Three-dimensional model generation method, system, computer device and storage medium

Abstract

This invention provides a method, system, computer device, and storage medium for generating three-dimensional models. The method includes: dividing a target object into regions based on model parameters to obtain multiple target regions, each target region including at least one regular region and at least one free-form region; sending all regular regions to a CAD engine to obtain a first three-dimensional modeling parameter set corresponding to the regular regions, containing boundary and dimension information; and sending all free-form regions to a neural image generation engine to obtain a second three-dimensional modeling parameter set corresponding to the free-form regions, containing three-dimensional mesh and texture information; and, based on manufacturability constraints and the first and second three-dimensional modeling parameter sets after coordinate alignment, combining the regular and free-form regions at their junctions to output a hybrid three-dimensional model. This invention balances model fidelity, parameter accuracy, and manufacturability.

Description

Technical Field

[0001] This invention belongs to the field of 3D modeling technology, and in particular relates to a 3D model generation method, system, computer equipment, and storage medium. Background Technology

[0002] Current 3D modeling solutions mainly include two unrelated technical systems: parametric CAD modeling and neuromorphic 3D generative modeling. Their shortcomings are: CAD modeling is suitable for regular geometric structures, but performs poorly on irregular shapes such as freeform structures, resulting in structures lacking realism and failing to achieve usable fidelity. Neuromorphic 3D generative modeling can quickly produce highly realistic textured organic irregular models, but the generated meshes cannot be parametrically edited, lack dimensional references, and cannot guarantee accurate and manufacturable parameters. For real-world modeling objects, which are typically hybrid models combining regular and freeform structures, using only one of the above methods often fails to simultaneously achieve fidelity, parametric accuracy, and manufacturability. Summary of the Invention

[0003] This invention addresses the technical problem in existing 3D modeling technologies that cannot simultaneously achieve model fidelity, parameter accuracy, and manufacturability, by providing a 3D model generation method, system, computer equipment, and storage medium.

[0004] In view of the above technical problems, embodiments of the present invention provide a method for generating a three-dimensional model, including: The system receives a 3D model generation instruction containing model parameters of a target object, and divides the target object into regions according to the model parameters to obtain multiple target regions, wherein the target regions include at least one regular region and at least one free-form region. All the regular regions are sent to the CAD engine to obtain a first three-dimensional modeling parameter set containing boundary information and size information corresponding to the regular regions, and all the free-form regions are sent to the neural image generation engine to obtain a second three-dimensional modeling parameter set containing three-dimensional mesh information and texture information corresponding to the free-form regions. Based on manufacturability constraint information, and the first and second 3D modeling parameter sets after coordinate alignment, the combination position of the regular region and the free-form region is processed to output a hybrid 3D model.

[0005] A 3D model generation system, comprising: The receiving module is used to receive a 3D model generation instruction containing model parameters of a target object, and to divide the target object into regions according to the model parameters to obtain multiple target regions, wherein the target regions include at least one regular region and at least one free-form region. The modeling module is used to send all the regular regions to the CAD engine to obtain a first three-dimensional modeling parameter set containing boundary information and size information corresponding to the regular regions, and to send all the free-form regions to the neural image generation engine to obtain a second three-dimensional modeling parameter set containing three-dimensional mesh information and texture information corresponding to the free-form regions. The model generation module is used to combine the regular region and the free-form region at their respective locations based on manufacturability constraint information, as well as the first and second 3D modeling parameter sets after coordinate alignment, and output a hybrid 3D model.

[0006] A computer device includes a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor implements the three-dimensional model generation method when executing the computer-readable instructions.

[0007] A computer-readable storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the three-dimensional model generation method.

[0008] In the three-dimensional model generation method of the present invention, a three-dimensional model generation instruction containing model parameters of a target object is first received. The target object is then divided into regions according to the model parameters to obtain multiple target regions, each including at least one regular region and at least one free-form region. All regular regions are then sent to a CAD engine to obtain a first three-dimensional modeling parameter set corresponding to the regular regions, containing boundary and dimension information. All free-form regions are then sent to a neural image generation engine to obtain a second three-dimensional modeling parameter set corresponding to the free-form regions, containing three-dimensional mesh and texture information. Finally, based on manufacturability constraint information and the first and second three-dimensional modeling parameter sets after coordinate alignment, the combination positions of the regular regions and the free-form regions are combined, and a hybrid three-dimensional model is output.

[0009] This invention divides the target object into regular and free-form regions based on model parameters, and models these regions in different ways. The regular regions are processed by a CAD engine to obtain a first 3D modeling parameter set containing boundary and dimensional information. The free-form regions are processed by a neural image generation engine to obtain a second 3D modeling parameter set containing 3D mesh and texture information. The first and second 3D modeling parameter sets are not converted to each other but are saved independently. Then, based on manufacturability constraints, the combined position of the first and second 3D modeling parameter sets after coordinate alignment is combined to ensure the manufacturability of the final hybrid 3D model. In this hybrid 3D model, the free-form regions possess high fidelity and photorealistic appearance, while the corresponding positions in the regular regions have editable parameters and accurate dimensions. Thus, this invention simultaneously achieves high model fidelity, parameter accuracy, and manufacturability. Attached Figure Description

[0010] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0011] Figure 1 This is a flowchart of a three-dimensional model generation method provided in an embodiment of the present invention.

[0012] Figure 2 This is a flowchart of step S300 of a three-dimensional model generation method provided in an embodiment of the present invention.

[0013] Figure 3 This is a schematic diagram of the structure of a three-dimensional model generation system provided in an embodiment of the present invention.

[0014] Figure 4 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] like Figure 1 As shown, an embodiment of the present invention provides a method for generating a three-dimensional model, including steps S100-S300: S100: Receive a 3D model generation instruction containing model parameters of a target object; divide the target object into regions according to the model parameters to obtain multiple target regions, each target region including at least one regular region and at least one free-form region; wherein, the input interface of the 3D model generation system can be used to receive a 3D model generation instruction containing model parameters of the target object, the aforementioned model parameters including but not limited to text information and reference images, wherein the text information refers to the characteristics of the target object described by natural language, the text information can be obtained by voice input and recognition, or by directly inputting text through a terminal; and the reference images can include one or more images, all of which need to be normalized to achieve a permissible resolution, and each reference image has a unique coded identifier.

[0017] Understandably, after receiving the 3D model generation command at the input interface, the 3D model generation system can determine whether the model parameters are complete through the controller. If incomplete, it can prompt for supplementation; if complete, it begins to divide the target object according to the model parameters. The system can determine the completeness of the model parameters based on preset completeness conditions. For example, preset completeness conditions can be set according to requirements, such as requiring the model parameters to include text information and reference images; or including multiple specified text information items, each with specific attributes; or including one or more reference images, as long as multiple specific attributes can be clearly identified in all reference images. These specific attributes can include size, scale, shape, etc.

[0018] In one embodiment, step S100, which involves dividing the target object into multiple target regions based on the model parameters, includes the following steps S101-S103: S101, extract text semantic labels from the text information of the model parameters; that is, when the model parameters include text information, it is necessary to extract text semantic labels from the text information. Specifically, the text information can be input into a preset semantic recognition model to obtain the text semantic labels output by the semantic recognition model. For example, when the target object is a combination of an animal figurine and a rectangular base with holes, if the text information is: "The cat figurine is placed on the rectangular base with holes", then the recognized text semantic labels can be "rectangular base" and "animal figurine". The training process of the semantic recognition model is as follows: obtain a sample dataset containing multiple text samples, each text sample is associated with a real semantic label; extract the semantic features of the text samples in the sample dataset based on the pre-trained model, predict the predicted semantic labels through the classification layer, calculate the error between the predicted semantic labels and the real semantic labels through the cross-entropy loss function, update the model parameters of the pre-trained model using the gradient descent algorithm for backpropagation, and obtain a semantic recognition model that can be used to recognize text semantic labels after iterative convergence.

[0019] S102, determine geometric feature information based on the text information and / or reference image in the model parameters. The geometric feature information includes at least one of the following: curvature distribution information of the target object, parameterizability, and manufacturing feature matching degree. Curvature distribution information refers to the statistical distribution characteristics of the principal curvature values ​​in the target region; the curvature values ​​of regular regions are concentrated within a constant or standard typical value range. Parameterizability refers to an index used to quantify how well the target region can be fully represented through a finite number of parameterized modeling operations. Manufacturing feature matching degree refers to the similarity between the target region and the processing features in a preset manufacturing feature library (which includes standard processing features such as holes, bosses, and chamfers).

[0020] S103, the system divides the target region into regular and free-form regions based on the text semantic tags and geometric feature information, and outputs the combination information between adjacent target regions. Specifically, since curvature distribution information can be used to determine whether the principal curvature of the target region is concentrated on basic regular surfaces such as planes, cylinders, and cones, the more regular and closer the distribution of the target region is to a standard surface, the higher the first regularity score corresponding to the curvature distribution information. Parameterizability is used to evaluate whether the target region can be completely reproduced through a small number of standard parameterization operations such as stretching, drilling, and chamfering; therefore, the higher the parameterizability, the higher the second regularity score corresponding to the parameterizability. Manufacturing feature matching degree is used to calculate the similarity between the target region and a standard machining feature library such as holes, bosses, chamfers, and threads; the higher the manufacturing feature matching degree, the closer it is to a regular manufacturing structure, and the higher its corresponding third regularity score. Simultaneously, text semantic tags can be optionally introduced as prior constraints; these tags are only used to fine-tune weights and correct deviations in the division of ambiguous regions. Next, the rule scores corresponding to the three geometric feature information items—namely, the first regularity score corresponding to curvature distribution information, the second regularity score corresponding to parameterizability, and the third regularity score corresponding to manufacturing feature matching degree—are each assigned a higher weight, while the text semantic label is assigned a lower weight. Then, the four scores are weighted and calculated to obtain a comprehensive score corresponding to the target region. If the comprehensive score is higher than a preset score threshold, the target region is determined to be a regular region (base, hole, chamfer, thread, etc., subsequently sent to the CAD engine); if the comprehensive score is lower than the preset score threshold, the target region is determined to be a free-form region (animal, curved surface, artistic shape, etc., subsequently sent to the neural image generation engine). The preset score threshold can be set according to requirements.

[0021] Understandably, the process of dividing regular and free-form regions based on the text semantic labels and geometric feature information can be performed by the classifier of the 3D model generation system. After the classifier outputs the morphological category label corresponding to each target region (the morphological category label is used to characterize whether the target region is a regular or free-form region), it also simultaneously outputs the combination information between adjacent target regions. The combination information is the interface specification between adjacent regions, which is used to record the boundary position, orientation, and boundary topology information of the joining position of adjacent regions. The combination information includes at least the interface identifier, the region identifiers on both sides of the interface, the combination type, and the interface boundary geometry. The combination information may also include data such as boundary normal, cross-sectional loop points, transition type, and fit tolerance. For example, the combination information can be used to characterize the fitting boundary information corresponding to the joining position between the bottom of the animal ornament and the top surface of the rectangular base. The interface specification (i.e., the combination information) established in addition to the region division results can provide a benchmark and unified constraint for the subsequent coordinate alignment and combination processing of different target regions. This interface specification serves as control data for subsequent geometric processing, rather than being saved only as model annotations.

[0022] In other embodiments, the process of dividing the regular region and the free-form region according to the text semantic label and the geometric feature information can also be performed by a trained preset classification model. That is, after the preset classification model inputs the text semantic label and the geometric feature information, it outputs the morphological category label corresponding to the target region and its corresponding confidence level. When the confidence level exceeds the preset confidence threshold (the preset confidence threshold can be set according to the requirements), the target region can be directly identified as the morphological category label corresponding to the confidence level. When the confidence level is lower than the preset confidence threshold, the target region is directly identified as the default preset label (the preset label is one of the preset morphological category labels), or the human is notified to confirm the morphological category label corresponding to the target region. In some embodiments, the training process of the preset classification model is as follows: A preset dataset is obtained, where each classification sample contains a true morphological label (corresponding to a morphological category label, which also represents whether the target region is a regular or free-form region). The preset dataset is fed into a binary classification neural network for forward propagation, outputting a single predicted score in the 0-1 interval. This score is then post-processed to obtain the predicted morphological label (corresponding to a morphological category label, which also represents whether the target region is a regular or free-form region) and the corresponding predicted confidence score for the current classification sample. The predicted morphological label is obtained, and the loss error between the predicted and true morphological labels is calculated. The network weights and bias parameters are then iteratively updated using a gradient descent algorithm to continuously optimize the model's classification accuracy and the reliability of the confidence score output. Training is iterated repeatedly until the model loss value stabilizes and the classification accuracy no longer significantly improves. Training is then terminated, and the network parameters are saved, ultimately resulting in a preset classification model that can simultaneously output morphological category labels and corresponding confidence scores.

[0023] S200: All the regular regions are sent to the CAD (Computer Aided Design) engine to obtain a first 3D modeling parameter set containing boundary and dimension information corresponding to the regular regions. All the freeform regions are then sent to the neural image generation engine to obtain a second 3D modeling parameter set containing 3D mesh and texture information corresponding to the freeform regions. The CAD engine and the neural image generation engine can model the regular regions and freeform regions sequentially (the order is not limited), or simultaneously. The regular regions and freeform regions can be routed to the CAD engine and the neural image generation engine respectively via a router.

[0024] The CAD engine is built upon an open-source geometric kernel, utilizing mature program interfaces for function calls. It integrates an intelligent scheduling module and a standardized tool call structure, automatically generating parametric modeling scripts based on external modeling knowledge resources. The CAD engine automatically fits the geometric primitive parameters of regular regions, executes the modeling script to generate topological boundary structures and editable dimensional parameters, completing the parametric 3D modeling of the regular regions. Geometric primitives are the basic, simple, and indivisible geometric units that constitute all models. Common types include: 2D primitives: points, lines, line segments, circles, arcs, ellipses, splines, rectangles, polygons, etc.; 3D primitives: planes, cylindrical surfaces, conical surfaces, spheres, tori, extruded bodies, solids of revolution, cuboids, spheres, cylinders, etc. Geometric primitive parameters are the unique set of parameters characterizing the shape, position, and orientation of the geometric primitive. Specifically, the process by which the CAD engine performs parametric 3D modeling to obtain a first 3D modeling parameter set containing boundary and dimension information corresponding to the regular region is as follows: The model parameters corresponding to the regular region transmitted by the router are received. The RANSAC algorithm is used to fit standard geometric primitives such as planar cylinders and holes within the region based on these model parameters, extracting the dimension and spatial position parameters corresponding to each primitive. A parametric modeling script containing operation sequences such as stretching, rotation, and drilling is generated based on the extracted primitive parameters. The script is then sent to the open-source geometric kernel for computation via a standardized tool call bridge, generating a complete boundary topology structure and simultaneously generating a feature tree with adjustable dimension parameters. The generated boundary topology information and feature tree dimension information are integrated and encapsulated to obtain the first 3D modeling parameter set. The first 3D modeling parameter set includes, but is not limited to, the following parameters: region identifier, region type, boundary information (such as model vertex coordinates, edge curve equations, surface geometric equations, face-edge-vertex topological relationships, entity closure determination information, version information, etc.), dimensional information (such as plane length and width, cylinder radius and height, hole diameter and depth, chamfer radius, and various feature spacings), spatial position constraint parameters (such as coordinate offsets of each primitive, rotation angle, symmetry constraints, and alignment constraints), and modeling operation parameters (such as extrusion depth, rotation angle, array number and spacing, Boolean operation type, feature generation order, and feature tree record).

[0025] The neural image generation engine is built upon a mature image-to-3D pre-trained model. It takes parametric images or text information of freeform regions as input and outputs a second 3D modeling parameter set containing 3D mesh and texture information end-to-end. The neural image generation engine supports asynchronous task scheduling, enabling it to run in parallel with the modeling tasks of a CAD engine, specifically for 3D modeling of freeform regions. Specifically, the process by which the neural image generation engine performs 3D modeling to obtain the second 3D modeling parameter set corresponding to the freeform region, containing 3D mesh and texture information, is as follows: It receives model parameters of the freeform region from the router; performs inference calculations based on these parameters to generate a 3D mesh structure adapted to the freeform surface shape; simultaneously generates texture information; and integrates, encapsulates, and outputs the 3D mesh information and texture information corresponding to the generated 3D mesh structure to obtain the second 3D modeling parameter set. The second set of 3D modeling parameters includes, but is not limited to, the following parameters: region identifier, region type, 3D mesh information (such as vertex space coordinates, triangle patch index, patch normal vector, mesh subdivision level, edge segmentation marker, etc.), and texture information (such as base color value, metallic value, surface roughness value, bump map data, transparency parameter, multi-view alignment marker, texture mapping coordinates, surface smoothing weight, detail relief texture data, etc.).

[0026] It should be noted that in this invention, after obtaining the first and second 3D modeling parameter sets, they are not converted to each other, but are stored separately in their original formats, forming a heterogeneous storage characteristic between the first and second 3D modeling parameter sets. After all target areas have been 3D modeled using the corresponding engines and the corresponding 3D modeling parameter sets have been stored, the process proceeds to step S300 for further processing.

[0027] In some embodiments, the neural image generation engine is deployed in the cloud, and the freeform region is sent to the cloud for remote processing by the cloud-based neural image generation engine. Specifically: the reference image and text information corresponding to the freeform region are encapsulated, a 3D generation task is submitted to the cloud, and the cloud returns a unique task identifier; the local terminal continuously polls the remote status interface connected to the cloud to query the task progress corresponding to the 3D generation task in real time until the remote status interface returns task completion information, then retrieves the remotely generated 3D data through the unique task identifier, performs unified format normalization processing on the 3D data, and decomposes and outputs a second 3D modeling parameter set containing 3D mesh information and texture information corresponding to the freeform region. In this way, by adopting remote asynchronous task scheduling, the local terminal does not need to deploy a high-computing neural image generation model, which can reduce the hardware configuration requirements of the local terminal and adapt to cloud-based batch modeling and multi-terminal collaborative design scenarios.

[0028] S300: Based on manufacturability constraint information and the first and second 3D modeling parameter sets after coordinate alignment, the combination positions of the regular region and the free-form region are combined, and a hybrid 3D model is output. Understandably, after outputting the hybrid 3D model, if downstream tools or services have fixed file format requirements, format normalization is needed. That is, the hybrid 3D model is converted to the corresponding fixed file format according to downstream requirements. For example, if downstream applications require 3D printing, the hybrid 3D model is converted to STL (STereoLithography) format. If downstream applications involve cross-software and cross-system 3D data exchange, it can be converted to the common STEP (Standard for the Exchange of Product model data) format.

[0029] Understandably, after outputting the hybrid 3D model, it is also necessary to detect watertightness defects such as holes, gaps, and overlapping surfaces in the hybrid 3D model, and then repair the watertightness defects to eliminate non-manifold geometry, so as to ensure that the hybrid 3D model after repair is a closed entity that meets the overall watertightness requirements and satisfies the requirements of additive printing.

[0030] In one embodiment, before step S300, that is, before the combination processing of the combination position of the regular region and the freeform region based on manufacturability constraint information and the first and second 3D modeling parameter sets after coordinate alignment, the method includes: performing coordinate alignment processing on the first and second 3D modeling parameter sets in the world coordinate system according to a preset coordinate alignment rule; the preset coordinate alignment rule includes at least one of the following three coordinate alignment methods: A. Obtain the bonding information between adjacent target regions, determine the boundary alignment parameters of the adjacent target regions based on the bonding information, and adjust the first 3D modeling parameter set and / or the second 3D modeling parameter set corresponding to the adjacent target regions based on the boundary alignment parameters; that is, the coordinate alignment method of this scheme is the boundary alignment method. First, read the bonding information of the adjacent target regions, extract the boundary alignment parameters in the bonding information (such as one or more of the following data in the bonding information: interface identifier, orientation and boundary topology information, region identifiers on both sides of the interface, bonding type, interface boundary geometry, boundary normal, cross-section loop point, transition type and fit tolerance, etc.), and adjust the coordinate position of the corresponding first 3D modeling parameter set or second 3D modeling parameter set based on the boundary alignment parameters to achieve boundary fitting and alignment between adjacent target regions.

[0031] B. Obtain the true size information of the target object from the model parameters, determine the scaling transformation coefficient of each target region based on the true size information, and adjust the first 3D modeling parameter set and / or the second 3D modeling parameter set based on the scaling transformation coefficient; the coordinate alignment method of this scheme is the scale scaling alignment method. First, extract the true size information of the target object, calculate the scaling transformation coefficient based on the correspondence between the true size information, and use the scaling transformation coefficient to scale and adjust the first 3D modeling parameter set and / or the second 3D modeling parameter set to unify the actual physical scale of all target regions.

[0032] C. Obtain the location data of landmarks in the reference image for model parameters, the first feature point data corresponding one-to-one with the landmarks in the first 3D modeling parameter set, and the second feature point data corresponding one-to-one with the landmarks in the second 3D modeling parameter set. Based on the correspondence between the location data of the landmarks, the first feature point data, and the second feature point data, perform coordinate transformation adjustment on the first 3D modeling parameter set and / or the second 3D modeling parameter set. The coordinate alignment method of this scheme is the landmark point matching alignment method. First, extract the location data, first feature point data, and second feature point data of the landmarks in the reference image. Relying on the one-to-one mapping relationship of the three sets of data, perform an overall coordinate transformation on the first 3D modeling parameter set and / or the second 3D modeling parameter set to complete the coarse alignment. After coarse alignment, the ICP (Iterative Closest Point) algorithm can be used for fine-tuning to improve the alignment accuracy.

[0033] In one embodiment, such as Figure 2 As shown, in step S300, the process of combining the regular region and the freeform region based on manufacturability constraint information and the first and second 3D modeling parameter sets after coordinate alignment includes the following steps S301-S303: S301, when the combination type of the combination position is the fusion type, obtain the combination information corresponding to the combination position between adjacent regular regions and free-form regions, and determine the transition boundary type according to the combination information; the fusion type refers to the combination type corresponding to the combination position between adjacent regular regions and free-form regions that need to be merged into one. Specifically, if the combination type of the combination position is a fusion type, after completing the coordinate alignment of the first 3D modeling parameter set and the second 3D modeling parameter set, the combination processing of the combination position of the regular region and the free-form region is performed based on the manufacturability constraint information. First, it is necessary to obtain the combination information of the combination position of the regular region and the free-form region, and determine the geometric relationship such as boundary gap, shape, and drop based on the combination information. Then, the transition boundary type is determined according to the above combination relationship. For example, if the boundary relationship indicates that the two sides are approximately coplanar and the gap is very small, the transition boundary type is determined to be a chamfer or rounded corner transition (e.g., the bottom of the animal ornament is connected to the plane of the rectangular base); if the boundary shape is ring-shaped and there is a height difference, the transition boundary type is determined to be a ring collar transition (e.g., a transition ring is placed around the cylindrical interface and the organic surface); if the boundary drop indicates that the difference in the shape of the two sides is large and a smooth connection is required, the transition boundary type is determined to be a loft transition (a loft skin is generated between the two boundary rings).

[0034] S302, starting from one side of the regular region at the joining position, after generating a transition geometric boundary corresponding to the transition boundary type, the free-form region at the joining position is bridged to the transition geometric boundary; a transition geometric boundary matching this type is constructed from one side of the regular region, and then the mesh boundary of the free-form region is connected to the transition geometric boundary to complete the bridging of the two sides; during execution, the boundary structure on one side of the regular region is first locally triangulated, and then a transition geometric boundary corresponding to the above transition boundary type is generated as a connecting bridge to realize the connection between the solid structure boundary on one side of the regular region and the texture mesh boundary on the side of the free-form region. At this time, at the joining position between adjacent regular regions and free-form regions, a continuous solid is formed between the regular region and the free-form region.

[0035] S303, the bridging joint position is constrained and verified according to the manufacturability constraint information. If the constraint verification passes, the joint is confirmed to be successful. Further, the manufacturability constraint information includes at least one of the following: watertightness constraint rules, minimum wall thickness constraint rules, meshless self-intersection rules, and draft constraint rules. In this embodiment, the manufacturability constraint information is used to verify the bridging joint position. If all constraint verifications pass, the region is determined to be successfully joined. That is, the manufacturability constraint information includes at least one of the following four types of verification rules, used to verify whether the geometry of the combined regular region and free-form region is compatible with the target manufacturing process: Watertightness constraint rule: The overall model after successful integration must be a closed manifold structure, without holes or non-manifold edges; Minimum wall thickness constraint rule: The wall thickness at all locations of the model shall not be less than the preset threshold of the corresponding manufacturing process. For example, the minimum wall thickness for FDM (Fused Deposition Modeling) process shall not be less than 0.8 mm. Meshless self-intersection rule: The model mesh does not have intersecting or overlapping surfaces; Draft constraint rules: Control the overhang angle of the model according to the manufacturing process requirements. The overhang angle shall not exceed the set threshold (the set threshold shall be set according to the requirements). If it exceeds the threshold, a support structure shall be added.

[0036] The 3D model generation system in this embodiment will verify the above rules one by one after the successfully combined hybrid 3D model. If all of them are met, it meets the manufacturability requirements and outputs the hybrid 3D model; if any one of them fails to meet the requirements, it is determined that it cannot be directly manufactured.

[0037] In one embodiment, after performing constraint verification on the bridged joint position based on the manufacturability constraint information in step S303, the method further includes the following steps S3031-S3032: S3031, when constraint verification fails, an adjustment operation is determined based on the verification failure information. The verification failure information includes the verification failure location and the verification failure type. In the above embodiment, after obtaining the verification failure type and location from the verification failure information upon constraint verification failure, the verification failure location is spatially associated with at least one of a free-form region, a regular region, or a combined location. The responsible region (at least one of a free-form region, a regular region, or a combined location) to which the verification failure belongs is first determined, and then the processing action corresponding to that responsible region is enabled. It should be noted that the spatial association between the verification failure location and at least one of a free-form region, a regular region, or a combined location can be performed by comparing the coordinates corresponding to the verification failure location with the coordinate intervals corresponding to each region. The region described by the coordinates of the verification failure location is the responsible region. When there are multiple verification failure locations, and different verification failure locations are distributed across regions, the verification failure locations corresponding to each responsible region can be recorded separately, and corresponding adjustment operations can be performed separately, or a main responsible region can be determined according to preset rules, and the adjustment operation can be performed only on that main responsible region in this instance.

[0038] The adjustment operation includes at least one of the following adjustment methods: Adjust the thickness, chamfer radius, overlap, or bridging width of the transition geometry boundary. For example, if the verification failure information includes a joint location, the joint location itself needs to be adjusted. Specifically, if the verification failure type at the joint location is that there is a gap, the watertightness constraint rule is not met, so increase the overlap or bridging width of the transition geometry boundary. If the verification failure type at the joint location is that the wall thickness at the joint location is insufficient, which does not meet the minimum wall thickness constraint rule, thicken the transition geometry boundary and increase the chamfer radius. If the verification failure type at the joint location is that the size corresponding to the transition geometry boundary is too large, resulting in mesh self-intersection, which does not meet the mesh-free self-intersection rule, reduce the overall size of the transition geometry boundary.

[0039] Adjust the 3D mesh information corresponding to the combination position in the second 3D modeling parameter set; that is, when the verification failure position in the verification failure information includes a free-form region, the free-form region needs to be adjusted. At this time, it is prohibited to use parameters unrelated to the free-form region (such as the first 3D modeling parameter set) or interface specifications (i.e., combination information) as adjustment methods. Instead, it is necessary to perform local processing on the part of the 3D mesh information associated with the combination position in the free-form region to achieve mesh regeneration. For example, the above processing method can be achieved by adjusting the mesh generation conditions or changing the manufacturing process (while loading the corresponding constraint configuration for the updated manufacturing process in the manufacturability constraint information to give the constraint verification conclusion without deviating from the process configuration). For example, if the verification failure type is a free-form region, and the corresponding verification failure type is the presence of damaged or intersecting mesh structures, then the local 3D mesh information corresponding to the combination position in the second 3D modeling parameter set is re-subdivided and encrypted to repair the damaged or intersecting mesh structures.

[0040] Adjust the boundary information corresponding to the combination position in the first three-dimensional modeling parameter set; for example, if the verification failure position includes a regular region and the verification failure type corresponding to the regular region is region mismatch, then fine-tune the boundary information corresponding to the combination position in the first three-dimensional modeling parameter set of the regular region to optimize the cooperation relationship between the two regions.

[0041] In the above embodiments, after completing the bridging and binding position constraint verification, subsequent processing is carried out based on the verification failure information, which includes the verification failure location and type. Spatial association is completed by comparing the coordinates of the verification failure location with the coordinate intervals of each region, and the responsibility area corresponding to the verification failure is accurately located. For multiple verification failure locations across regions, the corresponding adjustments can be recorded and executed separately for each region, or the main responsibility area can be selected for adjustment according to preset rules. Different adjustment methods are matched for different responsibility areas. For the binding position, the thickness of the transition geometric boundary, the chamfer radius, the overlap or bridging width can be adjusted to handle constraint problems such as gaps, insufficient wall thickness, and mesh self-intersection. For free-form regions, only the associated local 3D mesh information is locally processed and the mesh is regenerated. The use of irrelevant first 3D modeling parameter sets and interface specifications for repair is prohibited. The manufacturing process can also be changed and the corresponding constraint configuration can be loaded simultaneously. For regular regions, the corresponding boundary information of the first 3D modeling parameter set is finely adjusted to optimize the regional adaptation relationship. After the adjustment is completed, constraint verification is carried out. This allows for the localization of constraint verification failures at the model level to specific responsibility areas. Different responsibility areas only invoke adjustment methods adapted to their own data formats, preventing the use of irrelevant parameters for ineffective repairs. Constraint verification conclusions are always bound to manufacturing process configurations, achieving a correspondence between the location of verification failures, responsibility areas, and adjustment actions. This improves the matching degree between adjustment actions and actual failure areas, and avoids irrelevant parameters running idle.

[0042] S3032, the binding position after the adjustment operation is re-verified for constraints. After the constraint verification is passed, the binding is confirmed to be successful.

[0043] In this embodiment, when the constraint verification fails, the local three-dimensional mesh information of the second three-dimensional modeling parameter set at the joint position on one side of the transition geometric boundary or free-form region, or the boundary information of the first three-dimensional modeling parameter set at the joint position on one side of the regular region, needs to be automatically modified. Then, the constraint verification is carried out again. This process is repeated until the verification passes, and only then is the joint confirmed to be successful. It should be noted that the above-mentioned combination processing is only performed on the combination position where the combination type is fusion. When the combination type of the combination position is assembly type (assembly type refers to the combination type corresponding to the combination position between the regular area and the free form area where there is no need for fusion and the connection relationship is only an assembly relationship such as stacking, such as an ornament placed on a base), the combination processing in S301-S303 above is not performed. Instead, after coordinate alignment, the assembly can be completed directly without interference constraints, ensuring that there is no geometric interference between the regular area and the free form area, no continuous fitting boundary between the two, no transition geometric boundary is generated, and no single-entity Boolean merging is performed on the regular area and the free form area. Instead, the regular area and the free form area are positioned according to the combination information, and the object identities of different entities (regular area and free form area) are preserved. Only placement assembly is performed, and then only contact stability verification is performed. The contact stability verification includes support contact area verification (the effective contact area is not less than the minimum support area threshold of the process to avoid local single-point stress and tipping, and the minimum support area threshold of the process is set according to requirements), center of gravity projection verification (the vertical projection of the center of gravity of the overall model falls inside the contact support area, and there is no situation of center of gravity drifting outward and becoming unstable), overturning moment verification (the overturning moment generated by the model's own gravity is less than the anti-overturning moment corresponding to the maximum static friction force of the contact surface), no suspended interference verification (there are no local suspended or intersecting geometric structures after assembly), and load support thickness verification (the wall thickness of the model at the contact position meets the minimum wall thickness manufacturing constraint and will not break or deform after being subjected to force).

[0044] In this invention, the solution performs differentiated processing at the junction of fusion type and assembly type, using the junction information corresponding to the interface specification to directly participate in the combination processing of the 3D model. When the junction type is fusion type, the junction information of adjacent regular regions and free-form regions is obtained to determine the transition boundary type. A corresponding transition geometric boundary is generated from one side of the regular region, and the free-form region side is bridged to the transition geometric boundary. Then, constraint verification is completed based on manufacturability constraint information. Once the verification is passed, the junction is completed, ensuring that the integrated fusion geometry meets manufacturing requirements. When the junction type is assembly type, the above fusion junction processing is skipped. Placement assembly is completed under the condition of coordinate alignment and no geometric interference. No transition geometric boundary is generated, and no single-unit Boolean merging is performed. Positioning is completed based on the junction information, and the independent object identity of each region is preserved. Only contact stability verification is performed. Therefore, the same set of regional data (the first 3D modeling parameter set and the second 3D modeling parameter set) can output continuous entities or multi-entity assembly structures according to the actual combination relationship, effectively avoiding the assembly relationship being mistakenly processed as a permanent fusion relationship, so that the subsequent inspection rules can match the entity organization method of the model, and retain the component identity during the multi-object output process, which is convenient for subsequent processing of each component separately, taking into account both the manufacturability of fusion modeling and the entity independence of assembly modeling.

[0045] In this embodiment, regular regions and free-form regions are divided according to the model parameters in the target object, and the regular regions and free-form regions are modeled in different ways. The regular regions are sent to the CAD engine for processing to obtain a first 3D modeling parameter set containing boundary and size information, while the free-form regions are sent to the neural image generation engine for processing to obtain a second 3D modeling parameter set containing 3D mesh and texture information. The first and second 3D modeling parameter sets are not converted to each other but are saved independently. Then, based on manufacturability constraints, the combined position of the first and second 3D modeling parameter sets after coordinate alignment is combined to ensure the manufacturability of the final hybrid 3D model. In the hybrid 3D model, the free-form regions have high fidelity and photorealistic appearance, while the positions corresponding to the regular regions have editable parameters and accurate dimensions. Thus, this invention simultaneously considers model fidelity, parameter accuracy, and manufacturability. Furthermore, in the hybrid three-dimensional model obtained by the present invention, the first three-dimensional modeling parameters or the second three-dimensional modeling parameters can be adjusted or updated separately to locally regenerate or update regular or free-form regions, which can be integrated with the engineering capabilities of additive manufacturing.

[0046] In one embodiment, after step S300, that is, after outputting the hybrid 3D model, the following steps S400-S700 are further included: S400, perform multi-viewpoint rendering on the hybrid 3D model to obtain multiple rendering views corresponding to different viewpoints; that is, set up multiple virtual observation viewpoints on the hybrid 3D model and generate multiple corresponding rendering views by rendering from different angles.

[0047] S500: Based on the preset view specification information, structured difference identification is performed on all the rendered views to obtain the difference data corresponding to each target area. That is, all rendered views and the preset view specification information are input into the multimodal model for comparative analysis to generate the difference data corresponding to each target area. The difference data includes, but is not limited to, scale differences, position differences, orientation differences, and shape deviations. The preset view specification information consists of pre-saved target reference materials, serving as a benchmark for judging the appearance of the hybrid 3D model in the rendered view. It includes, but is not limited to, reference materials (standard renderings, design drawings, multi-angle images of reference objects, and shape description text) and constraint requirements (standard scale, outline, component position, shape, and other appearance specifications of the model from each viewpoint). The multimodal model compares the preset view specification information with the rendered views to extract the difference data corresponding to each target area in the rendered views. The training process of the multimodal model is as follows: training sample groups are collected. Each training sample group includes a reference standard image and its corresponding text description information, as well as multi-angle rendered images with deviations. The scale, position, orientation, and shape difference labels of each area corresponding to the training sample group are manually labeled. A dual-branch network is constructed: an image encoder extracts visual features from multi-angle rendered images and reference standard images; the other text encoder parses text features from text descriptions. The two feature dimensions are unified by a combination layer. Using manually labeled difference labels as the target, the error between the network's predicted difference data and the difference labels is calculated, and a loss is constructed for parameter updates. Gradient backpropagation iteratively updates the network weights, and the loop iterates until the loss converges, resulting in a multimodal model with a suspended chain.

[0048] S600, the target area whose difference data meets the preset difference standard is determined as the area to be corrected. The generation conditions of the second 3D modeling parameter set and / or the first 3D modeling parameter set of the area to be corrected are adjusted according to preset adjustment conditions. The preset adjustment conditions include: the difference data of the adjusted area to be corrected is reduced in more than half of the rendered views. The preset difference standard is a pre-set visual deviation judgment threshold used to measure whether the deviations in shape, size, position, etc., between the rendered view and the standard model exceed an acceptable range. For example, if the difference data of the target area is less than the preset difference standard, it means that the shape of the target area in the rendered view meets the standard and does not need adjustment; if the difference data of the target area is greater than or equal to the preset difference standard, it is determined that there is a significant deviation and is listed as an area to be corrected. The preset difference standard can be customized according to processing accuracy and appearance requirements; the higher the appearance requirements, the smaller the preset difference standard is set.

[0049] Specifically, target areas where the difference data exceeds the preset difference standard are selected as areas to be corrected. The first 3D modeling parameter set and / or the second 3D modeling parameter set corresponding to the area to be corrected are adjusted according to the adjustment target corresponding to the preset adjustment conditions. The adjustment target is to reduce the difference data of the area to be corrected in more than half of the rendering views, that is, to reduce the deviation in more than half of the rendering views. Specifically, if the area to be corrected is a regular area, since the size information, spatial position constraint parameters, etc. of the regular area are all included in the first 3D modeling parameter set, the shape of the regular area can be directly changed synchronously by modifying the size information, spatial position constraint parameters, etc. Therefore, at this time, only the size information, spatial position constraint parameters, etc. in the first 3D modeling parameter set need to be adjusted to achieve the adjustment of the difference data. If the region to be corrected is a free-form region, since the second 3D modeling parameter set output by the neural image generation engine contains 3D mesh and texture information, and the 3D mesh and texture information are inference outputs without directly editable size parameters, there are no independently adjustable driving parameters. Therefore, the generated second 3D modeling parameter set cannot be directly modified. To change the 3D mesh and texture information of the free-form region, the initial generation conditions input to the neural image generation engine (including reference images and text information) must be adjusted, allowing the neural image generation engine to re-output a new 3D modeling parameter set based on the adjusted new generation conditions. In this embodiment, only the regions to be corrected whose difference data meets the preset difference standard can be adjusted, while the remaining target regions can remain unchanged, reducing the amount of data processing.

[0050] The preset adjustment conditions are as follows: the difference data corresponding to the area to be corrected after adjustment is reduced in more than half of the rendered views; that is, the change in the difference data of the rendered views of all viewpoints before and after the adjustment is predicted, and the adjustment plan is adopted only when the difference data corresponding to more than half of the rendered views is reduced and the overall effect is improved; if the difference data corresponding to the rendered views of only a few viewpoints is optimized, but the difference data corresponding to the rendered views of most other viewpoints is increased, the adjustment plan is not adopted, so as to avoid global distortion caused by single-angle optimization.

[0051] S700 outputs an optimized 3D model when all difference data meet the preset difference criteria or the number of adjustments exceeds the preset threshold (the preset threshold can be set according to requirements). That is, after making effective adjustments that meet the preset adjustment conditions, multi-view rendering and difference identification are performed again, and the process is continuously iterated. When all difference data meet the preset difference criteria or the number of iterative adjustments reaches the preset threshold, the process terminates and finally outputs the optimized 3D model.

[0052] In this embodiment, differentiable rendering is not required, nor is gradient backpropagation performed. The rendered view is obtained only through conventional multi-view rendering. Based on the gap data corresponding to the intuitive visual deviation output by the multimodal model, the generation conditions of the second 3D modeling parameter set and / or the first 3D modeling parameter set are directly adjusted. There is no need to build a differential calculation and error backpropagation link, nor is there any need to change the underlying weights of the CAD engine and the neural image generation engine. Existing general tools can be directly reused, the overall architecture is lightweight, and the deployment difficulty is lower.

[0053] In one embodiment, after step S300, that is, after outputting the hybrid 3D model, the method further includes: Receive an update command containing update parameters for the target region; after receiving the update command, it is first necessary to determine whether the target region is a regular region or a free-form region.

[0054] When the target region is a regular region, the CAD engine regenerates the regular region according to the update parameters to obtain a third 3D modeling parameter set. Based on manufacturability constraint information, and the third 3D modeling parameter set and the second 3D modeling parameter set after coordinate alignment, the combination position of the freeform region and the regenerated regular region is processed, and a first updated 3D model is output. That is, if the target region is a regular region, the regular region is regenerated using the update parameters to obtain a third 3D modeling parameter set. In the above process, only the regular region is regenerated, and then the coordinate alignment is re-performed. The new combination position is re-verified according to the manufacturability constraint information, while the original freeform region in the hybrid 3D model remains unchanged, and the first updated 3D model is finally output. It should be noted that in this step, before re-aligning the coordinates and re-verifying the new combination position according to the manufacturability constraint information, it is necessary to compare the second 3D modeling parameter sets corresponding to the freeform region in the hybrid 3D model before the update of the regular region and the freeform region in the hybrid 3D model after the update of the regular region, respectively, to determine that the original freeform region in the hybrid 3D model remains unchanged. If, after comparison, it is confirmed that the second 3D modeling parameter set corresponding to the update rule area has changed, it indicates that the update process has touched data that should not be modified. The update should be rejected or marked as abnormal, and an update failure message corresponding to the update instruction should be generated. Conversely, if, after comparison, it is confirmed that the second 3D modeling parameter set corresponding to the update rule area is the same, it indicates that the update process has not touched data that should not be modified. Coordinate alignment can be re-performed, and the new bonding position can be re-verified based on manufacturability constraints.

[0055] When the target region is a freeform region, the neural image generation engine regenerates the freeform region according to the updated parameters to obtain a fourth 3D modeling parameter set. Based on manufacturability constraint information, and the fourth 3D modeling parameter set after coordinate alignment, and the first 3D modeling parameter set, the combination position of the regular region and the regenerated freeform region is processed, and a second updated 3D model is output. That is, if the target region is a freeform region, the freeform region is regenerated according to the update conditions to obtain a fourth 3D model parameter set. In the above process, only the freeform region is regenerated, the coordinate alignment is re-performed, and the new combination position is re-verified according to the manufacturability constraint information, while the original regular region in the hybrid 3D model remains unchanged, and the second updated 3D model is finally output. It should be noted that in this step, before re-aligning the coordinates and re-verifying the new combination position according to the manufacturability constraint information, it is necessary to compare the first 3D modeling parameter sets corresponding to the millet-shaped region in the hybrid 3D model before regenerating the freeform region and the regular region in the hybrid 3D model after regenerating the freeform region, respectively, to determine that the original regular region in the hybrid 3D model remains unchanged. If, after comparison, it is confirmed that the first 3D modeling parameter set corresponding to the update rule area has changed, it indicates that the update process has touched data that should not be modified. The update should be rejected or marked as abnormal, and an update failure message corresponding to the update instruction should be generated. Conversely, if, after comparison, it is confirmed that the first 3D modeling parameter set corresponding to the update rule area is the same, it indicates that the update process has not touched data that should not be modified. Coordinate alignment can be re-performed, and the new bonding position can be re-verified based on manufacturability constraints.

[0056] For example, if the hybrid 3D model consists of an animal figurine placed on a rectangular base, where the base is a regular region and the animal figurine is a free-form region, when it's necessary to modify the diameter of the through-hole on the rectangular base (e.g., changing the diameter from 5mm to 6mm), only the regular region (i.e., the rectangular base) needs to be locally updated and regenerated. The 3D mesh and texture information of the free-form region corresponding to the animal figurine are completely preserved and undisturbed. Afterward, only coordinate alignment and re-merging of the joint positions are needed to output the corresponding updated 3D model, achieving efficient local editing.

[0057] In this embodiment, only one type of target region in the hybrid 3D model needs to be regenerated, while the other type of target region can remain completely unchanged. Therefore, the two types of regions are decoupled and do not interfere with each other. There is no need to regenerate the entire hybrid 3D model, significantly saving computing power and time. This method offers high modification efficiency, low computational overhead, and avoids distortion of the existing design. Furthermore, after one type of target region is regenerated, the spatial position and orientation of the other type of target region remain unchanged throughout the process. Therefore, there is no need for overall realignment; instead, the original alignment and verification logic can be reused without repeating the entire process. This solution is flexible and easy to modify later.

[0058] Furthermore, after outputting the first or second updated 3D model, the differences between the models before and after the update can be compared. That is, the differences between the first or second updated 3D model and the hybrid 3D model can be compared, and it can be determined whether the difference falls within a regular region, a free-form region, or a combined region. The changes in all parameter sets before and after the update, as well as the correspondence between the difference and the region, can be recorded to retain the source records of parameter changes and 3D model ratio changes for subsequent reference or use.

[0059] This invention combines the advantages of modeling, simultaneously preserving the parametric structure corresponding to the first 3D modeling parameter set output by the CAD engine and the texture mesh corresponding to the second 3D modeling parameter set output by the neural image generation engine. This results in organically reproducing realistic textures, with precise dimensions for standard parts and support for parameter editing—an effect impossible with a single engine. Furthermore, the first and second 3D modeling parameter sets are mutually independent, eliminating the need to reverse-engineer the organic surfaces of freeform regions into solid models corresponding to regular regions, thus avoiding shape distortion caused by reverse modeling and preventing modeling inaccuracies. In addition, the regular regions retain complete boundary and dimensional information, allowing for dimensional tolerance annotation and ensuring model watertightness, enabling direct use in additive manufacturing such as 3D printing for processing and manufacturing. Secondly, this invention supports independent local editing, allowing for the regeneration and / or adjustment of only one type of target region while leaving the model portion corresponding to another type of target region untouched. This results in low computational load, stable modeling, and allows for iterative modification. This invention can complete appearance correction without differentiable rendering, has strong adaptability and robustness, strong engineering practicality, and wide applicability. It can be used for modeling needs where organic shapes and precise structures coexist, such as cultural and creative figurines, medical devices, film and television digital assets, and industrial composite parts.

[0060] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0061] In one embodiment, such as Figure 3 As shown, a three-dimensional model generation system is provided, including: The receiving module 11 is used to receive a three-dimensional model generation instruction containing model parameters of a target object, and to divide the target object into regions according to the model parameters to obtain multiple target regions, wherein the target regions include at least one regular region and at least one free-form region. Modeling module 12 is used to send all the regular regions to the CAD engine to obtain a first three-dimensional modeling parameter set containing boundary information and size information corresponding to the regular regions, and to send all the free-form regions to the neural image generation engine to obtain a second three-dimensional modeling parameter set containing three-dimensional mesh information and texture information corresponding to the free-form regions. The model generation module 13 is used to combine the combination positions of the regular region and the free-form region based on manufacturability constraint information, as well as the first three-dimensional modeling parameter set and the second three-dimensional modeling parameter set after coordinate alignment, and output a hybrid three-dimensional model.

[0062] In the 3D model generation system of this invention, regular regions and free-form regions are divided according to the model parameters in the target object. These regions are then modeled in different ways. The regular regions are sent to a CAD engine for processing to obtain a first 3D modeling parameter set containing boundary and dimension information. The free-form regions are sent to a neural image generation engine for processing to obtain a second 3D modeling parameter set containing 3D mesh and texture information. The first and second 3D modeling parameter sets are not converted to each other but are saved independently. Then, based on manufacturability constraints, the combined position of the first and second 3D modeling parameter sets after coordinate alignment is combined to ensure the manufacturability of the final hybrid 3D model. In this hybrid 3D model, the free-form regions possess high fidelity and photorealistic appearance, while the corresponding positions of the regular regions have editable parameters and accurate dimensions. Thus, this invention simultaneously achieves high model fidelity, parameter accuracy, and manufacturability.

[0063] Specific limitations regarding the 3D model generation system can be found in the limitations of the 3D model generation method described above, and will not be repeated here. The 3D model generation system may include a controller for executing the 3D model generation method. The various modules within the controller of the aforementioned 3D model generation system may be implemented entirely or partially through software, hardware, or a combination thereof. These modules may be embedded in hardware or independent of the processor in the computer device, or stored in software in the memory of the computer device, so that the processor can call and execute the operations corresponding to each module.

[0064] Furthermore, the 3D model generation system also includes a CAD engine and a neural image generation engine that communicate with the controller. Understandably, in this invention, the CAD engine and the neural image generation engine can be deployed entirely locally, entirely in the cloud, or one can be deployed locally and the other in the cloud; the CAD engine and the neural image generation engine can also be in different processes, hosts, or security domains.

[0065] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes a readable storage medium and internal memory. The readable storage medium stores an operating system, computer-readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the readable storage medium. The database stores data related to the 3D model generation method. The network interface communicates with external terminals via a network connection. When the computer-readable instructions are executed by the processor, a 3D model generation method is implemented. The readable storage medium provided in this embodiment includes both non-volatile and volatile readable storage media.

[0066] In one embodiment, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor executes the computer-readable instructions to implement the steps of the three-dimensional model generation method.

[0067] In one embodiment, one or more computer-readable storage media storing computer-readable instructions are provided. The computer-readable storage media provided in this embodiment include non-volatile readable storage media and volatile readable storage media. The computer-readable instructions stored on the readable storage media implement the steps of the three-dimensional model generation method when executed by one or more processors.

[0068] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a non-volatile readable storage medium or a volatile readable storage medium. When executed, these computer-readable instructions can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0069] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0070] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for generating a three-dimensional model, characterized in that, include: The system receives a 3D model generation instruction containing model parameters of a target object, and divides the target object into regions according to the model parameters to obtain multiple target regions, wherein the target regions include at least one regular region and at least one free-form region. All the regular regions are sent to the CAD engine to obtain a first three-dimensional modeling parameter set containing boundary information and size information corresponding to the regular regions, and all the free-form regions are sent to the neural image generation engine to obtain a second three-dimensional modeling parameter set containing three-dimensional mesh information and texture information corresponding to the free-form regions. Based on manufacturability constraint information, and the first and second 3D modeling parameter sets after coordinate alignment, the combination position of the regular region and the free-form region is combined to output a hybrid 3D model. The process of combining the regular region and the freeform region at their respective junctions, based on manufacturability constraint information and the first and second 3D modeling parameter sets after coordinate alignment, includes: When the combination type at the combination location is the fusion type, the combination information corresponding to the combination location between adjacent regular regions and free-form regions is obtained, and the transition boundary type is determined based on the combination information; the fusion type refers to the combination type corresponding to the combination location between adjacent regular regions and free-form regions that need to be merged into one. Starting from one side of the regular region at the bonding location, after generating a transition geometric boundary corresponding to the transition boundary type, the free-form region at the bonding location is bridged to the transition geometric boundary. The bonding positions after bridging are constrained and verified according to the manufacturability constraint information. After the constraint verification is passed, the bonding is confirmed to be successful.

2. The three-dimensional model generation method according to claim 1, characterized in that, The step of dividing the target object into regions based on the model parameters yields multiple target regions, including: Extract text semantic tags from the text information of the model parameters; Geometric feature information is determined based on textual information and / or reference images in the model parameters, wherein the geometric feature information includes at least one of curvature distribution information of the target object, parameterizability, and manufacturing feature matching degree. Based on the text semantic tags and the geometric feature information, the region is divided into regular regions and free-form regions, and the combination information between adjacent target regions is output.

3. The three-dimensional model generation method according to claim 1, characterized in that, Before combining the regular region and the freeform region at their respective locations based on manufacturability constraint information and the first and second 3D modeling parameter sets after coordinate alignment, the process includes: In the world coordinate system, the first 3D modeling parameter set and the second 3D modeling parameter set are aligned according to a preset coordinate alignment rule; the preset coordinate alignment rule includes at least one of the following coordinate alignment methods: Obtain the combination information between adjacent target regions, determine the boundary alignment parameters of adjacent target regions based on the combination information, and adjust the first 3D modeling parameter set and / or the second 3D modeling parameter set corresponding to the adjacent target regions based on the boundary alignment parameters; Obtain the true size information of the target object from the model parameters, determine the scaling transformation coefficient of each target region based on the true size information, and adjust the first three-dimensional modeling parameter set and / or the second three-dimensional modeling parameter set based on the scaling transformation coefficient; The system acquires the location data of landmarks in the reference image of the model parameters, the first feature point data corresponding to each landmark in the first 3D modeling parameter set, and the second feature point data corresponding to each landmark in the second 3D modeling parameter set. Based on the correspondence between the location data of the landmarks, the first feature point data, and the second feature point data, the system performs coordinate transformation adjustment on the first 3D modeling parameter set and / or the second 3D modeling parameter set.

4. The three-dimensional model generation method according to claim 1, characterized in that, After performing constraint verification on the bridged joint position based on the manufacturability constraint information, the method further includes: When constraint verification fails, an adjustment operation is determined based on the verification failure information. The adjustment operation includes: adjusting the thickness, chamfer radius, overlap or bridging width of the transition geometry boundary; and / or adjusting the 3D mesh information corresponding to the joint position in the second 3D modeling parameter set; and / or adjusting the boundary information corresponding to the joint position in the first 3D modeling parameter set. The combined position after the adjustment operation is re-verified for constraints. After the constraint verification is passed, the combination is confirmed to be successful.

5. The three-dimensional model generation method according to claim 1, characterized in that, The output hybrid 3D model also includes: The hybrid 3D model is rendered from multiple viewpoints to obtain multiple rendered views corresponding to different viewpoints. Based on the preset view specification information, the structured differences of all the rendered views are identified to obtain the difference data corresponding to each target area; The target area whose difference data meets the preset difference standard is determined as the area to be corrected. The generation conditions of the second three-dimensional modeling parameter set and / or the first three-dimensional modeling parameter set of the area to be corrected are adjusted according to the preset adjustment conditions. The preset adjustment conditions include: the difference data of the area to be corrected after adjustment is reduced in more than half of the rendered views. When all discrepancies meet the preset discrepancy standards or the number of adjustments exceeds the preset threshold, an optimized 3D model is output.

6. The three-dimensional model generation method according to claim 1, characterized in that, The output hybrid 3D model also includes: Receive an update command containing update parameters for the target region; When the target area is a regular area, the CAD engine is instructed to regenerate the regular area according to the updated parameters to obtain a third 3D modeling parameter set. Based on the manufacturability constraint information, and the third 3D modeling parameter set and the second 3D modeling parameter set after coordinate alignment, the combination position of the free-form area and the regenerated regular area is processed, and the first updated 3D model is output. When the target region is a free-form region, the neural image generation engine regenerates the free-form region according to the updated parameters to obtain a fourth 3D modeling parameter set. Based on manufacturability constraint information, and the fourth 3D modeling parameter set and the first 3D modeling parameter set after coordinate alignment, the combination position of the regular region and the regenerated free-form region is combined and processed to output a second updated 3D model.

7. A three-dimensional model generation system, characterized in that, include: The receiving module is used to receive a 3D model generation instruction containing model parameters of a target object, and to divide the target object into regions according to the model parameters to obtain multiple target regions, wherein the target regions include at least one regular region and at least one free-form region. The modeling module is used to send all the regular regions to the CAD engine to obtain a first three-dimensional modeling parameter set containing boundary information and size information corresponding to the regular regions, and to send all the free-form regions to the neural image generation engine to obtain a second three-dimensional modeling parameter set containing three-dimensional mesh information and texture information corresponding to the free-form regions. The model generation module is used to combine the junctions of the regular regions and the freeform regions based on manufacturability constraint information and the first and second 3D modeling parameter sets after coordinate alignment, and output a hybrid 3D model. The model generation module is also used to obtain the combination information corresponding to the junctions between adjacent regular regions and freeform regions when the combination type at the junction is a fusion type, and determine the transition boundary type based on the combination information. The fusion type refers to the combination type corresponding to the junction between adjacent regular regions and freeform regions that need to be merged into one. Starting from one side of the regular region at the junction, a transition geometric boundary corresponding to the transition boundary type is generated, and then the freeform region side of the junction is bridged to the transition geometric boundary. The bridged junction is constrained and verified based on the manufacturability constraint information, and the combination is confirmed to be successful after the constraint verification is passed.

8. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, characterized in that, When the processor executes the computer-readable instructions, it implements the three-dimensional model generation method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing computer-readable instructions, characterized in that, When the computer-readable instructions are executed by one or more processors, the one or more processors cause the three-dimensional model generation method as described in any one of claims 1 to 6 to be performed.

Images