle views must maintain visual style consistency with the first-person perspective image. In one specific implementation, the first-person perspective image is a hero shot scene image, representing a representative perspective of the scene's overall atmosphere and visual characteristics. In another specific implementation, the first-person perspective image can also be a frontal panoramic view or a three-quarter view of the scene; the specific perspective selection can be determined based on the scene type and preferences set in the creation requirements information.

[0040] In one specific implementation provided in this specification, an image generation model is invoked based on the creation requirements of the virtual scene to generate a first-person perspective image.

[0041] It should be clarified that this image generation model can be a general image generation model, that is, generating corresponding images based on multimodal information, or it can be a target image generation model that fits the creation requirements.

[0042] To facilitate understanding, this manual uses the generation requirements corresponding to images from different perspectives and the determination of the corresponding target image generation models as examples to explain the generation process of first-view images.

[0043] In one specific embodiment provided in this specification, during the construction of a virtual scene, different stages have varying requirements for the capabilities of the image generation model, and a single image generation model is unlikely to achieve optimal performance across all capability dimensions. Therefore, a mechanism is needed that can dynamically select a suitable image generation model based on the generation requirements of different stages.

[0044] In one specific embodiment provided in this specification, the method further includes: Based on the aforementioned viewpoint adjustment parameters, the generation requirement information corresponding to the reference viewpoint image is determined in the virtual scene; In the set of image generation models, the target image generation model corresponding to the generation requirement information is determined, wherein the set of image generation models includes at least one image generation model, and different image generation models have different image generation capabilities; A reference view image in the virtual scene is generated based on the target image generation model.

[0045] The image generation model set can be understood as a pre-built model library containing multiple image generation models, each with its own advantages and characteristics in a specific dimension of image generation capability. Image generation capability dimensions can include: aesthetic and stylistic expression capability (the visual appeal and artistic style of the generated image), spatial structure reasoning capability (the ability to understand and reason about three-dimensional spatial relationships and perspective structures), instruction compliance capability (the ability to accurately control the generated content according to text instructions), image cleanliness control capability (the level of control over noise, dirtiness, and artifacts in the generated image), and image editing and restoration capability (the ability to modify and optimize the quality of existing images). The capability profile of each image generation model can be evaluated and recorded in advance through benchmark testing and user feedback.

[0046] Generating requirement information can be understood as information that describes the specific requirements of a particular viewpoint image generation task on the model's capabilities, as determined by the creation requirement information. Different viewpoint image generation stages have different requirements for model capabilities: generating first-person perspective images (front view) requires higher capabilities in aesthetic and stylistic expression; generating second-person perspective images (top view) requires higher capabilities in spatial structure reasoning and instruction compliance; and image inpainting requires higher capabilities in controlling image clarity. Generation requirement information can include the required core capability dimensions, the priority and weight of each capability dimension, and other information.

[0047] The target image generation model can be understood as an image generation model that is best suited for the current generation task, matched from a set of image generation models based on the generation requirement information. The matching process can rank and select models based on the degree of matching between the model's capability profile and the generation requirement information, choosing the model with the highest matching degree as the target image generation model. In one specific implementation, the matching process can compare the capability dimension requirements in the generation requirement information with the capability profiles of each model, selecting the model that performs better in the required core capability dimensions.

[0048] In one specific embodiment provided in this specification, an image generation model set is maintained, including a first image generation model with strong aesthetic expression capabilities and a second image generation model with strong spatial reasoning and instruction compliance capabilities. During the generation of first-view images, based on the high priority requirement for aesthetic and stylistic expression capabilities in the generation requirement information, the first image generation model is selected as the target model; during the generation of second-view images, based on the high priority requirement for spatial reasoning and instruction compliance capabilities, the second image generation model is selected as the target model.

[0049] Optionally, a model performance monitoring mechanism can be introduced to track the generation quality, response time, success rate, and other operational metrics of each model in real time. When the performance metrics of a model decline, it can be automatically switched to a backup model. In practice, multiple factors such as capability matching, generation cost, and response speed can be comprehensively considered when selecting models, and a balance between quality and efficiency can be achieved through multi-objective optimization strategies.

[0050] Furthermore, when determining the generation model corresponding to images from different perspectives based on the image generation model set, different models may have their own inherent generation defects, such as spatial collapse in complex scenes, image noise and dirtiness, smearing or over-sharpening, etc. These defects will affect the quality of the generated images, requiring an automated defect compensation mechanism.

[0051] Accordingly, in one specific embodiment provided in this specification, the method further includes: Obtain a model defect profile library, which includes defect feature information corresponding to each image generation model; Accordingly, generating a reference viewpoint image in the virtual scene based on the target image generation model includes: Based on the model defect profile library, the target defect feature information corresponding to the target image generation model is determined; Based on the generation requirement information and the target defect feature information, an image generation prompt word is constructed. The image generation prompt word is input into the target image generation model to obtain a reference view image generated by the target image generation model.

[0052] The model defect profile library can be understood as a dataset recording the known defect features of each image generation model. These defect features describe potential quality issues that may arise during the generation process for a specific image generation model, including attributes such as the type, manifestation, triggering conditions, and severity of the defect. For example, the defect features of a particular image generation model might include: perspective distortion when generating complex spatial structures; pre-existing photographic noise in large areas of solid color; and smearing or over-sharpening in areas rich in detail. The model defect profile library can be built and updated through systematic benchmarking and quality analysis of each model, and can also be continuously improved by collecting user feedback and automated quality detection data.

[0053] Target defect feature information can be understood as defect description data extracted from the model defect profile library, corresponding to the current target image generation model. Target defect feature information indicates the specific defects that the currently used model may encounter during the generation process, providing a basis for constructing image generation prompts with defect compensation capabilities.

[0054] Image generation prompts can be understood as text instructions constructed by comprehensively considering generation requirements and target defect features, used to guide the target image generation model to generate a reference viewpoint image that meets the requirements. It should be clarified that the image generation prompts in this specification include three types of constraints: The first category is suppression constraints, which propose suppression requirements for known defects in the target model, such as "noise removal, suppression of particles and color spots, and elimination of artifacts". The second type is the original structure preservation constraint, which requires that the lighting, shadows and structural relationships of the original image be maintained during the generation process, such as "maintain the original lighting and shadow structure, and prohibit relighting". The third category is negative constraints, which list the types of defects that are not expected to occur in the form of an exclusion list, such as "no smearing feel, no over-sharpening, no waxy surface", etc.

[0055] The three types of constraints work together to guide the model to generate high-quality target images from three dimensions: positive guidance, structural protection, and negative exclusion.

[0056] In one specific embodiment provided in this specification, the model defect profile library records the known defect features of each image generation model. When a target model is invoked, the system automatically queries the defect profile library to obtain the defect feature information of the target model and constructs defect compensation prompts accordingly. The defect compensation prompts include three types of constraints: suppression constraints (such as noise reduction and cleaning, suppressing graininess, dirt spots, color noise, and artifacts), original structure preservation constraints (preserving the original lighting and shadow structure, prohibiting relighting), and negative constraints (an exclusionary list listing unwanted defect types). The system combines the defect compensation prompts with the generation requirement information and inputs them into the target image generation model.

[0057] Optionally, the strength of the three types of constraints can be dynamically adjusted based on the specific characteristics of the generation task. For example, when the generated content involves large areas of solid color, the constraint strength for suppressing noise and dirtiness is strengthened; when the generated content involves complex spatial structures, the constraint strength for preserving perspective and structural relationships is strengthened. In addition, a defect detection feedback mechanism can be introduced to automatically detect actual defects after the target image is generated, and feed the detection results back to the model defect profile library to continuously improve the defect profiles of each model.

[0058] According to a specific implementation method provided in this specification, proactive compensation for inherent defects in each image generation model is achieved through the construction of a model defect profile library and the automatic generation of defect compensation prompts. Before calling the target model, prompts containing three types of constraints—suppression, preservation, and negative—are automatically constructed based on its defect profile. Defect compensation information is integrated into the generation instructions, guiding the model to proactively avoid known defects during the generation process. It should be noted that the model defect profile library can be updated without modifying the system code as model versions are updated and new defects are discovered.

[0059] In one specific embodiment provided in this specification, a first-view image is obtained according to the above description.

[0060] Step 204: Based on the first perspective image and the creation requirement information, generate a second perspective image for the virtual scene.

[0061] In this context, second-view images can be understood as scene images viewed from a different angle than first-view images, focusing on showcasing the spatial layout information of the scene. While the first-view images primarily present the visual appearance, the second-view images supplement the scene's spatial structure information from another observational dimension, allowing the spatial relationships within the scene to be explicitly expressed. It's important to clarify that second-view images complement the first-view images, serving to define the spatial structure within the virtual scene.

[0062] In one specific embodiment, the second-view image is a top-down view, providing a bird's-eye view of the entire spatial layout from above, clearly demonstrating the spatial distribution of elements such as architectural structures, main objects, furniture, entrances, and passageways. The top-down view image can also carry semantic annotation information, numbering and labeling each object or area in the scene with directional information (such as east, south, west, and north), allowing spatial relationships to be structurally recorded and subsequently verified. In another specific embodiment, the second-view image can also be a side view or a cross-sectional view, suitable for displaying the spatial structure of specific scene types.

[0063] In one specific embodiment provided in this specification, a second perspective image, different from the first perspective image, is determined based on the first perspective image and creation requirement information.

[0064] In one specific implementation, the generation features of the second-view image are determined. Based on these features, a target image generation model corresponding to the second-view image is determined from the image generation model set, and defect feature information corresponding to the target image generation model is determined from the model defect profile library. On this basis, the first-view image, creation requirement information, and defect feature information are used as inputs to the target image generation model to obtain the second-view image output by the target image generation model.

[0065] Step 206: Analyze the second perspective image to determine the spatial structured attribute information of the virtual scene.

[0066] Spatial structured attribute information can be understood as structured data parsed from second-view images, used to describe the spatial topological relationships between various components within a virtual scene. Unlike second-view images in image form, spatial structured attribute information records the spatial relationships of the scene in a data format, facilitating automated processing and verification by computer programs. Spatial structured attribute information can include data in the following dimensions: Object identification and location data: Records the identification information of each object or region in the scene and its position in the scene coordinate system; Object relationship data describes the relative positions (e.g., "tables and chairs are located inside the room" and "door is located on the north wall") and adjacency relationships between objects; Entrance / exit and connectivity data, marking the location of each entrance / exit in the scene and its connected target areas; Movement data describes the primary and secondary movement paths of a character or object within the scene.

[0067] In one specific implementation, the spatial structured attribute information can be organized into a label layer with numbered semantic annotations, where each label item contains information such as number, semantic category, and location coordinates.

[0068] Step 208: Based on the first viewpoint image, the second viewpoint image, and the spatial structured attribute information, generate scene description data corresponding to the virtual scene, and render the virtual scene based on the scene description data.

[0069] Scene description data can be understood as a complete description of the virtual scene constructed based on first-view images, second-view images, and spatially structured attributes. The scene description data constrains and describes the virtual scene from multiple dimensions, ensuring that the final rendered virtual scene meets requirements in terms of visual appearance, spatial structure, and multi-view consistency.

[0070] It should be clarified that scene description data may include: Visual constraint information refers to the scene's visual style constraints determined by first-person perspective images, including reference standards for visual attributes such as color, lighting, and material. Spatial constraint information refers to scene spatial relationship constraints determined by second-view images and spatial structured attribute information, including constraint standards for spatial attributes such as object positions, entrance / exit relationships, and movement routes. A multi-view group is a set of images of the same scene presented from multiple different camera positions and angles. This set of multi-views maintains a consistent spatial structure and differs only in the viewing angle.

[0071] In one specific implementation, the multi-angle view group is presented in a nine-grid format, that is, the same scene is photographed from nine different perspectives, and the images from the nine perspectives together constitute a multi-angle appearance reference for the scene, providing a consistency constraint for subsequent scene rendering from different camera positions.

[0072] When generating scene description data for virtual scenes, it is necessary to consider both the visual appearance quality and the correctness of the spatial structure of the scene to ensure that the generated virtual scene has good visual appeal and maintains correctness and consistency in spatial relationships.

[0073] In one specific embodiment provided in this specification, the spatial structured attribute information includes at least one object with semantic annotation and movement structure; based on the first viewpoint image, the second viewpoint image, and the spatial structured attribute information, scene description data corresponding to the virtual scene is generated, including: Based on the first perspective image, determine the visual constraint information of the virtual scene; Based on the second-view image and at least one object with semantic annotation and movement structure, determine the spatial constraint information of each object in the virtual scene; Based on the visual constraint information and the spatial constraint information, scene description data corresponding to the virtual scene is generated.

[0074] The visual constraint information can be understood as scene visual appearance constraint data extracted from the first-view image, which is used to ensure that the visual style of the subsequently generated scene image is consistent with that of the first-view image.

[0075] It should be clarified that visual constraint information can include color distribution characteristics (such as dominant color tone, color saturation range, color contrast, etc.), lighting characteristics (such as light source direction, light intensity, shadow distribution, highlight areas, etc.), material texture characteristics (such as wood texture, stone texture, fabric texture, etc.), and overall atmosphere characteristics (such as warmth, coolness, light and dark contrast, etc.).

[0076] Building upon the above, methods for extracting visual constraint information may include feature encoding of the first-view image, mapping the image into a visual feature vector that characterizes the image's visual style attributes. In one specific implementation, the visual constraint information can be directly retained in the form of a reference image of the first-view image, serving as a reference standard for visual style in subsequent generation processes.

[0077] Spatial constraint information can be understood as scene spatial relationship constraint data extracted from second-view images and spatially structured attribute information, used to ensure that subsequently generated scene images maintain consistency with the determined spatial relationships in terms of spatial structure. It includes at least one object with semantic annotations and movement patterns.

[0078] It should be clarified that spatial constraint information can include constraints on the position of objects (the range of each object's position in the scene coordinate system), relative orientation constraints (the vertical, horizontal, and front-back relationships between objects), connectivity constraints (the connecting targets and passage directions of entrances and exits), movement constraints (the rationality of character movement paths), and scale constraints (the proportional relationships between objects). In one specific implementation, spatial constraint information can be organized into a set of constraint conditions, each of which describes a specific constraint that needs to be satisfied in the scene's spatial relationships.

[0079] In one specific embodiment provided in this specification, corresponding visual constraint information is extracted from the first-view image, including visual attributes such as the scene's color tone, lighting direction, shadow distribution, and material texture. From the second-view image, its corresponding semantic annotation content is determined, and spatial constraint information is generated based on the semantic annotations. This includes spatial attribute information such as the position coordinates, relative positional relationships, entrance / exit connections, and movement paths of each virtual object in the virtual scene. Based on this, a virtual scene is constructed according to the visual constraint information and spatial constraints.

[0080] Optionally, different weights can be assigned to visual and spatial constraints to adjust their influence based on specific application needs. For example, in artistic creation scenarios emphasizing visual expressiveness, the weight of visual constraints can be appropriately increased; in architectural visualization scenarios emphasizing spatial accuracy, the weight of spatial constraints can be appropriately increased. In practice, a constraint conflict detection mechanism can also be introduced. When visual and spatial constraints conflict (e.g., a certain lighting effect may affect the visual presentation of spatial relationships), automatic coordination is performed according to priority rules.

[0081] According to a specific implementation method provided in this specification, the generation process of scene description data is decomposed into two independent constraint dimensions: visual constraints and spatial constraints. This ensures that the scene description data is simultaneously constrained by both visual appearance and spatial structure. Visual constraint information guarantees the consistency of the generated result's visual style with the first-person perspective image, avoiding style drift or visual quality degradation during cross-viewpoint generation. Spatial constraint information guarantees the correctness of the generated result's spatial relationships, avoiding spatial consistency issues such as object position offsets, mirror reversals, or incorrect entrances / exits. These two types of constraint information complement each other, jointly constituting a comprehensive constraint system for the virtual scene.

[0082] Furthermore, to further improve the accuracy of virtual scene construction, this specification employs the following method to generate scene description data corresponding to the virtual scene based on the visual constraint information and the spatial constraint information, including: Based on the visual constraint information and the spatial constraint information, an initial virtual scene is generated; Obtain view adjustment parameters; Based on the perspective adjustment parameters, at least one third-view image corresponding to the initial virtual scene is determined; Based on each third-person perspective image, scene description data corresponding to the virtual scene is generated.

[0083] The initial virtual scene can be understood as an intermediate result of a scene constructed based on visual and spatial constraints. It satisfies both visual style and spatial structure requirements, but has not yet generated complete scene description data. The initial virtual scene can be understood as a scene prototype constructed under dual constraints, possessing basic visual appearance and spatial layout. The process of generating the initial virtual scene can involve inputting both visual and spatial constraints into a scene generation model, which then generates a scene image or scene description data that satisfies the constraints under the joint guidance of the two types of constraints.

[0084] Viewpoint adjustment parameters can be understood as defining the position and angle configuration of the camera when generating a third-person view image from the initial virtual scene. Viewpoint adjustment parameters can include multiple sets of camera configurations, each corresponding to a specific viewing angle. For example, viewpoint adjustment parameters can define nine different camera positions and angles, corresponding to front, left, right, back, top, bottom, and four oblique angles. In one specific implementation, viewpoint adjustment parameters can be stored as a preset parameter set, with an appropriate set selected based on the scene type and application requirements. In another specific implementation, viewpoint adjustment parameters can also be customized by the user according to specific needs.

[0085] Third-person perspective images can be understood as scene images generated from the initial virtual scene at specific angles based on perspective adjustment parameters. Each third-person perspective image presents the initial virtual scene from the corresponding camera position and angle. Since the initial virtual scene already satisfies visual and spatial constraints, the third-person perspective images generated from the initial virtual scene have a good foundation in visual style and spatial relationships. A multi-angle view group consists of multiple third-person perspective images, arranged in a nine-square grid. It should be noted that the generation method of third-person perspective images can refer to the generation method of first-person or second-person perspective images; this specification will not elaborate further.

[0086] In one specific embodiment provided in this specification, an initial virtual scene is generated based on visual constraint information and spatial constraint information. In this case, the generation quality of the initial virtual scene is determined, wherein the generation quality may be determined based on spatial verification conditions. Based on the generation quality of the initial virtual scene, at least one viewpoint adjustment parameter is further obtained, and at least one third-view image is obtained based on each viewpoint adjustment parameter. The virtual scene is generated based on each third-view image, the first-view image, and the second-view image.

[0087] Optionally, this specification may also include a consistency cross-check after generating the third-view images, comparing the appearance and position of the same virtual object in different third-view images to see if they are consistent, and correcting any inconsistencies. Optionally, different generation precision levels can be set according to the importance or frequency of use of the third-view images; for example, a more refined generation strategy can be used for frequently used perspectives, while a faster generation strategy can be used for infrequently used perspectives, thus balancing quality and efficiency.

[0088] In one specific implementation, the front view and top-down spatial structure diagram serve as a dual-condition input image generation model, generating initial virtual scene images that satisfy both visual and spatial constraints. Then, preset nine-grid perspective adjustment parameters are obtained, including nine different camera positions and angle configurations. Based on these nine sets of parameters, nine third-person perspective images are generated from the initial virtual scene. These nine images maintain the spatial structure unchanged, only altering the camera positions and angles, collectively forming a nine-grid multi-angle view group. This nine-grid multi-angle view group is stored as a core component of the scene description data.

[0089] According to a specific implementation method provided in this specification, an initial virtual scene satisfying dual constraints is first generated. Then, multiple third-view images are generated from the initial scene based on viewpoint adjustment parameters. Finally, these third-view images are organized into scene description data, realizing a progressive scene construction process from constraint fusion to multi-angle view generation. The initial virtual scene, as an intermediate step, encodes visual and spatial constraint information into a unified intermediate representation. Subsequent multi-angle view generation only requires changing the observation angle, eliminating the need to re-process the constraint fusion problem. This simplifies the multi-angle view generation process while ensuring consistency in visual style and spatial relationships among the third-view images.

[0090] In one specific embodiment provided in this specification, during the multi-view image generation process of a virtual scene, the complexity of images from different viewpoints varies. In view of this, this specification also provides an automatic triggering repair mechanism to ensure image quality while controlling the consumption of computing resources.

[0091] In one specific embodiment provided in this specification, the method further includes: Obtain the image complexity corresponding to the reference viewpoint image; If the image complexity meets the repair triggering condition, the reference viewpoint image is repaired.

[0092] Image complexity can be understood as a quantitative indicator measuring the complexity of image content from a reference perspective, reflecting the richness of visual information that needs to be generated and processed in the image. Image complexity can be evaluated from multiple dimensions, including the number of structural elements (the number of independent objects and architectural structures in the scene), device density (the density of furniture, decorations, and other devices in the scene), spatial depth (the number of front and back depth layers in the scene), and texture detail richness (the types and complexity of different material textures in the scene). In one specific implementation, image complexity can be automatically calculated by performing feature analysis on the image, such as by analyzing features like edge density, color distribution diversity, and texture complexity. In another specific implementation, image complexity can also be estimated based on scene characteristics (such as scene type, number of devices, and spatial scale) described in the creation requirements information.

[0093] The repair trigger condition can be understood as the criterion for determining whether repair processing of the reference viewpoint image is necessary. The repair trigger condition can be based on a threshold set for image complexity; when the image complexity exceeds a preset threshold, the repair process is triggered. Alternatively, the repair trigger condition can be based on the automatic evaluation result of image quality; when the image quality score is lower than a preset standard, repair is triggered. In one specific implementation, the repair trigger condition can comprehensively consider both image complexity and image quality; when the complexity is high and the quality score is low, repair is triggered. Repair operations can include denoising, structural repair, and image enhancement, and can be performed by an image processing model with strong image editing and repair capabilities.

[0094] In one specific embodiment provided in this specification, after generating each viewpoint image, the image complexity of that viewpoint image is calculated, including indicators such as the number of structural elements, device density, and spatial depth levels. When the image complexity exceeds a preset threshold, a repair phase is triggered, where a repair model with strong image editing capabilities performs denoising and structural repair processing on the reference viewpoint image. During the repair process, defect compensation prompts are also injected to ensure that the repair operation does not introduce new image defects.

[0095] Optionally, different levels of restoration strategies can be configured based on the image complexity. For moderately complex images, only lightweight denoising is performed; for highly complex images, a complete restoration process including structural restoration and image enhancement is executed. Furthermore, a restoration effect evaluation mechanism can be introduced to reassess the image quality after restoration. If the image quality still does not meet the preset standards, a second restoration is performed or the image is marked for manual review.

[0096] According to a specific implementation provided in this specification, intelligent scheduling of repair resources is achieved through image complexity assessment and gating of repair trigger conditions. For simple viewpoint images, the repair process is not triggered, saving unnecessary computational resource consumption; for complex viewpoint images, the repair process is automatically triggered to ensure image quality in complex scenes. Based on the complexity-based trigger mechanism, a good balance is achieved between image quality and computational efficiency, avoiding resource waste caused by performing repair operations on all images, and also avoiding quality problems caused by complex images not being repaired.

[0097] In one specific implementation provided in this specification, after generating a virtual scene and obtaining scene description data, it is necessary to verify the spatial consistency of the generated virtual scene under different perspectives to ensure that the scene assets will not have spatial continuity issues when referenced across shots in the future.

[0098] In one specific embodiment provided in this specification, after rendering the virtual scene, the method further includes: Based on the scene description data and the viewpoint adjustment parameters, at least one viewpoint image to be verified is determined in the virtual scene, and the spatial description information corresponding to each viewpoint image to be verified is determined. Verify the spatial description information corresponding to each viewpoint image to be verified, and obtain the verification results corresponding to each viewpoint image to be verified. If the verification results for each image from the viewpoint to be verified all meet the verification conditions, store the scene description data corresponding to the virtual scene. If the verification result corresponding to the image to be verified does not meet the verification conditions, the scene description data corresponding to the virtual scene is adjusted.

[0099] The image to be verified can be understood as a scene image determined based on the generated scene description data and viewpoint adjustment parameters, used to verify the spatial correctness of the virtual scene at a specific viewing angle. In this specification, the image to be verified can also be a first-view image, a second-view image, or a third-view image.

[0100] Viewpoint adjustment parameters can be understood as defining the camera's position and angle configuration, which may include parameters such as the camera's horizontal rotation angle, pitch angle, and offset from the scene center. By setting different viewpoint adjustment parameters, corresponding viewpoint images to be verified can be generated from multiple observation angles to be tested. For example, a reverse viewpoint at 180 degrees from the original main viewpoint can be set as the verification viewpoint to check whether the positional relationship of objects in the scene in the reverse direction is correct; a 90-degree side viewpoint can also be set to check whether the arrangement of objects on the side of the scene and the positions of entrances and exits meet expectations. Here, the camera can be understood as a virtual description deployed in a virtual scene.

[0101] It should be clarified that the number of images to be verified can be one or more, covering the key observation angles of the scene.

[0102] Spatial description information can be understood as descriptive information corresponding to different viewpoints to be verified in a generated virtual scene, used to describe the display content of that viewpoint in the virtual scene. It should be noted that the spatial description information includes at least the spatial location information of at least one object contained in the image of that viewpoint. The spatial description information corresponding to different viewpoint images collectively constitutes the spatial structured attribute information of the virtual scene.

[0103] Verification criteria can be understood as the standards used to determine whether the spatial relationships between objects in an image from a perspective to be verified are correct. Verification criteria can cover various types of spatial relationship checks, including the correctness of the relative positions of objects in a virtual scene, the correctness of entrance / exit locations and connectivity, the rationality of movement paths, checks for object mirror reversal, consistency of object size proportions, and the feasibility of camera positions. In one specific implementation, verification criteria may include the detection of the following seven types of spatial errors: left / right mirroring errors, object misalignment, missing or redundant objects, structural contradictions, errors in movement paths or entrances / exits, scale distortion, and infeasible camera positions. The verification process compares the image from the perspective to be verified with the spatial structured attribute information to determine whether the spatial relationships presented in the image are consistent with the description of the structured attribute information.

[0104] In this specification, the verification conditions can be understood to include at least one of a)-e): a) Mirror verification: Check for left-right mirroring errors between images from different viewpoints.

[0105] b) Misalignment check: Check whether the relative position of the same object is consistent in images from different viewpoints.

[0106] c) Entrance / Exit Verification: Check whether the position and orientation of the object information corresponding to the exit scene and the object information corresponding to the entrance scene in the virtual scene are accurate.

[0107] For example, check whether the position and orientation of doors, windows, and passageways in a virtual scene are correct from different perspectives.

[0108] d) Movement path verification: Check whether the character's movement path is continuous and reasonable in different viewpoints.

[0109] Here, "character" can be understood as a virtual character that can move around in the virtual scene later.

[0110] For example, based on the movement path of a virtual character in a virtual scene, it can be determined whether different perspectives are continuous and reasonable relative to the virtual character.

[0111] e) Scale verification: Check whether the positions and sizes of different virtual objects in the virtual scene are consistent.

[0112] For example, check whether the size proportions of furniture and devices are consistent in different viewpoints within a virtual scene.

[0113] In one specific implementation, spatial relationship verification uses semantic annotations in the top-view spatial structure diagram as a benchmark to automatically check the relative relationships of objects, entrances / exits, and movement paths in the generated frames. It specifically checks for seven types of spatial errors: left / right mirroring, object misalignment, missing or redundant objects, structural contradictions, incorrect movement paths or entrances / exits, scale distortion, and infeasible camera positions. If the verification fails, the erroneous local areas are regenerated until the verification passes.

[0114] Optionally, the verification conditions can be divided into sets of sub-conditions with different levels of stringency, and the corresponding level of verification standard can be selected according to the application requirements of the virtual scene. For example, for high-quality scene assets that need to be frequently referenced across shots, a stringent verification standard that includes all seven types of error detection can be adopted; for scenes that are only used for shooting from a fixed camera position, only a subset of spatial relationships related to the camera position can be verified. In specific implementation, an automated verification scoring mechanism can also be introduced to calculate a spatial consistency score for each image from the viewpoint to be verified. When the score is higher than a preset threshold, it is judged as passing the verification; when it is lower than the threshold, it is judged as failing, and the failed items and corresponding error types are fed back to the adjustment stage.

[0115] In one specific embodiment provided in this specification, verifying the spatial description information corresponding to each viewpoint image to be verified and obtaining the verification result corresponding to each viewpoint image to be verified includes: In each image to be verified, a first image to be verified and a second image to be verified, including the target object to be verified, are determined, and the first spatial description information corresponding to the first image to be verified and the second spatial description information corresponding to the second image to be verified are obtained, wherein the target object to be verified is any object in the virtual scene. Parse the first spatial description information and the second spatial description information to obtain the first spatial position information of the target object to be verified in the first view image to be verified and the second spatial position information of the target object to be verified in the second view image to be verified. By comparing the first spatial location information and the second spatial location information of the target object to be verified, the verification results corresponding to the first and second viewpoint images to be verified are obtained.

[0116] The image to be verified is determined based on the viewpoint adjustment parameters. It can be a first-view image, a second-view image, or a third-view image. It can also be an image from a different viewpoint than the first-view image, the second-view image, or the third-view image.

[0117] The target object to be verified is an object in a virtual scene that is included in both the first and second viewpoint images, and it is used to verify the accuracy of images from different viewpoints.

[0118] In one specific embodiment provided in this specification, a pair of viewpoint images to be verified are acquired to obtain a first viewpoint image and a second viewpoint image to be verified. The consistency of the target object to be verified in the first and second viewpoint images is determined. For example, information such as the object's spatial location and size is used to determine the rationality of the spatial structure of the first and second viewpoint images relative to the virtual scene.

[0119] According to a specific implementation method provided in this specification, a closed-loop mechanism from scene generation to spatial verification is formed by introducing a spatial verification step after the virtual scene is generated. The images to be verified examine the spatial relationships of the scene from angles different from the originally generated perspective, exposing spatial consistency issues that are difficult to detect from a single perspective. Spatial verification conditions are compared based on spatial structured attribute information, transforming the judgment of the correctness of spatial relationships into an automated verification process. When all images to be verified pass spatial verification, it indicates that the spatial relationships of the scene are correct from multiple angles, and the scene description data can be stored as a qualified scene asset. When there are perspectives that fail verification, the scene description data adjustment process is triggered to correct the parts with spatial errors, thereby ensuring that the finally stored scene asset has reliable spatial consistency.

[0120] In one specific embodiment provided in this specification, when it is determined, based on the spatial verification conditions, that a viewpoint to be verified does not meet the spatial verification conditions, the scene description data needs to be adjusted in a targeted manner to improve adjustment efficiency and reduce interference with the already verified parts. This specification uses the following method to adjust the scene description data corresponding to the virtual scene.

[0121] In one specific embodiment provided in this specification, adjusting the scene description data corresponding to the virtual scene includes: Obtain spatial description information corresponding to at least one target viewpoint image to be verified, wherein the verification result of the target viewpoint image to be verified does not satisfy the verification condition; Based on the second perspective image, at least one target space description information is adjusted until the target perspective image to be verified corresponding to each adjusted target space description information meets the verification conditions. Based on the adjusted target space description information, the scene description data corresponding to the virtual scene is updated.

[0122] The spatial description information corresponding to the target viewpoint image to be verified can be understood as the specific reasons or error types that cause the viewpoint image to fail to meet the spatial verification conditions, based on the display content of the viewpoint to be verified in the virtual scene.

[0123] In one specific embodiment provided in this specification, based on the second-view image, the spatial description information corresponding to the target view image to be verified is adjusted so that the adjusted spatial description information corresponding to the target view image to be verified satisfies the spatial structured attribute information of the virtual scene determined based on the second-view image. Accordingly, the scene description data corresponding to the virtual scene is updated.

[0124] In one specific embodiment provided in this specification, for ease of description, it may also involve acquiring spatial description information corresponding to at least one target image to be verified from a specific viewpoint, and determining adjustment factors based on the spatial description information. Based on the second perspective image and the adjustment factors, at least one target space description information is adjusted until the target perspective image to be verified corresponding to each adjusted target space description information meets the verification conditions. Based on the adjusted target space description information, the scene description data corresponding to the virtual scene is updated.

[0125] In one specific embodiment provided in this specification, the process of determining adjustment factors includes analyzing the specific manifestations of non-compliance with verification conditions, locating the location and type of spatial errors. For example, adjustment factors may be specific descriptions of spatial errors such as "the door's position is mirrored from the left to right from a reverse view" or "the display shelf is missing from a side view." Adjustment factors may also include an assessment of the severity of the error and a definition of the range of affected spatial elements. In one example, adjustment factors may be organized into a structured error report, including information such as the error type, error location, identification of affected objects, and suggested correction directions. In another example, adjustment factors may also include an analysis of the causes of the error, such as determining whether the spatial error is caused by inference bias in the image generation model or insufficient constraints on the conditional input.

[0126] Optionally, different adjustment strategies can be selected based on the type of adjustment factor. Specifically: For mirror-type errors, the image content of a specific area can be flipped to correct them; For errors involving missing objects, the missing objects can be added through local image completion. For positional offset errors, the positional relationships of objects can be corrected by locally redrawing.

[0127] Based on the above, the spatial verification process can be re-executed after adjustments, forming an iterative loop of "verification-adjustment-re-verification" until all images from the viewpoints to be verified meet the spatial verification conditions. Afterward, the virtual scene is rendered based on the updated scene description data corresponding to the virtual scene.

[0128] According to a specific implementation method provided in this specification, by first determining specific adjustment factors and then adjusting the scene description data accordingly, precise local correction rather than global reconstruction is achieved. By determining the adjustment factors in this way, the verified correct portions of the scene description data are preserved, avoiding new problems that might be introduced by global reconstruction. Furthermore, targeted adjustments for specific error types can be made to efficiently correct spatial relationships and reduce the number of adjustment iterations. The adjustment factors provide clear directional guidance for the adjustment operations, making the adjustment process targeted rather than blindly making random modifications, thus improving the efficiency and accuracy of spatial consistency correction.

[0129] In one specific embodiment provided in this specification, after obtaining the scene description data corresponding to the virtual scene, this specification also stores the virtual scene and the scene description data corresponding to the virtual scene to facilitate subsequent calls. For example, the scene identifier corresponding to the virtual scene is stored, and a correspondence between the scene identifier and the scene description data is established, thereby realizing the storage of the virtual scene. During subsequent calls, the requirement semantics are determined based on the creation requirement information. Based on the stored scene description data corresponding to the virtual scene, the scene semantics corresponding to different virtual scenes are determined, and the scene semantics matching the requirement semantics are determined, thereby directly obtaining the virtual scene corresponding to the creation requirement information. There is no need to perform the virtual scene generation operation again, thereby improving the generation efficiency of the virtual scene.

[0130] In one specific embodiment provided in this specification, after obtaining the creation requirement information, the text description and style parameters in the creation requirement information are input into a first image generation model, which generates a first-view image with better visual expressiveness. The first image generation model can be an image generation model with strong aesthetic and style expression capabilities to ensure that the first-view image meets the expected standards in visual quality. After generating the first-view image, the first-view image and the creation requirement information are input into a second image generation model, which uses the first-view image as a condition to generate a top-down view image of the virtual scene as the second-view image. The second image generation model can be an image generation model with strong spatial structure reasoning and instruction following capabilities to ensure that the top-down view accurately reflects the spatial layout of the scene. Subsequently, semantic segmentation and object detection are performed on the top-down view image to identify elements such as building structures, main scenery, and entrances / exits in the scene, and each element is assigned a number and semantic label to generate spatially structured attribute information with numbered semantic annotations and directional indicators. Finally, using the first-view and second-view images as dual-condition input image generation models, a set of multi-angle views is generated, which maintain the spatial structure but change the camera position and angle. The multi-angle views, together with visual constraint information and spatial constraint information, are combined into scene description data. Based on the scene description data, scene rendering is performed to obtain the final virtual scene.

[0131] Furthermore, during the generation process, a staged engine orchestration approach is adopted, following stages such as main view generation, top-down spatial structure diagram generation, and multi-angle view generation. Based on the performance of each image generation model across different capability dimensions, a relatively superior image generation model is selected for each stage. For example, in the main view generation stage, a model with strong aesthetic and stylistic expression capabilities is selected; in the top-down spatial structure diagram generation stage, a model with strong spatial structure reasoning and instruction compliance capabilities is selected. In addition, the system maintains a model defect profile library, recording known defect feature information of each image generation model. When calling the target model, defect compensation prompts are automatically injected based on its defect profile. These prompts include suppression constraints (such as noise reduction and cleaning, suppressing particles and color spots), original structure preservation constraints (such as maintaining the original lighting and shadow structure), and exclusionary negative constraints (such as excluding specific defect types).

[0132] According to a specific implementation method provided in this specification, by acquiring first-view and second-view images, and based on the technical approach of multi-view collaboration and structured constraints, the problems of lack of spatial reference and limited capabilities of a single model in virtual scene generation are solved. Furthermore, this specification uses requirement information to drive the generation of first-view images, determining the visual tone of the scene. Then, second-view images are generated based on the first-view images, supplementing the spatial layout information of the scene from different observation angles, thereby further clarifying the spatial structure corresponding to different objects in the virtual scene and laying the foundation for accurate virtual scene generation. By parsing the second-view images to obtain spatial structured attribute information, the spatial relationships in the images are transformed into computable structured data. Multi-dimensional information is fused to generate scene description data containing visual and spatial constraints, and the virtual scene is rendered accordingly. In this process, a progressive scene construction chain is formed from visual appearance to spatial structure to multi-dimensional constraints. The output of each step provides conditions and constraints for the next step, ensuring the information continuity and consistency between each link. The virtual scene construction method provided in this manual enables the generated virtual scene to have both visual appearance reference and spatial structure constraints. In subsequent scene rendering from different perspectives, the spatial relationship of the scene has explicit structured data as a constraint basis, avoiding problems such as mirror misalignment and object offset caused by relying solely on the model to infer spatial relationships. This effectively improves the accuracy of virtual scene generation and cross-view consistency.

[0133] The following is in conjunction with the appendix Figure 3 Taking the application of the virtual scene construction method provided in this specification in the virtual scene asset construction scenario as an example, the virtual scene construction method will be further explained. Among them, Figure 3 The flowchart of a virtual scene construction method provided in one embodiment of this specification is shown, which specifically includes the following steps.

[0134] Step 302: Obtain the virtual scene creation requirement information input by the user.

[0135] The creation requirements information includes parameters such as scene type, style description, and space scale.

[0136] For example, a user might enter: "An ancient Chinese tea room with a mahogany square table in the center, surrounded by four armchairs, a display shelf against the north wall, a door in the southeast corner, and a window on the west wall offering a view of a bamboo grove courtyard, bathed in warm-toned natural light."

[0137] Step 304: Determine the generation requirements for the main view based on the creation requirements information, and generate a first-person view image according to the generation requirements for the main view.

[0138] The generation requirements for the main view include core capability dimensions, namely aesthetic and style expression capabilities. Based on this, a first-person image generation model with a high score in classical style aesthetics is selected from the image generation model set as the target model for first-person perspective image generation. Simultaneously, a model defect profile library is queried to obtain the target defect feature information of the first-person image generation model, and an image generation instruction containing defect compensation prompts is constructed.

[0139] The generation instruction with defect compensation prompts is input into the first image generation model to generate a three-quarter view with better visual expression as the first-view image, which shows the overall style and spatial atmosphere of the tea room.

[0140] Step 306: Evaluate the image complexity of the main view to determine whether the repair trigger conditions are met.

[0141] If the complexity is low and the quality score meets the requirements, skip the repair phase and proceed directly to step 308.

[0142] If the complexity is high, the repair model will perform denoising and structural repair processing in step 310.

[0143] Step 308: Using the repaired main view as a condition and combining it with the creation requirement information, determine the generation requirement information for the top view. Then, based on the top view generation requirement information, generate a second-view image.

[0144] The core capability dimensions included in the requirements for generating the top-view image are spatial structure reasoning and instruction compliance. Based on this, a second image generation model with a higher score in spatial reasoning was selected from the image generation model set as the target model for second-view image generation.

[0145] Query the defect profile of the second image generation model, construct defect compensation prompts, and input the main view, creation requirements information and defect compensation prompts into the second image generation model to generate a top-down spatial structure map as a second-view image, showing the spatial distribution relationship of various objects in the tea room.

[0146] Step 310: Denoising and structural repair are performed using the repair model.

[0147] Step 312: Perform semantic parsing on the top-view spatial structure diagram to obtain spatial structured attribute information.

[0148] For example, identify elements such as building structures, main objects, furniture, entrances and exits in a scene, assign numbers and semantic labels to each element, mark location information, determine the main movement route of the character from the doorway through the passage to the square table, and generate spatial structured attribute information with numbered semantic labels and location markers.

[0149] Step 314: Determine visual constraint information based on the main view, determine spatial constraint information based on the top-view spatial structure diagram and spatial structured attribute information, and generate a nine-grid multi-angle view group accordingly.

[0150] Visual constraint information can include, for example, color tone, lighting direction, and material texture.

[0151] Spatial constraint information can include, for example, object position, relative orientation, connectivity, and movement paths.

[0152] For example, using the front view and the top view spatial structure diagram as two conditions, the preset nine-grid perspective adjustment parameters (including nine different camera positions and angle configurations) are obtained, and nine third-view images are generated that keep the spatial structure unchanged and only change the camera positions and angles, forming a nine-grid multi-angle view group.

[0153] Step 316: Combine the nine-grid multi-angle view group, visual constraint information, and spatial constraint information into scene description data.

[0154] Step 318: Verify the scene description data according to the spatial verification conditions.

[0155] For example, based on scene description data and viewpoint adjustment parameters, three viewpoint images to be verified (reverse view, side view, and interior view) are generated. The spatial relationships in each viewpoint image are then verified one by one based on spatial verification conditions. The checks include seven types of spatial errors: left-right mirroring, object misalignment, missing or redundant objects, structural contradictions, incorrect movement paths or entrances / exits, scale distortion, and infeasible camera positions.

[0156] If all images from the viewpoints to be verified meet the spatial verification conditions, then the scene description data corresponding to the virtual scene is stored as a qualified scene asset.

[0157] If there are images of the viewpoint that do not meet the spatial verification conditions, the adjustment factors that do not meet the conditions are identified, the scene description data is adjusted accordingly based on the adjustment factors, and the verification process is repeated until all of them pass.

[0158] Step 320: Render the final virtual scene based on the scene description data.

[0159] The virtual scene has complete visual appearance references, spatial structure constraints, and multi-angle consistency constraints. It can be used as a unified scene asset in the subsequent short drama storyboard production for cross-camera reference, ensuring cross-camera scene consistency.

[0160] In one specific embodiment provided in this specification, the process of rendering a virtual scene may be as follows: based on scene description data, determine the video storyboard frames corresponding to different perspectives in the virtual scene, and then construct the virtual scene.

[0161] The aforementioned virtual scene construction method achieves a fully automated scene construction process, from inputting creation requirements to outputting structured scene assets. The entire process integrates multiple technical mechanisms, including phased model orchestration, automatic injection of defect compensation prompts, multi-dimensional constraint fusion, and automatic spatial relationship verification. The generated multi-dimensional scene board (main view + top-down spatial structure diagram + nine-grid multi-angle view) exhibits good quality in terms of visual expressiveness, spatial structural correctness, and cross-perspective consistency, effectively supporting the scene consistency requirements of subsequent video storyboarding and video generation stages.

[0162] Corresponding to the above method embodiments, this specification also provides a method for constructing game scenes.

[0163] Figure 4 A flowchart illustrating a method for constructing a virtual game scene according to an embodiment of this specification is shown.

[0164] Step 402: Obtain the creation requirement text of the game virtual scene, and generate a first-person view image of the game virtual scene based on the creation requirement text.

[0165] Step 404: Based on the main view image and the creation requirement text, generate a top-down view image for the game virtual scene.

[0166] Step 406: Analyze the top-down view image to determine at least one object to be constructed in the game virtual scene with semantic annotations and movement structure.

[0167] Step 408: Based on the main view image, the top view image, and each object to be constructed in the game virtual scene with semantic annotations and movement structure, generate scene description data corresponding to the game virtual scene, and render the game virtual scene based on the scene description data.

[0168] It should be clarified that the game virtual scene construction method provided in this specification applies to the aforementioned virtual scene construction method. Its implementation method is as described above, and will not be repeated here.

[0169] Corresponding to the above method embodiments, this specification also provides embodiments of a virtual scene construction apparatus. Figure 5 A schematic diagram of a virtual scene construction apparatus according to one embodiment of this specification is shown. Figure 5 As shown, the device includes: The acquisition unit 502 is configured to acquire the creation requirement information of the virtual scene and generate a first-view image of the virtual scene based on the creation requirement information.

[0170] The generation unit 504 is configured to generate a second-view image for the virtual scene based on the first-view image and the creation requirement information.

[0171] The parsing unit 506 is configured to parse the second perspective image and determine the spatial structured attribute information of the virtual scene.

[0172] The rendering unit 508 is configured to generate scene description data corresponding to the virtual scene based on the first view image, the second view image and the spatial structured attribute information, and render the virtual scene based on the scene description data.

[0173] Furthermore, rendering unit 508 is also configured as follows: Based on the scene description data and the viewpoint adjustment parameters, at least one viewpoint image to be verified is determined in the virtual scene, and the spatial description information corresponding to each viewpoint image to be verified is determined. Verify the spatial description information corresponding to each viewpoint image to be verified, and obtain the verification results corresponding to each viewpoint image to be verified. If the verification results for each image from the viewpoint to be verified all meet the verification conditions, store the scene description data corresponding to the virtual scene. If the verification result corresponding to the image to be verified does not meet the verification conditions, the scene description data corresponding to the virtual scene is adjusted.

[0174] Furthermore, rendering unit 508 is further configured as follows: In each image to be verified, a first image to be verified and a second image to be verified, including the target object to be verified, are determined, and the first spatial description information corresponding to the first image to be verified and the second spatial description information corresponding to the second image to be verified are obtained, wherein the target object to be verified is any object in the virtual scene. Parse the first spatial description information and the second spatial description information to obtain the first spatial position information of the target object to be verified in the first view image to be verified and the second spatial position information of the target object to be verified in the second view image to be verified. By comparing the first spatial location information and the second spatial location information of the target object to be verified, the verification results corresponding to the first and second viewpoint images to be verified are obtained.

[0175] Furthermore, rendering unit 508 is further configured as follows: Obtain spatial description information corresponding to at least one target viewpoint image to be verified, wherein the verification result of the target viewpoint image to be verified does not satisfy the verification condition; Based on the second perspective image, at least one target space description information is adjusted until the target perspective image to be verified corresponding to each adjusted target space description information meets the verification conditions. Based on the adjusted target space description information, the scene description data corresponding to the virtual scene is updated.

[0176] Furthermore, the generating unit 504 is further configured as follows: Based on the first-view image, the second-view image, and the spatial structured attribute information, scene description data corresponding to the virtual scene is generated, including: Based on the first perspective image, determine the visual constraint information of the virtual scene; Based on the second-view image and at least one object with semantic annotation and movement structure, determine the spatial constraint information of each object in the virtual scene; Based on the visual constraint information and the spatial constraint information, scene description data corresponding to the virtual scene is generated.

[0177] Furthermore, the generating unit 504 is further configured as follows: Based on the visual constraint information and the spatial constraint information, an initial virtual scene is generated; Obtain view adjustment parameters; Based on the perspective adjustment parameters, at least one third-view image corresponding to the initial virtual scene is determined; Based on each third-person perspective image, scene description data corresponding to the virtual scene is generated.

[0178] Furthermore, the creation requirement information includes viewpoint adjustment parameters; Acquisition unit 502 is also configured as follows: Based on the aforementioned viewpoint adjustment parameters, the generation requirement information corresponding to the reference viewpoint image is determined in the virtual scene; In the set of image generation models, the target image generation model corresponding to the generation requirement information is determined, wherein the set of image generation models includes at least one image generation model, and different image generation models have different image generation capabilities; A reference view image in the virtual scene is generated based on the target image generation model.

[0179] Furthermore, the acquisition unit 502 is also configured as follows: Obtain a model defect profile library, which includes defect feature information corresponding to each image generation model; Accordingly, generating a reference viewpoint image in the virtual scene based on the target image generation model includes: Based on the model defect profile library, the target defect feature information corresponding to the target image generation model is determined; Based on the generation requirement information and the target defect feature information, an image generation prompt word is constructed. The image generation prompt word is input into the target image generation model to obtain a reference view image generated by the target image generation model.

[0180] Furthermore, the acquisition unit 502 is also configured as follows: Obtain the image complexity corresponding to the reference viewpoint image; If the image complexity meets the repair triggering condition, the reference viewpoint image is repaired.

[0181] The above is an illustrative scheme of a virtual scene construction device according to this embodiment. It should be noted that the technical solution of this virtual scene construction device and the technical solution of the virtual scene construction method described above belong to the same concept. For details not described in detail in the technical solution of the virtual scene construction device, please refer to the description of the technical solution of the virtual scene construction method described above.

[0182] Figure 6 A structural block diagram of a computing device provided according to one embodiment of this specification is shown.

[0183] The computing device 600 includes: Memory 610 and processor 620; The memory 610 is used to store computer programs / instructions, and the processor 620 is used to execute the computer programs / instructions, which, when executed by the processor 620, implement the steps of the virtual scene construction method.

[0184] In one or more embodiments of this specification, the computing device can be understood as an integrated smart terminal, including but not limited to a server, desktop computer, PC (Personal Computer), all-in-one model machine, mobile phone, tablet computer or other portable smart terminal, etc., and the computing device may have the model described in the above embodiments of this application pre-installed.

[0185] Specifically, this computing device can pre-install various types of models, including but not limited to models in natural language processing, visual processing, speech processing, code processing, and multimodal task processing, thus providing diverse model selection. In different product forms, this computing device can support one or more model usage methods, including but not limited to model training, model invocation, model fine-tuning, model deployment, model inference, and application. In some product forms, this computing device also supports model management, including but not limited to multi-type model management (supporting the management of discriminative, generative, and other model types), model version control (supporting the control of different model versions), and model evaluation (evaluating model performance and effectiveness based on model evaluation tools). In other product forms, this computing device can also create applications based on models, providing API (Application Programming Interface) calling capabilities. Users can call models into created applications through the API interface, and application management tools are also provided to manage and monitor the applications.

[0186] Furthermore, the computing device can also include data management (supporting the creation and management of model tuning datasets), a training center (providing abundant training resources to help users learn and master AI (Artificial Intelligence) technology), and basic control capabilities (providing enterprise-level basic control capabilities to ensure the security and efficient operation of the system). Through the above functions, it provides a comprehensive and integrated device for AI development, training, deployment, and application.

[0187] Figure 7 A structural block diagram of an electronic device provided according to one embodiment of this specification is shown.

[0188] A memory 710 and a processor 720 are connected via a bus 730; The memory 710 is used to store computer programs / instructions, and the processor 720 is used to execute the computer programs / instructions, which, when executed by the processor 720, implement the steps of the method.

[0189] Specifically, the components of the electronic device 700 include, but are not limited to, a memory 710 and a processor 720. The processor 720 is connected to the memory 710 via a bus 730, and the database 750 is used to store data.

[0190] Electronic device 700 also includes access device 740, which enables electronic device 700 to communicate via one or more networks 760. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. Access device 740 may include one or more of any type of wired or wireless network interface (e.g., network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, Wi-MAX (Worldwide Interoperability for Microwave Access) interface, Ethernet interface, Universal Serial Bus (USB) interface, cellular network interface, Bluetooth interface, Near Field Communication (NFC) interface, and so on.

[0191] In one embodiment of this specification, the above-described components of the electronic device 700 and Figure 7 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 7 The block diagram of the electronic device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0192] Electronic device 700 can be any type of stationary or mobile electronic device, including mobile computers or mobile electronic devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable electronic devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary electronic devices such as desktop computers or personal computers (PCs). Electronic device 700 can also be a mobile or stationary server.

[0193] The above is an illustrative scheme of an electronic device according to this embodiment. It should be noted that the technical solution of this electronic device and the technical solution of the virtual scene construction method described above belong to the same concept. For details not described in detail in the technical solution of the electronic device, please refer to the description of the technical solution of the virtual scene construction method described above.

[0194] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the virtual scene construction method described above.

[0195] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the virtual scene construction method described above belong to the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the virtual scene construction method described above.

[0196] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the virtual scene construction method described above.

[0197] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the virtual scene construction method described above belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the virtual scene construction method described above.

[0198] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0199] The computer program / instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0200] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0201] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0202] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A method for constructing a virtual scene, comprising: Obtain the creation requirement information of the virtual scene, and generate a first-view image of the virtual scene based on the creation requirement information; Based on the first perspective image and the creation requirement information, a second perspective image for the virtual scene is generated; Analyze the second-view image to determine the spatial structured attribute information of the virtual scene; Based on the first perspective image, the second perspective image, and the spatial structured attribute information, scene description data corresponding to the virtual scene is generated, and the virtual scene is rendered based on the scene description data.

2. The method as described in claim 1, wherein after rendering the virtual scene, the method further includes: Based on the scene description data and the viewpoint adjustment parameters, at least one viewpoint image to be verified is determined in the virtual scene, and the spatial description information corresponding to each viewpoint image to be verified is determined. Verify the spatial description information corresponding to each viewpoint image to be verified, and obtain the verification results corresponding to each viewpoint image to be verified. If the verification results for each image from the viewpoint to be verified all meet the verification conditions, store the scene description data corresponding to the virtual scene. If the verification result corresponding to the image to be verified does not meet the verification conditions, the scene description data corresponding to the virtual scene is adjusted.

3. The method as described in claim 2, verifying the spatial description information corresponding to each viewpoint image to be verified, and obtaining the verification result corresponding to each viewpoint image to be verified, includes: In each image to be verified, a first image to be verified and a second image to be verified, including the target object to be verified, are determined, and the first spatial description information corresponding to the first image to be verified and the second spatial description information corresponding to the second image to be verified are obtained, wherein the target object to be verified is any object in the virtual scene. Parse the first spatial description information and the second spatial description information to obtain the first spatial position information of the target object to be verified in the first view image to be verified and the second spatial position information of the target object to be verified in the second view image to be verified. By comparing the first spatial location information and the second spatial location information of the target object to be verified, the verification results corresponding to the first and second viewpoint images to be verified are obtained.

4. The method as described in claim 2, wherein adjusting the scene description data corresponding to the virtual scene includes: Obtain spatial description information corresponding to at least one target viewpoint image to be verified, wherein the verification result of the target viewpoint image to be verified does not satisfy the verification condition; Based on the second perspective image, at least one target space description information is adjusted until the target perspective image to be verified corresponding to each adjusted target space description information meets the verification conditions. Based on the adjusted target space description information, the scene description data corresponding to the virtual scene is updated.

5. The method as described in claim 1, wherein the spatial structured attribute information includes at least one object with semantic annotation and a movement structure; Based on the first-view image, the second-view image, and the spatial structured attribute information, scene description data corresponding to the virtual scene is generated, including: Based on the first perspective image, determine the visual constraint information of the virtual scene; Based on the second-view image and at least one object with semantic annotation and movement structure, determine the spatial constraint information of each object in the virtual scene; Based on the visual constraint information and the spatial constraint information, scene description data corresponding to the virtual scene is generated.

6. The method as described in claim 5, wherein generating scene description data corresponding to the virtual scene based on the visual constraint information and the spatial constraint information includes: Based on the visual constraint information and the spatial constraint information, an initial virtual scene is generated; Obtain view adjustment parameters; Based on the perspective adjustment parameters, at least one third-view image corresponding to the initial virtual scene is determined; Based on each third-person perspective image, scene description data corresponding to the virtual scene is generated.

7. The method according to any one of claims 1 to 6, wherein the creation requirement information includes viewpoint adjustment parameters; The method further includes: Based on the aforementioned viewpoint adjustment parameters, the generation requirement information corresponding to the reference viewpoint image is determined in the virtual scene; In the set of image generation models, the target image generation model corresponding to the generation requirement information is determined, wherein the set of image generation models includes at least one image generation model, and different image generation models have different image generation capabilities; A reference view image in the virtual scene is generated based on the target image generation model.

8. The method of claim 7, further comprising: Obtain a model defect profile library, which includes defect feature information corresponding to each image generation model; Accordingly, generating a reference viewpoint image in the virtual scene based on the target image generation model includes: Based on the model defect profile library, the target defect feature information corresponding to the target image generation model is determined; Based on the generation requirement information and the target defect feature information, an image generation prompt word is constructed, and the image generation prompt word is input into the target image generation model to obtain a reference view image generated by the target image generation model.

9. The method of claim 8, further comprising: Obtain the image complexity corresponding to the reference viewpoint image; If the image complexity meets the repair triggering condition, the reference viewpoint image is repaired.

10. A method for constructing a game scene, comprising: Obtain the creation requirement text of the game virtual scene, and generate a first-person view image of the game virtual scene based on the creation requirement text; Based on the main view image and the creation requirement text, a top-down view image for the game virtual scene is generated; Analyze the top-down view image to identify at least one object to be constructed in the game virtual scene, which has semantic annotations and a movement structure. Based on the main view image, the top view image, and each object to be constructed in the game virtual scene with semantic annotations and movement structure, scene description data corresponding to the game virtual scene is generated, and the game virtual scene is rendered based on the scene description data.

11. A virtual scene construction device, comprising: The acquisition unit is configured to acquire the creation requirement information of the virtual scene and generate a first-view image of the virtual scene based on the creation requirement information; The generation unit is configured to generate a second-view image for the virtual scene based on the first-view image and the creation requirement information. The parsing unit is configured to parse the second-view image and determine the spatial structured attribute information of the virtual scene; The rendering unit is configured to generate scene description data corresponding to the virtual scene based on the first view image, the second view image, and the spatial structured attribute information, and to render the virtual scene based on the scene description data.

12. A computing device, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 10.

13. An electronic device, comprising: A memory and a processor, the memory and the processor being connected via a bus; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 10.

14. A computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 10.

15. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 10.

Images