A free-form surface meshing design method for open-hole buildings based on generative adversarial networks based on attention enhancement

Abstract

The present invention discloses a mesh division design method for a free-form surface of a building with holes based on an attention-enhanced generative adversarial network. The method comprises the following steps: step S1, obtaining a free-form surface of a building with holes that needs to be meshed, and constructing a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation and a free-form surface semantic mask representation according to the free-form surface of the building with holes; step S2, inputting the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation into a pre-trained attention-enhanced generative adversarial network model, and generating a mesh structure of the free-form surface with holes through the attention-enhanced generative adversarial network model; step S3, extracting mesh nodes and topological relationships in the mesh structure of the free-form surface generated by the attention-enhanced generative adversarial network model, and then restoring it to a three-dimensional space, so as to finally obtain a three-dimensional mesh structure of the free-form surface with holes.

Description

Technical Field

[0001] The present invention belongs to the field of civil engineering and computer deep learning application technology, and in particular relates to a free-form surface meshing design method for open-hole buildings based on an attention-enhanced generative adversarial network. Background Art

[0002] In recent years, architectural free-form surface structures have been widely used in large-span space buildings such as stadiums, large airports, and exhibition halls due to their rich shapes. However, the diversity and irregularity of free-form surface shapes make reasonable meshing design a research problem, especially due to the need to open holes on the surface due to practical use. The meshing is more difficult.

[0003] At present, new data-driven paradigms such as artificial intelligence are profoundly changing various fields, and structural intelligent design methods based on deep learning image generation algorithms have made significant progress. Taking architectural design as an example, the use of generative adversarial network technology can reduce the dependence on the experience of engineering designers and simplify the complex process of explicit programming that must consider all surface characteristics. However, for free-form surfaces with openings, the quality of local meshing at the openings of generative adversarial networks is still insufficient. In addition, existing design data and experience are difficult to repeat and effectively apply, and this problem needs to be solved urgently. Therefore, it is urgent to develop intelligent design methods that can learn from existing design experience and generate new designs. Summary of the invention

[0004] Purpose of the invention: The present invention aims to solve the problem that existing meshing technologies have poor meshing capabilities in processing free-form surfaces of buildings with openings, and proposes a meshing design method for free-form surfaces of buildings with openings based on attention-enhanced generative adversarial networks. This method can effectively process the meshing near the openings of free-form surfaces of buildings, thereby improving the overall meshing quality of free-form surfaces of buildings. At the same time, this method greatly avoids the code defects caused by explicit programming and the cumbersome process of manual mesh optimization and adjustment.

[0005] Technical solution, in order to solve the above-mentioned technical problems, the present invention provides a free-form surface meshing design method for open-hole buildings based on a generative adversarial network with enhanced attention, the method comprising the following steps:

[0006] Step S1, obtaining a free-form surface of an open-hole building that needs to be meshed, and constructing a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation according to the free-form surface of the open-hole building;

[0007] Step S2, inputting the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation into a pre-trained attention-enhanced generative adversarial network model, and generating an open-hole free-form surface mesh structure through the attention-enhanced generative adversarial network model;

[0008] Step S3: extracting the mesh nodes and topological relationships of the open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model and restoring it to a three-dimensional space to finally obtain a three-dimensional open hole free-form surface mesh structure.

[0009] Furthermore, in step S1, the method for constructing the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the free-form surface semantic mask representation is:

[0010] (1) Constructing a free-form surface curvature cloud map representation: Using a color map with 3 color channels, the free-form surface of the building with openings is input using the battery function in the Grasshopper software. The maximum principal curvature value of each point on the free-form surface is automatically calculated and the corresponding RGB value is obtained to obtain a free-form surface curvature cloud map representation of the free-form surface of the building with openings.

[0011] (2) Constructing the elevation cloud map representation of the free-form surface: Using a color map with 3 color channels, the free-form surface of the building with openings is input using the battery function in the Grasshopper software. The elevation value of each point on the free-form surface is automatically calculated and the corresponding RGB value is obtained to obtain the elevation cloud map representation of the free-form surface of the building with openings.

[0012] (3) Constructing semantic mask: It is represented by a grayscale image with a color channel number of 1, and the free-form surface and the opening area of ​​the building are represented at the same time. The surface itself is represented by a grayscale value of 0, and the opening area is represented by a grayscale value of 142.

[0013] Furthermore, in step S2, the construction process of the pre-trained attention-enhanced generative adversarial network model specifically includes the following:

[0014] (1) Obtaining the free-form surface of the opening building and its corresponding free-form surface grid structure;

[0015] (2) constructing a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation corresponding to the free-form surface of the opening building in step (1);

[0016] (3) The free-form surface curvature cloud map representation, free-form surface elevation cloud map representation, and free-form surface semantic mask representation are used as the input of the model, and the corresponding open hole free-form surface mesh structure is used as the model output;

[0017] (4) dividing the data in step (3) into a training data set and a test data set according to the form of input and output data pairs;

[0018] (5) Based on the variant of the generative adversarial network model GridGAN, its model input channel and model structure are reconstructed so that the new generative adversarial network model simultaneously inputs the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation, and includes an attention mechanism. The new generative adversarial network model including the attention mechanism is named GridGAN-AE;

[0019] (6) training the generative adversarial network model GridGAN-AE containing the attention mechanism in step (5) by using the training set consisting of the curvature cloud map representation of the open-hole free-form surface, the elevation cloud map representation of the free-form surface, the semantic mask representation of the free-form surface, and the grid structure of the open-hole free-form surface in step (4);

[0020] (7) The generative adversarial network model GridGAN-AE with attention mechanism trained in step (5) is tested and evaluated by using the test set consisting of the free-form surface curvature cloud map representation, free-form surface elevation cloud map representation, free-form surface semantic mask representation and free-form surface mesh structure with holes in step (4). The model that meets the requirements of computer vision evaluation indicators and free-form surface mesh structure geometry evaluation indicators is used as the final generative adversarial network model GridGAN-AE with attention mechanism.

[0021] Furthermore, in step (5), the model input channel and model structure are reconstructed as follows:

[0022] (1) The operation of reconstructing the model input channel is as follows: The reconstruction of the model input channel specifically involves modifying the model generator and discriminator; the generator adds a new image input interface, and the number of model input channels is changed from 6 to 7. At this point, the generator input is expressed as formula (1), where B size is the number of images processed by the generator at a time, 7 represents the number of model input channels, and H img and W img They represent the height and width of the image respectively. The modification of the discriminator is the same as that of the generator. A new image input interface is added. The number of model input channels is changed from 9 to 10. The input of the discriminator of the model is expressed as formula (2).

[0023] (B size ,7,H img ,W img )(1)

[0024] (B size ,10,H img ,W img )(2)

[0025] (2) The operations to reconstruct the model structure are as follows: adding an attention mechanism to enhance the model’s ability to learn the correlations between different discrete spatial regions and the local features that must be focused on for autonomous learning; specifically, in the generator, the self-attention layer is configured between the output of the 5th-layer convolutional feature extraction network and the input of the residual network; in the discriminator, the discriminator contains two sub-discriminators with exactly the same structure, and a self-attention layer is inserted after the 2nd-layer convolutional network of each sub-discriminator.

[0026] Furthermore, in step (7), the specific calculation formulas of the computer vision evaluation index and the free-form surface mesh structure geometric evaluation index are as follows:

[0027] (1) The evaluation index based on computer vision is shown in formula (3), μ real and μ gen are the mean vector of the real free-form surface mesh image of the open-hole building and the mean vector of the free-form surface mesh structure image of the open-hole building generated by the generative adversarial network model GridGAN-AE containing the attention mechanism, X and X respectively. m They respectively represent the covariance matrix of the real free-form surface mesh image of the open-hole building and the covariance matrix of the free-form surface mesh structure image of the open-hole building generated by the generative adversarial network model GridGAN-AE containing the attention mechanism. Tr represents the trace of the matrix, that is, the sum of the diagonal elements of the matrix. In formula (3), the first part Represents the Euclidean distance between two mean vectors, which is used to evaluate the difference in the center points of the two sets of image features. The second part evaluates the difference in their distribution shapes;

[0028]

[0029] When the FID value is greater than 60, it is considered that the mesh generation effect does not meet the requirements; when the FID is below 60, it is considered that the mesh generation effect based on computer vision evaluation meets the requirements, and the free-form surface mesh structure geometric evaluation index is further performed;

[0030] (2) The geometric evaluation index of the free-form surface mesh structure includes the mean value index of the member length and the mesh shape quality index, where the member length is the length of the line connecting two mesh nodes in the free-form surface mesh structure, and the mesh shape refers to the basic geometric unit that constitutes the free-form surface mesh structure, including quadrilateral mesh basic units and triangular mesh basic units;

[0031] (2.1) Mean length of the member Calculation formula and mean square deviation of member length δ L The calculation formulas are shown in equations (4) and (5), where n is the total number of rods, L i represents the length of the i-th rod;

[0032]

[0033]

[0034] (2.2) The free-form surface grid structure of the open-hole building is composed of n r quadrilateral grid basic units and n t The shape quality coefficient of the i-th quadrilateral mesh basic unit is and the shape quality coefficient of the jth triangular mesh basic unit See equations (6) and (7) respectively, where i = 1, 2, ..., n r , j=1,2,…,n t , A, B, C and D represent the vertices of the basic unit, S ΔABC , S ΔADC , S ΔADB and S ΔCBD are the areas of the triangles formed by the corresponding nodes of A, B, C and D, l AB , l AD , l AC , l BC , l CD is the side length of the connecting rod corresponding to the vertex, and The change pattern of is consistent, and its value range is (0,1];

[0035]

[0036]

[0037] (2.3) After obtaining the shape quality coefficients of the quadrilateral mesh basic unit and the triangular mesh basic unit, calculate the average shape quality coefficient of the free-form surface mesh structure of the entire opening building and mean square error δ q , the calculation formulas are respectively formula (8) and formula (9);

[0038]

[0039]

[0040] (2.4) The comprehensive index for geometric evaluation of free-form surface mesh structure is formula (10), which considers both the mean index of member length and the mesh shape quality index, where J is the comprehensive index, α and β are the mean square error of member length δ L and shape quality coefficient mean square error δ qThe corresponding weight coefficients are: when J is greater than the preset value, the grid generation result is considered to be unsatisfactory; when J is less than the preset value, the grid generation result is satisfactory;

[0041] J=αδ L +βδ q (10).

[0042] Furthermore, in step S3, the open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model is restored to a three-dimensional space by extracting the mesh nodes and topological relationships therein, so as to obtain a three-dimensional open hole free-form surface mesh structure in the following specific steps:

[0043] (1) The coordinate values ​​of the grid nodes in the open free-form surface grid structure image generated by the attention-enhanced generative adversarial network model are converted into key-value pairs {K:V pixel} format, where V pixel Stored in the form of two-dimensional Cartesian coordinates, representing the coordinate values ​​of the grid nodes in the grid structure image of the open hole free surface, K is V pixel The corresponding sequence number value;

[0044] (2) The n topological connection relationships of the mesh nodes in the open free-form surface mesh structure image generated by the attention-enhanced generative adversarial network model are listed in C connect The form of storage is as follows: i It can be further expressed as formula (12), which represents the connection relationship between the i-th two nodes, n p ∈K,n k ∈K, and n p ≠n k ;

[0045] C connect =[C1,C2,…,C i ,…,C n ] (11)

[0046] C i =[n p ,n k ] (12)

[0047] (3) Coordinate alignment: using the coordinates N of the lower left corner of the free-form surface mesh structure with the hole in step S3 LB (x LB ,y LB ) and the coordinates of the upper right corner point N RT (x RT ,y RT ) and the coordinates of the lower left corner of the corresponding real opening free-form surface and the coordinates of the upper right corner Calculate the translation coefficient [M x ,M y ] and the scaling factor [S x ,S y ]:

[0048] (3.1) Calculate N LB (x LB ,y LB )and The translation coefficient [M x ,M y ],in,

[0049] (3.2) Using the translation coefficient [M x ,M y ] Calculate the N after translation RT (x RT ,y RT ) pixel coordinates and expressed as in,

[0050] (3.3) Calculate the scaling factor using formula (13);

[0051]

[0052] (4) Obtain the true values ​​of the plane coordinate nodes of the free-form mesh structure with holes {K:V real}:

[0053] (4.1) The coordinate values ​​of the mesh nodes in the mesh structure image of the hole free-form surface in step (1) {K:V pixel Substitute into formula (14) in turn to obtain the coordinate values ​​of the grid nodes after coordinate translation and alignment. Among them, x i and i V pixel The coordinate value of and for The coordinate value of

[0054]

[0055] (4.2) The coordinate values ​​of the grid nodes after coordinate translation and alignment Substituting into equation (15) in turn, we can obtain the true values ​​of the plane coordinate nodes of the free-form mesh structure with holes {K:V real},in, and V real The coordinate value of

[0056]

[0057] (5) Replace {K:V real The mesh nodes are projected onto the corresponding free-form surface with real openings, and the C connect The n mesh node topological connection relationships saved in the [3D] grid are used to connect the mesh nodes that have been projected onto the real open hole free-form surface, and the final three-dimensional open hole free-form surface mesh structure can be obtained.

[0058] In addition, the present invention also proposes a free-form surface meshing design system for open-hole buildings based on attention enhancement generative adversarial network, and the system includes the following modules:

[0059] (1) A data acquisition module is used to acquire the free-form surface of the building with openings that needs to be meshed, and to construct a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation based on the free-form surface of the building with openings;

[0060] (2) a model construction module, which is used to input the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation into a pre-trained attention-enhanced generative adversarial network model, and generate an open-hole free-form surface mesh structure through the attention-enhanced generative adversarial network model;

[0061] (3) A scheme design module is used to extract the mesh nodes and topological relationships of the open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model, and then restore it to three-dimensional space to finally obtain a three-dimensional open hole free-form surface mesh structure.

[0062] In addition, the present invention proposes a computer program product, including a computer program, which, when executed by a processor, implements any one of the attention-enhanced generative adversarial network methods for meshing free-form surfaces of open-hole buildings.

[0063] In addition, the present invention proposes a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement any one of the attention-enhanced generative adversarial network methods for meshing free-form surfaces of open-hole buildings.

[0064] In addition, the present invention proposes an electronic device, comprising a processor, a storage, and a computer program stored in the storage and executable on the processor, characterized in that when the processor executes the program, the program implements any one of the attention-enhanced generative adversarial network methods for meshing free-form surfaces of open-hole buildings.

[0065] Beneficial effects: Compared with the prior art, the present invention adopts the above technical solution to have the following beneficial technical effects:

[0066] (1) This paper proposes a method for meshing free-form surfaces of open-hole buildings based on a generative adversarial network with attention enhancement. This method combines generative adversarial networks with an attention mechanism to solve the problem of poor local mesh quality in the meshing design of open-hole free-form surfaces.

[0067] (2) This method is a data-driven artificial intelligence method. Its underlying logic is to mine and learn the deep rules of existing architectural free-form surfaces and their corresponding grid structure data, so as to achieve effective grid division of given new unknown surfaces. This method makes full use of existing data and empirical knowledge, is a new paradigm for grid division, and has strong practical significance.

[0068] (3) This method combines curvature cloud maps, elevation cloud maps and semantic masks, uses the attention mechanism, and uses artificial intelligence methods to quickly generate the grid structure division results of the free-form surface of the open-hole building, effectively solving the problem of grid division quality near the opening of the free-form surface of the open-hole building. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0070] Figure 1 A schematic diagram of the process of a design method for meshing free-form surfaces of open-hole buildings using a generative adversarial network based on attention enhancement provided by an embodiment of the present invention;

[0071] Figure 2 The visualization process of obtaining the free-form surface grid structure of the open-hole building provided in the embodiment of this article;

[0072] Figure 3 A diagram of obtaining a semantic mask representation of a free-form surface of a building with a hole provided in an embodiment of this article;

[0073] Figure 4 A schematic diagram of the composition of the model "input-output" pair of a generative adversarial network model GridGAN-AE based on attention enhancement provided in an embodiment of this article;

[0074] Figure 5 A schematic diagram of the input composition of the generator and discriminator of the attention-enhanced generative adversarial network model GridGAN-AE provided in the embodiments of this article;

[0075] Figure 6A schematic diagram of a generator and an attention layer addition of a generative adversarial network model GridGAN-AE based on attention enhancement provided in an embodiment of this article;

[0076] Figure 7 A schematic diagram of a process of restoring a free-form surface mesh structure image with holes provided in the embodiments of this article to a free-form surface mesh structure with holes in three-dimensional space;

[0077] Figure 8 A schematic diagram of module connections of a free-form surface meshing design system for open-hole buildings based on a generative adversarial network with enhanced attention provided in an embodiment of this article;

[0078] Fig. 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this article. DETAILED DESCRIPTION

[0079] In order to make the technical solution, advantages and ultimate purpose of the present invention clearer, the present invention will be further explained and illustrated in conjunction with the accompanying drawings and specific implementation methods.

[0080] like Figure 1 As shown, the present invention proposes a free-form surface meshing design method for open-hole buildings based on a generative adversarial network with attention enhancement, and the method comprises the following steps:

[0081] Step S1, obtaining a free-form surface of an open-hole building that needs to be meshed, and constructing a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation according to the free-form surface of the open-hole building;

[0082] Step S2, inputting the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation into a pre-trained attention-enhanced generative adversarial network model, and generating an open-hole free-form surface mesh structure through the attention-enhanced generative adversarial network model;

[0083] Step S3: extracting the mesh nodes and topological relationships of the open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model and restoring it to a three-dimensional space to finally obtain a three-dimensional open hole free-form surface mesh structure.

[0084] Furthermore, in step S1, the method for constructing the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the free-form surface semantic mask representation is:

[0085] (1) Constructing a free-form surface curvature cloud map representation: Using a color map with 3 color channels, the free-form surface of the building with openings is input using the battery function in the Grasshopper software. The maximum principal curvature value of each point on the free-form surface is automatically calculated and the corresponding RGB value is obtained to obtain a free-form surface curvature cloud map representation of the free-form surface of the building with openings.

[0086] (2) Constructing the elevation cloud map representation of the free-form surface: Using a color map with 3 color channels, the free-form surface of the building with openings is input using the battery function in the Grasshopper software. The elevation value of each point on the free-form surface is automatically calculated and the corresponding RGB value is obtained to obtain the elevation cloud map representation of the free-form surface of the building with openings.

[0087] (3) Constructing semantic mask: It is represented by a grayscale image with a color channel number of 1, and the free-form surface and the opening area of ​​the building are represented at the same time. The surface itself is represented by a grayscale value of 0, and the opening area is represented by a grayscale value of 142.

[0088] Furthermore, in step S2, the construction process of the pre-trained attention-enhanced generative adversarial network model specifically includes the following:

[0089] (1) Obtaining the free-form surface of the opening building and its corresponding free-form surface grid structure;

[0090] (2) constructing a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation corresponding to the free-form surface of the opening building in step (1);

[0091] (3) The free-form surface curvature cloud map representation, free-form surface elevation cloud map representation, and free-form surface semantic mask representation are used as the input of the model, and the corresponding open hole free-form surface mesh structure is used as the model output;

[0092] (4) dividing the data in step (3) into a training data set and a test data set according to the form of input and output data pairs;

[0093] (5) Based on the variant of the generative adversarial network model GridGAN, its model input channel and model structure are reconstructed so that the new generative adversarial network model simultaneously inputs the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation, and includes an attention mechanism. The new generative adversarial network model including the attention mechanism is named GridGAN-AE;

[0094] (6) training the generative adversarial network model GridGAN-AE containing the attention mechanism in step (5) by using the training set consisting of the curvature cloud map representation of the open-hole free-form surface, the elevation cloud map representation of the free-form surface, the semantic mask representation of the free-form surface, and the grid structure of the open-hole free-form surface in step (4);

[0095] (7) The generative adversarial network model GridGAN-AE with attention mechanism trained in step (5) is tested and evaluated by using the test set consisting of the free-form surface curvature cloud map representation, free-form surface elevation cloud map representation, free-form surface semantic mask representation and free-form surface mesh structure with holes in step (4). The model that meets the requirements of computer vision evaluation indicators and free-form surface mesh structure geometry evaluation indicators is used as the final generative adversarial network model GridGAN-AE with attention mechanism.

[0096] Furthermore, in step (5), the model input channel and model structure are reconstructed as follows:

[0097] (1) The operation of reconstructing the model input channel is as follows: The reconstruction of the model input channel specifically involves modifying the model generator and discriminator; the generator adds a new image input interface, and the number of model input channels is changed from 6 to 7. At this point, the generator input is expressed as formula (1), where B size is the number of images processed by the generator at a time, 7 represents the number of model input channels, and H img and W img They represent the height and width of the image respectively. The modification of the discriminator is the same as that of the generator. A new image input interface is added. The number of model input channels is changed from 9 to 10. The input of the discriminator of the model is expressed as formula (2).

[0098] (B size ,7,H img ,W img )(1)

[0099] (B size ,10,H igg ,W img )(2)

[0100] (2) The operations to reconstruct the model structure are as follows: adding an attention mechanism to enhance the model’s ability to learn the correlations between different discrete spatial regions and the local features that must be focused on for autonomous learning; specifically, in the generator, the self-attention layer is configured between the output of the 5th-layer convolutional feature extraction network and the input of the residual network; in the discriminator, the discriminator contains two sub-discriminators with exactly the same structure, and a self-attention layer is inserted after the 2nd-layer convolutional network of each sub-discriminator.

[0101] Furthermore, in step (7), the specific calculation formulas of the computer vision evaluation index and the free-form surface mesh structure geometric evaluation index are as follows:

[0102] (1) The evaluation index based on computer vision is shown in formula (3), μ real and μ gen are the mean vector of the real free-form surface mesh image of the open-hole building and the mean vector of the free-form surface mesh structure image of the open-hole building generated by the generative adversarial network model GridGAN-AE containing the attention mechanism, X and X respectively. m They respectively represent the covariance matrix of the real free-form surface mesh image of the open-hole building and the covariance matrix of the free-form surface mesh structure image of the open-hole building generated by the generative adversarial network model GridGAN-AE containing the attention mechanism. Tr represents the trace of the matrix, that is, the sum of the diagonal elements of the matrix. In formula (3), the first part Represents the Euclidean distance between two mean vectors, which is used to evaluate the difference in the center points of the two sets of image features. The second part evaluates the difference in their distribution shapes;

[0103]

[0104] When the FID value is greater than 60, it is considered that the mesh generation effect does not meet the requirements; when the FID is below 60, it is considered that the mesh generation effect based on computer vision evaluation meets the requirements, and the free-form surface mesh structure geometric evaluation index is further performed;

[0105] (2) The geometric evaluation index of the free-form surface mesh structure includes the mean value index of the member length and the mesh shape quality index, where the member length is the length of the line connecting two mesh nodes in the free-form surface mesh structure, and the mesh shape refers to the basic geometric unit that constitutes the free-form surface mesh structure, including quadrilateral mesh basic units and triangular mesh basic units;

[0106] (2.1) Mean length of the member Calculation formula and mean square deviation of member length δ L The calculation formulas are shown in equations (4) and (5), where n is the total number of rods, L i represents the length of the i-th rod;

[0107]

[0108]

[0109] (2.2) The free-form surface grid structure of the open-hole building is composed of n r quadrilateral grid basic units and n t The shape quality coefficient of the i-th quadrilateral mesh basic unit is and the shape quality coefficient of the jth triangular mesh basic unit See equations (6) and (7) respectively, where i = 1, 2, ..., n r , j=1,2,…,n t , A, B, C and D represent the vertices of the basic unit, S ΔABC , S ΔADC , S ΔADB and S ΔCBD are the areas of the triangles formed by the corresponding nodes of A, B, C and D, l AB , l AD , l AC , l BC , l CD is the side length of the connecting rod corresponding to the vertex, and The change pattern of is consistent, and its value range is (0,1];

[0110]

[0111]

[0112] (2.3) After obtaining the shape quality coefficients of the quadrilateral mesh basic unit and the triangular mesh basic unit, calculate the average shape quality coefficient of the free-form surface mesh structure of the entire opening building and mean square error δ q , the calculation formulas are respectively formula (8) and formula (9);

[0113]

[0114]

[0115] (2.4) The comprehensive index for geometric evaluation of free-form surface mesh structure is formula (10), which considers both the mean index of member length and the mesh shape quality index, where J is the comprehensive index, α and β are the mean square error of member length δ L and shape quality coefficient mean square error δ q The corresponding weight coefficients are: when J is greater than the preset value, the grid generation result is considered to be unsatisfactory; when J is less than the preset value, the grid generation result is satisfactory;

[0116] J=αδ L +βδ q (10).

[0117] Furthermore, in step S3, the open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model is restored to a three-dimensional space by extracting the mesh nodes and topological relationships therein to obtain a three-dimensional open hole free-form surface mesh structure. The specific steps are as follows:

[0118] (1) The coordinate values ​​of the grid nodes in the open free-form surface grid structure image generated by the attention-enhanced generative adversarial network model are converted into key-value pairs {K:V pixel} format, where V pixel Stored in the form of two-dimensional Cartesian coordinates, representing the coordinate values ​​of the grid nodes in the grid structure image of the open hole free surface, K is V pixel The corresponding sequence number value;

[0119] (2) The n topological connection relationships of the mesh nodes in the open free-form surface mesh structure image generated by the attention-enhanced generative adversarial network model are listed in C connect The form of storage is as follows: i It can be further expressed as formula (12), which represents the connection relationship between the i-th two nodes, n p ∈K,n k ∈K, and n p ≠n k ;

[0120] C cnnnect =[C1,C2,…,C i ,…,C n ] (11)

[0121] C i =[n p ,n k ] (12)

[0122] (3) Coordinate alignment: using the coordinates N of the lower left corner of the free-form surface mesh structure with the hole in step S3 LB (x LB ,y LB ) and the coordinates of the upper right corner point N RT (x RT ,y RT ) and the coordinates of the lower left corner of the corresponding real opening free-form surface and the coordinates of the upper right corner Calculate the translation coefficient [M x ,M y ] and the scaling factor [S x ,S y ]:

[0123] (3.1) Calculate N LB (x LB ,y LB )and The translation coefficient [M x ,M y ],in,

[0124] (3.2) Using the translation coefficient [M x ,M y ] Calculate the N after translation RT (x RT ,y RT ) pixel coordinates and expressed as in,

[0125] (3.3) Calculate the scaling factor using formula (13);

[0126]

[0127] (4) Obtain the true values ​​of the plane coordinate nodes of the free-form mesh structure with holes {K:V real}:

[0128] (4.1) The coordinate values ​​of the mesh nodes in the mesh structure image of the hole free-form surface in step (1) {K:V pixel Substitute into formula (14) in turn to obtain the coordinate values ​​of the grid nodes after coordinate translation and alignment. Among them, x i and i V pixel The coordinate value of and for The coordinate value of

[0129]

[0130] (4.2) The coordinate values ​​of the grid nodes after coordinate translation and alignment Substituting into equation (15) in turn, we can obtain the true values ​​of the plane coordinate nodes of the free-form mesh structure with holes {K:V real},in, and V real The coordinate value of

[0131]

[0132] (5) Replace {K:V real The mesh nodes are projected onto the corresponding free-form surface with real openings, and the C connect The n mesh node topological connection relationships saved in the [3D] grid are used to connect the mesh nodes that have been projected onto the real open hole free-form surface, and the final three-dimensional open hole free-form surface mesh structure can be obtained.

[0133] In addition, the present invention also proposes a free-form surface meshing design system for open-hole buildings based on attention enhancement generative adversarial network, and the system includes the following modules:

[0134] (1) A data acquisition module is used to acquire the free-form surface of the building with openings that needs to be meshed, and to construct a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation based on the free-form surface of the building with openings;

[0135] (2) a model construction module, which is used to input the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation into a pre-trained attention-enhanced generative adversarial network model, and generate an open-hole free-form surface mesh structure through the attention-enhanced generative adversarial network model;

[0136] (3) A scheme design module is used to extract the mesh nodes and topological relationships of the open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model, and then restore it to three-dimensional space to finally obtain a three-dimensional open hole free-form surface mesh structure.

[0137] Among them, Fig. 9 As shown, the physical structure of an electronic device of the present invention may include the following components: a processor 910, a communication interface 920, a memory 930, and a communication bus 940. These components communicate with each other through the communication bus 940. The processor 910 can access the logic instructions in the memory 930 to execute a design method for meshing free-form surfaces of open-hole buildings based on an attention-enhanced generative adversarial network, the method comprising:

[0138] Obtain the free-form surface of the building with openings that needs to be meshed, and construct a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation based on the free-form surface of the building with openings;

[0139] Inputting the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation into a pre-trained attention-enhanced generative adversarial network model, and generating an open-hole free-form surface mesh structure through the attention-enhanced generative adversarial network model;

[0140] The open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model is restored to the three-dimensional space by extracting the mesh nodes and topological relationships therein, so as to finally obtain the three-dimensional open hole free-form surface mesh structure.

[0141] In addition, the logic instructions in the above-mentioned memory 930 can be implemented by a software function module and can be sold or used as an independent product. The logic instructions can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, containing several instructions, so that a computer device (which can be a personal computer, a server or a network device, etc.) performs all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium may include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media capable of storing program code.

[0142] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a free-form surface meshing design method for an open-hole building based on an attention-enhanced generative adversarial network provided in the above methods. The method includes:

[0143] Obtain the free-form surface of the building with openings that needs to be meshed, and construct a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation based on the free-form surface of the building with openings;

[0144] Inputting the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation into a pre-trained attention-enhanced generative adversarial network model, and generating an open-hole free-form surface mesh structure through the attention-enhanced generative adversarial network model;

[0145] The open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model is restored to the three-dimensional space by extracting the mesh nodes and topological relationships therein, so as to finally obtain the three-dimensional open hole free-form surface mesh structure.

[0146] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, a method for meshing a free-form surface of an open-hole building based on an attention-enhanced generative adversarial network is implemented. The method comprises:

[0147] Obtain the free-form surface of the building with openings that needs to be meshed, and construct a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation based on the free-form surface of the building with openings;

[0148] Inputting the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation into a pre-trained attention-enhanced generative adversarial network model, and generating an open-hole free-form surface mesh structure through the attention-enhanced generative adversarial network model;

[0149] The open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model is restored to the three-dimensional space by extracting the mesh nodes and topological relationships therein, so as to finally obtain the three-dimensional open hole free-form surface mesh structure.

[0150] The above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them. Although the present invention is described in detail through the above embodiments, it should be understood by those skilled in the art that they can still modify the technical solutions described in the above embodiments, or perform equivalent replacements on some of the technical features. These modifications or replacements do not change the essence of the corresponding technical solutions, nor do they deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A free-form surface meshing design method for open-hole buildings based on attention-enhanced generative adversarial networks, characterized in that: The method comprises the following steps: Step S1, obtaining a free-form surface of an open-hole building that needs to be meshed, and constructing a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation according to the free-form surface of the open-hole building; Step S2, inputting the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation into a pre-trained attention-enhanced generative adversarial network model, and generating an open-hole free-form surface mesh structure through the attention-enhanced generative adversarial network model; Step S3, extracting the mesh nodes and topological relationships of the open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model and restoring it to a three-dimensional space to finally obtain a three-dimensional open hole free-form surface mesh structure; In step S2, the construction process of the pre-trained attention-enhanced generative adversarial network model specifically includes the following: (1) Obtaining the free-form surface of the opening building and its corresponding free-form surface grid structure; (2) constructing a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation corresponding to the free-form surface of the opening building in step (1); (3) The free-form surface curvature cloud map representation, free-form surface elevation cloud map representation, and free-form surface semantic mask representation are used as the input of the model, and the corresponding open hole free-form surface mesh structure is used as the model output; (4) dividing the data in step (3) into a training data set and a test data set according to the form of input and output data pairs; (5) Based on the variant of the generative adversarial network model GridGAN, its model input channel and model structure are reconstructed so that the new generative adversarial network model simultaneously inputs the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation, and includes an attention mechanism. The new generative adversarial network model including the attention mechanism is named GridGAN-AE; (6) training the generative adversarial network model GridGAN-AE containing the attention mechanism in step (5) by using the training set consisting of the curvature cloud map representation of the open-hole free-form surface, the elevation cloud map representation of the free-form surface, the semantic mask representation of the free-form surface, and the grid structure of the open-hole free-form surface in step (4); (7) The generative adversarial network model GridGAN-AE with attention mechanism trained in step (5) is tested and evaluated by using the test set consisting of the curvature cloud map representation of the free-form surface with holes, the elevation cloud map representation of the free-form surface, the semantic mask representation of the free-form surface, and the mesh structure of the free-form surface with holes in step (4). The model that meets the requirements of the computer vision evaluation index and the free-form surface mesh structure geometry evaluation index is used as the final generative adversarial network model GridGAN-AE with attention mechanism; In step (5), the model input channel and model structure are reconstructed as follows: (5.1) The operation of reconstructing the model input channel is as follows: The reconstruction of the model input channel is specifically to modify the generator and discriminator of the model; the generator adds a new image input interface, and the number of model input channels is changed from 6 to 7. At this point, the generator input is expressed as formula (1), where B size is the number of images processed by the generator at a time, 7 represents the number of model input channels, and H img and W img Represent the height and width of the image respectively; the modification of the discriminator is the same as that of the generator, with the addition of a new image input interface, the number of model input channels is changed from 9 to 10, and the input of the discriminator of the model is expressed as formula (2); (B size ,7,H img ,W img )(1) (B size ,10,H img ,W img )(2) (5.2) The operations to reconstruct the model structure are: adding an attention mechanism to enhance the model's ability to learn the correlation between different discrete spatial regions and autonomously learn the local characteristics of focus; specifically, in the generator, the self-attention layer is configured between the output of the 5th layer convolutional feature extraction network and the input of the residual network; in the discriminator, the discriminator contains two sub-discriminators with exactly the same structure, and the self-attention layer is inserted after the 2nd layer convolutional network of each sub-discriminator.

2. The method for meshing free-form surfaces of open-hole buildings based on attention-enhanced generative adversarial networks according to claim 1 is characterized in that: In step S1, the method for constructing the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the free-form surface semantic mask representation is as follows: (1) Constructing a free-form surface curvature cloud map representation: Using a color map with 3 color channels, the free-form surface of the open-hole building is input using the battery function in the Grasshopper software. The maximum principal curvature value of each point on the free-form surface is automatically calculated, and the corresponding RGB value is obtained to obtain the free-form surface curvature cloud map representation of the free-form surface of the open-hole building; (2) Constructing the elevation cloud map representation of the free-form surface: Using a color map with 3 color channels, the free-form surface of the building with openings is input using the battery function in the Grasshopper software. The elevation value of each point on the free-form surface is automatically calculated, and the corresponding RGB value is obtained to obtain the elevation cloud map representation of the free-form surface of the building with openings. (3) Constructing semantic mask: It is represented by a grayscale image with a color channel number of 1, and the free-form surface and the opening area of ​​the building are represented at the same time. The surface itself is represented by a grayscale value of 0, and the opening area is represented by a grayscale value of 142.

3. The method for meshing free-form surfaces of open-hole buildings based on attention-enhanced generative adversarial networks according to claim 1 is characterized in that: In step (7), the specific calculation formulas of the computer vision evaluation index and the free-form surface mesh structure geometric evaluation index are as follows: (1) The evaluation index based on computer vision is shown in formula (3), μ real and μ gen are the mean vector of the real free-form surface mesh image of the open-hole building and the mean vector of the free-form surface mesh structure image of the open-hole building generated by the generative adversarial network model GridGAN-AE containing the attention mechanism, X and X respectively. m They respectively represent the covariance matrix of the real free-form surface mesh image of the open-hole building and the covariance matrix of the free-form surface mesh structure image of the open-hole building generated by the generative adversarial network model GridGAN-AE containing the attention mechanism. Tr represents the trace of the matrix, that is, the sum of the diagonal elements of the matrix. In formula (3), the first part Represents the Euclidean distance between two mean vectors, which is used to evaluate the difference in the center points of the two sets of image features. The second part evaluates the difference in their distribution shapes; When the FID value is greater than 60, it is considered that the mesh generation effect does not meet the requirements; when the FID is below 60, it is considered that the mesh generation effect based on computer vision evaluation meets the requirements, and the free-form surface mesh structure geometric evaluation index is further performed; (2) The geometric evaluation index of the free-form surface mesh structure includes the mean value index of the member length and the mesh shape quality index, where the member length is the length of the line connecting two mesh nodes in the free-form surface mesh structure, and the mesh shape refers to the basic geometric unit that constitutes the free-form surface mesh structure, including quadrilateral mesh basic units and triangular mesh basic units; (2.1) Mean length of the member Calculation formula and mean square deviation of member length δ L The calculation formulas are shown in equations (4) and (5), where n is the total number of rods, L i represents the length of the i-th rod; (2.2) The free-form surface grid structure of the open-hole building is composed of n r quadrilateral grid basic units and n t The shape quality coefficient of the i-th quadrilateral mesh basic unit is and the shape quality coefficient of the jth triangular mesh basic unit See equations (6) and (7) respectively, where i = 1, 2, ..., n r ,j=1,2,...,n t , A, B, C and D represent the vertices of the basic unit, S ΔABC , S ΔADC , S ΔADB and S ΔCBD are the areas of the triangles formed by the corresponding nodes of A, B, C and D, l AB , l AD , l AC , l BC , l CD is the side length of the connecting rod corresponding to the vertex, and The change pattern of is consistent, and its value range is (0,1]; (2.3) After obtaining the shape quality coefficients of the quadrilateral mesh basic unit and the triangular mesh basic unit, calculate the average shape quality coefficient of the free-form surface mesh structure of the entire opening building and mean square error δ q , the calculation formulas are formula (8) and formula (9) respectively; (2.4) The comprehensive index for geometric evaluation of free-form surface mesh structure is formula (10), which considers both the mean index of member length and the mesh shape quality index, where J is the comprehensive index, α and β are the mean square error of member length δ L and shape quality coefficient mean square error δ q The corresponding weight coefficients are: when J is greater than the preset value, the grid generation result is considered to be unsatisfactory; when J is less than the preset value, the grid generation result is satisfactory; J=ad L +βδ q (10)。 4. The method for meshing free-form surfaces of open-hole buildings based on attention-enhanced generative adversarial networks according to claim 1 is characterized in that: In step S3, the open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model is restored to a three-dimensional space by extracting the mesh nodes and topological relationships therein to obtain a three-dimensional open hole free-form surface mesh structure. The specific steps are as follows: (1) The coordinate values ​​of the grid nodes in the open free-form surface grid structure image generated by the attention-enhanced generative adversarial network model are converted into key-value pairs {K:V pixel } format, where V pixel Stored in the form of two-dimensional Cartesian coordinates, representing the coordinate values ​​of the grid nodes in the grid structure image of the open hole free surface, K is V pixel The corresponding sequence number value; (2) The n topological connection relationships of the mesh nodes in the open free-form surface mesh structure image generated by the attention-enhanced generative adversarial network model are listed in C connect The form of storage is as follows: i It can be further expressed as formula (12), which represents the connection relationship between the i-th two nodes, n p ∈K,n k ∈K, and n p ≠n k ; C connect =[C1,C2,…,C i ,…,C n ] (11) C i =[n p ,n k ] (12) (3) Coordinate alignment: using the coordinates N of the lower left corner of the free-form surface mesh structure with the hole in step S3 LB (x LB ,y LB ) and the coordinates of the upper right corner point N RT (x RT ,y RT ) and the coordinates of the lower left corner of the corresponding real opening free-form surface and the coordinates of the upper right corner Calculate the translation coefficient [M x ,M y ] and the scaling factor [S x ,S y ]: (3.1) Calculate N LB (x LB ,y LB )and The translation coefficient [M x ,M y ],in, (3.2) Using the translation coefficient [M x ,M y ] Calculate the N after translation RT (x RT ,y RT ) pixel coordinates and expressed as in, (3.3) Calculate the scaling factor using formula (13); (4) Obtain the true values ​​of the plane coordinate nodes of the free-form mesh structure with holes {K:V real }: (4.1) The coordinate values ​​of the mesh nodes in the mesh structure image of the hole free-form surface in step (1) {K:V pixel Substitute into formula (14) in turn to obtain the coordinate values ​​of the grid nodes after coordinate translation and alignment. Among them, x i and i V pixel The coordinate value of and for The coordinate value of (4.2) The coordinate values ​​of the grid nodes after coordinate translation and alignment Substituting into equation (15) in turn, we can obtain the true values ​​of the plane coordinate nodes of the free-form mesh structure with holes {K:V real },in, and V real The coordinate value of (5) Replace {K:V real The mesh nodes are projected onto the corresponding free-form surface with real openings, and the C connect The n mesh node topological connection relationships saved in the [3D] grid are used to connect the mesh nodes that have been projected onto the real open hole free-form surface, and the final three-dimensional open hole free-form surface mesh structure can be obtained.

5. A free-form surface meshing design system for open-hole buildings based on attention-enhanced generative adversarial networks, characterized in that: The system includes the following modules: A data acquisition module is used to acquire the free-form surface of the building with openings that needs to be meshed, and to construct a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation based on the free-form surface of the building with openings; A model building module, used for inputting the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation into a pre-trained attention-enhanced generative adversarial network model, and generating an open-hole free-form surface mesh structure through the attention-enhanced generative adversarial network model; Among them, the construction process of the pre-trained attention-enhanced generative adversarial network model specifically includes the following: (1) Obtaining the free-form surface of the opening building and its corresponding free-form surface grid structure; (2) constructing a free-form surface curvature cloud map representation, a free-form surface elevation cloud map representation, and a free-form surface semantic mask representation corresponding to the free-form surface of the opening building in step (1); (3) The free-form surface curvature cloud map representation, free-form surface elevation cloud map representation, and free-form surface semantic mask representation are used as the input of the model, and the corresponding open hole free-form surface mesh structure is used as the model output; (4) dividing the data in step (3) into a training data set and a test data set according to the form of input and output data pairs; (5) Based on the variant of the generative adversarial network model GridGAN, its model input channel and model structure are reconstructed so that the new generative adversarial network model simultaneously inputs the free-form surface curvature cloud map representation, the free-form surface elevation cloud map representation and the semantic mask representation, and includes an attention mechanism. The new generative adversarial network model including the attention mechanism is named GridGAN-AE; (6) training the generative adversarial network model GridGAN-AE containing the attention mechanism in step (5) by using the training set consisting of the curvature cloud map representation of the open-hole free-form surface, the elevation cloud map representation of the free-form surface, the semantic mask representation of the free-form surface, and the grid structure of the open-hole free-form surface in step (4); (7) The generative adversarial network model GridGAN-AE with attention mechanism trained in step (5) is tested and evaluated by using the test set consisting of the curvature cloud map representation of the free-form surface with holes, the elevation cloud map representation of the free-form surface, the semantic mask representation of the free-form surface, and the mesh structure of the free-form surface with holes in step (4). The model that meets the requirements of the computer vision evaluation index and the free-form surface mesh structure geometry evaluation index is used as the final generative adversarial network model GridGAN-AE with attention mechanism; In step (5), the model input channel and model structure are reconstructed as follows: (5.1) The operation of reconstructing the model input channel is as follows: The reconstruction of the model input channel is specifically to modify the generator and discriminator of the model; the generator adds a new image input interface, and the number of model input channels is changed from 6 to 7. At this point, the generator input is expressed as formula (1), where B size is the number of images processed by the generator at a time, 7 represents the number of model input channels, and H img and W img Represent the height and width of the image respectively; the modification of the discriminator is the same as that of the generator, with the addition of a new image input interface, the number of model input channels is changed from 9 to 10, and the input of the discriminator of the model is expressed as formula (2); (B size ,7,H img ,W img )(1) (B size ,10,H img ,W img )(2) (5.2) The operations for reconstructing the model structure are: adding an attention mechanism to enhance the model's ability to learn the correlation between different discrete spatial regions and autonomously learn the local characteristics of focus; specifically, in the generator, the self-attention layer is configured between the output of the 5th layer convolutional feature extraction network and the input of the residual network; in the discriminator, the discriminator contains two sub-discriminators with exactly the same structure, and the self-attention layer is inserted after the 2nd layer convolutional network of each sub-discriminator; The scheme design module is used to extract the mesh nodes and topological relationships of the open hole free-form surface mesh structure generated by the attention-enhanced generative adversarial network model, and then restore it to the three-dimensional space to finally obtain the three-dimensional open hole free-form surface mesh structure.

6. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for designing free-form surface meshing of open-hole buildings based on a generative adversarial network with attention enhancement is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program is executed by the processor to implement the free-form surface meshing design method of open-hole buildings based on the attention-enhanced generative adversarial network as described in any one of claims 1 to 4.

8. An electronic device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the program implements the free-form surface meshing design method for open-hole buildings based on the attention-enhanced generative adversarial network as described in any one of claims 1 to 4.

Images