Model generation method and apparatus

Abstract

Provided in the present application is a model generation method. The method comprises: acquiring, from a historical model database, a historical model file and a historical model description file corresponding to the historical model file; training a dedicated large language model by using, as training data pairs, the historical model file and the historical model description file corresponding to the historical model file, wherein the dedicated large language model is based on a general large language model; and receiving a query request from a user, and the dedicated large language model outputting, on the basis of the query request, a model file that matches the query request.

Description

Model generation method and apparatus Technical Field

[0001] This application mainly relates to the field of industrial digitalization, and in particular to a model generation method and apparatus. Background Technology

[0002] For customized production, due to the diversity of customization needs, products need to be redesigned according to these needs. After receiving a customized order, the product designer needs to redesign the product, a process that consumes a significant amount of the designer's workload and time, and is prone to errors and omissions.

[0003] Historical design documents include configuration information, process information, simulation verification results, and 3D models. As a form of knowledge accumulation, historical design documents are often used as the foundation for customized designs. However, due to issues such as incompatible formats, lost details, changes and upgrades, poor design reusability, and the inapplicability of outdated designs, the reuse efficiency of historical design documents is low.

[0004] Currently, some software tools can automatically generate new design files based on customized requirements using existing design files. However, these software tools use feature-based modeling techniques to automatically create parametric feature-related geometries, but this approach still requires fine-tuning of the entire structure, thus demanding a significant amount of work and time from designers.

[0005] Summary of the Invention

[0006] To address the aforementioned technical problems, this application provides a model generation method and apparatus to quickly and accurately generate customized model files.

[0007] To achieve the above objectives, this application proposes a model generation method, the method comprising:

[0008] Retrieve historical model files and corresponding historical model description files from the historical model database;

[0009] A dedicated large language model is trained using historical model files and corresponding historical model description files as training data. The dedicated large language model is based on a general large language model.

[0010] Upon receiving a user's query request, the dedicated large language model outputs a model file that matches the query request.

[0011] To address this, by creating a dedicated large language model based on a general large language model, the computational and reasoning capabilities of the dedicated large language model can be utilized to output model files that match query requests. This improves the speed and accuracy of model generation by the dedicated large language model, enables automatic generation and adjustment of semantic-based models, and significantly reduces the workload and time for designers.

[0012] Optionally, the dedicated large language model outputting a model file matching the query request includes: the dedicated large language model calling a knowledge graph to output a model file matching the query request. Therefore, the dedicated large language model can utilize a knowledge graph to provide more accurate and richer search results.

[0013] Optionally, receiving a user's query request and having the dedicated large language model output a model file matching the query request includes: receiving a query image input by the user, extracting features from the query image using a neural network model, generating a semantic description, and having the dedicated large language model output a model file matching the query request based on the semantic description. This implements image modality input.

[0014] Optionally, the dedicated large language model outputting model files matching the query request includes: the dedicated large language model outputting multiple model files matching the query request. Therefore, by using a dedicated large language model to output multiple model files matching the query request, the flexibility of model generation is improved.

[0015] Optionally, the method further includes: receiving user feedback on the output model file, and updating the dedicated large language model based on the user feedback. Therefore, updating the dedicated large language model based on user feedback can better meet user needs and improve the user experience.

[0016] Optionally, the method further includes: receiving a first model file and operation instructions, parsing the first model file, and the dedicated large language model generating a second model file based on the parsed first model file and the operation instructions. This allows designers to focus more on design concepts and creativity without spending excessive time and effort on tedious parameter adjustments, greatly simplifying the design process, improving design efficiency, and providing a better user experience.

[0017] This application also proposes a model generation apparatus, the apparatus comprising:

[0018] The acquisition module retrieves historical model files and corresponding historical model description files from the historical model database.

[0019] The training module uses historical model files and corresponding historical model description files as training data to train a dedicated large language model, which is based on a general large language model.

[0020] The inference module receives user query requests, and the dedicated large language model outputs a model file that matches the query request.

[0021] This application also proposes an electronic device including a processor, a memory, and instructions stored in the memory, wherein the instructions, when executed by the processor, implement the method described above.

[0022] This application also proposes a computer-readable storage medium having computer instructions stored thereon, which, when executed, perform the methods described above.

[0023] This application also proposes a computer program product, including a computer program that, when executed by a processor, implements the method described above. Attached Figure Description

[0024] The accompanying drawings are intended only to illustrate and explain this application and do not limit the scope of this application.

[0025] Figure 1 is a flowchart of a method for generating a three-dimensional model according to an embodiment of this application;

[0026] Figure 2 is an architecture diagram of a modeling system according to an embodiment of this application;

[0027] Figure 3 is a schematic diagram of a user query model file according to an embodiment of this application;

[0028] Figure 4 is a schematic diagram of a three-dimensional model generation apparatus according to an embodiment of this application;

[0029] Figure 5 is a schematic diagram of an electronic device according to an embodiment of this application.

[0030] Figure Label Explanation: 100 3D Model Generation Method; 110-130 Steps; 21 User Input Unit; 22 Historical Model Database; 23 Data Processing Unit; 231 User Interface; 232 Data Preprocessing Module; 233 Model Training Module; 234 Model Fine-tuning Module; 235 Text Processing Module; 236 Instruction Management Module; 24 Dedicated Large Language Model; 25 Knowledge Graph; 26 Modeling Unit; 27 Model File; 30 User; 31 User Interface; 32 Dedicated Large Language Model; 321 Internal Training Data; 322 External Training Data; 323 Training Module; 33 Matching and Ranking Module; 34 Recommendation Module; 400 3D Model Generation Device; 410 Acquisition Module; 420 Training Module; 430 Inference Module; 500 Electronic Device; 510 Processor; 520 Memory. Detailed Implementation

[0031] To provide a clearer understanding of the technical features, objectives, and effects of this application, specific embodiments of this application will now be described with reference to the accompanying drawings.

[0032] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein, and therefore this application is not limited to the specific embodiments described below.

[0033] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0034] This application proposes a model generation method. Figure 1 is a flowchart of a three-dimensional model generation method 100 according to an embodiment of this application. As shown in Figure 1, method 100 includes:

[0035] Step 110: Obtain the historical model file and the corresponding historical model description file from the historical model database;

[0036] A historical model database is a database that stores historical model data, which can include historical model files and corresponding historical model description files. Historical model files can be CAD files, CAE files, or digital assets, and can be parsed and loaded into 3D models. Historical model description files can include object configuration information, process information, simulation verification information, etc. By accessing the historical model database, historical model files and their corresponding historical model description files can be retrieved.

[0037] Step 120: Use the historical model file and the historical model description file corresponding to the historical model file as training data to train a dedicated large language model. The dedicated large language model is based on the general large language model.

[0038] Model description information, explanatory information, and 3D modeling task information can be parsed from historical model description files. Historical model description files may also include metadata, such as the object name, material, author, and creation time of the 3D model. Using the model description information as a prompt as input, a specific 3D model can be generated, or an existing model can be modified as output. The input and output can be used as training data pairs to train a specialized large language model. This specialized large language model is based on a general large language model. A general large language model refers to a large-scale natural language processing model with broad language understanding and generation capabilities. These models use deep learning techniques, through large amounts of training data and powerful computing capabilities, to understand and generate various expressions of human language. General large language models can be used for various natural language processing tasks, such as machine translation, text summarization, speech recognition, and dialogue systems. They can perform semantic understanding, syntactic analysis, and contextual understanding on input text and generate output that conforms to language rules and semantic logic. By constructing a specialized large language model based on a general large language model, the computational and reasoning capabilities of the specialized large language model can be utilized, thereby improving the speed and accuracy of the generated model.

[0039] Before training a dedicated large language model, the training data can be preprocessed to improve its quality and compatibility. Preprocessing of training data can include geometry repair, polygon reduction, and file format conversion. During 3D scanning or modeling, incomplete scan data or missing elements in the model may occur. Geometry repair techniques can fill in missing parts or reconstruct shapes based on existing geometric information and algorithms to obtain a complete 3D model. Polygon reduction refers to the process of simplifying a 3D model by reducing the number of polygons. The detail and complexity of a model are often determined by the number of polygons; too many polygons can lead to increased model file size and slower rendering speeds. Polygon reduction can reduce model complexity and improve performance. Common polygon reduction methods include mesh simplification, edge collapse, and approximate surfaces.

[0040] After creating a dedicated large language model, it can be fine-tuned. This includes adjusting hyperparameters such as the learning rate, batch size, and regularization strength. Fine-tuning can be done manually or automatically. Automatic fine-tuning can employ grid search or random search. After fine-tuning, the model is validated. Performance can be evaluated using a single validation set or through cross-validation. Common evaluation metrics include accuracy, precision, recall, or F1 score.

[0041] Step 130: Receive the user's query request. The dedicated large language model outputs a model file that matches the query request.

[0042] The user's query request describes the target 3D model. Using the user's query request as input, the dedicated large language model infers from the input query request and outputs a model file that matches the query request. Therefore, by creating a dedicated large language model based on a general large language model, the computational and reasoning capabilities of the dedicated large language model can be utilized to output a model file that matches the query request, improving the speed and accuracy of model generation by the dedicated large language model.

[0043] In some embodiments of this application, the dedicated large language model outputting a model file matching the query request includes: the dedicated large language model calling a knowledge graph to output a model file matching the query request. A knowledge graph is a method for organizing and representing knowledge in a graphical structure. It is a semantic network used for storing, representing, and reasoning about knowledge, composed of entities and relationships between entities. The construction of knowledge graphs is typically based on manual or automated knowledge extraction and representation techniques. It can extract structured and semi-structured information from various data sources and represent it in the form of entities and relationships. Dedicated large language models can leverage knowledge graphs to provide more accurate and richer search results.

[0044] In some embodiments of this application, receiving a user's query request and having a dedicated large language model output a model file matching the query request includes: receiving a query image input by the user, extracting features from the query image using a neural network model, generating a semantic description, and having the dedicated large language model output a model file matching the query request based on the semantic description. Specifically, the user can input a product photo through a user interface. After image processing, the product photo is input into a convolutional neural network or a recurrent neural network for feature extraction, and a semantic description is generated based on the feature extraction. The speech description is then sent as input to the dedicated large language model for inference. This enables the input of an image modality.

[0045] In some embodiments of this application, the dedicated large language model outputs model files matching the query request, including multiple model files matching the query request. For example, the dedicated large language model can generate four model files: A, B, C, and D. Each model has corresponding metrics, and the system sorts the four models based on these metrics. Therefore, by using the dedicated large language model to output multiple model files matching the query request, the flexibility of model generation is improved.

[0046] In some embodiments of this application, the method further includes: receiving user feedback on the output model file, and updating the dedicated large language model based on the user feedback. For example, if a user gives a high score to an output result of the dedicated large language model, this score can serve as positive feedback for that input-output, and that input-output can be retained in the dedicated large language model. Conversely, if a user gives a low score to an output result of the dedicated large language model, this score can serve as negative feedback for that input-output, and that input-output may be removed from the dedicated large language model. Therefore, updating the dedicated large language model based on user feedback can better meet user needs and improve the user experience.

[0047] In some embodiments of this application, the method further includes: receiving a first model file and operation instructions, parsing the first model file, and generating a second model file based on the parsed first model file and operation instructions using a dedicated large language model. Specifically, the first model file can be input by the user or output by the dedicated large language model. The first model file is parsed to extract geometric information, material information, and texture information, etc. The user can operate and edit the 3D model through the user interface, such as selecting, translating, scaling, rotating, and adjusting materials. Based on user operations and the first model file, design and optimization suggestions can be generated, such as object movement, material, and lighting suggestions, and the suggestion results can be displayed to the user in real time. The user can intuitively see the design and optimization results and output the corresponding 3D model file. Real-time rendering and preview functions can be used to display the suggestion results to the user in real time.

[0048] As a non-restrictive example, a designer could describe "the length of the component increases by 50%, and the width decreases by 20%." The system can automatically parse the increase and decrease values ​​and adjust the model's parameters accordingly. This demonstrates that semantic-based automatic model adjustment eliminates the need for designers to manually modify each parameter. Instead, they can guide the adjustment process through natural language descriptions, enabling quick and accurate adjustments. Designers can focus more on design concepts and creativity without spending excessive time and effort on tedious parameter adjustments, greatly simplifying the design process, improving efficiency, and providing a better user experience.

[0049] Figure 2 is an architecture diagram of a modeling system according to an embodiment of this application. As shown in Figure 2, the architecture includes a user input unit 21, a historical model database 22, a data processing unit 23, a dedicated large language model 24, a knowledge graph 25, a modeling unit 26, and a model file 27. The data processing unit 23 includes a user interface 231, a data preprocessing module 232, a model training module 233, a model fine-tuning module 234, a text processing module 235, and an instruction management module 236.

[0050] Modeling unit 26 can be modeling software. Data processing unit 23 can call the API of modeling unit 26 to achieve bidirectional communication. Scripts can be written to enable data processing unit 23 to receive input text and convert it into instructions or commands that modeling unit 26 can recognize. Data processing unit 23 can send a 3D model to modeling unit 26 to generate corresponding response text. Data processing unit 23 can also be embedded in modeling unit 26 as a plugin or extension. Data processing unit 23 can also be integrated into modeling unit 26 as an application.

[0051] The data processing module 23 retrieves historical model files and historical model description files from the historical model database 22. The data preprocessing module 232 preprocesses the historical model files and historical model description files. The model training module 233 uses the preprocessed data to train the model, generating a dedicated large language model 24 based on a general large language model. The model fine-tuning module 234 fine-tunes the dedicated large language model. The text processing module 235 processes the text and generates response text. The instruction management module 236 parses the user's operation instructions.

[0052] After the user (designer) inputs a model description through the user input unit 21, the user interface 231 reads the model description. The model description is input into the dedicated large language model 24, which can call the knowledge graph 25 to output a response text and output a model file through the modeling unit 26.

[0053] Figure 3 is a schematic diagram of a user querying model files according to an embodiment of this application. As shown in Figure 3, user 30 inputs a query request through user interface 31, which is then input into a dedicated large language model 32. The dedicated large language model 32 is trained by training module 323 using internal training data 321 and external training data 322. Internal training data 321 may be internal design drawings and descriptions in a drawing database within the company's local area network. External training data 322 may be external design drawings and descriptions in a drawing database outside the company's local area network. The dedicated large language model 32 outputs multiple model files, and matching and sorting module 33 matches and sorts the output multiple model files. Recommendation module 34 makes recommendations based on the matching and sorting results, and the user provides feedback on the recommendation results. The user feedback is input into the dedicated large language model 32, and the dedicated large language model 32 is updated based on the user feedback.

[0054] The embodiments of this application propose a model generation method. By creating a dedicated large language model based on a general large language model, the computational and reasoning capabilities of the dedicated large language model can be utilized to output a model file that matches the query request. This improves the speed and accuracy of model generation from the dedicated large language model, realizes automatic generation and adjustment of semantic-based models, and greatly reduces the workload and time of designers.

[0055] This application also proposes a model generation apparatus. Figure 4 is a schematic diagram of a three-dimensional model generation apparatus 400 according to an embodiment of this application. As shown in Figure 4, the apparatus 400 includes:

[0056] Module 410 retrieves historical model files and corresponding historical model description files from the historical model database;

[0057] Training module 420 uses historical model files and corresponding historical model description files as training data to train a dedicated large language model, which is based on a general large language model.

[0058] The inference module 430 receives user query requests, and the dedicated large language model outputs a model file that matches the query request.

[0059] This application also proposes an electronic device 500. Figure 5 is a schematic diagram of an electronic device 500 according to an embodiment of this application. As shown in Figure 5, the electronic device 500 includes a processor 510 and a memory 520, wherein the memory 520 stores instructions, wherein when the instructions are executed by the processor 510, they implement the method 100 described above.

[0060] This application also proposes a computer-readable storage medium having computer instructions stored thereon, which, when executed, perform the method 100 described above.

[0061] This application also proposes a computer program product, including a computer program that, when executed by a processor, performs the method 100 described above.

[0062] Some aspects of the methods and apparatus of this application can be executed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The aforementioned hardware or software may be referred to as a "data block," "module," "engine," "unit," "component," or "system." The processor may be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DAPDs), programmable logic devices (PLCs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, or combinations thereof. Furthermore, aspects of this application may manifest as a computer product residing in one or more computer-readable media, including computer-readable program code. For example, computer-readable media may include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, magnetic tapes, etc.), optical discs (e.g., compact discs (CDs), digital multifunction discs (DVDs), etc.), smart cards, and flash memory devices (e.g., cards, sticks, key drives, etc.).

[0063] Flowcharts are used herein to illustrate the operations performed by the method according to embodiments of this application. It should be understood that the preceding operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations may be added to these processes, or one or more steps may be removed from them.

[0064] It should be understood that although this specification is described according to various embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0065] The above description is merely an illustrative embodiment of this application and is not intended to limit the scope of this application. Any equivalent changes, modifications, and combinations made by those skilled in the art without departing from the concept and principles of this application shall fall within the scope of protection of this application.

[0066] In this patent application, nouns and pronouns relating to people are not limited to specific genders.

Claims

1. A model generation method (100), characterized in that, The method (100) includes: Obtain historical model files and corresponding historical model description files from the historical model database (110); A dedicated large language model is trained using historical model files and historical model description files corresponding to the historical model files as training data. The dedicated large language model is based on the general large language model (120). The dedicated large language model receives a user's query request and outputs a model file (130) that matches the query request.

2. The method (100) according to claim 1, characterized in that, The dedicated large language model outputs a model file that matches the query request according to the query request, including: the dedicated large language model calls the knowledge graph to output a model file that matches the query request according to the query request.

3. The method (100) according to claim 1, characterized in that, Receiving a user's query request, and the dedicated large language model outputting a model file matching the query request according to the query request includes: receiving a query image input by the user, using a neural network model to extract features from the query image and generate a semantic description, and the dedicated large language model outputting a model file matching the query request according to the semantic description.

4. The method (100) according to claim 1, characterized in that, The dedicated large language model outputs model files that match the query request according to the query request, including: the dedicated large language model outputs multiple model files that match the query request according to the query request.

5. The method (100) according to claim 1, characterized in that, The method (100) further includes: receiving user feedback on the output model file and updating the dedicated large language model based on the user feedback.

6. The method (100) according to claim 1, characterized in that, The method (100) further includes: receiving a first model file and an operation instruction, parsing the first model file, and the dedicated large language model generating a second model file based on the parsed first model file and the operation instruction.

7. A model generation apparatus (400), characterized in that, The device (400) includes: The acquisition module (410) acquires historical model files and historical model description files corresponding to the historical model files from the historical model database; The training module (420) uses historical model files and historical model description files corresponding to the historical model files as training data to train a dedicated large language model, wherein the dedicated large language model is based on a general large language model; The reasoning module (430) receives the user's query request, and the dedicated large language model outputs a model file that matches the query request.

8. An electronic device (500) comprising a processor (510), a memory (520) and instructions stored in the memory (520), wherein the instructions, when executed by the processor (510), implement the method (100) as claimed in any one of claims 1-6.

9. A computer-readable storage medium having stored thereon computer instructions that, when executed, perform the method (100) according to any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, performs the method (100) of any one of claims 1-6.

Images