A method, device and equipment for self-training of a three-dimensional modeling model, and a storage medium

Abstract

The application discloses a three-dimensional modeling model self-training method and device, equipment and storage medium, the method comprises the steps of obtaining a first modeling instruction; input the first modeling instruction into a large language model, and combine the calling functions of all target basic modeling modules through the large language model to obtain a modeling program for constructing a target building object; execute the modeling program to obtain a three-dimensional model of the target building object; train the large language model with the modeling program and the three-dimensional model as a set of training data, and obtain a three-dimensional modeling model capable of outputting a corresponding modeling program based on an input three-dimensional model; the application can enable the model to obtain training data and perform training, improve the efficiency of training a large model in the field of architecture, and reduce the labor cost.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, device, and storage medium for self-training of a 3D modeling model. Background Technology

[0002] In training large-scale models in the architectural field, a large amount of training data is required. This data can come from the internet, be generated programmatically, or be from publicly available academic datasets. However, acquiring training data is very costly and requires a significant amount of time. For example, technicians might scrape data from the DeepCAD dataset online, or they might write programs to use it as training data. Furthermore, generating architectural data programmatically is extremely inefficient and unsuitable for training models in the architectural field.

[0003] Therefore, how to efficiently and cost-effectively train 3D modeling models without relying on additional manual data has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, device, and storage medium for self-training of 3D modeling models, which can efficiently and cost-effectively train 3D modeling models without relying on additional manual data.

[0005] According to one aspect of the present invention, a method for self-training a 3D modeling model is provided, the method comprising: Obtain a first modeling instruction; wherein the first modeling instruction includes a function that calls one or more target basic modeling modules selected from a preset building modeling library; The first modeling instruction is input into the large language model, and the calling functions of all target basic modeling modules are combined and processed by the large language model to obtain a modeling program for constructing the target building object; The modeling procedure is executed to obtain a three-dimensional model of the target building object; The modeling program and the 3D model are used as a set of training data to train the large language model, resulting in a 3D modeling model that can output the corresponding modeling program based on the input 3D model.

[0006] Optionally, after inputting the first modeling instruction into the large language model and combining the calling functions of all target basic modeling modules through the large language model to obtain a modeling program for constructing the target building object, the method further includes: Determine whether the modeling program can be executed normally. If not, send the program error message to the large language model to instruct the large language model to regenerate the modeling program. Determine whether the 3D model obtained after the normal execution of the modeling program is a complete contour model. If not, send the model error information to the large language model to instruct the large language model to regenerate the modeling program. Determine whether the 3D model with a complete contour volume model contains all target basic modeling modules. If not, send module error information to the large language model to instruct the large language model to regenerate the modeling program.

[0007] Optionally, executing the modeling procedure to obtain a three-dimensional model of the target building object includes... Replace all geometric values ​​contained in the modeling program with variable parameters; Multiple new geometric values ​​are randomly generated for each variable parameter according to the preset variable value range, thereby obtaining multiple new modeling programs; Each of the resulting modeling procedures is executed sequentially to create multiple three-dimensional models for the target building object.

[0008] Optionally, executing the modeling procedure to obtain a three-dimensional model of the target building object includes: Obtain a second modeling instruction; wherein the second modeling instruction includes multiple modeling programs for different building objects; The second modeling instruction is input into the large language model, and the large language model is used to learn the multiple modeling programs to obtain a modeling program for constructing new building objects; Each modeling procedure is executed sequentially to create multiple 3D models for different architectural objects.

[0009] Optionally, the step of training the large language model using the modeling program and the 3D model as a set of training data to obtain a 3D modeling model capable of outputting the corresponding modeling program based on the input 3D model includes: The three-dimensional models in each training data are traversed sequentially, and the modal information parsed from the currently traversed three-dimensional model is input into the large language model to obtain the prediction modeling program; wherein, the modal information includes at least one of the following: image, point cloud information, geometric information, and topological information; The prediction modeling procedure is executed to obtain a prediction model, and the similarity between the prediction model and the currently traversed 3D model is calculated to obtain a similarity score. The large language model is adjusted based on the similarity score to obtain the three-dimensional modeling model.

[0010] Optionally, the step of calculating the similarity score between the predicted model and the currently traversed 3D model includes: The first total number of each preset geometric feature in the prediction model is calculated, and the second total number of each preset geometric feature in the currently traversed 3D model is calculated; wherein, the preset geometric features include: number of vertices, number of edges, number of faces, and number of rings; The similarity score of the first total quantity and the second total quantity is calculated using a similarity algorithm.

[0011] To achieve the above objectives, the present invention also provides a device for self-training a 3D modeling model, the device comprising: An acquisition module is used to acquire a first modeling instruction; wherein the first modeling instruction includes a function to call one or more target basic modeling modules selected from a preset building modeling library; A construction module is used to input the first modeling instruction into the large language model, and to combine the calling functions of all target basic modeling modules through the large language model to obtain a modeling program for constructing the target building object; The execution module is used to execute the modeling program to obtain a three-dimensional model of the target building object; The training module is used to train the large language model by using the modeling program and the 3D model as a set of training data, so as to obtain a 3D modeling model that can output the corresponding modeling program based on the input 3D model.

[0012] Optionally, the device further includes a determination module, used for: Determine whether the modeling program can be executed normally. If not, send the program error message to the large language model to instruct the large language model to regenerate the modeling program. Determine whether the 3D model obtained after the normal execution of the modeling program is a complete contour model. If not, send the model error information to the large language model to instruct the large language model to regenerate the modeling program. Determine whether the 3D model with a complete contour volume model contains all target basic modeling modules. If not, send module error information to the large language model to instruct the large language model to regenerate the modeling program.

[0013] To achieve the above objectives, the present invention also provides a computer device, which specifically includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the self-training method for three-dimensional modeling described above.

[0014] To achieve the above objectives, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the self-training method for three-dimensional modeling described above.

[0015] The present invention provides a method, apparatus, device, and storage medium for self-training of a 3D modeling model. This involves: acquiring a first modeling instruction; inputting the first modeling instruction into a large language model; combining the call functions of all target basic modeling modules through the large language model to obtain a modeling program for constructing a target building object; executing the modeling program to obtain a 3D model of the target building object; and training the large language model with the modeling program and the 3D model as a set of training data to obtain a 3D modeling model capable of outputting a corresponding modeling program based on the input 3D model. This improves the efficiency of training large models in the architectural field and reduces labor costs. Attached Figure Description

[0016] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a schematic diagram of an optional process for the self-training method of the 3D modeling model provided in Embodiment 1; Figure 2 This is a schematic diagram of an example modeling program provided in Implementation Example 1; Figure 3 A schematic diagram of data generalization provided in Example 1; Figure 4 This is a schematic diagram of the system architecture provided in Embodiment 1; Figure 5 A schematic diagram of an optional component structure of the device for self-training of a 3D modeling model provided in Embodiment 2; Figure 6 This is a schematic diagram of an optional hardware structure for the computer device provided in Embodiment 3. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0018] Example 1 This invention provides a method for self-training of a 3D modeling model, such as... Figure 1 As shown, the method specifically includes the following steps: S101: Obtain a first modeling instruction; wherein the first modeling instruction includes a function that calls one or more target basic modeling modules selected from a preset building modeling library.

[0019] In this embodiment, the preset architectural modeling library is a commonly used modeling library in 3D modeling design, such as the CADquery modeling library and the Build123d modeling library. Preferably, the CADquery modeling library is used in this embodiment. The calling functions of the target basic modeling module are the modeling APIs in the CADquery modeling library. The APIs in the CADquery modeling library are traversed sequentially to determine the APIs used to build the modeling program. The APIs in the CADquery modeling library include: basic geometry creation API, sketching API, transformation operation API, Boolean operation API, feature operation API, selection and filtering API, array operation API, sketch constraint API, position API, etc. The user selects the API to be used from these APIs used to build the 3D modeling program and generates the first modeling instruction, that is, the description statement of the target architectural object that the user wants to build. For example, generating a 3D modeling program for doors and windows based on CADquery, which includes basic geometry creation API and transformation operation API.

[0020] S102: Input the first modeling instruction into the large language model, and combine the calling functions of all target basic modeling modules through the large language model to obtain a modeling program for constructing the target building object.

[0021] In this embodiment, the large language model is either deepseek-r1 or qwen3-coder. The large language model parses the natural language input by the user and uses the parsed information to call the API in the CADquery modeling library to generate a 3D modeling program.

[0022] S103: Execute the modeling procedure to obtain a three-dimensional model of the target building object.

[0023] In this embodiment, the 3D modeling program is run on the 3D modeling platform to generate a 3D model of the corresponding target building object and obtain the modal information of the 3D model. The modal information includes images, point cloud information, brep geometry information, model topology information, etc. related to the 3D model.

[0024] S104: The modeling program and the 3D model are used as a set of training data to train the large language model, so as to obtain a 3D modeling model that can output the corresponding modeling program based on the input 3D model.

[0025] In this embodiment, all qualified 3D modeling programs and their corresponding 3D models are used as training data to train the large language model. During training, the modal information corresponding to the 3D model is used as input data for the large language model. The large language model parses this modal information, then generates a predictive modeling program based on the API corresponding to the modal information, and determines the similarity between the predicted model generated by the predictive modeling program and the 3D model. Finally, the large language model is adjusted based on this similarity to obtain the 3D model. The modal information includes images, point cloud information, BREP volume geometry information, model topology information, etc., related to the 3D model.

[0026] In this embodiment, a first modeling instruction is obtained; the first modeling instruction is input into a large language model, and the calling functions of all target basic modeling modules are combined and processed by the large language model to obtain a modeling program for constructing a target building object; the modeling program is executed to obtain a three-dimensional model of the target building object; the modeling program and the three-dimensional model are used as a set of training data to train the large language model, thereby obtaining a three-dimensional modeling model that can output the corresponding modeling program based on the input three-dimensional model, which improves the efficiency of training large models in the field of architecture and reduces labor costs.

[0027] Specifically, after inputting the first modeling instruction into the large language model in step S102, and combining the calling functions of all target basic modeling modules through the large language model to obtain a modeling program for constructing the target building object, the method further includes: Step A1: Determine whether the modeling program can be executed normally. If not, send the program error message to the large language model to instruct the large language model to regenerate the modeling program. Step A2: Determine whether the 3D model obtained after the normal execution of the modeling program is a complete contour model. If not, send the model error information to the large language model to instruct the large language model to regenerate the modeling program. Step A3: Determine whether the 3D model with a complete contour model contains all target basic modeling modules. If not, send the module error information to the large language model to instruct the large language model to regenerate the modeling program.

[0028] In this embodiment, the 3D modeling program generated by the large language model undergoes three levels of verification. The first level verifies the executability of the 3D modeling program. If the program fails to run correctly, a syntax error or logic error message is generated. This message, along with the first modeling instruction, is input to the large language model, and the 3D modeling program is regenerated. The second level verifies the 3D model generated after the program runs successfully. It determines whether the brep model is a complete and closed 3D geometry. If the geometry is incomplete, a missing information message is input to the large language model along with the first modeling instruction, and the 3D modeling program is regenerated. The third level verifies the 3D modeling program corresponding to the complete geometry. It determines whether the program contains the target API desired by the user. If the target API is not included, a message is generated, and the first modeling instruction is input to the large language model, and the 3D modeling program is regenerated. If any of the three verification methods fails, the 3D modeling program is regenerated, and the new program undergoes three levels of verification until a qualified 3D modeling program is generated that passes all three levels of verification. By using a large language model to iterate data and generate qualified 3D modeling programs, manual data processing is reduced and efficiency is improved.

[0029] In addition, when the number of verifications reaches a preset threshold, even multiple iterations cannot generate a qualified 3D modeling program. At this point, the system will retrieve relevant examples of the API that the user wants from the CADquery modeling library. These relevant examples will generate a project document. This project document and the first modeling instruction will be input into the large language model at the same time. The large language model will learn from the examples in the project document and regenerate a qualified 3D modeling program for that API after learning.

[0030] Specifically, in step S103, executing the modeling procedure to obtain a three-dimensional model of the target building object includes: Step B1: Replace all geometric values ​​contained in the modeling program with variable parameters; Step B2: Randomly generate multiple new geometric values ​​for each variable parameter according to the preset variable value range, thereby obtaining multiple new modeling programs; Step B3: Execute each of the obtained modeling procedures in sequence to form multiple three-dimensional models for the target building object.

[0031] In this embodiment, the CADquery modeling library contains over two hundred APIs related to 3D modeling. When generating a 3D modeling program based on the first modeling instruction, typically ten sample data points are generated. However, this amount of data is too small to train a powerful large-scale model. Therefore, data generalization is required, following steps B1 to B3 above. That is, all geometric values ​​in the modeling program are replaced with new geometric values ​​that meet the requirements, thereby obtaining a new modeling program. Through data generalization, a large amount of program data is obtained, increasing the amount of training data. For example, as... Figure 2 The image shows an example of a modeling program that includes multiple geometric values; Figure 3 For example Figure 2 The example shown demonstrates the process of data generalization, such as... Figure 3 As shown, it will be as follows Figure 2 The geometric values ​​are extracted and replaced with the corresponding parameter variables. Finally, the parameter variables are assigned new values ​​to obtain a new modeling program.

[0032] Specifically, in step S103, executing the modeling procedure to obtain a three-dimensional model of the target building object includes: Step C1: Obtain the second modeling instruction; wherein the second modeling instruction includes multiple modeling programs for different building objects; Step C2: Input the second modeling instruction into the large language model, and learn the multiple modeling programs through the large language model to obtain a modeling program for constructing new building objects; Step C3: Execute each of the obtained modeling procedures in sequence to form multiple 3D models for different architectural objects.

[0033] In this embodiment, generalizing the 3D modeling program according to steps B1 to B3 will yield a large amount of program data. However, these programs only change the numerical dimensions, not the structure. Therefore, it is necessary to obtain program data with different structures. Following steps C1 to C3, a modeling program for a new structure can be obtained based on the original modeling program. The user obtains multiple 3D modeling programs with different APIs and constructs a second modeling instruction. This instruction includes the APIs from these multiple different 3D modeling programs. Natural language statements containing these APIs and the corresponding 3D modeling programs are simultaneously input into a large language model. The large language model uses a few-shot learning strategy to learn the usage of each input API and automatically combines these APIs to obtain a 3D modeling program including multiple APIs. These multiple 3D modeling programs with different APIs can be multiple different basic modeling programs or 3D modeling programs generalized from basic modeling programs. Complex 3D modeling programs are verified and run after verification to obtain more complex 3D models. Combining multiple APIs to obtain more complex 3D modeling programs increases the complexity of the 3D model, thereby enriching the diversity of the training data.

[0034] Specifically, step S104, which involves training the large language model using the modeling program and the 3D model as a set of training data to obtain a 3D modeling model capable of outputting the corresponding modeling program based on the input 3D model, includes: Step D1: Iterate through the 3D models in each training data set sequentially, and input the modal information parsed from the currently iterated 3D model into the large language model to obtain the prediction modeling program; wherein, the modal information includes at least one of the following: image, point cloud information, geometric information, and topological information; Step D2: Execute the prediction modeling program to obtain the prediction model, and calculate the similarity between the prediction model and the currently traversed 3D model to obtain a similarity score; Step D3: Adjust the large language model according to the similarity score to obtain the three-dimensional modeling model.

[0035] Furthermore, step D2, which involves calculating the similarity score between the predicted model and the currently traversed 3D model, includes: Step D21: Calculate the first total number of each preset geometric feature in the prediction model, and calculate the second total number of each preset geometric feature in the currently traversed 3D model; wherein, the preset geometric features include: number of vertices, number of edges, number of faces, and number of rings; Step D22: Calculate the similarity score of the first total quantity and the second total quantity using a similarity algorithm.

[0036] In this embodiment, during the training of the large language model, each qualified 3D modeling program is run sequentially to obtain the 3D model and modal information. The modal information is then input into the large language model to obtain the predictive modeling program. First, the predictive modeling program is verified to run normally. If it does, the predictive modeling program is run to obtain the predictive model. The topological information of the predictive model is compared with the topological information of the 3D model: the first total number of each preset geometric feature in the predictive model is calculated, and the second total number of each preset geometric feature in the 3D model is also calculated. A similarity algorithm is used to calculate the similarity score between the first and second total numbers. When the similarity score is higher than a preset score threshold, it indicates that the predictive model and the 3D model are highly similar, and no processing of the large language model is required. When the similarity score is less than or equal to the preset score threshold, it indicates that the difference between the predictive model and the 3D model is significant, and the large language model needs to be adjusted accordingly based on the similarity score to obtain the 3D modeling model. The reasons for using topological information for similarity comparison are as follows: The key to Brep / CAD is not just geometry (point coordinates, curves, faces), but also topological structure (adjacency between vertices / edges / loops / faces / volumes, boundary relationships, loop / hole / shell structures, face orientation, coplanarity of edges, etc.); using only geometric distance (e.g., L2 point cloud, Chamfer, IoU) ignores "connectivity" and "validity" (e.g., redundant / missing edges, non-manifolds, holes, incorrect face order); topological similarity as a comparison can encourage the model to learn to generate legitimate and structurally similar Brep (e.g., the same number of faces / holes, the same topological type: with holes vs. without holes, the same loop structure, etc.). Furthermore, in addition to comparing similarity based on the topological information of the 3D model, comparisons are also made based on multiple dimensions such as size information, geometric distance information, and the geometric integrity of the Brep volume, to obtain a 3D modeling model that can output the corresponding modeling program based on the input 3D model.

[0037] In this embodiment, as Figure 4 As shown, Deepseek-r1 / Qwen3-coder is used as the large language model for generating 3D modeling programs. The first modeling instruction is input into the large language model to generate 3D modeling programs, which are then validated through multiple iterations to obtain qualified 3D modeling programs. Qwen2.5 VL 3b / 7b is used as the training model to parameterize the qualified 3D modeling programs, resulting in a large amount of program data with the same structure but different dimensions. Few-shot learning is then performed on multiple qualified 3D modeling programs to obtain complex program data with different structures. The large language model is trained using the training data to obtain a 3D modeling model that can output the corresponding modeling program based on the input 3D model. Specifically, the large language model for generating 3D modeling programs and the training model can be the same model.

[0038] In this embodiment, a first modeling instruction is obtained; the first modeling instruction is input into a large language model, and the large language model combines the call functions of all target basic modeling modules to obtain a modeling program for constructing the target building object; the modeling program is executed to obtain a three-dimensional model of the target building object; the modeling program and the three-dimensional model are used as a set of training data to train the large language model, resulting in a three-dimensional modeling model that can output the corresponding modeling program based on the input three-dimensional model; training and enhancement are performed when the large model does not require additional data. Among the three conditions for training the large model, namely data, algorithm, and computing power, the dependence on data is directly reduced, the efficiency of training large models in the construction field is improved, and the labor cost is reduced.

[0039] Example 2 This invention provides a device for self-training of a 3D modeling model, such as... Figure 5 As shown, the device specifically includes the following components: The acquisition module 501 is used to acquire a first modeling instruction; wherein the first modeling instruction includes a calling function for one or more target basic modeling modules selected from a preset building modeling library; The construction module 502 is used to input the first modeling instruction into the large language model, and to combine the calling functions of all target basic modeling modules through the large language model to obtain a modeling program for constructing the target building object; Execution module 503 is used to execute the modeling program to obtain a three-dimensional model of the target building object; The training module 504 is used to train the large language model by using the modeling program and the 3D model as a set of training data, so as to obtain a 3D modeling model that can output the corresponding modeling program based on the input 3D model.

[0040] Specifically, the device further includes a judgment module, used for: Determine whether the modeling program can be executed normally. If not, send the program error message to the large language model to instruct the large language model to regenerate the modeling program. Determine whether the 3D model obtained after the normal execution of the modeling program is a complete contour model. If not, send the model error information to the large language model to instruct the large language model to regenerate the modeling program. Determine whether the 3D model with a complete contour volume model contains all target basic modeling modules. If not, send module error information to the large language model to instruct the large language model to regenerate the modeling program.

[0041] Specifically, the execution module 503 is used for: Replace all geometric values ​​contained in the modeling program with variable parameters; Multiple new geometric values ​​are randomly generated for each variable parameter according to the preset variable value range, thereby obtaining multiple new modeling programs; Each of the resulting modeling procedures is executed sequentially to create multiple three-dimensional models for the target building object.

[0042] Specifically, the execution module 503 is used for: Obtain a second modeling instruction; wherein the second modeling instruction includes multiple modeling programs for different building objects; The second modeling instruction is input into the large language model, and the large language model is used to learn the multiple modeling programs to obtain a modeling program for constructing new building objects; Each modeling procedure is executed sequentially to create multiple 3D models for different architectural objects.

[0043] Specifically, the training module 504 is used for: The three-dimensional models in each training data are traversed sequentially, and the modal information parsed from the currently traversed three-dimensional model is input into the large language model to obtain the prediction modeling program; wherein, the modal information includes at least one of the following: image, point cloud information, geometric information, and topological information; The prediction modeling procedure is executed to obtain a prediction model, and the similarity between the prediction model and the currently traversed 3D model is calculated to obtain a similarity score. The large language model is adjusted based on the similarity score to obtain the three-dimensional modeling model.

[0044] Furthermore, the training module 504 is also used for: The first total number of each preset geometric feature in the prediction model is calculated, and the second total number of each preset geometric feature in the currently traversed 3D model is calculated; wherein, the preset geometric features include: number of vertices, number of edges, number of faces, and number of rings; The similarity score of the first total quantity and the second total quantity is calculated using a similarity algorithm.

[0045] Example 3 This embodiment also provides a computer device, such as a smartphone, tablet computer, laptop computer, desktop computer, rack server, blade server, tower server, or cabinet server (including a standalone server or a server cluster composed of multiple servers), etc., capable of executing programs. Figure 6As shown, the computer device 60 in this embodiment includes, but is not limited to, a memory 601 and a processor 602 that are communicatively connected to each other via a system bus. It should be noted that... Figure 6 Only a computer device 60 with components 601-602 is shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0046] In this embodiment, the memory 601 (i.e., the readable storage medium) includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 601 may be an internal storage unit of the computer device 60, such as the hard disk or memory of the computer device 60. In other embodiments, the memory 601 may also be an external storage device of the computer device 60, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 60. Of course, the memory 601 may include both the internal storage unit and the external storage device of the computer device 60. In this embodiment, the memory 601 is typically used to store the operating system and various application software installed on the computer device 60. In addition, the memory 601 may also be used to temporarily store various types of data that have been output or will be output.

[0047] In some embodiments, processor 602 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. This processor 602 is typically used to control the overall operation of computer device 60.

[0048] Specifically, in this embodiment, the processor 602 is used to execute the program of the self-training method for the 3D modeling model stored in the memory 601. When the program of the self-training method for the 3D modeling model is executed, it performs the following steps: Obtain a first modeling instruction; wherein the first modeling instruction includes a function that calls one or more target basic modeling modules selected from a preset building modeling library; The first modeling instruction is input into the large language model, and the calling functions of all target basic modeling modules are combined and processed by the large language model to obtain a modeling program for constructing the target building object; The modeling procedure is executed to obtain a three-dimensional model of the target building object; The modeling program and the 3D model are used as a set of training data to train the large language model, resulting in a 3D modeling model that can output the corresponding modeling program based on the input 3D model.

[0049] For a detailed description of the above method steps, please refer to Example 1. This example will not be repeated here.

[0050] Example 4 This embodiment also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, server, app store, etc., which stores a computer program. When the computer program is executed by a processor, it implements the following method steps: Obtain a first modeling instruction; wherein the first modeling instruction includes a function that calls one or more target basic modeling modules selected from a preset building modeling library; The first modeling instruction is input into the large language model, and the calling functions of all target basic modeling modules are combined and processed by the large language model to obtain a modeling program for constructing the target building object; The modeling procedure is executed to obtain a three-dimensional model of the target building object; The modeling program and the 3D model are used as a set of training data to train the large language model, resulting in a 3D modeling model that can output the corresponding modeling program based on the input 3D model.

[0051] For a detailed description of the above method steps, please refer to the first embodiment. This embodiment will not repeat the details here.

[0052] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0053] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0054] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0055] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for self-training a 3D modeling model, characterized in that, The method includes: Obtain a first modeling instruction; wherein the first modeling instruction includes a function that calls one or more target basic modeling modules selected from a preset building modeling library; The first modeling instruction is input into the large language model, and the calling functions of all target basic modeling modules are combined and processed by the large language model to obtain a modeling program for constructing the target building object; The modeling procedure is executed to obtain a three-dimensional model of the target building object; The modeling program and the 3D model are used as a set of training data to train the large language model, resulting in a 3D modeling model that can output the corresponding modeling program based on the input 3D model.

2. The method for self-training a 3D modeling model according to claim 1, characterized in that, After inputting the first modeling instruction into the large language model and combining the calling functions of all target basic modeling modules through the large language model to obtain a modeling program for constructing the target building object, the method further includes: Determine whether the modeling program can be executed normally. If not, send the program error message to the large language model to instruct the large language model to regenerate the modeling program. Determine whether the 3D model obtained after the normal execution of the modeling program is a complete contour model. If not, send the model error information to the large language model to instruct the large language model to regenerate the modeling program. Determine whether the 3D model with a complete contour volume model contains all target basic modeling modules. If not, send module error information to the large language model to instruct the large language model to regenerate the modeling program.

3. The method for self-training a 3D modeling model according to claim 1, characterized in that, The execution of the modeling program to obtain a three-dimensional model of the target building object includes... Replace all geometric values ​​contained in the modeling program with variable parameters; Multiple new geometric values ​​are randomly generated for each variable parameter according to the preset variable value range, thereby obtaining multiple new modeling programs; Each of the resulting modeling procedures is executed sequentially to create multiple three-dimensional models for the target building object.

4. The method for self-training a 3D modeling model according to claim 1, characterized in that, The execution of the modeling procedure to obtain a three-dimensional model of the target building object includes: Obtain a second modeling instruction; wherein the second modeling instruction includes multiple modeling programs for different building objects; The second modeling instruction is input into the large language model, and the large language model is used to learn the multiple modeling programs to obtain a modeling program for constructing new building objects; Each modeling procedure is executed sequentially to create multiple 3D models for different architectural objects.

5. The method for self-training a 3D modeling model according to claim 1, characterized in that, The step of training the large language model using the modeling program and the 3D model as a set of training data to obtain a 3D modeling model that can output the corresponding modeling program based on the input 3D model includes: The three-dimensional models in each training data are traversed sequentially, and the modal information parsed from the currently traversed three-dimensional model is input into the large language model to obtain the prediction modeling program; wherein, the modal information includes at least one of the following: image, point cloud information, geometric information, and topological information; The prediction modeling procedure is executed to obtain a prediction model, and the similarity between the prediction model and the currently traversed 3D model is calculated to obtain a similarity score. The large language model is adjusted based on the similarity score to obtain the three-dimensional modeling model.

6. The method for self-training a 3D modeling model according to claim 5, characterized in that, The step of calculating the similarity score between the predicted model and the currently traversed 3D model includes: The first total number of each preset geometric feature in the prediction model is calculated, and the second total number of each preset geometric feature in the currently traversed 3D model is calculated; wherein, the preset geometric features include: number of vertices, number of edges, number of faces, and number of rings; The similarity score of the first total quantity and the second total quantity is calculated using a similarity algorithm.

7. A device for self-training a 3D modeling model, characterized in that, The device includes: An acquisition module is used to acquire a first modeling instruction; wherein the first modeling instruction includes a function to call one or more target basic modeling modules selected from a preset building modeling library; A construction module is used to input the first modeling instruction into the large language model, and to combine the calling functions of all target basic modeling modules through the large language model to obtain a modeling program for constructing the target building object; The execution module is used to execute the modeling program to obtain a three-dimensional model of the target building object; The training module is used to train the large language model by using the modeling program and the 3D model as a set of training data, so as to obtain a 3D modeling model that can output the corresponding modeling program based on the input 3D model.

8. The apparatus for self-training of a 3D modeling model according to claim 7, characterized in that, The device further includes a judgment module for: Determine whether the modeling program can be executed normally. If not, send the program error message to the large language model to instruct the large language model to regenerate the modeling program. Determine whether the 3D model obtained after the normal execution of the modeling program is a complete contour model. If not, send the model error information to the large language model to instruct the large language model to regenerate the modeling program. Determine whether the 3D model with a complete contour volume model contains all target basic modeling modules. If not, send module error information to the large language model to instruct the large language model to regenerate the modeling program.

9. A computer device, the computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images