3D Curve Frame Generation Method, Device, Computer Equipment and Readable Storage Medium

Abstract

The present application relates to a method, apparatus, computer device, and readable storage medium for generating a three-dimensional curve frame. The method includes: obtaining initial image data; performing feature encoding on the initial image data to obtain initial image features corresponding to the initial image data; based on the initial image features, performing distribution mapping on the sampled noise data to obtain a potential representation of the three-dimensional curve frame corresponding to the initial image data; performing feature decoding on the potential representation of the three-dimensional curve frame to obtain the three-dimensional curve frame structure and curve potential representation corresponding to the initial image data; performing feature decoding on the curve potential representation to obtain the three-dimensional curve frame curve corresponding to the initial image data; and generating a target three-dimensional curve frame corresponding to the initial image data based on the three-dimensional curve frame structure and the three-dimensional curve frame curve. Using this method can improve the generation accuracy of the three-dimensional curve frame.

Description

Technical Field

[0001] The present application relates to the technical field of generative models, and particularly to a method, apparatus, computer device, computer-readable storage medium, and computer program product for generating a three-dimensional curve frame. Background Art

[0002] In the field of computer-aided design, as an important expression form of preliminary design, the three-dimensional wireframe has an intuitive display effect on the structure and layout of an object. The three-dimensional wireframe is a three-dimensional shape representation method, and the geometric shape of a three-dimensional object is represented by a network structure composed of continuous lines and discrete topological connections.

[0003] When generating a three-dimensional wireframe in the prior art, it is usually based on a deep learning method to generate a corresponding three-dimensional wireframe according to the three-dimensional object model in point cloud or image data. However, when dealing with sparse or incomplete input data, the existing methods are prone to generate incomplete or structurally incorrect three-dimensional wireframes, and the existing methods are also limited to regular three-dimensional object models or simple geometric shapes, and there are large errors in the three-dimensional curve frames generated for three-dimensional object models with complex curves, resulting in the problem of low accuracy in generating three-dimensional curve frames. Summary of the Invention

[0004] Based on this, it is necessary to provide a method, apparatus, computer device, computer-readable storage medium, and computer program product for generating a three-dimensional curve frame that can improve the accuracy of generating a three-dimensional curve frame for the above technical problems.

[0005] In a first aspect, the present application provides a method for generating a three-dimensional curve frame, including:

[0006] Obtain initial image data;

[0007] Perform feature encoding on the initial image data to obtain initial image features corresponding to the initial image data;

[0008] Based on the initial image features, perform distribution mapping on the sampled noise data to obtain a potential representation of the three-dimensional curve frame corresponding to the initial image data;

[0009] Perform feature decoding on the potential representation of the three-dimensional curve frame to obtain the structure of the three-dimensional curve frame and the potential representation of the curve corresponding to the initial image data;

[0010] Perform feature decoding on the potential representation of the curve to obtain the three-dimensional curve frame curve corresponding to the initial image data;

[0011] Based on the structure of the three-dimensional curve frame and the three-dimensional curve frame curve, generate a target three-dimensional curve frame corresponding to the initial image data.

[0012] Second aspect, the present application also provides a three-dimensional curve frame generation device, including:

[0013] An acquisition module, configured to acquire initial image data;

[0014] An encoding module, configured to perform feature encoding on the initial image data to obtain initial image features corresponding to the initial image data;

[0015] A distribution mapping module, configured to perform distribution mapping on the sampled noise data based on the initial image features to obtain a three-dimensional curve frame latent representation corresponding to the initial image data;

[0016] A structure decoding module, configured to perform feature decoding on the three-dimensional curve frame latent representation to obtain a three-dimensional curve frame structure and a curve latent representation corresponding to the initial image data;

[0017] A curve decoding module, configured to perform feature decoding on the curve latent representation to obtain a three-dimensional curve frame curve corresponding to the initial image data;

[0018] A wireframe generation module, configured to generate a target three-dimensional curve frame corresponding to the initial image data based on the three-dimensional curve frame structure and the three-dimensional curve frame curve.

[0019] Third aspect, the present application also provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0020] Acquire initial image data;

[0021] Perform feature encoding on the initial image data to obtain initial image features corresponding to the initial image data;

[0022] Perform distribution mapping on the sampled noise data based on the initial image features to obtain a three-dimensional curve frame latent representation corresponding to the initial image data;

[0023] Perform feature decoding on the three-dimensional curve frame latent representation to obtain a three-dimensional curve frame structure and a curve latent representation corresponding to the initial image data;

[0024] Perform feature decoding on the curve latent representation to obtain a three-dimensional curve frame curve corresponding to the initial image data;

[0025] Generate a target three-dimensional curve frame corresponding to the initial image data based on the three-dimensional curve frame structure and the three-dimensional curve frame curve.

[0026] Fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0027] Acquire initial image data;

[0028] Perform feature encoding on the initial image data to obtain the initial image features corresponding to the initial image data;

[0029] Based on the initial image features, perform distribution mapping on the sampled noise data to obtain the potential representation of the three-dimensional curve box corresponding to the initial image data;

[0030] Perform feature decoding on the potential representation of the three-dimensional curve box to obtain the three-dimensional curve box structure and the potential representation of the curve corresponding to the initial image data;

[0031] Perform feature decoding on the potential representation of the curve to obtain the three-dimensional curve box curve corresponding to the initial image data;

[0032] Based on the three-dimensional curve box structure and the three-dimensional curve box curve, generate the target three-dimensional curve box corresponding to the initial image data.

[0033] In a fifth aspect, the present application also provides a computer program product, including a computer program, which when executed by a processor implements the following steps:

[0034] Obtain the initial image data;

[0035] Perform feature encoding on the initial image data to obtain the initial image features corresponding to the initial image data;

[0036] Based on the initial image features, perform distribution mapping on the sampled noise data to obtain the potential representation of the three-dimensional curve box corresponding to the initial image data;

[0037] Perform feature decoding on the potential representation of the three-dimensional curve box to obtain the three-dimensional curve box structure and the potential representation of the curve corresponding to the initial image data;

[0038] Perform feature decoding on the potential representation of the curve to obtain the three-dimensional curve box curve corresponding to the initial image data;

[0039] Based on the three-dimensional curve box structure and the three-dimensional curve box curve, generate the target three-dimensional curve box corresponding to the initial image data.

[0040] The above three-dimensional curve frame generation method, device, computer device, computer-readable storage medium, and computer program product obtain initial image features by performing feature encoding on initial image data. Based on the initial image features, distribution mapping is performed on the sampled noise data to obtain a potential representation of the three-dimensional curve frame corresponding to the initial image data, so that the potential representation of the three-dimensional curve frame can integrate the curve frame structure and curve feature information of the three-dimensional curve frame in the initial image data, ensuring the data integrity of the three-dimensional curve frame represented by the potential representation of the three-dimensional curve frame. Since constructing a three-dimensional curve frame requires the dependency relationship between the curve frame structure and the curve of the curve frame, by performing feature decoding on the potential representation of the three-dimensional curve frame, the curve frame structure and curve potential representation of the reconstructed initial image data are obtained synchronously, avoiding the separate processing of encoding and decoding reconstruction of the three-dimensional curve frame structure based on the initial image data and encoding and decoding reconstruction of the curve of the curve frame. Compared with the separate processing (encoding and decoding reconstruction) of the initial image data in the curve frame structure and the curve of the curve frame, by encoding the initial image data to obtain initial image features, and then synchronously decoding the potential representation of the three-dimensional curve frame corresponding to the initial image features to obtain the reconstructed three-dimensional curve frame structure and curve potential representation, the accuracy of the three-dimensional curve frame structure and curve potential representation is ensured. Furthermore, when decoding and reconstructing the curve potential representation, the accuracy of the three-dimensional curve frame curve is ensured, improving the generation accuracy of the three-dimensional curve frame. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] To more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments of the present application or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other related drawings can be obtained based on these drawings.

[0042] Figure 1 It is an application environment diagram of the three-dimensional curve frame generation method in an embodiment;

[0043] Figure 2 It is a flowchart of the three-dimensional curve frame generation method in an embodiment;

[0044] Figure 3 It is a training diagram of the wireframe processing model in an embodiment;

[0045] Figure 4 It is a schematic diagram of the standard curve frame curve in an embodiment;

[0046] Figure 5 It is a training diagram of the curve processing model in an embodiment;

[0047] Figure 6Schematic diagram for training the distribution mapping model in an embodiment

[0048] Figure 7 Schematic diagram of the three-dimensional curve box generation model in an embodiment

[0049] Figure 8 Structural block diagram of the three-dimensional curve box generation device in an embodiment

[0050] Figure 9 Internal structure diagram of a computer device in an embodiment

[0051] Figure 10 Internal structure diagram of a computer device in another embodiment Detailed implementation manners

[0052] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] The three-dimensional curve box generation method provided by the embodiments of the present application can be applied to an application environment as shown in Figure 1 . Among them, the terminal 102 communicates with the server 104 through a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or can be placed in the cloud or other network servers. The server 104 can obtain the initial image data through the terminal 102; the server 104 performs feature encoding on the initial image data to obtain the initial image features corresponding to the initial image data; the server 104 performs distribution mapping on the sampled noise data based on the initial image features to obtain the three-dimensional curve box latent representation corresponding to the initial image data; the server 104 performs feature decoding on the three-dimensional curve box latent representation to obtain the three-dimensional curve box structure and curve latent representation corresponding to the initial image data; the server 104 performs feature decoding on the curve latent representation to obtain the three-dimensional curve box curve corresponding to the initial image data; the server 104 generates the target three-dimensional curve box corresponding to the initial image data based on the three-dimensional curve box structure and the three-dimensional curve box curve. Among them, the terminal 102 can be but is not limited to various personal computers, laptop computers, smart phones, tablet computers, etc. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0054] In an exemplary embodiment, as shown in Figure 2 , a three-dimensional curve box generation method is provided. Taking the method applied to the server 104 in Figure 1 as an example, the method includes the following steps:

[0055] Step 202: Obtain initial image data.

[0056] Step 204: Perform feature encoding on the initial image data to obtain initial image features corresponding to the initial image data.

[0057] Among them, the initial image data refers to data with three-dimensional images, such as point cloud data, image data, etc. The point cloud data can be sparse point cloud data or incomplete point cloud data, and the image data can be sketches, drawings, etc. The initial image features refer to feature vectors of a specific dimension obtained by encoding the initial image data.

[0058] Exemplarily, in response to a three-dimensional curve frame generation request sent by a terminal, the server obtains initial image data, which may include one or more types of data with three-dimensional images such as point cloud data and image data. The three-dimensional image is, for example, a curve frame image representing a three-dimensional object. The server encodes the initial image data, which can call the corresponding image encoder according to the data type of the initial image data (such as point cloud data, image data, etc.) to perform feature encoding on the three-dimensional image in the initial image data to obtain initial image features of a specific dimension. Generally, the initial image features are 1024-dimensional feature vectors.

[0059] Step 206: Based on the initial image features, perform distribution mapping on the sampled noise data to obtain a three-dimensional curve frame latent representation corresponding to the initial image data.

[0060] Among them, the sampled noise data can be random noise that follows a normal distribution, such as Gaussian noise, for providing data sampling. Distribution mapping refers to the process of mapping one distribution (such as a simple distribution: standard normal distribution) to a target data distribution (such as the distribution of complex data like images and texts). The three-dimensional curve frame latent representation is identification data representing the three-dimensional curve frame, serving as data specifying the three-dimensional curve frame for feature decoding, representing the wireframe information of the three-dimensional curve frame, and containing information such as the curve frame structure and curve frame curves.

[0061] Exemplarily, the server obtains noise data of a preset dimension, generally Gaussian noise. The preset dimension corresponding to the obtained noise data is the same as the feature dimension corresponding to the initial image features, such as 1024 dimensions. Then, the server uses the shape of the three-dimensional curve frame corresponding to the initial image features as a constraint condition, maps the normal distribution corresponding to the sampled noise data to a target data distribution that satisfies the constraint condition, samples the target data distribution, and obtains a three-dimensional curve frame latent representation corresponding to the initial image data.

[0062] Step 208: Perform feature decoding on the potential representation of the three-dimensional curve frame to obtain the three-dimensional curve frame structure and curve potential representation corresponding to the initial image data.

[0063] Among them, the three-dimensional curve frame structure refers to the information describing the connection relationship of the curves of the curve frame, such as the topological structure corresponding to the initial image data. The curve potential representation is the data specifying the curves of the curve frame for feature decoding, and can be the features representing the geometric shapes of the curves of the curve frame, such as wavy geometric shapes, arc-shaped geometric shapes, etc.

[0064] Exemplarily, the server invokes a pre-trained target wireframe processing model, which is used to decode and reconstruct the input potential representation of the three-dimensional curve frame into a three-dimensional curve frame structure and a curve potential representation. The decoding network (decoder) in the target wireframe processing model can be pre-trained. Then the server inputs the potential representation of the three-dimensional curve frame into the trained decoding network in the target wireframe processing model, and performs feature decoding on the potential representation of the three-dimensional curve frame through the decoding network to obtain the three-dimensional curve frame structure and curve potential representation corresponding to the initial image. The target wireframe processing model also includes an encoder, which can encode the input three-dimensional curve frame structure and curve potential representation into a potential representation of the three-dimensional curve frame. The encoder is also used to train the decoder during the training process of the target wireframe processing model with the potential representation of the three-dimensional curve frame encoded by the encoder, so that the decoder can decode the potential representation of the three-dimensional curve frame to obtain an accurate three-dimensional curve frame structure and curve potential representation.

[0065] In an exemplary embodiment, the dimension of the potential representation of the three-dimensional curve frame can be adjusted according to different input data, such as adjusting according to the data volume of the input data, or adjusting according to the data type of the input data.

[0066] Step 210: Perform feature decoding on the curve potential representation to obtain the three-dimensional curve frame curves corresponding to the initial image data.

[0067] Among them, the three-dimensional curve frame curves refer to the curve lines with geometric shapes. The curve potential representation can be decoded to obtain the three-dimensional curve frame curves.

[0068] Exemplarily, after obtaining the latent representation of the curve, the server invokes a pre-trained target curve processing model, which is used to decode the input latent representation of the curve into a three-dimensional curve frame curve. It can be that the decoding network in the target curve decoding model is pre-trained. Then the server inputs the latent representation of the curve into the decoding network (decoder) in the target curve processing model, and decodes the latent representation of the curve into the corresponding three-dimensional curve frame curve through the decoding network. The three-dimensional curve frame curve includes multiple types, and a straight line also belongs to a type of curve. The target curve processing model also includes an encoder, which can encode the input three-dimensional curve frame curve into a latent representation of the curve. The encoder is also used to train the decoder during the training process of the target curve processing model with the latent representation of the curve encoded by the encoder, so that the decoder can accurately decode the latent representation of the curve to obtain the three-dimensional curve frame curve. The target curve decoding model can be a network model integrated with the target image processing model or independent network models respectively.

[0069] Step 212: Generate a target three-dimensional curve frame corresponding to the initial image data based on the three-dimensional curve frame structure and the three-dimensional curve frame curve.

[0070] Exemplarily, the three-dimensional curve frame structure includes the endpoint coordinates and adjacent relationships of the three-dimensional curve frame curve. The server determines the adjacent three-dimensional curve frame curves according to the adjacent relationships and connects the adjacent three-dimensional curve frame curves according to the endpoint coordinates to obtain the target three-dimensional curve frame corresponding to the initial image data. For example, if it is determined according to the adjacent relationships that endpoint A1 of curve A is adjacent to endpoint B1 of curve B and endpoint C1 of curve C, then curves A, B, and C are connected according to the endpoint coordinates of endpoint A1 and A2 of curve A, endpoint B1 and B2 of curve B, and endpoint C1 and C2 of curve C.

[0071] In the above three-dimensional curve frame generation method, the initial image features are obtained by performing feature encoding on the initial image data. Based on the initial image features, distribution mapping is performed on the sampled noise data to obtain the potential representation of the three-dimensional curve frame corresponding to the initial image data, so that the potential representation of the three-dimensional curve frame can integrate the curve frame structure and curve feature information of the three-dimensional curve frame in the initial image data, ensuring the data integrity of the three-dimensional curve frame represented by the potential representation of the three-dimensional curve frame. Since constructing a three-dimensional curve frame requires the dependency relationship between the curve frame structure and the curve of the curve frame, by performing feature decoding on the potential representation of the three-dimensional curve frame, the curve frame structure and the curve potential representation of the reconstructed initial image data are obtained synchronously, avoiding the separate processing of encoding and decoding reconstruction of the three-dimensional curve frame structure based on the initial image data and encoding and decoding reconstruction of the curve of the curve frame. Compared with the separate processing (encoding and decoding reconstruction) of the initial image data for the curve frame structure and the curve of the curve frame, by encoding the initial image data to obtain the initial image features, and then synchronously decoding the potential representation of the three-dimensional curve frame corresponding to the initial image features to obtain the reconstructed three-dimensional curve frame structure and curve potential representation, the accuracy of the three-dimensional curve frame structure and curve potential representation is ensured. Furthermore, when decoding and reconstructing the curve potential representation, the accuracy of the three-dimensional curve frame curve is ensured, improving the generation accuracy of the three-dimensional curve frame.

[0072] In an exemplary embodiment, the three-dimensional curve frame generation method further includes:

[0073] Obtain the first training curve frame structure and the first training curve frame curve corresponding to the first training three-dimensional curve frame;

[0074] Perform feature encoding on the first training curve frame structure to obtain the first training structure features, and perform feature encoding on the first training curve frame curve to obtain the first training curve potential representation;

[0075] Fuse the first training structure features and the first training curve potential representation to obtain the first training curve frame potential representation;

[0076] Input the first training curve frame potential representation into the initial wireframe processing model to obtain the first predicted curve frame structure and the first predicted curve potential representation corresponding to the first training three-dimensional curve frame;

[0077] Based on the first training curve frame structure, the first training curve potential representation, the first predicted curve frame structure, and the first predicted curve potential representation, adjust the model parameters of the initial wireframe processing model until the convergence condition is met to obtain the target wireframe processing model;

[0078] Step 208, perform feature decoding on the potential representation of the three-dimensional curve frame to obtain the curve frame structure and the curve potential representation of the initial image data corresponding to the three-dimensional curve frame, including:

[0079] Input the potential representation of the three-dimensional curve frame into the target wireframe processing model to obtain the three-dimensional curve frame structure and curve potential representation corresponding to the initial image data.

[0080] Among them, the first training three-dimensional curve frame is the training data for training the initial wireframe processing model, represented in the form of a three-dimensional curve frame. The first training curve frame structure refers to the topological structure corresponding to the first training three-dimensional curve frame. The first training curve frame curve refers to the curve lines in the first training three-dimensional curve frame, including curve lines of different geometric shapes. The first training structure feature is a feature vector representing the first training curve frame structure. The first training curve potential representation is data representing the geometric shape of the first training curve frame curve. The first training curve frame potential representation refers to the potential representation of the three-dimensional curve frame used to train the initial wireframe processing model. The first predicted curve frame structure refers to the curve frame structure predicted (decoded) by the initial wireframe processing model based on the first training curve frame potential representation. The first predicted curve potential representation is the curve potential representation predicted (decoded) by the initial wireframe processing model based on the first training curve frame potential representation.

[0081] Exemplarily, an initial wireframe processing model to be trained is pre-deployed in the server for generating a three-dimensional curve frame structure and curve potential representation. Before training the initial wireframe processing model, the server obtains the first training curve frame structure and the first training curve frame curve corresponding to the first training three-dimensional curve frame, performs feature encoding on the first training curve frame curve to obtain the first training curve potential representation, which can be to use the encoder in the trained target curve processing model to perform feature encoding on the first training curve frame curve to obtain the first training curve potential representation. Then, feature extraction is performed on the first training curve frame structure. For example, the first training curve frame structure can be converted into a data list, which can be to generate an adjacency list according to each vertex in the topological network corresponding to the first training curve frame structure (i.e., the endpoints of the first training curve frame curve) and the adjacent vertices respectively connected to each vertex (i.e., the endpoints of the adjacent first training curve frame curves in the first training three-dimensional curve frame), and generate endpoint coordinate data according to the endpoint coordinates (three-dimensional coordinates of the endpoints) of each endpoint of the first training curve frame curve. According to the data list and endpoint coordinate data converted from the first training curve frame structure, the first training structure feature is obtained. To ensure a unified vertex arrangement distribution for the topological structures of different wireframes, the adjacency list can be sorted using breadth-first search. The server fuses the first training structure feature and the first training curve potential representation, which can be to splice the first training structure feature and the first training curve potential representation to obtain the spliced first training curve frame potential representation. The first training curve frame potential representation is used to train the initial wireframe processing model.

[0082] The server inputs the first training curve box latent representation into the initial wireframe processing model, and obtains the first predicted curve box structure and the first predicted curve latent representation corresponding to the first training three-dimensional curve box. Then, based on the loss between the first training curve box structure and the first predicted curve box structure, and the loss between the first training curve latent representation and the first predicted curve latent representation, the server iteratively adjusts the model parameters of the initial wireframe processing model until the convergence condition is met, and obtains the trained target wireframe processing model. The initial wireframe processing model and the target wireframe processing model can be variational autoencoder models.

[0083] In the process of the server generating the target three-dimensional curve box corresponding to the initial image data, after obtaining the three-dimensional curve box latent representation corresponding to the initial image data, the server inputs the three-dimensional curve box latent representation into the target wireframe processing model, and obtains the three-dimensional curve box structure and the curve latent representation corresponding to the initial image data.

[0084] In this embodiment, by training the initial wireframe processing model with the first training curve box latent representation obtained by fusing the first training structural features and the first training curve latent representation, the target wireframe processing model realizes the synchronous decoding of the three-dimensional curve box structure and the curve latent representation, ensuring the accuracy of the three-dimensional curve box structure and the curve latent representation. Moreover, converting the three-dimensional curve box structure (topological structure) into an adjacency list simplifies the representation of the topological structure, reduces redundant information, thereby reducing the computational amount of the wireframe processing model and improving the processing efficiency of the wireframe processing model.

[0085] In an exemplary embodiment, the initial wireframe processing model includes an encoder and a decoder; inputting the first training curve box latent representation into the initial wireframe processing model to obtain the first predicted curve box structure and the first predicted curve latent representation corresponding to the first training three-dimensional curve box includes:

[0086] Input the first training curve box latent representation into the encoder in the initial wireframe processing model to obtain an intermediate training curve box latent representation;

[0087] Input the intermediate training curve box latent representation into the decoder in the initial wireframe processing model to obtain the first predicted curve box structure and the first predicted curve latent representation corresponding to the first training three-dimensional curve box;

[0088] Inputting the three-dimensional curve box latent representation into the target wireframe processing model to obtain the three-dimensional curve box structure and the curve latent representation corresponding to the initial image data includes:

[0089] Input the three-dimensional curve box latent representation into the decoder in the target wireframe processing model to obtain the three-dimensional curve box structure and the curve latent representation corresponding to the initial image data.

[0090] Exemplarily, the initial wireframe processing model includes an encoder and a decoder. During the training process of the initial wireframe processing model, specifically, the encoder and decoder in the initial wireframe processing model are trained. The server inputs the first training curve frame latent representation into the encoder in the initial wireframe processing model to obtain an intermediate training curve frame latent representation. The intermediate training curve frame latent representation is a fixed-length wireframe latent representation obtained after encoding by the encoder. For example, if it is a fixed length of 64x16, then the encoder in the initial wireframe processing model can be understood as being used to encode the input wireframe latent representation in a fixed-length dimension. Then the server inputs the intermediate training curve frame latent representation into the decoder in the initial wireframe processing model for decoding to obtain the first predicted curve frame structure and the first predicted curve latent representation corresponding to the first training three-dimensional curve frame.

[0091] The server optimizes the encoder in the initial wireframe processing model according to the loss between the first training curve frame structure and the first predicted curve frame structure, and optimizes the decoder in the initial wireframe processing model according to the loss between the first training curve latent representation and the first predicted curve latent representation until the convergence condition is met to obtain the target wireframe processing model.

[0092] During the process of the server generating the target three-dimensional curve frame corresponding to the initial image data, after obtaining the three-dimensional curve frame latent representation corresponding to the initial image data, the server inputs the three-dimensional curve frame latent representation into the decoder in the target wireframe processing model for decoding to obtain the three-dimensional curve frame structure and the curve latent representation corresponding to the initial image data.

[0093] In an exemplary embodiment, as Figure 3 shown, a training schematic diagram of a wireframe processing model is provided. Figure 3 The input in is the first training three-dimensional curve frame. The arrow indicates the first training curve frame structure (i.e., the topological network in the figure) and the first training curve frame curve corresponding to the first training three-dimensional curve frame. Through the trained target curve processing model, the first training curve frame curve is encoded to obtain the first training curve latent representation, as Figure 3 the Nx12 rectangular block in, representing the curve frame latent representation of N wireframe curves; the first training curve frame structure is converted into an adjacency list (such as Figure 3 the Nx48 rectangular block in) and endpoint coordinate data (such as Figure 3 the Nx3x128 rectangular block in) to obtain the first training structure feature. Then the first training structure feature and the first training curve latent representation are fused to obtain the first training curve frame latent representation, as Figure 3 the NxCw rectangular block Zcurve in, and the first training curve frame latent representation is input into the encoder of the wireframe processing model, as Figure 3the Cross Attention model therein, where the rectangular block of 64xCw represents the input vector Query of the Cross Attention model, obtaining an intermediate training curve box latent representation with a fixed length of 64x16, as Figure 3 the rectangular block of 64x16 therein. Then, the intermediate training curve box latent representation is input into the decoder in the initial wireframe processing model, as Figure 3 the Cross Attention model and self Attention model pointed to by the intermediate training curve box latent representation in

[0094] In this embodiment, encoding the curve through the cross-attention mechanism can handle more complex and diverse curve geometries, ensuring the generation accuracy of the three-dimensional curve box.

[0095] In an exemplary embodiment, the three-dimensional curve box generation method further includes:

[0096] Obtaining the third training curve box curve and the third training curve latent representation;

[0097] Inputting the third training curve box curve into the encoder in the initial curve processing model to obtain a third predicted curve latent representation;

[0098] Inputting the third predicted curve latent representation into the decoder in the initial curve processing model to obtain a third predicted curve box curve;

[0099] Based on the third training curve latent representation, the third predicted curve latent representation, the third training curve box curve, and the third predicted curve box curve, adjusting the model parameters of the initial curve processing model until the convergence condition is met to obtain the target curve processing model;

[0100] Step 210, performing feature decoding on the curve latent representation to obtain a three-dimensional curve box curve corresponding to the initial image data, including:

[0101] Input the curve latent representation into the decoder in the target curve processing model to obtain the three-dimensional curve frame curve corresponding to the initial image data.

[0102] Among them, the third training curve frame curve is the wireframe curve used to train the initial curve processing model. The third training curve latent representation is the true curve latent representation obtained by pre-encoding the third training curve frame curve. The third predicted curve latent representation is the curve latent representation obtained by encoding the third training curve frame curve through the encoder of the initial curve processing model. The third predicted curve frame curve is the wireframe curve obtained by decoding the third predicted curve latent representation through the decoder of the initial curve processing model.

[0103] Exemplarily, an initial curve processing model to be trained is pre-deployed in the server. The curve processing model can encode the input three-dimensional curve frame curve into a curve latent representation, and decode and reconstruct the three-dimensional curve frame curve from the encoded curve latent representation. The initial curve processing model includes an encoder and a decoder. During the training process of the initial curve processing model, specifically, the encoder and decoder in the initial curve processing model are trained.

[0104] During the training process of the initial curve processing model, the server obtains the third training curve frame curve and the third training curve latent representation, inputs the third training curve frame curve into the encoder in the initial curve processing model for feature encoding to obtain the third predicted curve latent representation. Then, the third predicted curve latent representation is input into the decoder in the initial curve processing model to obtain the third predicted curve frame curve.

[0105] The server iteratively optimizes the encoder in the initial curve processing model according to the loss between the third training curve latent representation and the third predicted curve latent representation, and iteratively optimizes the decoder in the initial curve processing model according to the loss between the third training curve frame curve and the third predicted curve frame curve until the convergence condition is met to obtain the target curve processing model. The mean squared error loss can be used to optimize the encoder and decoder in the initial curve processing model. The initial curve processing model and the target curve processing model can be variational autoencoder models.

[0106] During the process of the server generating the target three-dimensional curve frame corresponding to the initial image data, after obtaining the curve latent representation corresponding to the three-dimensional curve frame latent representation, the curve latent representation is input into the decoder in the target curve processing model for decoding to obtain the three-dimensional curve frame curve corresponding to the initial image data.

[0107] In this embodiment, by training the encoder and decoder in the initial curve processing model, the accuracy of the encoded curve latent representation and the accuracy of the decoded curve frame curve are ensured.

[0108] In an exemplary embodiment, obtaining the third training curve box curve includes:

[0109] Obtaining an initial curve box curve;

[0110] Based on a preset starting point and a preset ending point, unify the positions of the starting points and ending points of each curve in the initial curve box curve to obtain a standard curve box curve;

[0111] Perform discretization processing on the standard curve box curve to obtain the third training curve box curve.

[0112] Exemplarily, the server obtains the initial curve box curve, and performs geometric transformations on each curve in the initial curve box curve according to the starting point coordinates of the preset starting point and the ending point coordinates of the preset ending point, such as translation, rotation, and scaling operations, so that the starting point (i.e., the starting endpoint) of each curve coincides with the preset starting point, and the ending point (i.e., the ending endpoint) of each curve coincides with the preset ending point, to obtain each curve with unified starting point position and ending point position, and use each curve with unified starting point position and ending point position as the standard curve box curve. The schematic diagram of the standard curve box curve is as Figure 4 shown, Figure 4 In the left figure in [reference], it shows each curve before the position unification process, and in the right figure, it shows each curve after the starting point position and ending point position are unified. The starting points and ending points of each curve are fixed at a unified position (preset starting point coordinates [-1, 0, 0] and preset ending point coordinates [1, 0, 0]).

[0113] Then the server divides each standard curve box curve into 256 discrete sampling points to obtain the third training curve box curve.

[0114] In this embodiment, through standardization processing, the starting point and the ending point are fixed at a unified position, eliminating redundant information caused by different scales and directions of different curves, thereby enhancing the training stability of the initial curve processing model.

[0115] In an exemplary embodiment, as Figure 5 shown, a training schematic diagram of a curve processing model is provided. Figure 5 In [reference], the input curve is the third training curve box curve after position unification and discretization processing. The third training curve box curve is converted into a feature vector of 256, such as Figure 5 the 256xC rectangular block in [reference], and then the feature vector of the third training curve box curve is downsampled, such as Figure 5 the 64xC rectangular block in [reference], which is used as the input vector Query (query) of the encoder (Cross Attention model) in the initial curve processing model. Among them, Figure 5The K and V therein represent the input vector Key and the input vector Value of the encoder (Cross Attention of the cross - attention model).

[0116] Input the feature vector of the third training curve box curve into the encoder in the initial curve processing model (such as Figure 5 the Cross Attention on the left in Figure 5 the cross - attention model). Through the encoder, perform feature encoding on the third training curve box curve to obtain the latent representation of the third predicted curve, and then use a one - dimensional convolutional layer (such as Figure 5 the residual network in Figure 5 ResNet1D) to compress the dimension of the latent representation of the third predicted curve to obtain the final latent representation of the third predicted curve (such as

[0117] In an exemplary embodiment, the three - dimensional curve box generation method further includes:

[0118] Obtain the second training structural feature and the second training curve latent representation corresponding to the second training three - dimensional curve box;

[0119] Fuse the second training structural feature and the second training curve latent representation to obtain the second training curve box latent representation corresponding to the second training three - dimensional curve box;

[0120] Input the second training curve box latent representation and the sampled noise data into the initial distribution mapping model to obtain the predicted curve box latent representation corresponding to the second training three - dimensional curve box;

[0121] Based on the second training curve box latent representation and the predicted curve box latent representation, adjust the model parameters of the initial distribution mapping model until the convergence condition is met to obtain the target distribution mapping model;

[0122] Step 206, perform distribution mapping on the sampled noise data based on the initial wireframe latent representation to obtain the three - dimensional curve box latent representation corresponding to the initial image data, including:

[0123] Input the initial wireframe latent representation and the sampled noise data into the target distribution mapping model to obtain the three - dimensional curve box latent representation corresponding to the initial image data.

[0124] Among them, the second training three-dimensional curve frame is input data for providing second training structural features and second training curve latent representations to train the initial distribution mapping model, and is represented in the form of a three-dimensional curve frame. The second training structural features are structural features obtained according to the curve frame structure corresponding to the second training three-dimensional curve frame. The second training curve latent representation is a curve latent representation obtained by encoding the wireframe curve of the second training three-dimensional curve frame.

[0125] Exemplarily, an initial distribution mapping model to be trained is also deployed in the server. Before training the initial distribution mapping model, the server obtains the second training structural features and the second training curve latent representation corresponding to the second training three-dimensional curve frame. The second training curve latent representation can be obtained by encoding the wireframe curve of the second training three-dimensional curve frame through the encoder of the target curve processing model. The second training structural features and the second training curve latent representation are fused to obtain the second training curve frame latent representation corresponding to the second training three-dimensional curve frame. The second training curve frame latent representation can be obtained by encoding the fused latent representation by the encoder of the target wireframe processing model according to a fixed length after fusing the second training structural features and the second training curve latent representation. Then, the second training curve frame latent representation and the sampled noise data are input into the initial distribution mapping model to obtain the predicted curve frame latent representation corresponding to the second training three-dimensional curve frame. According to the loss between the second training curve frame latent representation and the predicted curve frame latent representation, the model parameters of the initial distribution mapping model are adjusted until the convergence condition is met, and the target distribution mapping model is obtained. Specifically, the target distribution mapping model can refer to a flow matching model. The flow matching model includes a velocity field represented by a neural network. The velocity field is used to describe the instantaneous velocity of the latent representation and converts Gaussian noise (sampled noise data) into the target data distribution by solving an ordinary differential equation. During the optimization process of the flow matching model, the flow matching model can be optimized by minimizing the error of the velocity field to obtain the optimized flow matching model, that is, the target distribution mapping model.

[0126] In the process of the server generating the target three-dimensional curve frame corresponding to the initial image data, after obtaining the initial wireframe latent representation, the initial wireframe latent representation is input into the distribution mapping model, and the distribution mapping model outputs the three-dimensional curve frame latent representation corresponding to the initial image data. In an exemplary embodiment, as Figure 6 shown, a training schematic diagram of the distribution mapping model is provided. Figure 6The input in it is the second training three-dimensional curve box. Obtain the second training curve box structure and the second training curve box curve corresponding to the second training three-dimensional curve box respectively. Convert the second training curve box structure into the second training structural feature, input the second training curve box curve into the encoder in the target curve processing model for encoding, and obtain the second training curve latent representation. Fuse the second training structural feature and the second training curve latent representation, input the fused latent representation into the encoder in the target wireframe processing model for encoding according to a fixed length, and obtain the second training curve box latent representation corresponding to the second training three-dimensional curve box (such as Figure 6 the 64x16 rectangular block in ). Then input the second training curve box latent representation and the sampled noise data (a normal distribution with a mean of 0 and a standard deviation of 1 , denoted as N(0, 1)) into the initial distribution mapping model, and obtain the predicted curve box latent representation corresponding to the second training three-dimensional curve box (such as Figure 6 the dashed box rectangle in ). According to the loss LOSS between the second training line initial latent representation and the predicted curve box latent representation, optimize the initial distribution mapping model, which can be to optimize the velocity field vt to obtain the target distribution mapping model.

[0127] In an exemplary embodiment, the flow matching model supports unconditional and conditional generation. In unconditional generation, the flow matching model generates three-dimensional wireframes from Gaussian noise; in conditional generation, the input image data or point cloud data is processed by a pre-trained feature extractor (such as DINOv2 (an unsupervised learning method based on the ViT model) and PointNet++ (a neural network for classification and segmentation tasks of irregular point cloud data)), and is input as a constraint condition to control the flow matching model to generate three-dimensional wireframes that meet the constraint conditions.

[0128] In this embodiment, through the application of the latent flow matching model, the generation process is made smoother and more continuous, avoiding the errors in discrete generation methods, ensuring the high precision and diversity of the generation results, and being able to support unconditional and conditional generation, with stronger adaptability.

[0129] In an exemplary embodiment, as Figure 7 shown, a schematic diagram of a three-dimensional curve box generation model is provided. Take the initial image data as the input, and the initial image data includes sparse point cloud data, incomplete point cloud data, sketch images, etc. According to the encoder corresponding to the initial image data, perform feature encoding on the initial image data to obtain the initial wireframe latent representation corresponding to the initial image data; input the initial wireframe latent representation and Gaussian noise into the flow matching model (target distribution mapping model), and output the three-dimensional curve box latent representation corresponding to the initial image data through the flow matching model, such as Figure 7The 64x16 rectangular block Zw in the middle; the decoder in the target wireframe processing model decodes the features of the potential representation of the three-dimensional curve frame to obtain the three-dimensional curve frame structure corresponding to the initial image data (such as Figure 7 the reticular topological structure in the middle) and the curve potential representation (such as Figure 7 the Nx12 rectangular block Zcurve in the middle); the decoder in the target curve processing model decodes the features of the curve potential representation to obtain the three-dimensional curve frame curve corresponding to the initial image data; then, according to the three-dimensional curve frame structure and the three-dimensional curve frame curve, the target three-dimensional curve frame corresponding to the initial image data is generated.

[0130] Among them, the target curve processing model can be a curve variational autoencoder, which is used to encode various types of geometric curves into curve potential representations during the training process and can adapt to the geometric features of different curves, such as straight lines, circles, and Bézier curves. The target wireframe processing model can be a frame variational autoencoder, which is used to combine the curve potential representation (curve geometric information) generated by the curve variational autoencoder with topological information during the training process to generate a global fixed-length three-dimensional curve frame potential representation for reconstructing a complete three-dimensional wireframe through subsequent encoding. Flow matching maps Gaussian noise to the target data distribution (i.e., the distribution of the target wireframe potential representation) through the velocity field of a neural network based on the three-dimensional curve frame potential representation generated by the frame variational autoencoder to output the three-dimensional curve frame potential representation.

[0131] Among them, during the decoding process of the target curve processing model, the curve potential representation is input into the upsampling convolutional network, and it is transformed into a continuous three-dimensional curve by using the cross-attention mechanism and the multi-layer perceptron. During the decoding process of the target wireframe processing model, a set of learnable query vectors (such as Figure 3 the rectangular block MxCw representing the input vector Query in the middle) perform cross-attention calculation with the three-dimensional curve frame potential representation (such as Figure 3 the 64x16 rectangular block in the middle), extract the features of each curve, and generate the corresponding adjacency list, endpoint coordinates, and curve potential representation.

[0132] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown according to the indications of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear indication in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0133] Based on the same inventive concept, an embodiment of the present application further provides a three-dimensional curve frame generation device for implementing the three-dimensional curve frame generation method involved above. The implementation solution provided by this device for solving problems is similar to the implementation solution described in the above method. Therefore, the specific limitations in one or more embodiments of the three-dimensional curve frame generation device provided below can refer to the limitations on the three-dimensional curve frame generation method in the above text, and will not be elaborated here.

[0134] In an exemplary embodiment, as Figure 8 shown, a three-dimensional curve frame generation device 800 is provided, including: an acquisition module 802, an encoding module 804, a distribution mapping module 806, a structure decoding module 808, a curve decoding module 810, and a wireframe generation module 812, where:

[0135] The acquisition module 802 is configured to acquire initial image data;

[0136] The encoding module 804 is configured to perform feature encoding on the initial image data to obtain an initial wireframe latent representation corresponding to the initial image data;

[0137] The distribution mapping module 806 is configured to perform distribution mapping on the sampled noise data based on the initial image features to obtain a three-dimensional curve frame latent representation corresponding to the initial image data;

[0138] The structure decoding module 808 is configured to perform feature decoding on the three-dimensional curve frame latent representation to obtain a three-dimensional curve frame structure and a curve latent representation corresponding to the initial image data;

[0139] The curve decoding module 810 is configured to perform feature decoding on the curve latent representation to obtain a three-dimensional curve frame curve corresponding to the initial image data;

[0140] The wireframe generation module 812 is configured to generate a target three-dimensional curve frame corresponding to the initial image data based on the three-dimensional curve frame structure and the three-dimensional curve frame curve.

[0141] In an exemplary embodiment, the three-dimensional curve frame generation device 800 is further configured to obtain the first training curve frame structure and the first training curve frame curve corresponding to the first training three-dimensional curve frame; perform feature encoding on the first training curve frame structure to obtain the first training structure feature, and perform feature encoding on the first training curve frame curve to obtain the first training curve latent representation; fuse the first training structure feature and the first training curve latent representation to obtain the first training curve frame latent representation; input the first training curve frame latent representation into the initial wireframe processing model to obtain the first predicted curve frame structure and the first predicted curve latent representation corresponding to the first training three-dimensional curve frame; based on the first training curve frame structure, the first training curve latent representation, the first predicted curve frame structure, and the first predicted curve latent representation, adjust the model parameters of the initial wireframe processing model until the convergence condition is met to obtain the target wireframe processing model; the structure decoding module 808 is further configured to input the three-dimensional curve frame latent representation into the target wireframe processing model to obtain the three-dimensional curve frame structure and the curve latent representation corresponding to the initial image data.

[0142] In an exemplary embodiment, the initial wireframe processing model includes an encoder and a decoder. The three-dimensional curve frame generation device 800 is further configured to input the first training curve frame latent representation into the encoder in the initial wireframe processing model to obtain the intermediate training curve frame latent representation; input the intermediate training curve frame latent representation into the decoder in the initial wireframe processing model to obtain the first predicted curve frame structure and the first predicted curve latent representation corresponding to the first training three-dimensional curve frame; the structure decoding module 808 is further configured to input the three-dimensional curve frame latent representation into the decoder in the target wireframe processing model to obtain the three-dimensional curve frame structure and the curve latent representation corresponding to the initial image data.

[0143] In an exemplary embodiment, the three-dimensional curve frame generation device 800 is further configured to obtain the third training curve frame curve and the third training curve latent representation; input the third training curve frame curve into the encoder in the initial curve processing model to obtain the third predicted curve latent representation; input the third predicted curve latent representation into the decoder in the initial curve processing model to obtain the third predicted curve frame curve; based on the third training curve latent representation, the third predicted curve latent representation, the third training curve frame curve, and the third predicted curve frame curve, adjust the model parameters of the initial curve processing model until the convergence condition is met to obtain the target curve processing model; the curve decoding module 810 is further configured to input the curve latent representation into the decoder in the target curve processing model to obtain the three-dimensional curve frame curve corresponding to the initial image data.

[0144] In an exemplary embodiment, the obtaining module 802 is further configured to obtain the initial curve frame curve; based on a preset starting point and a preset ending point, unify the positions of the starting points and ending points of each curve in the initial curve frame curve to obtain a standard curve frame curve; and perform a discretization process on the standard curve frame curve to obtain a third training curve frame curve.

[0145] In an exemplary embodiment, the three-dimensional curve frame generating device 800 is further configured to obtain the second training structural feature and the second training curve latent representation corresponding to the second training three-dimensional curve frame; fuse the second training structural feature and the second training curve latent representation to obtain the second training curve frame latent representation corresponding to the second training three-dimensional curve frame; input the second training curve frame latent representation and the sampled noise data into an initial distribution mapping model to obtain the predicted curve frame latent representation corresponding to the second training three-dimensional curve frame; adjust the model parameters of the initial distribution mapping model based on the second training curve frame latent representation and the predicted curve frame latent representation until a convergence condition is satisfied to obtain a target distribution mapping model; and the three-dimensional curve frame generating device 800 is further configured to input the initial wireframe latent representation and the sampled noise data into the target distribution mapping model to obtain the three-dimensional curve frame latent representation corresponding to the initial image data.

[0146] Each module in the above three-dimensional curve frame generating device can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor in the computer device in hardware form or be independent of it, or can be stored in the memory in the computer device in software form, so that the processor can call and execute the operations corresponding to the above respective modules.

[0147] In an exemplary embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 9 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as three-dimensional wireframes. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. The computer program, when executed by the processor, implements a three-dimensional curve frame generation method.

[0148] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structural diagram may be as shown in Figure 10 . The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a three-dimensional curve box generation method. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse, etc.

[0149] Those skilled in the art can understand that Figures 9 - 10 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0150] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0151] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0152] In an embodiment, a computer program product is provided, including a computer program. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0153] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0154] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., and are not limited thereto.

[0155] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this application.

[0156] The above-described embodiments merely represent several implementation manners of this application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent of this application. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of this application shall be subject to the appended claims.

Claims

1. A method for generating a three-dimensional curve frame, characterized in that, The method includes: Obtaining initial image data; Performing feature encoding on the initial image data to obtain an initial image feature corresponding to the initial image data; Based on the initial image feature, performing distribution mapping on sampled noise data to obtain a three-dimensional curve box latent representation corresponding to the initial image data; Performing feature decoding on the three-dimensional curve box latent representation to obtain a three-dimensional curve box structure and a curve latent representation corresponding to the initial image data; Performing feature decoding on the curve latent representation to obtain a three-dimensional curve box curve corresponding to the initial image data; Based on the three-dimensional curve box structure and the three-dimensional curve box curve, generating a target three-dimensional curve box corresponding to the initial image data.

2. The method according to claim 1, wherein The method further includes: Obtaining a first training curve box structure and a first training curve box curve corresponding to a first training three-dimensional curve box; Performing feature encoding on the first training curve box structure to obtain a first training structure feature, and performing feature encoding on the first training curve box curve to obtain a first training curve latent representation; Fusing the first training structure feature and the first training curve latent representation to obtain a first training curve box latent representation; Inputting the first training curve box latent representation into an initial wireframe processing model to obtain a first predicted curve box structure and a first predicted curve latent representation corresponding to the first training three-dimensional curve box; Based on the first training curve box structure, the first training curve latent representation, the first predicted curve box structure, and the first predicted curve latent representation, adjusting the model parameters of the initial wireframe processing model until a convergence condition is satisfied to obtain a target wireframe processing model; The performing feature decoding on the three-dimensional curve box latent representation to obtain a three-dimensional curve box structure and a curve latent representation corresponding to the initial image data includes: Inputting the three-dimensional curve box latent representation into the target wireframe processing model to obtain a three-dimensional curve box structure and a curve latent representation corresponding to the initial image data.

3. The method according to claim 2, wherein The initial wireframe processing model includes an encoder and a decoder; The inputting the first training curve box latent representation into the initial wireframe processing model to obtain a first predicted curve box structure and a first predicted curve latent representation corresponding to the first training three-dimensional curve box includes: Inputting the first training curve box latent representation into the encoder in the initial wireframe processing model to obtain an intermediate training curve box latent representation; Inputting the intermediate training curve box latent representation into the decoder in the initial wireframe processing model to obtain a first predicted curve box structure and a first predicted curve latent representation corresponding to the first training three-dimensional curve box; The inputting the three-dimensional curve box latent representation into the target wireframe processing model to obtain a three-dimensional curve box structure and a curve latent representation corresponding to the initial image data includes: Inputting the three-dimensional curve box latent representation into the decoder in the target wireframe processing model to obtain a three-dimensional curve box structure and a curve latent representation corresponding to the initial image data.

4. The method according to claim 1, characterized in that, The method further includes: Obtaining a third training curve box curve and a third training curve latent representation; Input the third training curve frame curve into the encoder in the initial curve processing model to obtain the third predicted curve latent representation; Input the third predicted curve latent representation into the decoder in the initial curve processing model to obtain the third predicted curve frame curve; Based on the third training curve latent representation, the third predicted curve latent representation, the third training curve frame curve, and the third predicted curve frame curve, adjust the model parameters of the initial curve processing model until the convergence condition is met to obtain the target curve processing model; The feature decoding of the curve latent representation to obtain the three-dimensional curve frame curve corresponding to the initial image data includes: Input the curve latent representation into the decoder in the target curve processing model to obtain the three-dimensional curve frame curve corresponding to the initial image data.

5. The method according to claim 4, characterized in that, The obtaining of the third training curve frame curve includes: Obtain the initial curve frame curve; Based on the preset starting point and preset ending point, unify the positions of the starting points and ending points of each curve in the initial curve frame curve to obtain the standard curve frame curve; Perform discrete processing on the standard curve frame curve to obtain the third training curve frame curve.

6. The method according to claim 1, characterized in that The method further includes: Obtain the second training structure feature and the second training curve latent representation corresponding to the second training three-dimensional curve frame; Fuse the second training structure feature and the second training curve latent representation to obtain the second training curve frame latent representation corresponding to the second training three-dimensional curve frame; Input the second training curve frame latent representation and the sampled noise data into the initial distribution mapping model to obtain the predicted curve frame latent representation corresponding to the second training three-dimensional curve frame; Based on the second training curve frame latent representation and the predicted curve frame latent representation, adjust the model parameters of the initial distribution mapping model until the convergence condition is met to obtain the target distribution mapping model; The distribution mapping of the sampled noise data based on the initial image feature to obtain the three-dimensional curve frame latent representation corresponding to the initial image data includes: Input the initial image feature and the sampled noise data into the target distribution mapping model to obtain the three-dimensional curve frame latent representation corresponding to the initial image data.

7. A three-dimensional curve frame generation device, characterized in that, The device includes: An acquisition module for acquiring initial image data; An encoding module for performing feature encoding on the initial image data to obtain the initial image feature corresponding to the initial image data; A distribution mapping module for performing distribution mapping on the sampled noise data based on the initial image feature to obtain the three-dimensional curve frame latent representation corresponding to the initial image data; A structure decoding module for performing feature decoding on the three-dimensional curve frame latent representation to obtain the three-dimensional curve frame structure and the curve latent representation corresponding to the initial image data; A curve decoding module for performing feature decoding on the curve latent representation to obtain the three-dimensional curve frame curve corresponding to the initial image data; A wireframe generation module for generating the target three-dimensional curve frame corresponding to the initial image data based on the three-dimensional curve frame structure and the three-dimensional curve frame curve.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

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

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images