Server structure model generation method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a server structure model generation method and device, electronic equipment and a storage medium. The method can be used in artificial intelligence technology field. The method comprises the following steps: acquiring component size information and component material information corresponding to each component in a to-be-modeled server; inputting the component size information and the component material information of each component into a generator of a first specific task generative adversarial network to obtain a three-dimensional view model corresponding to the to-be-modeled server; performing projection processing on the three-dimensional view model to obtain two-dimensional views of the three-dimensional view model corresponding to each surface; the two-dimensional views of all the faces are input into a generator of a second specific task generative adversarial network, a three-dimensional voxel model corresponding to the to-be-modeled server is obtained, and the generator of the second generative adversarial network is used for converting the two-dimensional views into the three-dimensional voxel model; and performing model optimization on the three-dimensional voxel model to obtain a mainboard view corresponding to the to-be-modeled server. The method is used for achieving the effect of improving the modeling efficiency.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular, to a method, device, electronic device, and storage medium for generating a server structure model. Background Art

[0002] The overall server modeling mainly relies on computer-aided design (CAD, Computer-Aided Design) software and tools. After years of development, these methods have achieved a high degree of maturity and accuracy, and are widely used in the design and manufacturing of servers and entire cabinets.

[0003] Currently, the main way to model the overall server through computer-aided design software is that the user inputs the component size information of the server components in the computer-aided design software, and uses the computer-aided design software to independently model the server components according to the component size information of the server components input by the user to obtain the component structure model corresponding to the server components; then, the user combines the multiple component structure models of the server in the computer-aided design software according to the actual assembly sequence and positional relationship to obtain the overall structure model of the server, that is, the server structure model.

[0004] However, the existing methods for modeling the overall server have the problem of low modeling efficiency. Summary of the Invention

[0005] Embodiments of this application provide a method, device, electronic device, and storage medium for generating a server structure model to solve the problem of low overall server modeling efficiency.

[0006] In a first aspect, embodiments of this application provide a method for generating a server structure model, including:

[0007] Obtain the component size information and component material information corresponding to each component in the server to be modeled;

[0008] Input the component size information and component material information of each component into the generator of the first specific task generative adversarial network. In the generator, based on the learned assembly relationship between components, obtain the three-dimensional view model corresponding to the server to be modeled according to the component size information and component material information, and the three-dimensional view model conforms to the assembly relationship between components;

[0009] Perform projection processing on the three-dimensional view model to obtain two-dimensional views corresponding to each face of the three-dimensional view model;

[0010] Input the two-dimensional views of each surface into the generator of the second specific-task generative adversarial network to obtain a three-dimensional voxel model corresponding to the server to be modeled. The generator of the second generative adversarial network is used to convert the two-dimensional views into a three-dimensional voxel model;

[0011] Optimize the three-dimensional voxel model to obtain a motherboard view corresponding to the server to be modeled.

[0012] In a possible implementation, optimizing the three-dimensional voxel model to obtain a motherboard view corresponding to the server to be modeled includes:

[0013] Input the three-dimensional voxel model into the generator of the information bottleneck generative adversarial network. In the generator of the information bottleneck generative adversarial network, extract the three-dimensional features of the three-dimensional voxel model, determine the latent space representation of the three-dimensional features, and perform reconstruction optimization on the latent space representation to obtain a motherboard view corresponding to the server to be modeled.

[0014] In a possible implementation, optimizing the three-dimensional voxel model to obtain a motherboard view corresponding to the server to be modeled includes:

[0015] Among the multiple three-dimensional voxel models corresponding to the server to be modeled, determine the optimal three-dimensional voxel model;

[0016] According to the generator of the information bottleneck generative adversarial network, optimize the optimal three-dimensional voxel model to obtain a motherboard view corresponding to the server to be modeled.

[0017] In a possible implementation, the first specific-task generative adversarial network and the second specific-task generative adversarial network are jointly trained in the following manner:

[0018] Obtain training samples, which include a training server model, as well as training component size information, training component material information, and training component assembly relationship information corresponding to each training component in the training server;

[0019] Input the training component size information, training component material information, and training component assembly relationship information of each training component into the generator of the first specific-task generative adversarial network for feature extraction processing to obtain a training three-dimensional view model corresponding to the training server;

[0020] Perform projection processing on the training three-dimensional view model corresponding to the training server to obtain training two-dimensional views corresponding to each surface of the training three-dimensional view;

[0021] Input the training two-dimensional views of each surface into the generator of the second specific-task generative adversarial network for dimension conversion processing to obtain a training three-dimensional voxel model corresponding to the training server;

[0022] Input the trained 3D view model and the trained server model into the discriminator of the first specific task generative adversarial network to obtain first loss information, and input the trained 3D voxel model and the trained server model into the discriminator of the second specific task generative adversarial network to obtain second loss information;

[0023] Train the generator of the first specific task generative adversarial network and the generator of the second specific task generative adversarial network according to the first loss information and the second loss information.

[0024] In a possible implementation manner, before inputting the component size information and the component material information of each component into the generator of the first specific task generative adversarial network, the method further includes:

[0025] Obtain available computing resources;

[0026] Determine that the available computing resources do not meet the preset computing resource requirement conditions.

[0027] In a possible implementation manner, the method further includes:

[0028] If the available computing resources meet the preset computing resource requirement conditions, input the component size information and the component material information of each component into the generator of the orthogonal frequency spectrum space generative adversarial network for 3D modeling processing to obtain a 3D initial model corresponding to the server to be modeled;

[0029] Optimize the 3D initial model to obtain a motherboard view corresponding to the server to be modeled.

[0030] In a possible implementation manner, the orthogonal frequency spectrum space generative adversarial network is trained through the following method:

[0031] Input the training component size information, the training component material information, and the training component assembly relationship information into the generator of the orthogonal frequency spectrum space generative adversarial network for 3D modeling processing to obtain a training 3D initial model corresponding to the training modeling server;

[0032] Input the training 3D initial model and the training server model into the discriminator of the orthogonal frequency spectrum space generative adversarial network to obtain third loss information;

[0033] Train the generator of the orthogonal frequency spectrum space generative adversarial network according to the third loss information.

[0034] In a second aspect, an embodiment of the present application provides a server structure model generation device, including:

[0035] An acquisition module, configured to acquire the component size information and the component material information respectively corresponding to each component in the server to be modeled;

[0036] A first obtaining module, configured to input the component size information and component material information of each component into the generator of the first specific task generative adversarial network. In the generator, based on the learned assembly relationships between components, a three-dimensional view model corresponding to the server to be modeled is obtained according to the component size information and component material information, and the three-dimensional view model conforms to the assembly relationships between components;

[0037] A second obtaining module, configured to perform projection processing on the three-dimensional view model to obtain two-dimensional views corresponding to each surface of the three-dimensional view model;

[0038] A third obtaining module, configured to input the two-dimensional views of each surface into the generator of the second specific task generative adversarial network to obtain a three-dimensional voxel model corresponding to the server to be modeled. The generator of the second generative adversarial network is used to convert the two-dimensional views into a three-dimensional voxel model;

[0039] A fourth obtaining module, configured to optimize the three-dimensional voxel model to obtain a motherboard view corresponding to the server to be modeled.

[0040] In a third aspect, an embodiment of the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0041] The memory stores computer-executable instructions;

[0042] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the first aspect and / or various possible implementation manners of the first aspect as above.

[0043] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect and / or various possible implementation manners of the first aspect as above.

[0044] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the first aspect and / or various possible implementation manners of the first aspect as above.

[0045] The server structure model generation method, device, electronic device, and storage medium provided by the embodiments of the present application obtain the component size information and component material information corresponding to each component in the server to be modeled; input the component size information and component material information of each component into the generator of the first specific task generative adversarial network. In the generator, based on the learned assembly relationships between components, a three-dimensional view model corresponding to the server to be modeled is obtained according to the component size information and component material information, and the three-dimensional view model conforms to the assembly relationships between components; perform projection processing on the three-dimensional view model to obtain two-dimensional views corresponding to each face of the three-dimensional view model; input the two-dimensional views of each face into the generator of the second specific task generative adversarial network to obtain a three-dimensional voxel model corresponding to the server to be modeled. The generator of the second generative adversarial network is used to convert the two-dimensional view into a three-dimensional voxel model; perform model optimization on the three-dimensional voxel model to obtain the main board view corresponding to the server to be modeled. By learning the assembly relationships between components through the first generative adversarial network, a three-dimensional view model based on the assembly relationships is generated, then the two-dimensional projection of the three-dimensional view model is converted into a three-dimensional voxel model through the second generative adversarial network, and then the three-dimensional voxel model is optimized through an optimization algorithm to generate the main board view of the server to be modeled, achieving the effect of quickly generating a model with assembly relationships and improving the efficiency of server modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present application and used together with the specification to explain the principles of the present application.

[0047] Figure 1 It is a flowchart of the server structure model generation method provided by the embodiments of the present application;

[0048] Figure 2 It is a flowchart of another server structure model generation method provided by the embodiments of the present application;

[0049] Figure 3 It is a step diagram of another server structure model generation method provided by the embodiments of the present application;

[0050] Figure 4 It is a model processing flowchart of another server structure model generation method provided by the embodiments of the present application;

[0051] Figure 5 It is a schematic structural diagram of the server structure model generation device provided by the embodiments of the present application;

[0052] Figure 6 It is a schematic structural diagram of the electronic device provided by the embodiments of the present application.

[0053] Through the above-mentioned drawings, specific embodiments of the present application have been shown, and there will be a more detailed description hereinafter. These drawings and the written description are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. Detailed Description of Specific Embodiments

[0054] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0055] First, the terms related to the present application are explained:

[0056] Server structure model: It refers to a model used to describe and represent the various components inside the server and their mutual relationships, which can include physical dimensions and geometric shapes, and can also contain information such as material properties, assembly relationships, and electrical connections.

[0057] In the prior art, when modeling a server through computer-aided design software, only parts can be output for corresponding requirements, and there is no assembly relationship and combination between the parts, and a structured three-dimensional model with an assembly relationship cannot be output, and multiple design prototypes cannot be quickly generated according to requirements. There is a technical problem that a model with an assembly relationship cannot be quickly generated and the modeling efficiency is low.

[0058] The server structure model generation method, device, electronic device, and storage medium provided by the present application use a first specific task generative adversarial network to generate a three-dimensional view model based on the assembly relationship, and then use a second specific task generative adversarial network to convert the two-dimensional projection of the three-dimensional view model into a three-dimensional voxel model, and optimize the three-dimensional voxel model to generate the motherboard view of the server to be modeled, thus solving the technical problems of inability to quickly generate a model with an assembly relationship and low modeling efficiency.

[0059] The following uses specific embodiments to detail the technical solution of the present application and how the technical solution of the present application solves the above technical problems. These several specific embodiments below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0060] Figure 1 It is a flowchart of the server structure model generation method provided by the embodiments of the present application. The execution subject of this method can be a server or other servers, and no special limitation is made here in this embodiment. For exampleFigure 1 As shown in Figure 1 , the method may include:

[0061] S101. Obtain the component size information and component material information corresponding to each component in the server to be modeled.

[0062] Among them, the server to be modeled may refer to the entire server that needs to perform 3D modeling during the design, optimization, or manufacturing process. The server to be modeled includes components to be modeled such as the main board, radiator, power module, etc.

[0063] The component size information may refer to the geometric size data of each component of the server, which may include basic dimensions such as length, width, and height, as well as detailed dimensions such as hole positions and interface positions. The format of the component size information may be JSON (JavaScript Object Notation), CSV (Comma-Separated Values), etc. It is also possible to convert structured data into forms such as vectors and tensors suitable for processing by deep learning models. In the embodiments of this application, the component size information may also be accompanied by component name information. For example, main board: length 30 cm, width 24.5 cm, thickness 0.3 cm.

[0064] The component material information may refer to the material types used for each component of the server and their physical properties, such as density, strength, thermal conductivity, etc.

[0065] The method for obtaining the component size information and component material information corresponding to each component in the server to be modeled may include extracting the size information and material information of each component from the design document of the server to be modeled.

[0066] S102. Input the component size information and component material information of each component into the generator of the first specific task generative adversarial network. In the generator, based on the learned assembly relationships between components, obtain the 3D view model corresponding to the server to be modeled according to the component size information and component material information, and the 3D view model conforms to the assembly relationships between components.

[0067] Among them, the generator of the first specific task generative adversarial network may refer to the generator of the generative adversarial network that can generate a 3D view model according to the component size information and component material information. In the embodiments of this application, the first specific task generative adversarial network may be StyleGAN (Style -Based Generative Adversarial Network). After receiving the structured component size information and component material information, StyleGAN uses a residual network to extract the features of the input data, combines the learned assembly relationships between components, and generates a 3D view model through the generator.

[0068] Among them, in the embodiments of the present application, before inputting the component size information and component material information of each component into the generator of the first specific task generative adversarial network, the method may further include:

[0069] Obtain available computing resources;

[0070] Determine that the available computing resources do not meet the preset computing resource requirement conditions.

[0071] Among them, the available computing resources may refer to the hardware and software resources available for executing computing tasks in the current system, such as computing power, storage space, and network bandwidth.

[0072] The method for obtaining available computing resources may include using monitoring tools to obtain various available computing resources in the current system.

[0073] The computing resource requirement conditions may refer to the conditions set according to the demand for computing resources of different models used. For example, the computing resource requirement condition may be that the available amount of memory is at least 50G.

[0074] In the embodiments of the present application, when the available computing resources do not meet the preset computing resource requirement conditions, first generate a three-dimensional view model through the first specific task generative adversarial network, and then the second specific task generative adversarial network generates a three-dimensional voxel model according to the two-dimensional views of each surface of the three-dimensional view model, so as to indirectly complete the generation of the model, reduce the demand for available computing resources, and can also generate the model relatively quickly when the available computing resources cannot meet the computing resource requirement conditions.

[0075] Among them, in the embodiments of the present application, the method may further include:

[0076] If the available computing resources meet the preset computing resource requirement conditions, input the component size information and component material information of each component into the generator of the orthogonal frequency spectrum space generative adversarial network for three-dimensional modeling processing to obtain a three-dimensional initial model corresponding to the server to be modeled;

[0077] Optimize the three-dimensional initial model to obtain a motherboard view corresponding to the server to be modeled.

[0078] Among them, the generator of the orthogonal spectral space generative adversarial network can refer to the generator of the generative adversarial network that can directly generate a three-dimensional model based on the component size information and component material information. In the embodiments of the present application, the orthogonal spectral space generative adversarial network can be FT-GAN (Fourier Transform GAN). After receiving the structured component size information and component material information, FT-GAN uses the Fourier transform to convert the spatial domain data into spectral domain data, extracts spectral features, and then uses a convolutional neural network and a residual network to combine the spatial domain features and spectral features to form a comprehensive feature representation. Then, the generator generates a three-dimensional initial model based on the comprehensive feature representation and the learned assembly relationships between components.

[0079] In the embodiments of the present application, when the available computing resources meet the preset computing resource requirement conditions, a three-dimensional initial model can be directly generated through the orthogonal spectral space generative adversarial network, and the model generation can be completed more quickly when the available computing resources are sufficient.

[0080] S103. Perform projection processing on the three-dimensional view model to obtain two-dimensional views corresponding to each face of the three-dimensional view model.

[0081] Among them, the method of performing projection processing on the three-dimensional view model to obtain two-dimensional views corresponding to each face of the three-dimensional view model can include performing orthogonal projection on the front, left, and top surfaces of the three-dimensional view model in sequence to obtain the three views of the three-dimensional view model, and can also perform orthogonal projection on the six surfaces of the three-dimensional view model in sequence to obtain two-dimensional views of all surfaces.

[0082] S104. Input the two-dimensional views of each face into the generator of the second specific task generative adversarial network to obtain a three-dimensional voxel model corresponding to the to-be-modeled server. The generator of the second generative adversarial network is used to convert the two-dimensional view into a three-dimensional voxel model.

[0083] Among them, the generator of the second specific task generative adversarial network can refer to the generator of the generative adversarial network that can generate a three-dimensional voxel model based on the two-dimensional view. In the embodiments of the present application, the second specific task generative adversarial network can be Pix2Vox. After receiving the two-dimensional view, Pix2Vox uses a convolutional neural network to extract the features of the two-dimensional view, and then converts the two-dimensional view into a three-dimensional voxel model through the generator.

[0084] Among them, in the embodiments of the present application, the first specific task generative adversarial network and the second specific task generative adversarial network can be jointly trained in the following manner:

[0085] Obtain training samples, where the training samples include a training server model, as well as the training component size information, training component material information, and training component assembly relationship information corresponding to each training component in the training server;

[0086] Input the training component size information, training component material information, and training component assembly relationship information of each training component into the generator of the first specific task generative adversarial network for feature extraction processing to obtain the training 3D view model corresponding to the training server;

[0087] Perform projection processing on the training 3D view model corresponding to the training server to obtain the training 2D views corresponding to each face of the training 3D view;

[0088] Input the training 2D views of each face into the generator of the second specific task generative adversarial network for dimension conversion processing to obtain the training 3D voxel model corresponding to the training server;

[0089] Input the training 3D view model and the training server model into the discriminator of the first specific task generative adversarial network to obtain the first loss information, and input the training 3D voxel model and the training server model into the discriminator of the second specific task generative adversarial network to obtain the second loss information;

[0090] Train the generator of the first specific task generative adversarial network and the generator of the second specific task generative adversarial network according to the first loss information and the second loss information.

[0091] Among them, the training server model can refer to the server whole machine model with an assembly structure. The training component size information can refer to the geometric dimension data of each component in the training server. The training component material information can refer to the material types and their physical properties used for each component in the training server. The training component assembly relationship information can refer to the assembly relationship information such as the relative positions, connection methods, and constraint conditions between each component in the training server.

[0092] The training process of the first specific task generative adversarial network is as follows. After receiving the training component size information, training component material information, and training component assembly relationship information, the first specific task generative adversarial network uses a residual network to extract the features of the input data, then generates a training 3D view model through the generator, and then inputs the training 3D view model and the training server model into the discriminator of the first specific task generative adversarial network to obtain the first loss information representing the deviation value between the two, and reversely guides the training of the generator of the first specific task generative adversarial network according to the first loss information.

[0093] The training process of the second specific task generative adversarial network is as follows. After receiving the training 2D views, the second specific task generative adversarial network uses a convolutional neural network to extract the features of the training 2D views, then converts the training 2D views into a training 3D voxel model through a generator, and then inputs the training 3D voxel model and the training server model into the discriminator of the second specific task generative adversarial network to obtain the second loss information representing the deviation value between the two. According to the second loss information, the generator of the second specific task generative adversarial network is trained in a reverse direction.

[0094] After model training, an independent validation set can also be used to evaluate the generation quality of the model. The generated 3D models are evaluated through visual inspection and automated metrics such as shape similarity and structural integrity, and it is determined whether the model needs to be retrained according to the evaluation results.

[0095] Among them, in the embodiments of the present application, the orthogonal spectral space generative adversarial network can be trained in the following manner:

[0096] Input the training component size information, training component material information, and training component assembly relationship information into the generator of the orthogonal spectral space generative adversarial network for 3D modeling processing to obtain the training 3D initial model corresponding to the training modeling server;

[0097] Input the training 3D initial model and the training server model into the discriminator of the orthogonal spectral space generative adversarial network to obtain the third loss information;

[0098] According to the third loss information, the generator of the orthogonal spectral space generative adversarial network is trained.

[0099] Among them, the training process of the orthogonal spectral space generative adversarial network is as follows. After receiving the training component size information, training component material information, and training component assembly relationship information, the orthogonal spectral space generative adversarial network uses Fourier transform to convert the training spatial domain data into training spectral domain data, extracts the training spectral features, then uses a convolutional neural network and a residual network to combine the training spatial domain features and the training spectral features to form a training comprehensive feature representation. Then, the generator generates a training 3D initial model based on the training comprehensive feature representation, and then inputs the training 3D initial model and the training server model into the discriminator of the orthogonal spectral space generative adversarial network to obtain the third loss information representing the deviation value between the two. According to the third loss information, the generator of the orthogonal spectral space generative adversarial network is trained in a reverse direction.

[0100] S105. Optimize the 3D voxel model to obtain the motherboard view corresponding to the server to be modeled.

[0101] Among them, the motherboard view can refer to a view extracted from a three-dimensional voxel model that specifically shows the motherboard structure and its connection relationship with other components, and can directly reflect information such as the layout of the motherboard, the interface position, and the connection method.

[0102] Among them, in the embodiments of the present application, the method for optimizing the three-dimensional voxel model to obtain the motherboard view corresponding to the server to be modeled may include:

[0103] Input the three-dimensional voxel model into the generator of the information bottleneck generative adversarial network. In the generator of the information bottleneck generative adversarial network, extract the three-dimensional features of the three-dimensional voxel model, determine the latent space representation of the three-dimensional features, and perform reconstruction optimization on the latent space representation to obtain the motherboard view corresponding to the server to be modeled.

[0104] Among them, the generator of the information bottleneck generative adversarial network can refer to the generator of the generative adversarial network that can optimize the three-dimensional voxel model. In the embodiments of the present application, the information bottleneck generative adversarial network can be 3D VAE (3D Variational Auto-Encoders, three-dimensional variational auto-encoder), 3D ShapeNets (three-dimensional shape network), etc. The method for optimizing the three-dimensional voxel model may include inputting the three-dimensional voxel model into the generator of 3D VAE. In the generator of 3D VAE, extract the three-dimensional features of the three-dimensional voxel model through a convolutional neural network, determine the latent space representation of the three-dimensional features, and perform reconstruction optimization on the latent space representation to obtain a preliminarily optimized three-dimensional voxel model. Then input the preliminarily optimized three-dimensional voxel model into the generator of 3D ShapeNets, use the convolutional neural network to further refine the three-dimensional features, and then generate a high-precision three-dimensional voxel model. The format of the high-precision three-dimensional voxel model can be STJ (stereolithography, stereolithography), OBJ (Object File Format, object file), etc., and obtain the motherboard view corresponding to the server to be modeled according to the high-precision three-dimensional voxel model.

[0105] Among them, in the embodiments of the present application, the method for optimizing the three-dimensional voxel model to obtain the motherboard view corresponding to the server to be modeled may include:

[0106] Determine the optimal three-dimensional voxel model among the multiple three-dimensional voxel models corresponding to the server to be modeled;

[0107] According to the generator of the information bottleneck generative adversarial network, optimize the model of the optimal three-dimensional voxel model to obtain the motherboard view corresponding to the server to be modeled.

[0108] Among them, after the input component dimension information and component material information are processed by the generators of the first specific task generative adversarial network and the second specific task generative adversarial network, multiple 3D voxel models can be obtained, representing all assembly possibilities.

[0109] The method for determining the optimal 3D voxel model may include presenting multiple 3D voxel models to the user. The user views and selects them according to actual needs, and determines the model that best meets the requirements among the multiple 3D voxel models as the optimal 3D voxel model. It is also possible to preset the scoring criteria and rules for the 3D voxel models in advance, and take the model with the highest score among the multiple 3D voxel models as the optimal 3D voxel model.

[0110] The server structure model generation method provided by the embodiments of this application uses the generator of the first specific task generative adversarial network to generate a 3D view model based on the assembly relationship, and then uses the generator of the second specific task generative adversarial network to convert the 2D projection of the 3D view model into a 3D voxel model. After optimizing the 3D voxel model, a motherboard view of the server to be modeled is generated. This technical means solves the technical problem of being unable to quickly generate a model with an assembly relationship and having a low modeling efficiency.

[0111] Figure 2 It is a schematic flowchart of another server structure model generation method provided by the embodiments of this application. Figure 3 It is a step diagram of another server structure model generation method provided by the embodiments of this application. Figure 4 It is a model processing flowchart of another server structure model generation method provided by the embodiments of this application. As shown in Figure 2 、 3 、4, this method includes:

[0112] S201. Input the dimension information of each component in the server into the StyleGAN3 model. The StyleGAN3 model obtains a 3D view model according to the dimension information of each component.

[0113] Among them, the method of inputting into the StyleGAN3 model includes that the user enters the server AI (Artificial Intelligence) modeling system, and inputs the configuration data containing the dimension information of each component into the StyleGAN3 model through the API (Application Programming Interface).

[0114] Before inputting into the StyleGAN3 model, it is also possible to judge the computing resources of the current system. If the computing resources meet the preset computing resource requirement conditions, that is, the computing resources are sufficient, then it is input into the StyleGAN3 model. If the computing resources do not meet the preset computing resource requirement conditions, that is, the computing resources are insufficient, then it is input into the OSSGAN (Orthogonal Spectrum Space Generative Adversarial Network) model, and the OSSGAN model is the OSSGAN model that has learned the assembly relationship between components.

[0115] The StyleGAN3 model is the StyleGAN3 model that has learned the assembly relationship between components.

[0116] S202. Perform projection processing on the three-dimensional view model to obtain two-dimensional views corresponding to each face of the three-dimensional view model.

[0117] Among them, the method of performing projection processing on the three-dimensional view model to obtain two-dimensional views corresponding to each face of the three-dimensional view model can include performing orthogonal projection from the front, left, and top of the three-dimensional view model in sequence to obtain the three views of the three-dimensional view model, and it is also possible to perform orthogonal projection on the six faces of the three-dimensional view model in sequence to obtain two-dimensional views of all faces.

[0118] S203. Input the two-dimensional view into the Pix2Vox model, and the Pix2Vox model obtains a three-dimensional voxel model according to the two-dimensional view.

[0119] Among them, the method by which the Pix2Vox model obtains a three-dimensional voxel model according to the two-dimensional view can include that after receiving the two-dimensional view, the Pix2Vox uses a convolutional neural network to extract the features of the two-dimensional view, and then converts the two-dimensional view into a three-dimensional voxel model through a generator.

[0120] S204. Display multiple three-dimensional voxel models to the user, and the user determines the optimal three-dimensional voxel model.

[0121] Among them, the method by which the user determines the optimal three-dimensional voxel model can include that the user views and selects multiple three-dimensional voxel models according to actual needs, and determines the model that best meets the needs among the multiple three-dimensional voxel models as the optimal three-dimensional voxel model.

[0122] S205. Input the optimal three-dimensional voxel model into the 3D VAE model and the 3D ShapeNets model in sequence, and the 3D VAE model and the 3D ShapeNets model refine the optimal three-dimensional voxel model to obtain a high-precision three-dimensional model.

[0123] Among them, the method for optimizing the optimal three-dimensional voxel model may include inputting the optimal three-dimensional voxel model into the generator of the 3D VAE. In the generator of the 3D VAE, three-dimensional features of the three-dimensional voxel model are extracted through a convolutional neural network, and a latent space representation of the three-dimensional features is determined. Then, the latent space representation is reconstructed and optimized to obtain a preliminarily optimized optimal three-dimensional voxel model. Next, the preliminarily optimized optimal three-dimensional voxel model is input into the generator of the 3D ShapeNets, and the convolutional neural network is used to further refine the three-dimensional features to generate a high-precision three-dimensional voxel model.

[0124] S206. Obtain the main board view corresponding to the server according to the high-precision three-dimensional model.

[0125] Among them, the main board view may refer to a view extracted from the three-dimensional voxel model that specifically shows the structure of the main board and its connection relationship with other components, and can directly reflect information such as the layout of the main board, the position of the interfaces, and the connection method.

[0126] After obtaining the main board view corresponding to the server, it can also be displayed to the user at the front end.

[0127] The high-precision three-dimensional model and the main board view can be saved to the background database, and the models in the server AI modeling system are updated accordingly.

[0128] Another server structure model generation method provided by the embodiments of the present application can sequentially process the dimension information of each component in the server through the StyleGAN3 model, the Pix2Vox model, the 3D VAE model, and the 3D ShapeNets model to obtain a high-precision three-dimensional model, and then obtain the main board view corresponding to the server according to the high-precision three-dimensional model, solving the technical problem of being unable to quickly generate a model with an assembly relationship and having a low modeling efficiency.

[0129] Figure 5 It is a schematic structural diagram of a server structure model generation device provided by the embodiments of the present application. As Figure 5 shown, the server structure model generation device 50 includes: an acquisition module 501, a first obtaining module 502, a second obtaining module 503, a third obtaining module 504, and a fourth obtaining module 505. Among them:

[0130] The acquisition module 501 is used to acquire the component dimension information and component material information corresponding to each component in the server to be modeled;

[0131] The first obtaining module 502 is configured to input the component size information and component material information of each component into the generator of the first specific task generative adversarial network. In the generator, based on the learned assembly relationships between components, a three-dimensional view model corresponding to the server to be modeled is obtained according to the component size information and component material information, and the three-dimensional view model conforms to the assembly relationships between components;

[0132] The second obtaining module 503 is configured to perform projection processing on the three-dimensional view model to obtain two-dimensional views corresponding to each surface of the three-dimensional view model;

[0133] The third obtaining module 504 is configured to input the two-dimensional views of each surface into the generator of the second specific task generative adversarial network to obtain a three-dimensional voxel model corresponding to the server to be modeled, and the generator of the second generative adversarial network is used to convert the two-dimensional view into a three-dimensional voxel model;

[0134] The fourth obtaining module 505 is configured to optimize the three-dimensional voxel model to obtain a main board view corresponding to the server to be modeled.

[0135] In the embodiment of the present application, the first obtaining module 502 may further be configured to:

[0136] Obtain available computing resources;

[0137] Determine that the available computing resources do not meet the preset computing resource requirement conditions.

[0138] In the embodiment of the present application, the first obtaining module 502 may further be configured to:

[0139] If the available computing resources meet the preset computing resource requirement conditions, input the component size information and component material information of each component into the generator of the orthogonal frequency spectrum space generative adversarial network for three-dimensional modeling processing to obtain a three-dimensional initial model corresponding to the server to be modeled;

[0140] Optimize the three-dimensional initial model to obtain a main board view corresponding to the server to be modeled.

[0141] In the embodiment of the present application, the third obtaining module 504 may further be configured to:

[0142] Obtain training samples, where the training samples include a training server model, and training component size information, training component material information, and training component assembly relationship information corresponding to each training component in the training server;

[0143] Input the training component size information, training component material information, and training component assembly relationship information of each training component into the generator of the first specific task generative adversarial network for feature extraction processing to obtain a training three-dimensional view model corresponding to the training server;

[0144] Perform projection processing on the training three-dimensional view model corresponding to the training server to obtain training two-dimensional views corresponding to each face of the training three-dimensional view;

[0145] Input the training two-dimensional views of each face into the generator of the second specific task generative adversarial network for dimensionality conversion processing to obtain the training three-dimensional voxel model corresponding to the training server;

[0146] Input the training three-dimensional view model and the training server model into the discriminator of the first specific task generative adversarial network to obtain the first loss information, and input the training three-dimensional voxel model and the training server model into the discriminator of the second specific task generative adversarial network to obtain the second loss information;

[0147] Train the generator of the first specific task generative adversarial network and the generator of the second specific task generative adversarial network according to the first loss information and the second loss information.

[0148] In the embodiment of the present application, the third obtaining module 504 can also be used for:

[0149] Input the training component size information, training component material information, and training component assembly relationship information into the generator of the orthogonal frequency spectrum space generative adversarial network for three-dimensional modeling processing to obtain the training three-dimensional initial model corresponding to the training modeling server;

[0150] Input the training three-dimensional initial model and the training server model into the discriminator of the orthogonal frequency spectrum space generative adversarial network to obtain the third loss information;

[0151] Train the generator of the orthogonal frequency spectrum space generative adversarial network according to the third loss information.

[0152] In the embodiment of the present application, the fourth obtaining module 505 can also be used for:

[0153] Input the three-dimensional voxel model into the generator of the information bottleneck generative adversarial network. In the generator of the information bottleneck generative adversarial network, extract the three-dimensional features of the three-dimensional voxel model, determine the latent space representation of the three-dimensional features, and perform reconstruction optimization on the latent space representation to obtain the motherboard view corresponding to the server to be modeled.

[0154] In the embodiment of the present application, the fourth obtaining module 505 can also be used for:

[0155] Determine the optimal three-dimensional voxel model among the multiple three-dimensional voxel models corresponding to the server to be modeled;

[0156] Optimize the optimal three-dimensional voxel model according to the generator of the information bottleneck generative adversarial network to obtain the motherboard view corresponding to the server to be modeled.

[0157] As can be seen from the above, the server structure model generation device according to the embodiment of the present application includes an acquisition module 501 for acquiring the component size information and component material information corresponding to each component in the server to be modeled; a first obtaining module 502 for inputting the component size information and component material information of each component into the generator of the first specific task generative adversarial network. In the generator, based on the learned assembly relationship between components, a three-dimensional view model corresponding to the server to be modeled is obtained according to the component size information and component material information, and the three-dimensional view model conforms to the assembly relationship between components; a second obtaining module 503 for performing projection processing on the three-dimensional view model to obtain two-dimensional views corresponding to each surface of the three-dimensional view model; a third obtaining module 504 for inputting the two-dimensional views of each surface into the generator of the second specific task generative adversarial network to obtain a three-dimensional voxel model corresponding to the server to be modeled, and the generator of the second generative adversarial network is used to convert the two-dimensional view into a three-dimensional voxel model; a fourth obtaining module 505 for optimizing the three-dimensional voxel model to obtain a main board view corresponding to the server to be modeled. Thus, the embodiment of the present application can generate a three-dimensional view model based on the assembly relationship through the first specific task generative adversarial network, and then convert the two-dimensional projection of the three-dimensional view model into a three-dimensional voxel model through the second specific task generative adversarial network, and optimize the three-dimensional voxel model to generate the main board view of the server to be modeled, thereby solving the technical problem of being unable to quickly generate a model with an assembly relationship and having a low modeling efficiency.

[0158] Figure 6 This is a schematic structural diagram of the electronic device provided by the present application. As Figure 6 shown, the electronic device 60 provided in this embodiment includes at least one processor 601 and a memory 602. Optionally, the device 60 further includes a communication component 603. Among them, the processor 601, the memory 602, and the communication component 603 are connected through a bus 604.

[0159] In a specific implementation process, at least one processor 601 executes the computer execution instructions stored in the memory 602, so that at least one processor 601 executes the above method.

[0160] For the specific implementation process of the processor 601, reference can be made to the above method embodiment, and its implementation principle and technical effect are similar, so details are not described here in this embodiment.

[0161] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU for short), or may also be other general-purpose processors, digital signal processors (DSP for short), application specific integrated circuits (ASIC for short), etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the method disclosed in combination with the invention can be directly implemented by the execution of the hardware processor, or can be implemented by the combination of hardware and software modules in the processor.

[0162] The memory may include a random access memory (RAM), and may also include a non-volatile memory (NVM), such as at least one disk memory.

[0163] The bus may be an industry standard architecture (ISA) bus, a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, the bus in the drawings of this application is not limited to only one bus or one type of bus.

[0164] This application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0165] This application also provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the processor executes the computer-executable instructions, the above method is implemented.

[0166] The above-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk. The readable storage medium can be any available medium accessible by a general-purpose or special-purpose computer.

[0167] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an Application Specific Integrated Circuits (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in a device.

[0168] The division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Additionally, the couplings or direct couplings or communication connections shown or discussed between each other can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0169] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0170] In addition, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0171] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in each embodiment of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks or optical discs that can store program codes.

[0172] Those of ordinary skill in the art will understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including those of the above method embodiments; and the foregoing storage medium includes: various media such as ROM, RAM, magnetic disks, or optical discs that can store program codes.

[0173] Finally, it should be noted that: After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily think of other embodiments of the present invention. The present invention is intended to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include common general knowledge or conventional technical means in the technical field not disclosed by the present invention. It is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A method for generating a server structure model, characterized in that: include: Obtain component size information and component material information corresponding to each component in the server to be modeled; Inputting the component size information and component material information of each component into a generator of a first specific task generative adversarial network, wherein the generator obtains a three-dimensional view model corresponding to the server to be modeled according to the component size information and the component material information based on the learned assembly relationship between the components, and the three-dimensional view model conforms to the assembly relationship between the components; Performing projection processing on the three-dimensional view model to obtain two-dimensional views of each surface of the three-dimensional view model; Inputting the two-dimensional views of each face into a generator of a second specific task generative adversarial network to obtain a three-dimensional voxel model corresponding to the server to be modeled, wherein the generator of the second generative adversarial network is used to convert the two-dimensional view into a three-dimensional voxel model; The three-dimensional voxel model is optimized to obtain a mainboard view corresponding to the server to be modeled.

2. The method according to claim 1, characterized in that The step of optimizing the three-dimensional voxel model to obtain a mainboard view corresponding to the server to be modeled includes: The three-dimensional voxel model is input into a generator of an information bottleneck generative adversarial network. In the generator of the information bottleneck generative adversarial network, the three-dimensional features of the three-dimensional voxel model are extracted, the latent space representation of the three-dimensional features is determined, and the latent space representation is reconstructed and optimized to obtain a motherboard view corresponding to the server to be modeled.

3. The method according to claim 1 or 2, characterized in that: The step of optimizing the three-dimensional voxel model to obtain a mainboard view corresponding to the server to be modeled includes: Determining an optimal three-dimensional voxel model among a plurality of three-dimensional voxel models corresponding to the server to be modeled; A generator of an adversarial network is generated according to the information bottleneck, and the optimal three-dimensional voxel model is optimized to obtain a mainboard view corresponding to the server to be modeled.

4. The method according to claim 1 or 2, characterized in that: The first specific task generation adversarial network and the second specific task generation adversarial network are obtained by joint training in the following manner: Acquire a training sample, wherein the training sample includes a training server model, and training component size information, training component material information, and training component assembly relationship information corresponding to each training component in the training server; Inputting the training component size information, the training component material information and the training component assembly relationship information of each training component into the generator of the first specific task generative adversarial network for feature extraction processing to obtain a training 3D view model corresponding to the training server; Projecting the training three-dimensional view model corresponding to the training server to obtain training two-dimensional views of each face corresponding to the training three-dimensional view; Inputting the training two-dimensional views of each face into the generator of the second specific task generative adversarial network for dimension conversion processing to obtain a training three-dimensional voxel model corresponding to the training server; Inputting the training 3D view model and the training server model into a discriminator of the first specific task generative adversarial network to obtain first loss information, and inputting the training 3D voxel model and the training server model into a discriminator of the second specific task generative adversarial network to obtain second loss information; According to the first loss information and the second loss information, the generator of the first specific task generative adversarial network and the generator of the second specific task generative adversarial network are trained.

5. The method according to claim 1 or 2, characterized in that: Before inputting the component size information and component material information of each component into the generator of the first specific task generative adversarial network, the method further includes: Obtain available computing resources; It is determined that the available computing resources do not meet a preset computing resource requirement condition.

6. The method according to claim 5, characterized in that The method further comprises: If the available computing resources meet the preset computing resource requirement conditions, the component size information and component material information of each component are input into the generator of the orthogonal spectrum space generative adversarial network for three-dimensional modeling processing to obtain a three-dimensional initial model corresponding to the server to be modeled; The three-dimensional initial model is optimized to obtain a mainboard view corresponding to the server to be modeled.

7. The method according to claim 6, characterized in that The orthogonal spectrum space generative adversarial network is trained in the following way: Inputting the training component size information, the training component material information and the training component assembly relationship information into the generator of the orthogonal spectrum space generative adversarial network for three-dimensional modeling processing to obtain a training three-dimensional initial model corresponding to the training modeling server; Inputting the training three-dimensional initial model and the training server model into the discriminator of the orthogonal spectrum space generative adversarial network to obtain third loss information; According to the third loss information, the generator of the orthogonal spectrum space generative adversarial network is trained.

8. A server structure model generating device, characterized in that: include: An acquisition module is used to acquire component size information and component material information corresponding to each component in the server to be modeled; A first obtaining module is used to input the component size information and component material information of each component into a generator of a first specific task generative adversarial network, in which, based on the learned assembly relationship between the components, a three-dimensional view model corresponding to the server to be modeled is obtained according to the component size information and the component material information, and the three-dimensional view model conforms to the assembly relationship between the components; A second obtaining module is used to perform projection processing on the three-dimensional view model to obtain two-dimensional views of each surface of the three-dimensional view model; A third obtaining module is used to input the two-dimensional views of each face into a generator of a second specific task generative adversarial network to obtain a three-dimensional voxel model corresponding to the server to be modeled, and the generator of the second generative adversarial network is used to convert the two-dimensional view into a three-dimensional voxel model; The fourth obtaining module is used to optimize the three-dimensional voxel model to obtain a mainboard view corresponding to the server to be modeled.

9. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

Images