A space intelligent geometry topology and texture collaborative generation method and system

Abstract

The application relates to a space intelligent geometry topology and texture collaborative generation method and system, and relates to the technical field of space intelligence and computer graphics, which comprises the following steps: dividing a three-dimensional surface discretization representation into local surface blocks; constructing a multi-modal feature for each local surface block; predicting a geometry structure generation parameter of a current local surface block based on a geometry generation flow; generating a physical constraint modulation signal for indicating a weight of a texture fracture tendency between the current local surface block and an adjacent local surface block under the condition of increasing or decreasing; under the constraint of the physical constraint modulation signal, predicting a texture structure generation parameter of the current local surface block; determining a texture fusion mapping parameter of the current local surface block; and determining three-dimensional generation data based on the geometry structure generation parameter and the texture fusion mapping parameter of each local surface block. Therefore, the geometry feature at the shared boundary directly influences the weight of the texture fracture tendency, and it is ensured that the joint appears at the boundary with reasonable geometry features.

Description

Technical Field

[0001] This application relates to the fields of spatial intelligence and computer graphics technology, and in particular to a method and system for co-generating spatial intelligent geometric topology and texture. Background Technology

[0002] With the rapid development of spatial intelligence, embodied AI, and technologies such as virtual reality, augmented reality, and digital twins, the fundamental role of 3D environmental assets in intelligent systems is becoming increasingly prominent. How to efficiently and automatically generate 3D environmental assets that simultaneously possess geometric accuracy and visual consistency has become an important research direction in the field of spatial intelligence.

[0003] Currently, most 3D generation methods primarily focus on modeling and generating geometric topology, with texture mapping treated as a separate post-processing step. Generally, after the geometric model is generated, additional texture unwrapping (UVUnwrapping) algorithms or manual tools are used to parameterize the model surface. Existing automated UV algorithms are mostly based on global optimization or heuristic rules, solving for the UV coordinates of the entire model in one go by defining a global objective function (such as minimizing overall texture stretching or maximizing texture space utilization). These global optimization algorithms have an inherent flaw—a tendency towards averaging the global optimization objective. Because the algorithm pursues overall optimality rather than local rationality, large smooth areas dominate the objective function. The algorithm prioritizes the texture quality of large smooth surfaces, while the alignment accuracy of small, high-curvature areas (such as edges and transition surfaces) accounts for a very small proportion of the global objective. The alignment accuracy at edges is easily sacrificed, causing texture seams to deviate from geometric feature lines. Overall, UV algorithms struggle to adaptively adjust to local geometric characteristics, and texture region division often does not conform to physical structure or spatial semantic features.

[0004] Therefore, how to improve the problem of geometric and texture separation in the 3D generation process is a key research topic for those skilled in the art. Summary of the Invention

[0005] In a first aspect, this application provides a spatial intelligent geometric topology and texture co-generation method, the method comprising: dividing a three-dimensional surface discretization representation into local surface blocks with topological connections to obtain a set of discrete local surface blocks; constructing multimodal features for each local surface block, the multimodal features including a geometric information statistical summary, texture mapping prior conditions, and spatial relationship information with neighboring blocks; predicting geometric structure generation parameters of the current local surface block based on a geometric generation flow; and generating a physical constraint modulation signal based on the geometric structure generation parameters of the current local surface block, the physical constraint modulation signal being used to indicate increasing or decreasing the relationship between the current local surface block and neighboring local surface blocks. The text describes a process involving several steps: First, it assigns weights to the texture breakage tendency between local surface patches. Second, it defines texture structure generation parameters for the current local surface patch based on the texture generation flow, including texture mapping coordinates and texture breakage tendency weights for shared boundaries with adjacent patches. Third, it determines texture fusion mapping parameters for the current local surface patch based on its geometric and texture structure generation parameters, including aligned texture mapping coordinates, seam decisions, and texture island affiliation identifiers. Finally, it determines 3D generated data based on the geometric and texture fusion mapping parameters of each local surface patch in the discrete local surface patch set.

[0006] The spatial intelligent geometric topology and texture co-generation method provided in this application, on the one hand, determines a physical constraint modulation signal that actively increases or decreases the texture breakage tendency weight at shared boundaries based on the geometric structure generation parameters, and constrains the prediction process of texture structure generation parameters based on this physical constraint modulation signal. In this way, the geometric features at the shared boundary (such as curvature, normal, and geometric feature lines) directly affect the texture breakage tendency weight, thereby ensuring that texture seams appear at reasonable geometric boundaries during the block-by-block generation process, overcoming the defect of existing global UV optimization algorithms where the alignment accuracy of seams at edges is sacrificed due to averaging tendency.

[0007] On the other hand, unlike the separate processing method of "first generating complete geometry and then independently performing UV unwrapping" in the prior art, this application uses local surface blocks as the smallest generation unit. After the geometry generation stream outputs the geometry generation parameters, it immediately constrains the texture generation stream through physical constraint modulation signals. This ensures that the prediction process of texture mapping is always guided by the geometry state. In other words, it adopts a collaborative mechanism of "generating geometry and making decisions on texture at the same time". This allows the placement decision of texture seams to be driven by the geometry structure, rather than independent optimization in the posterior, thus avoiding semantic conflicts between geometry and texture from the generation source.

[0008] Secondly, this application also provides a spatial intelligent geometric topology and texture collaborative generation system, the system comprising: a discretized 3D surface acquisition and local surface block construction module, used to divide the discretized representation of the 3D surface into local surface blocks with topological connections, obtaining a set of discrete local surface blocks; a multimodal encoding module, used to construct multimodal features for each local surface block, the multimodal features including geometric information statistical summary, texture mapping prior conditions, and spatial relationship information with adjacent blocks; a geometric topology generation flow module, used to predict the geometric structure generation parameters of the current local surface block based on the geometric generation flow; and a physical constraint module, used to generate a physical constraint modulation signal based on the geometric structure generation parameters of the current local surface block, the physical constraint modulation signal being used to indicate increasing or decreasing the current local surface block. The system includes: a texture breakage tendency weight between a block and adjacent local surface blocks; a texture mapping generation stream module, used to predict the texture structure generation parameters of the current local surface block based on the texture generation stream under the constraint of the physical constraint modulation signal, the texture structure generation parameters including texture mapping coordinates and texture breakage tendency weights shared with adjacent blocks; a texture fusion module, used to determine the texture fusion mapping parameters of the current local surface block based on the geometric structure generation parameters and texture structure generation parameters, the texture fusion mapping parameters including aligned texture mapping coordinates, seam decisions, and texture island attribution identifiers; and a 3D asset determination module, used to determine the 3D generation data system based on the geometric structure generation parameters and texture fusion mapping parameters of each local surface block in the discrete local surface block set.

[0009] Thirdly, this application also provides a spatial intelligent geometric topology and texture co-generation apparatus, including a unit for performing any one of the spatial intelligent geometric topology and texture co-generation methods in the first aspect.

[0010] Fourthly, this application also provides a computer storage medium that can store multiple instructions, which are adapted to be loaded and executed by a processor using any one of the spatial intelligent geometric topology and texture co-generation methods in the first aspect.

[0011] Fifthly, embodiments of this application also provide a computer program product containing instructions, which, when run on an electronic device, causes the electronic device to execute any one of the spatial intelligent geometric topology and texture co-generation methods in the first aspect.

[0012] In a sixth aspect, embodiments of this application also provide a chip module, including a transceiver component and a chip, wherein the chip is used to execute any one of the spatial intelligent geometric topology and texture co-generation methods in the first aspect.

[0013] In a seventh aspect, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the method described in any one of the first aspects is performed when the processor executes the program.

[0014] It is understood that the spatial intelligent geometric topology and texture co-generation device, computer storage medium, computer program, computer program product, chip system, and electronic device provided above are all used to execute the method shown in any implementation of the first aspect of the embodiments of this application. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects in the corresponding methods, and will not be repeated here. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method flow for a spatial intelligent geometric topology and texture co-generation method provided in an embodiment of this application; Figure 2 This is a schematic flowchart of a method for merging a local surface block with an adjacent block according to an embodiment of this application; Figure 3 This is a schematic flowchart of another method for merging a local surface block with an adjacent block provided in an embodiment of this application; Figure 4 This is a schematic diagram of the architecture of a spatial intelligent geometric topology and texture co-generation system provided in an embodiment of this application; Figure 5 This is a schematic diagram of the architecture of another spatial intelligent geometric topology and texture co-generation system provided in the embodiments of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described below in conjunction with the accompanying drawings.

[0017] It should be noted that the embodiments of this application use an electronic device as an example to illustrate the execution subject of the spatial intelligent geometric topology and texture co-generation method provided in this application. This electronic device can also be understood as the spatial intelligent geometric topology and texture co-generation device shown in the embodiments of this application. In the embodiments of this application, the electronic device can be a microprocessor or computer for executing program code, etc. Any electronic device that can be used to execute the method provided in the embodiments of this application falls within the protection scope of the embodiments of this application, and this application does not impose any limitations. For example, the electronic device can be a desktop computer, a portable laptop, a mobile terminal, a 32-bit microprocessor, or a 64-bit microprocessor, etc., and the embodiments of this application do not limit this.

[0018] Example 1:

[0019] Please see Figure 1 , Figure 1 This document presents a flowchart illustrating a method for co-generating spatial intelligent geometric topology and texture, as provided in an embodiment of this application. Figure 1 As shown, the spatial intelligent geometry topology and texture co-generation method includes the following steps: Phase 1 (S101-S102): Preprocessing S101, the electronic device discretizes the three-dimensional surface representation into local surface blocks with topological connections, and obtains a set of discrete local surface blocks.

[0020] In this embodiment, the three-dimensional surface discretization representation refers to the discretized data form of the three-dimensional surface, which can be a complete surface model or incrementally acquired partial surface data. When new data arrives, the discretization representation is dynamically expanded and new local surface blocks are constructed.

[0021] In some possible implementations, the electronic device can receive three-dimensional surface input, which includes, but is not limited to: a three-dimensional mesh model, point cloud reconstruction results, implicit surface function discretization sampling results, and initial geometric descriptions from procedurally generated or simulated environments. The electronic device first performs discretization processing on the continuous three-dimensional surface, and then divides the surface into multiple local surface blocks based on a preset or adaptive strategy.

[0022] In the embodiments of this application, each local surface block includes, but is not limited to: a set of vertices or sampling points within the block, and a description of the boundary relationship with adjacent surface blocks. In some possible implementations, it may also include one or more of the following: local topological connectivity (connection relationships between vertices within the current block), local normal distribution, and curvature information (statistical distribution of normal directions and curvature values ​​of each vertex within the current block).

[0023] The construction of local surface blocks can be based on rule partitioning, octree segmentation, curvature-driven segmentation, or learning-based adaptive segmentation. However, this invention does not limit the specific segmentation algorithm, but emphasizes that local surface blocks are the unified semantic units generated subsequently.

[0024] As an example, the electronic device divides the surface into blocks based on a curvature-driven segmentation method, specifically including: detecting geometric feature regions based on local curvature changes, the geometric feature regions including at least one of the following: angles (curvature discontinuous, infinite, and abrupt changes in normal direction, such as an angle greater than 30°), facets (curvature changes within a certain range, and normal changes within a certain range, such as 15°~30°), and transition zones (curvature changes continuously, and the angle of the normal changes gradually); using points in non-geometric feature regions as seed points, performing region growth in regions with gentle curvature changes, and automatically terminating growth when encountering geometric feature regions, preset curvature abrupt change regions, or preset normal abrupt change regions at the growth boundary, and taking the continuous region formed by growth as a single local surface block, with the boundary of the local surface block aligned with the boundary of the geometric feature region.

[0025] In the embodiments of this application, the block-level topological relationship description of the local surface block includes a set of adjacent block indices. In some possible implementations, it may also include, but is not limited to, at least one of the following: shared boundary length, shared boundary curve parameters, prior normal angles between adjacent blocks, and prior geometric transition types (smooth / turning).

[0026] In some other possible implementations, if the preset boundary feature line appears inside a local surface block, the electronic device can also trigger the re-division of the local block, transforming the preset boundary feature line into a new inter-block boundary.

[0027] S102, the electronic device constructs multimodal features for each local surface block.

[0028] The multimodal features include a statistical summary of geometric information, prior conditions for texture mapping, and spatial relationship information with neighboring blocks. In this embodiment, the electronic device constructs multimodal features for each local surface block.

[0029] In some other possible implementations, the aforementioned multimodal features may also include intra-block vertex index information and face index information, used to describe the intra-block topology.

[0030] In this embodiment, the aforementioned geometric information statistical summary includes at least one of vertex relative position index, normal distribution, and curvature statistics, used to characterize the statistical distribution features of the local surface block geometry. The normal distribution may include, but is not limited to, one or more of the following: mean normal, variance normal, and histogram distribution of normal directions at vertices or sampling points within the block. The curvature statistics may include, but is not limited to, one or more of the following: mean curvature, variance curvature, curvature extrema, and curvature distribution graph at vertices or sampling points within the block.

[0031] The aforementioned texture mapping prior conditions include at least one of the following: initial mapping reference, texture direction constraint, and texture continuity flag, used to provide prior constraints for texture generation. The texture mapping prior conditions serve as input constraints for texture prediction. The initial mapping reference is a coarse UV coordinate (provided if there is a prior, otherwise empty or a default value). As the initial input or prior condition of the texture generation stream, it can help the network converge faster or generate more reasonable UV coordinates in subsequent texture prediction. Without a prior, the network can predict from scratch. Texture direction constraints are generally U-axis direction vectors, V-axis direction vectors, or other preferred directions of the texture coordinate axes. This paper does not limit these. As constraints, they can guide the U and V directions of the UV coordinates output by the texture generation stream to align with the preset directions in subsequent texture prediction (S105), avoiding texture distortion. They can also maintain the consistency of texture direction between the current block and adjacent blocks (e.g., the U-axis direction should be approximately the same) during subsequent UV coordinate alignment (S106), avoiding distortion after alignment. The texture continuity flag is information specified by the user as indicating whether a certain area is continuous or broken, to indicate an increase in the texture breakage tendency weight (or it can also be used as a mandatory indication, in which seam decisions are made based on this mandatory indication regardless of the value of the breakage tendency weight in subsequent seam generation steps).

[0032] The aforementioned spatial relationship information between adjacent blocks may include, but is not limited to, one or more of the following: normal angle, boundary sharing ratio, and historical generation status. The historical generation status includes at least one of the following: the determined geometric structure generation parameters of adjacent blocks, texture structure generation parameters, texture island fusion decision results, and generation order identifier.

[0033] In this embodiment, unlike the prior art which only encodes geometric information for local blocks, this invention constructs a geometric-texture co-encoding feature vector for each local surface block. For example, its form can be represented as: Vpatch={Gfeat,Tfeat,Cfeat}, where G_feat is a statistical summary of geometric information, T_feat (texture mapping prior conditions), and C_feat (spatial relationship information with neighboring blocks, which can also be understood as contextual features). Compared to the prior art, this multimodal feature encoding process ensures that each local surface block carries the joint semantics of "geometric shape at this location," "how the texture at this location should be mapped," and "contextual information of neighboring blocks" before entering the generation network, providing prior conditions for subsequent co-generation.

[0034] Phase 2 (S103-S106): Block-by-block collaborative generation (executed iteratively for each local surface block) S103, the electronic device predicts the geometric structure generation parameters of the current local surface block based on the geometry generation flow.

[0035] For example, the above geometry generation flow is implemented based on a deep generation model, which is used to predict the geometric state of local surface blocks block by block. The deep generation model includes any one of Transformer, diffusion model, generative adversarial network, variational autoencoder or autoregressive model. The present invention does not limit the specific type of deep generation model.

[0036] In the embodiments of this application, the geometric structure generation parameters include, but are not limited to, at least one of the following: vertex position offset adjustment amount, intra-block vertex normal vector, curvature level identifier (low curvature, medium curvature, high curvature), and geometric feature line identifier, used to represent the specific geometric shape of the local surface block. In some other possible implementations, it may also include generating a refined structure of intra-block topological connections (to refine the connection relationship and merging strategy of intra-block patches).

[0037] Specifically, the geometric information statistical summary of the current local surface block extracted in step S102 is used as the input feature of the geometry generation flow. The vertex relative position index in the geometric information statistical summary is used to help the generation model understand the spatial layout of vertices within the block and predict the vertex position adjustment amount. The normal distribution is used to help the model understand the original geometry of the current block and predict reasonable vertex position adjustment amount. It can also be used to help predict the vertex normal vector in the geometry generation parameters. The curvature statistics are used to help predict reasonable vertex position adjustment amount. It can also be used to help predict the curvature level identifier in the geometry generation parameters.

[0038] For example, the curvature level identifier mentioned above, serving as a block-level geometric complexity label for a local surface block, can be directly determined based on the curvature statistics recorded in the geometric information statistical summary; alternatively, it can be determined by inferring the updated vertex positions based on the predicted vertex position offset adjustment, inferring the actual curvature statistics based on the updated vertex positions, and determining the curvature level identifier based on the inferred actual curvature statistics. This curvature level identifier can guide the in-block texture generation strategy in subsequent steps; a higher curvature level and higher complexity allow for more texture stretching within the block, while smooth areas are not permitted.

[0039] The vertex normal vectors within the block can be directly predicted based on the normal statistical features recorded in the geometric information statistical summary (the model learns the mapping from normal statistical features to specific normal vectors based on the original data); or a coarse prediction can be made based on the normal statistical features recorded in the geometric information statistical summary, and then the vertex positions can be updated based on the predicted vertex position adjustment amount, and fine-tuned using derived calculations based on the updated vertex positions to ensure self-consistency.

[0040] In the embodiments of this application, geometric feature lines are determined based on curvature differences and / or the included angle of the normals. It should be noted that a block boundary refers to the shared boundary between adjacent local surface blocks, determined by the block division strategy of the local surface blocks. It may be an artificial cutting line inside a smooth surface, or it may coincide with a geometric feature line. Geometric feature lines, on the other hand, refer to edges or turning points on a three-dimensional surface where curvature changes abruptly or the normal direction changes drastically; they are geometric properties of the object itself.

[0041] For example, geometric feature lines include one or more of the following: object edges, high-curvature ridges, and abrupt changes in normals. To determine whether a boundary is a geometric feature line, if the boundary is at the edge's end, the electronic device determines it to be a geometric feature line. If the boundary is not at the edge's end, and the adjacent blocks sharing the boundary have completed the determination of the geometric structure generation parameters in S103, the electronic device can determine whether it is a geometric feature line based on the curvature difference and the included angle of the normals on both sides of the boundary between the current local surface block and the adjacent block. For example, if the curvature difference on both sides of the boundary is greater than or equal to a preset high-curvature threshold, and / or the included angle of the normals on both sides of the boundary is greater than or equal to a preset included angle threshold, then it is determined to be a geometric feature line, and thus a geometric feature line identifier is marked. It should be noted that if the boundary is not at the edge's end, and the adjacent blocks sharing the boundary have not completed the determination of the geometric structure generation parameters in S103, the electronic device first marks the state of the geometric feature line of that boundary as pending. When the adjacent blocks execute S103, the status of the boundary as a geometric feature line is re-evaluated, and a consistency update is performed.

[0042] S104, the electronic device generates a physical constraint modulation signal based on the geometric structure generation parameters of the current local surface block. The physical constraint modulation signal is used to indicate whether to increase or decrease the texture breakage tendency weight between the current local surface block and the adjacent local surface blocks.

[0043] For example, the physical constraint modulation signal is a scaling factor, which is used to multiply the output value of the feature channel (hereinafter referred to as "feature channel") associated with texture breakage in the texture generation stream to amplify or reduce the activation intensity of the feature channel. Specifically, when the scaling factor is greater than 1, the output value of the feature channel is amplified, causing the texture breakage tendency weight after activation function mapping to tend towards the high breakage tendency range; when the scaling factor is less than 1, the output value of the feature channel is reduced, causing the texture breakage tendency weight after activation function mapping to tend towards the low breakage tendency range; when the scaling factor is equal to 1, the output value of the feature channel remains unchanged. This strategy ensures that the interior of each local surface block is geometrically as continuous and morphologically consistent as possible.

[0044] In this embodiment of the application, the electronic device generates a physical constraint modulation signal based on the geometric structure parameters of the current local surface block, which can be in the following manner: method 1, method 2, or a combination of the two methods.

[0045] Method 1 (based on boundary-level attributes): The electronic device generates a boundary reference modulation signal based on the properties of the shared boundary between the current local surface block and its neighboring blocks. Specifically, if it is determined that the shared boundary between the current local surface block and its neighboring blocks belongs to a geometric feature line, a first modulation signal that enhances the tendency of texture breakage is generated; otherwise, a second modulation signal that suppresses the tendency of texture breakage is generated.

[0046] For example, if the modulation signal is a scaling factor, then the first modulation signal is greater than 1 and the second modulation signal is less than 1. For example, if the modulation signal is a modulation factor, the first modulation signal is 1.5 and the second modulation signal is 0.5, or it can be other suitable values, which are not limited in this article.

[0047] It should be noted that this approach can require the geometric feature lines of the shared boundary to have explicit values ​​rather than undetermined values. This can also be understood as requiring the first adjacent block to have completed the geometric state prediction of S103 before the current local surface block performs the prediction. Alternatively, for the geometric feature lines of the shared boundary to have undetermined values, this can be determined as a third modulation signal. This third modulation signal neither enhances nor suppresses the texture breakage tendency weight; for example, for the scaling factor, this third modulation signal is set to 1.0.

[0048] Compared with existing global UV optimization algorithms where the seam position is uniformly solved by the overall objective function and is difficult to adaptively adjust in combination with local geometry, this scheme directly uses the curvature difference and normal angle at each shared boundary as the decision basis. This ensures that the texture seam is accurately aligned with the geometric feature line and that the texture coordinates of the smooth surface are continuous. It fundamentally avoids the problem of seams appearing in the middle of the smooth surface or missing seams at the geometric feature line, and realizes adaptive matching between texture region division and local geometric features.

[0049] Method 2 (based on block-level curvature level): The electronic device generates a block-level reference modulation signal (the influence range of which is the shared boundary between the current local surface block and each adjacent block) based on the curvature level identifier of the current local surface block. Specifically, a fourth modulation signal (strong suppression signal) is generated when the curvature level is low, strongly suppressing texture breakage tendency; a fifth modulation signal (weak suppression or neutral signal) is generated when the curvature level is medium, weakly suppressing or neutralizing texture breakage tendency; and a sixth modulation signal (enhanced signal) is generated when the curvature level is high, enhancing texture breakage tendency. The fourth modulation signal may be equal to or unequal to the second modulation signal, the sixth modulation signal may be equal to or unequal to the first modulation signal, and the fifth modulation signal may be equal to or unequal to the third modulation signal; this paper does not impose any restrictions on these aspects.

[0050] For example, the modulation signal is a scaling factor. The scaling factor corresponding to the low curvature level (the fourth modulation signal) is less than 1, the scaling factor corresponding to the medium curvature level (the fifth modulation signal) is equal to 1, and the scaling factor corresponding to the high curvature level (the sixth modulation signal) is greater than 1. For example, the scaling factor corresponding to the low curvature level is 0.5, the scaling factor corresponding to the medium curvature level is 1.0, and the scaling factor corresponding to the high curvature level is 1.5.

[0051] Using this method 2, the high curvature region seam accuracy, which is sacrificed due to the averaging tendency in traditional global UV optimization, is transformed into a block-level modulation strategy that actively enhances or reduces the fracture tendency weight based on the block-level curvature level. This gives the boundaries of high curvature blocks (such as near edges and turning surfaces) a higher fracture priority, while the boundaries of low curvature blocks (such as smooth surfaces) are forced to be continuous.

[0052] Combination 1 of Method 1 and Method 2: Electronic devices can also combine modulation signals generated in two ways to determine the final physical constraint modulation signal, for example, by using a combination strategy of providing basic modulation at the block level curvature level and boundary-level attributes to provide boundary differentiation.

[0053] For example, the modulation signal is the scaling factor, and the final target scaling factor of the current local surface block is the weighted statistical value of the scaling sub-factors determined by method 1 and method 2 (e.g., weighted summation or weighted product, which is not limited in this document).

[0054] In some other possible implementations, the combination of Method 1 and Method 2 can also be embodied as Combination 2: when neighboring blocks have not completed the prediction of geometric structure generation parameters, the above physical constraint modulation signal is determined based on the curvature level attribute of the current block itself (i.e., Method 2 is adopted); when neighboring blocks have completed the prediction of geometric structure generation parameters, a refined decision is made based on the shared boundary attribute (i.e., Method 1 is adopted). Alternatively, it can also be embodied as Combination 3: by default, Method 2 is used for coarse decision-making, and after neighboring blocks are generated, Method 1 is used for refined correction.

[0055] By employing a combined approach, two modulation signal sources are simultaneously introduced: a "boundary reference modulation signal" and a "block-level reference modulation signal." This ensures that the generation of the physical constraint modulation signal no longer relies on a single geometric information. The boundary reference modulation signal, based on the geometric feature line attributes of shared boundaries, reflects the local transition relationships between blocks; the block-level reference modulation signal, based on the curvature level of the current block, reflects the overall curvature within the block. This combination ensures that even if one information source contains noise or misjudgments (such as the normal calculation at the boundary being affected by sampling, or the curvature statistics within the block being affected by the block segmentation method), the other information source can still provide effective modulation guidance, thereby improving the robustness of joint decision-making.

[0056] On the other hand, uniform modulation based solely on block-level curvature (such as enhancing fractures at all boundaries of high-curvature blocks) may lead to erroneous fractures even at smooth boundaries, while modulation based solely on boundary attributes will not work when neighboring blocks are not generated. The combined approach, by coordinating decisions between block-level reference modulation signals and boundary reference modulation signals, can achieve fine-grained control in specific scenarios: for low-curvature blocks, even if boundary attributes are misjudged, the block-level signal can still strongly suppress fractures, preventing unreasonable seams on smooth surfaces; for high-curvature blocks, the block-level signal provides basic enhancement, while the boundary signal further differentiates and adjusts, causing fractures at edges and aligning smooth boundaries.

[0057] It should be noted that the physical constraint modulation signal can be generated in a physical constraint module independent of the texture generation stream and then input into the texture generation stream; or, a physical constraint attention gate can be set in the texture generation stream, and the physical constraint modulation signal can be generated in the physical constraint attention gate.

[0058] S105, under the constraint of the physical constraint modulation signal, the electronic device predicts the texture structure generation parameters of the current local surface block based on the texture generation flow.

[0059] In this embodiment, the texture structure generation parameters include texture mapping coordinates and texture breakage tendency weights between adjacent blocks. The texture mapping coordinates refer to the coordinate mapping of a local surface block in the texture parameter space, and the texture breakage tendency weights represent the degree of need to generate texture seams between adjacent blocks; the larger the weight, the greater the need to generate seams.

[0060] In this embodiment, while generating the geometric structure block by block, the electronic device infers the corresponding texture structure generation parameters in real time based on the generated parameters of the currently generated geometric structure. This makes texture mapping an integral part of the geometric generation process, rather than an independent post-processing step after the geometric model is fully generated. Specifically, in terms of timing, for each local surface block, its geometric structure is generated first, and then its texture structure is predicted based on the geometric structure. Both are completed sequentially in the same block-by-block loop.

[0061] In this embodiment, the texture generation stream is implemented based on a deep generation model, which includes any one of Transformer, diffusion model, generative adversarial network, variational autoencoder or autoregressive model. This invention does not limit the specific type of deep generation model, but may require it to have the following functions: able to receive texture priors as conditional input, able to receive physical constraint modulation signals or receive geometric structure generation parameters as conditional input, feature channels can be modulated by modulation signals, and output texture parameters of local surface blocks in units of local surface blocks.

[0062] For example, the aforementioned physical constraint modulation signal is a scaling factor, which acts on the output value of the feature channel (the scaling factor is used to multiply the output value of the feature channel related to texture breakage in the texture generation network). The original output value of the feature channel represents the original texture breakage tendency weight. When the scaling factor is greater than 1, the original output value of the feature channel can be amplified, increasing the texture breakage tendency weight after activation function mapping, i.e., enhancing the breakage tendency; when the scaling factor is less than 1, the original output value of the feature channel can be reduced, decreasing the texture breakage tendency weight after activation function mapping (approaching 0), i.e., suppressing the breakage tendency.

[0063] In some other possible implementations, the texture generation stream is also used to check whether the texture within a block remains continuous, and if not, to correct it so that the texture within a local surface block remains continuous.

[0064] In some other possible implementations, after predicting the texture structure generation parameters of the current local surface block (S105) and before determining the texture fusion mapping parameters of the current local surface block (S106), the electronic device performs joint consistency correction on the geometric structure generation parameters and texture structure generation parameters of the current local surface block, and determines the corrected geometric structure generation parameters and texture structure generation parameters as the state to be fused (which can also be understood as the final prediction result of the geometric structure generation parameters and texture structure generation parameters of the current local surface block). Among them, the joint consistency correction includes at least one of the following: when the boundary of the current local surface block contains geometric feature lines, verify whether the texture breakage tendency weight at the geometric feature line is greater than or equal to a preset breakage threshold. If not, correct the texture breakage tendency weight to maintain the continuity trend. When there is preset severe conflict data between the geometric structure generation parameters and texture structure generation parameters of the current local surface block (for example, the geometric structure generation parameters mark a shared boundary as a geometric feature line, but the corresponding texture breakage tendency weight is less than a preset minimum threshold), roll back the texture mapping prediction result and resample.

[0065] S106, the electronic device determines the texture fusion mapping parameters of the current local surface block based on the geometric structure generation parameters and texture structure generation parameters of the current local surface block.

[0066] The aforementioned texture fusion mapping parameters include aligned texture mapping coordinates, seam decisions, and texture island affiliation identifiers.

[0067] In some possible implementations, such as Figure 2 As shown, step S106 specifically includes: S1061, the electronic device determines whether there are adjacent local surface blocks that meet the basic fusion conditions based on the difference in the normal angle at the shared boundary between the current local surface block and the adjacent local surface blocks, the topological continuity, and the texture breakage tendency weight, and identifies the adjacent local surface blocks that meet the basic fusion conditions as the target adjacent blocks.

[0068] In this embodiment of the application, the basic fusion conditions include: the included angle of the normals is less than a preset included angle threshold, the topological relationship is continuous, and the texture breakage tendency weight corresponding to the shared boundary is lower than a preset breakage threshold.

[0069] In some other possible implementations, the basic fusion conditions mentioned above may include, but are not limited to, at least one of the following: whether the geometric curvature levels are consistent or similar (similar means that the difference in curvature levels does not exceed one level), and whether the texture direction constraints are compatible.

[0070] This basic fusion condition is adopted to determine whether there are target neighboring blocks that meet the basic fusion condition by comprehensively considering three dimensions: the difference in normal angle at the shared boundary, topological continuity, and texture breakage tendency weight. This makes the fusion decision no longer dependent on a single indicator. The difference in normal angle reflects the degree of geometric transition between blocks, topological continuity ensures that the two sides of the boundary can be aligned, and the texture breakage tendency weight provides prior information for network prediction. The three factors are combined to determine whether the fusion condition is met only when the blocks are geometrically smooth and continuous, topologically connected, and the network does not break. This avoids misjudgments caused by relying on a single indicator.

[0071] S1062, if the aforementioned target adjacent block exists, the electronic device aligns the texture parameters of the current local surface block with the target adjacent block, and generates texture seams at the corresponding boundaries for other adjacent blocks that do not meet the fusion conditions, thereby obtaining the texture fusion mapping parameters of the current local surface block.

[0072] In this embodiment, texture parameter alignment includes texture mapping coordinate alignment and texture island affiliation alignment. It is understood that performing texture parameter alignment means that the seam decision at the corresponding boundary is to not generate a seam.

[0073] In some possible implementations, such as Figure 3 As shown, if the aforementioned target adjacent block exists, in S1062, the electronic device aligns the texture parameters of the current local surface block to the target adjacent block, specifically including: S10621, when the number of target adjacent blocks is one, align the texture parameters of the current local surface block to the target adjacent block.

[0074] S10622, when there are two or more target adjacent blocks, if the two or more target adjacent blocks belong to the same target texture island, then the current local surface block is aligned with the texture parameters of the overall texture space of the target texture island so that the texture coordinates at the shared boundaries between the current local surface block and the two or more target adjacent blocks are simultaneously continuous.

[0075] S10623, when there are two or more target adjacent blocks and the two or more target adjacent blocks belong to at least two different texture islands, count the number of target adjacent blocks contained in each target texture island, and determine the target texture island containing the most target adjacent blocks as the priority texture island, and align the texture parameters of the current local surface block to the overall texture space of the priority texture island.

[0076] S10624 If there are two or more candidate target texture islands with the largest number of equals, the priority of each candidate target texture island is determined based on at least one of the shared boundary length, normal angle, and texture generation order, and the current local surface block is aligned with the overall texture space of the candidate target texture island with the highest priority.

[0077] For target adjacent blocks in other target texture islands that are not identified as the priority texture island, the current local surface block does not perform texture parameter alignment to them, but generates texture seams at the corresponding shared boundaries.

[0078] Thus, a hierarchical arbitration rule is adopted: when there is only one target neighbor block, alignment is performed directly; when multiple target neighbor blocks belong to the same texture island, alignment is performed towards the overall texture space of that texture island, ensuring that all shared boundaries are simultaneously continuous; when multiple target neighbor blocks belong to different texture islands, alignment is preferentially performed towards the texture island containing the most target neighbor blocks, and if the number is the same, priority is determined based on boundary length, normal angle, or generation order. This hierarchical arbitration mechanism avoids the problems of indecisiveness or random decision-making in the case of multiple neighbor conflicts, making the fusion result deterministic and predictable.

[0079] The process of aligning texture parameters includes: aligning texture mapping coordinates and aligning texture island affiliations.

[0080] Texture mapping coordinate alignment is used to ensure continuous texture coordinates at shared boundaries where texture seams should not be generated. Texture island affiliation alignment means using the same texture island identifier (texture region identifier), and local surface patches with the same texture island identifier share the same UV continuum region.

[0081] Specifically, based on the corresponding point pairs on the shared boundary that need to be preserved, rigid body or affine transformation parameters are calculated from the UV space of the current local surface patch to the UV space of the target adjacent patch or the target texture island. The UV coordinates of all vertices of the current patch are mapped through this transformation, making the UV coordinates continuous at the shared boundary. After UV alignment is completed using this method, it can be ensured that in subsequent texture color processing, the current local surface patch will sample colors from adjacent regions of the texture image, thus ensuring the natural continuity of texture colors at the source without the need for additional color fusion post-processing.

[0082] Thus, texture parameter alignment includes texture mapping coordinate alignment and texture island affiliation alignment. Coordinate alignment ensures UV coordinate continuity at shared boundaries, achieving seamless texture sampling. Texture island affiliation alignment adds the current block to the target texture island (the basic unit for organizing texture mapping), achieving texture continuity. This dual alignment ensures complete integration of the current block and its target neighbors at both geometric (UV coordinates) and semantic (texture island affiliation) levels.

[0083] S1063, if there is no target adjacent block, then generate texture seams at each boundary of the current local surface block and start new texture islands for the current local surface block, thereby obtaining the texture fusion mapping parameters of the current local surface block.

[0084] In this embodiment, initiating a new texture island means that the texture mapping coordinates and texture island affiliation identifier of the current local surface block are determined based on the new texture island. Furthermore, initiating a new texture island also means that the seam decision between the current local surface block and its corresponding adjacent blocks is to generate a seam. Specifically, for common vertices at the boundary, they no longer correspond to the same point in the UV texture space, but are split into two different UV coordinates.

[0085] In this embodiment of the application, after the electronic device completes the prediction of the geometric structure generation parameters and texture fusion mapping parameters of the current local surface block, it performs the following propagation operations: updates the context feature cache of adjacent blocks; and broadcasts the texture region attribution information of the current block to the adjacent blocks.

[0086] Phase 3: Determine the 3D generation data: S107, the electronic device determines the 3D generation data based on the geometric structure generation parameters and texture fusion mapping parameters of each local surface block in the discrete local surface block set.

[0087] As an example, for static data scenarios, after all local surface blocks are generated and fused one by one, the electronic device summarizes and assembles the geometric structure generation parameters and texture fusion mapping parameters of each block to output complete 3D model data.

[0088] As another example, for dynamic incremental scenarios, since the generation process of geometry and texture in this application is itself a gradual construction process of 3D data, in dynamic incremental scenarios, electronic devices will continuously collect new data, generate the geometry and texture of new local surface blocks in real time, and can immediately merge its data into the 3D generation data and output it after generating a block, or can perform a unified merging after generating a preset number of blocks, add them to the 3D generation data and output them.

[0089] In some possible implementations, the electronic device determines the 3D generated data, including: after all local surface blocks are generated, the electronic device also performs consistency verification on the generated 3D asset to be verified. When a local anomaly is detected, it backtracks to the corresponding local surface block's co-generation stage (stage two), triggering local regeneration or parameter adjustment, and then performs consistency verification again, outputting the verified 3D asset data. This consistency verification includes at least one of the following: verifying whether the mesh topology connectivity is complete and free of self-intersections or illegal structures; verifying whether texture coordinates have overlapping, inversion, or discontinuity anomalies; verifying whether the texture seam position matches the geometric feature line; verifying whether there are unreasonable texture breaks within the smooth geometric region. Regions that do not meet the conditions will be marked and re-enter stage two. Finally, a high-fidelity 3D environment asset is output. For example, if there is no seam at the geometric feature line, a seam is actively added at that position or the seam is backtracked and regenerated with the feature line offset. If the distance between the seam position and the feature line is greater than a threshold, the UV coordinates are adjusted to move the seam to the feature line.

[0090] The spatial intelligent geometric topology and texture co-generation method provided in this application, on the one hand, the geometric features (such as curvature, normal, and geometric feature lines) at the shared boundary directly affect the texture breakage tendency weight, thereby ensuring that the texture seam appears at the reasonable boundary of the geometric features during the block generation process, overcoming the defect of the existing global UV optimization algorithm that sacrifices the alignment accuracy of the seam at the edge due to the averaging tendency.

[0091] On the other hand, using local surface blocks as the smallest generation unit, after the geometric generation stream outputs the geometric structure generation parameters, it immediately constrains the texture generation stream through physical constraint modulation signals, ensuring that the texture mapping prediction process is always guided by the geometric state. This collaborative mechanism of "generating geometry while making texture decisions" allows the placement decision of texture seams to be driven by the geometric structure, avoiding semantic conflicts between geometry and texture from the generation source.

[0092] Furthermore, in the application scenarios of this application (spatial intelligence, embodied AI, real-time generation), the existing global UV unwrapping method requires the geometric model to be complete and determined. If new data is added each time, the entire UV layout needs to be recalculated, which is difficult to adapt to the needs of generative and dynamically expanding scenarios. However, the block-by-block generation method provided by this application generates geometry and texture step by step. It can allocate new texture regions (new UV islands) to newly discovered surface blocks or attempt to merge them with existing regions (if the geometry is continuous) without recalculating existing regions, thus adapting to the needs of generative and dynamically expanding scenarios.

[0093] The following combination Figure 4 This application describes a spatial intelligent geometric topology and texture co-generation system, which includes: The discretized 3D surface acquisition and local surface block construction module 401 is used to divide the discretized representation of the 3D surface into local surface blocks with topological connections to obtain a set of discrete local surface blocks; The multimodal coding module 402 is used to construct multimodal features for each local surface patch. The multimodal feature encoding includes geometric information statistical summary, texture mapping prior conditions, and spatial relationship information with neighboring patches. The geometry topology generation flow module 403 is used to predict the geometry generation parameters of the current local surface block based on the geometry generation flow. The physical constraint module 404 is used to generate a physical constraint modulation signal based on the geometric structure generation parameters of the current local surface block. The physical constraint modulation signal is used to indicate whether to increase or decrease the texture breakage tendency weight between the current local surface block and the adjacent local surface blocks. The texture mapping generation stream module 405 is used to predict the texture structure generation parameters of the current local surface block based on the texture generation stream under the constraint of the physical constraint modulation signal. The texture structure generation parameters include texture mapping coordinates and texture breakage tendency weights that share boundaries with adjacent blocks. The texture fusion module 406 is used to determine the texture fusion mapping parameters of the current local surface block based on the geometric structure generation parameters and texture structure generation parameters of the current local surface block. The texture fusion mapping parameters include aligned texture mapping coordinates, seam decision, and texture island affiliation identifier. The 3D asset determination module 407 is used to determine the 3D generation data system based on the geometric generation parameters and texture fusion mapping parameters of each local surface patch in the discrete local surface patch set. In some possible implementations, such as Figure 5As shown, the system also includes a geometric texture joint consistency correction module 408, which is used to perform joint consistency correction on the geometric structure generation parameters and texture structure generation parameters of the current local surface block, and determine the corrected geometric structure generation parameters and texture structure generation parameters as the state to be fused.

[0094] In some possible implementations, the aforementioned 3D asset determination module 407 specifically includes: a 3D asset consistency verification module 4071, used to perform consistency verification on the generated 3D asset to be verified after all local surface blocks have been generated. When a local anomaly is detected, it backtracks to the corresponding local surface block generation stage, triggers local regeneration or parameter adjustment, and then performs consistency verification again. It also includes an output module 4072, used to output the verified 3D asset data.

[0095] In this application, any of the implementation methods mentioned in the method embodiments are also applicable to the spatial intelligent geometric topology and texture co-generation system provided in this application. For specific execution steps and related terminology, please refer to the description of the foregoing method embodiments, which will not be detailed here.

[0096] Currently, research on 3D generation has made some progress in terms of geometric modeling accuracy, diversity control, and large-scale data-driven modeling. Existing technologies typically employ 3D representations based on meshes, point clouds, or implicit surfaces, combined with deep learning models to achieve automatic generation of complex 3D structures. To address the computational and storage pressure brought by high-resolution 3D meshes, some technical solutions introduce discretization methods, dividing continuous 3D surfaces into several local regions or surface blocks and serializing and encoding them to improve the efficiency and stability of geometric generation. However, existing 3D generation technologies still have significant shortcomings in the coordinated consistency between geometric structure generation and surface texture mapping, making it difficult to meet the comprehensive requirements of spatial intelligent systems for visual-physical consistency. This is mainly reflected in the following aspects.

[0097] The processes of geometric topology generation and texture mapping are separate.

[0098] Most existing 3D generation methods primarily focus on modeling and generating geometric topology, with texture mapping typically treated as a separate post-processing step. After the geometric model is generated, additional texture unwrapping algorithms (UVUnwrapping) or manual tools are often needed to parameterize the texture of the model surface. This "geometry first, texture later" approach lacks unified modeling during the generation stage, resulting in no inherent generational constraints between the geometric structure and texture attributes, making it difficult to guarantee their spatial semantic consistency.

[0099] Local geometric discrete units lack the ability to express texture semantics.

[0100] In existing 3D generation methods based on local blocks or fragments, the discretized basic units are typically used only to describe local geometry or topological connections. Their feature representations focus on geometric information such as vertex positions, normal directions, or adjacency relationships, without incorporating texture mapping parameters into a unified feature modeling framework. This results in a lack of explicit constraints on texture continuity and boundaries between local surface blocks during the generation process, easily leading to unreasonable texture breaks or splicing artifacts on the model surface.

[0101] Texture seams are difficult to align with geometric feature lines.

[0102] Since texture unfolding is typically independent of the geometry generation process, existing technologies struggle to ensure that texture seam locations align with edges, high-curvature regions, or topological transitions in the geometry. In spatial intelligence applications, this inconsistency can lead to discrepancies between visual perception and physical geometry. For example, textured regions that appear as distinct boundaries in visual sensors may correspond to smooth surfaces in physical models, thus reducing the accuracy of the agent's understanding of its environment.

[0103] The automatic texture unwrapping results in severe fragmentation and a lack of structural constraints.

[0104] Existing automated UV unwrapping algorithms are mostly based on global optimization or heuristic rules. When faced with complex geometries or large-scale models, they tend to generate a large number of scattered texture islands, increasing rendering overhead and reducing texture utilization efficiency. Especially in automatically generated scenes, these algorithms struggle to adaptively adjust to local geometric characteristics, resulting in texture region partitioning that does not conform to physical structure or spatial semantic features.

[0105] Compared with existing 3D generation technologies that mainly employ separate processing of geometry generation and texture mapping or rely on post-processing texture unfolding, this invention has significant advantages in terms of geometry-texture generation consistency, spatial semantic rationality, degree of automation, and spatial intelligent adaptability.

[0106] 1. Significantly improved consistency in geometry and texture generation. This invention uses local surface blocks as the basic generation unit, and simultaneously infers the geometric topology and texture mapping attributes during the generation stage. This avoids the structural fragmentation problem caused by independently performing texture unrolling after geometric generation in traditional processes. By explicitly constraining the texture generation process through geometric states, it ensures that the texture seam positions are consistent with the geometric feature lines, thus maintaining consistency between the generated model in terms of visual appearance and physical structure.

[0107] 2. Enhanced semantic integrity of locally generated units This invention constructs a multimodal joint representation for each local surface block, incorporating geometric features and texture parameters, enabling the smallest unit in the generation process to simultaneously possess both "morphological description" and "texture mapping semantics." Compared to traditional local generation methods that only model geometry, this invention significantly improves the semantic integrity of the local generation results and effectively reduces the probability of texture misalignment, stretching, and discontinuities.

[0108] 3. The quality and efficiency of automatic texture unwrapping are significantly improved. By employing an adaptive texture blending strategy based on topological continuity, this invention automatically constructs structurally sound, continuous texture regions during the generation process, avoiding the creation of numerous fragmented texture islands. The generated 3D models can be used directly without manual intervention or independent UV unwrapping tools, significantly improving the automation level and rendering efficiency of 3D asset generation.

[0109] 4. Enhanced spatial perception consistency and application adaptability The 3D assets generated by this invention maintain a high degree of consistency between geometric structure and visual texture, effectively avoiding the problem of mismatch between visual perception results and physical geometric models in spatial intelligent systems. This characteristic makes the generated results more suitable for applications with high requirements for spatial consistency, such as robot simulation, spatial perception training, virtual-real fusion, and digital twins.

[0110] 5. Improved system stability and scalability This invention employs a modular collaborative generation architecture, where functional modules interact through local surface block state flows. When local anomalies occur, block-level backtracking and regeneration can be achieved without requiring overall model reconstruction, thereby improving system stability. Furthermore, this method is not limited to specific discretization methods, possessing good scalability and versatility, and can be adapted to various 3D representations and generation models.

[0111] This application also provides a spatial intelligent geometric topology and texture co-generation device, including a processor, which is used to execute any one of the spatial intelligent geometric topology and texture co-generation methods in the above method embodiments.

[0112] This application also provides a computer storage medium that can store multiple instructions. These instructions are adapted to be loaded and executed by a processor using the spatial intelligent geometric topology and texture co-generation method provided in this application. For details of the execution process, please refer to the specific description of the method embodiments shown above, which will not be elaborated here.

[0113] This application also provides a computer program product containing instructions that, when run on an electronic device, cause the electronic device to execute the method steps of the method embodiments shown above.

[0114] This application also provides a chip module, including a transceiver component and a chip, wherein the chip is used to execute the method steps of the above-described method embodiments.

[0115] It is understood that the spatial intelligent geometric topology and texture co-generation system, spatial intelligent geometric topology and texture co-generation device, computer storage medium, computer program, computer program product, and chip provided above are all used to execute the method shown in any implementation of the corresponding aspect of the embodiments of this application. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects in the corresponding method, and will not be detailed here.

[0116] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes the processes of the embodiments of the above methods.

[0117] The term "at least one" in this application refers to one or more items. "More than one item" means two or more items. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship. Furthermore, it should be understood that although the terms "first," "second," etc., may be used to describe objects in this application, these objects should not be limited to these terms. These terms are only used to distinguish the objects from each other.

[0118] The terms “including” and “having” mentioned above, and any variations thereof, are intended to cover non-exclusive inclusion.

[0119] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for co-generating spatial intelligent geometric topology and texture, characterized in that, The method includes: The three-dimensional surface is discretized and divided into local surface blocks with topological connections, resulting in a set of discrete local surface blocks; Multimodal features are constructed for each local surface patch, including a statistical summary of geometric information, prior conditions for texture mapping, and spatial relationship information with neighboring patches; Based on the geometry generation flow, predict the geometry generation parameters of the current local surface block; Based on the geometric structure generation parameters of the current local surface block, a physical constraint modulation signal is generated. The physical constraint modulation signal is used to indicate the modulation of the output value of the feature channel related to texture breakage in the texture generation stream, so as to increase or decrease the output value of the feature channel related to texture breakage. The output value of the feature channel related to texture breakage is the original texture breakage tendency weight. The texture breakage tendency weight is used to represent the degree of need for generating texture seams between the current local surface block and adjacent blocks. Under the constraint of the physical constraint modulation signal, based on the texture generation flow, the texture structure generation parameters of the current local surface block are predicted. The texture structure generation parameters include texture mapping coordinates and texture breakage tendency weights that share boundaries with adjacent blocks. Based on the geometric structure generation parameters and texture structure generation parameters of the current local surface block, the texture fusion mapping parameters of the current local surface block are determined. The texture fusion mapping parameters include aligned texture mapping coordinates, seam decisions, and texture island affiliation identifiers. The 3D generation data is determined based on the geometric structure generation parameters and texture fusion mapping parameters of each local surface block in the discrete local surface block set.

2. The method according to claim 1, characterized in that, The process of discretizing the three-dimensional surface representation into local surface blocks with topological connections includes: Geometric feature regions are detected based on local curvature changes, and the geometric feature regions include at least one of edges, facets, and transition zones. Using points in non-geometric feature regions as seed points, region growth is performed in regions with gentle curvature changes. When the growth boundary encounters the geometric feature region, a preset curvature abrupt change region, or a preset normal abrupt change region, the growth is automatically terminated. The continuous region formed by the growth is taken as a single local surface block, and the boundary of the local surface block is aligned with the boundary of the geometric feature region.

3. The method as described in claim 1, characterized in that, The geometric information statistical summary includes at least one of vertex relative position index, normal distribution, and curvature statistics, used to characterize the statistical distribution features of the local surface block geometry; The texture mapping prior conditions include at least one of the following: initial mapping reference, texture direction constraint, and texture continuity flag, which are used to provide prior constraints for texture generation. The geometric structure generation parameters also include at least one of the following: vertex position offset adjustment, vertex normal vector within the block, block-level curvature level identifier, and geometric feature line identifier, used to represent the specific geometric shape of the local surface block.

4. The method according to any one of claims 1 to 3, characterized in that, The generation of a physical constraint modulation signal based on the geometric structure generation parameters of the current local surface block includes: The physical constraint modulation signal is determined based on the boundary reference modulation signal and / or the block-level reference modulation signal; The boundary reference modulation signal is generated based on the properties of the shared boundary between the current local surface block and its neighboring blocks. When the shared boundary is a geometric feature line, the boundary reference modulation signal is a first modulation signal that enhances the tendency of texture breakage; otherwise, it is a second modulation signal that suppresses the tendency of texture breakage. The geometric feature line is determined based on the curvature difference and / or the angle between the normals. The block-level reference modulation signal is generated based on the curvature level identifier of the current local surface block. When the curvature level identifier is low, the block-level reference modulation signal is a fourth modulation signal that strongly suppresses the tendency of texture breakage. When the curvature level identifier is medium, the block-level reference modulation signal is a fifth modulation signal that weakly suppresses or neutralizes the tendency of texture breakage. When the curvature level identifier is high, the block-level reference modulation signal is a sixth modulation signal that enhances the tendency of texture breakage.

5. The method according to any one of claims 1 to 3, characterized in that, The step of determining the texture fusion mapping parameters of the current local surface patch based on the geometric structure generation parameters and texture structure generation parameters of the current local surface patch includes: Based on the difference in the normal angle at the shared boundary between the current local surface block and the adjacent local surface block, the topological continuity, and the texture breakage tendency weight, it is determined whether there are adjacent local surface blocks that meet the basic fusion conditions, and the adjacent local surface blocks that meet the basic fusion conditions are identified as target adjacent blocks. If the target adjacent block exists, the texture parameters of the current local surface block are aligned with those of the target adjacent block, and texture seams are generated at the corresponding boundaries of other adjacent blocks that do not meet the fusion conditions; if the target adjacent block does not exist, texture seams are generated at each boundary of the current local surface block, and new texture islands are started for the current local surface block; thereby obtaining the texture fusion mapping parameters of the current local surface block.

6. The method as described in claim 5, characterized in that; If the target neighboring block exists, then aligning the texture parameters of the current local surface block to the target neighboring block includes: If the number of target neighboring blocks is one, the current local surface block is aligned with the target neighboring block in terms of texture parameters; If there are two or more target adjacent blocks, and if the two or more target adjacent blocks belong to the same target texture island, then the current local surface block is aligned with the texture parameters of the overall texture space of the target texture island so that the texture coordinates at each shared boundary between the current local surface block and the two or more target adjacent blocks are simultaneously continuous. When there are two or more target adjacent blocks, and the two or more target adjacent blocks belong to at least two different texture islands, the number of target adjacent blocks contained in each target texture island is counted, and the target texture island containing the most target adjacent blocks is determined as the priority texture island. The current local surface block is then aligned with the overall texture space of the priority texture island using texture parameters. If there are two or more candidate target texture islands with the largest number of equals, the priority of each candidate target texture island is determined based on at least one of the shared boundary length, normal angle, and texture generation order, and the current local surface block is aligned with the overall texture space of the candidate target texture island with the highest priority. For target adjacent blocks in other target texture islands that are not identified as the priority texture island, the current local surface block does not perform texture parameter alignment to them, but generates texture seams at the corresponding shared boundaries; The process of aligning texture parameters includes: aligning texture mapping coordinates and aligning texture island affiliations.

7. The method according to any one of claims 1 to 3, characterized in that, After predicting the texture structure generation parameters of the current local surface patch and before determining the texture fusion mapping parameters of the current local surface patch, the method further includes: Perform joint consistency correction on the geometric structure generation parameters and texture structure generation parameters of the current local surface block, and determine the corrected geometric structure generation parameters and texture structure generation parameters as the state to be fused. The joint consensus correction includes at least one of the following: If the boundary of the current local surface block contains geometric feature lines, verify whether the texture breakage tendency weight at the geometric feature lines is greater than or equal to a preset breakage threshold. If not, adjust the texture breakage tendency weight to maintain the continuity trend. If there is a pre-defined serious conflict between the geometric structure generation parameters and the texture structure generation parameters of the current local surface block, the texture mapping prediction result is rolled back and resampled.

8. The method according to any one of claims 1 to 3, characterized in that, The determination of the three-dimensional generated data includes: After all local surface blocks have been generated, the generated 3D assets to be verified are subjected to consistency verification. When a local anomaly is detected, the process is backtracked to the corresponding local surface block generation stage, triggering local regeneration or parameter adjustment, and then consistency verification is performed again to determine the 3D asset data that has passed the verification. The consistency verification includes at least one of the following: Verify that the mesh topology connections are complete and free of self-intersections or illegal structures. Verify that there are no anomalies such as overlapping, reversal, or discontinuity in texture coordinates; Verify that the texture seam position matches the geometric feature line; Verify whether there are unreasonable texture breaks within the smooth geometric region.

9. A spatial intelligent geometric topology and texture collaborative generation system, characterized in that, The system includes: The discretized 3D surface acquisition and local surface block construction module is used to divide the discretized representation of the 3D surface into local surface blocks with topological connections, thereby obtaining a set of discrete local surface blocks; A multimodal coding module is used to construct multimodal features for each local surface patch. The multimodal features include a geometric information statistical summary, texture mapping prior conditions, and spatial relationship information with neighboring patches. The geometry topology generation flow module is used to predict the geometry generation parameters of the current local surface block based on the geometry generation flow. The physical constraint module is used to generate a physical constraint modulation signal based on the geometric structure generation parameters of the current local surface block. The physical constraint modulation signal is used to indicate the modulation of the output value of the feature channel related to texture breakage in the texture generation stream, so as to increase or decrease the output value of the feature channel related to texture breakage. The output value of the feature channel related to texture breakage is the original texture breakage tendency weight, which is used to represent the degree of need for generating texture seams between the current local surface block and adjacent blocks. The texture mapping generation stream module is used to predict the texture structure generation parameters of the current local surface block based on the texture generation stream under the constraint of the physical constraint modulation signal. The texture structure generation parameters include texture mapping coordinates and texture breakage tendency weights that share boundaries with adjacent blocks. The texture fusion module is used to determine the texture fusion mapping parameters of the current local surface block based on the geometric structure generation parameters and texture structure generation parameters of the current local surface block. The texture fusion mapping parameters include aligned texture mapping coordinates, seam decisions, and texture island affiliation identifiers. The 3D asset determination module is used to determine the 3D generation data system based on the geometric structure generation parameters and texture fusion mapping parameters of each local surface block in the discrete local surface block set.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, which, when executed, performs the method according to any one of claims 1 to 8.

Images