Three-dimensional model acquisition method and device, computer equipment and storage medium

Abstract

The embodiment of the invention discloses a three-dimensional model obtaining method and device, computer equipment and a storage medium, and belongs to the technical field of computers. The method comprises the following steps: identifying the fineness of a first three-dimensional model to obtain a first fineness; based on the first fineness, encoding the first three-dimensional model to obtain a model feature corresponding to the first fineness; predicting a model feature corresponding to the second fineness based on the model feature corresponding to the first fineness; and decoding the model features corresponding to the second fineness to obtain a second three-dimensional model. According to the method and the device, a three-dimensional model optimization mode is realized, the fineness of the three-dimensional model can be improved, the three-dimensional model does not need to be edited manually, time consumed for obtaining the high-fineness three-dimensional model is saved, and the generation efficiency of the three-dimensional model can be improved while the fineness of the three-dimensional model is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technologies, and in particular, to a method, an apparatus, a computer device, and a storage medium for obtaining a three-dimensional model. Background Art

[0002] With the development of computer technologies, constructing a three-dimensional model has gradually become a relatively mature technology, and three-dimensional models are required in fields such as 3D (three Dimensions) printing, games, and architecture. In related technologies, in order to obtain a high-quality three-dimensional model, it is usually necessary for a human to manually generate a three-dimensional model through a model generation tool, but generating a three-dimensional model in this way requires a large amount of time, thereby resulting in low efficiency in generating three-dimensional models. Summary of the Invention

[0003] The embodiments of the present application provide a method, an apparatus, a computer device, and a storage medium for obtaining a three-dimensional model, which can improve the efficiency of generating a three-dimensional model. The technical solution is as follows:

[0004] On the one hand, a method for obtaining a three-dimensional model is provided. The method includes:

[0005] Identifying the fineness of a first three-dimensional model to obtain a first fineness, where the first fineness is used to represent the fineness degree of the first three-dimensional model;

[0006] Encoding the first three-dimensional model based on the first fineness to obtain a model feature corresponding to the first fineness;

[0007] Predicting a model feature corresponding to a second fineness based on the model feature corresponding to the first fineness, where the second fineness is higher than the first fineness;

[0008] Decoding the model feature corresponding to the second fineness to obtain a second three-dimensional model.

[0009] On the other hand, a device for obtaining a three-dimensional model is provided. The device includes:

[0010] An identification module, configured to identify the fineness of a first three-dimensional model to obtain a first fineness, where the first fineness is used to represent the fineness degree of the first three-dimensional model;

[0011] An encoding module, configured to encode the first three-dimensional model based on the first fineness to obtain a model feature corresponding to the first fineness;

[0012] A prediction module, configured to predict a model feature corresponding to a second fineness based on the model feature corresponding to the first fineness, where the second fineness is higher than the first fineness;

[0013] A decoding module, configured to decode the model features corresponding to the second fineness to obtain a second 3D model.

[0014] In a possible implementation manner, the prediction module is configured to process the model features corresponding to the first fineness through a fineness improvement model to obtain the model features corresponding to the second fineness. The fineness improvement model is used to predict the model features of the next fineness based on the input model features, and the second fineness is the next fineness of the first fineness.

[0015] In another possible implementation manner, the prediction module is configured to predict the model features corresponding to the second fineness based on the model features corresponding to the first fineness and control information, where the control information is used to describe the shape or color of the 3D model to be generated.

[0016] In another possible implementation manner, the prediction module is configured to encode the control information to obtain control features; fuse the control features and the model features corresponding to the first fineness to obtain a first fusion feature; and predict the model features corresponding to the second fineness based on the first fusion feature.

[0017] In another possible implementation manner, there are m fineness levels between the second fineness and the first fineness, where m is a positive integer greater than 0; the prediction module is configured to start from the first fineness and predict the model features corresponding to the next fineness based on the model features corresponding to the previous fineness.

[0018] In another possible implementation manner, there are m third fineness levels between the second fineness and the first fineness, where m is a positive integer greater than 0; the prediction module is configured to predict the model features corresponding to the first third fineness based on the model features corresponding to the first fineness; when the model features corresponding to the nth third fineness are obtained currently, predict the model features corresponding to the next fineness of the nth third fineness based on the model features corresponding to the first fineness to the nth third fineness, where n is a positive integer not greater than m.

[0019] In another possible implementation manner, when the model features corresponding to the nth third fineness are obtained currently, the prediction module is configured to fuse the model features corresponding to the first fineness to the nth third fineness to obtain a second fusion feature; and predict the model features corresponding to the next fineness of the nth third fineness based on the second fusion feature.

[0020] In another possible implementation manner, the encoding module is configured to sample from the contour of the first three-dimensional model based on the density corresponding to the first fineness to obtain point cloud data; and encode the point cloud data to obtain the model features corresponding to the first fineness.

[0021] In another possible implementation manner, the recognition module is configured to obtain a perspective image of the first three-dimensional model; encode the perspective image to obtain image features; and classify the image features to obtain the first fineness.

[0022] In another possible implementation manner, the perspective image includes images of the first three-dimensional model from multiple perspectives; or,

[0023] the perspective image includes at least one of a depth map of the first three-dimensional model, a normal map of the first three-dimensional model, or a coordinate map of the first three-dimensional model.

[0024] In another possible implementation manner, the decoding module is configured to decode the model features corresponding to the second fineness to obtain distance field information, where the distance field information indicates the distance from a point in the target space to the contour of the three-dimensional model; and generate the second three-dimensional model in the target space based on the distance field information.

[0025] On the other hand, a computer device is provided, where the computer device includes a processor and a memory, and at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the three-dimensional model acquisition method as described in the above aspect.

[0026] On the other hand, a computer-readable storage medium is provided, where at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to implement the operations performed by the three-dimensional model acquisition method as described in the above aspect.

[0027] In still another aspect, a computer program product is provided, including a computer program, and the computer program, when executed by a processor, implements the operations performed by the three-dimensional model acquisition method as described in the above aspect.

[0028] In the solution provided by the embodiments of the present application, by identifying the fineness of the first three-dimensional model to be optimized, encoding the model features corresponding to the first fineness according to the first fineness of the first three-dimensional model, using the model features corresponding to the first fineness to predict the model features of a higher fineness, and then decoding to obtain the second three-dimensional model of a higher fineness. In this way, a method for optimizing the three-dimensional model is realized, which can improve the fineness of the three-dimensional model, eliminate the need for manual editing of the three-dimensional model, save the time required to obtain a three-dimensional model with high fineness, and improve the generation efficiency of the three-dimensional model while improving the fineness of the three-dimensional model. Moreover, this three-dimensional model optimization method can be applied to any fineness, and can optimize the three-dimensional model corresponding to any fineness into the three-dimensional model corresponding to a higher fineness, improving the applicability of optimizing the three-dimensional model. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0030] Figure 1 is a structural block diagram of a computer system provided by an embodiment of the present application;

[0031] Figure 2 is a flowchart of a method for obtaining a three-dimensional model provided by an embodiment of the present application;

[0032] Figure 3 is a flowchart of another method for obtaining a three-dimensional model provided by an embodiment of the present application;

[0033] Figure 4 is a flowchart of yet another method for obtaining a three-dimensional model provided by an embodiment of the present application;

[0034] Figure 5 is a flowchart of yet another method for obtaining a three-dimensional model provided by an embodiment of the present application;

[0035] Figure 6 is a flowchart of yet another method for obtaining a three-dimensional model provided by an embodiment of the present application;

[0036] Figure 7 is a flowchart of yet another method for obtaining a three-dimensional model provided by an embodiment of the present application;

[0037] Figure 8 is a flowchart of a method for identifying fineness provided by an embodiment of the present application;

[0038] Figure 9It is a flowchart for obtaining predicted point cloud data provided by an embodiment of the present application;

[0039] Figure 10 It is a schematic diagram of point cloud data corresponding to a fineness provided by an embodiment of the present application;

[0040] Figure 11 It is a schematic diagram of a three-dimensional model provided by an embodiment of the present application;

[0041] Figure 12 It is a schematic structural diagram of a three-dimensional model acquisition device provided by an embodiment of the present application;

[0042] Figure 13 It is a schematic structural diagram of a terminal provided by an embodiment of the present application;

[0043] Figure 14 It is a schematic structural diagram of a server provided by an embodiment of the present application. Detailed implementation manners

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will further describe the embodiments of the present application in detail with reference to the accompanying drawings.

[0045] The terms "first", "second", "third", "fourth", etc. used in the present application may be used herein to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the present application, the first fineness may be referred to as the second fineness, and similarly, the second fineness may be referred to as the first fineness.

[0046] The terms "at least one", "multiple", "each", "any one" used in the present application, at least one includes one, two, or more than two, multiple includes two or more than two, and each refers to each one of the corresponding multiple, and any one refers to any one of the multiple. For example, multiple fineness levels include 3 fineness levels, and each refers to each of these 3 fineness levels, and any one refers to any one of these 3 fineness levels, which can be the first fineness level, or the second fineness level, or the third fineness level.

[0047] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data for analysis, stored data, displayed data, etc.), and signals involved in the present application are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions. For example, the three-dimensional models, control information, etc. involved in the present application are obtained under full authorization.

[0048] The 3D model acquisition method provided by the embodiments of the present application is executed by a computer device. Optionally, the computer device is a terminal or a server. Optionally, the server is an independent physical server, or a server cluster composed of multiple physical servers. Optionally, the terminal is a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal, etc., but is not limited thereto.

[0049] Figure 1 is a structural block diagram of a computer system provided by the embodiments of the present application. Refer to Figure 1 As shown in the figure, the computer system includes a terminal 101 and a server 102.

[0050] The terminal 101 is configured to send a first 3D model to the server 102. The server 102 is configured to receive the first 3D model, optimize the first 3D model according to the method provided by the embodiments of the present application to improve the fineness of the 3D model, and obtain a second 3D model with higher fineness, and send the second 3D model to the terminal 101. The terminal 101 is configured to receive the second 3D model so as to be able to display the second 3D model.

[0051] In some embodiments, an application 111 provided by the server 102 is installed and run on the terminal 101. Optionally, the application 111 is any application for generating a 3D model. The terminal 101 is a terminal used by a user. The user uses the terminal 101 to log in to the application 111 based on an object to display a model display interface, so as to display a 3D model through the model display interface.

[0052] The terminal 101 is connected to the server 102 through a wireless network or a wired network. The server 102 is configured to provide background services for the application 111. Exemplarily, the server 102 includes a processor and a memory. The memory further includes a receiving module and a sending module. The receiving module is configured to receive requests sent by the terminal 101, such as model optimization requests, model acquisition requests, etc.; the processor is configured to respond to the received requests; the sending module is configured to send responses to the terminal 101, such as sending an updated 3D model to the terminal 101. Optionally, the server 102 undertakes the main computing work, and the terminal 101 undertakes the secondary computing work; or, the server 102 undertakes the secondary computing work, and the terminal 101 undertakes the main computing work.

[0053] In a possible implementation, the terminal 101 is used to import a first 3D model to be optimized in a model display interface, and in response to an optimization operation on the first 3D model, send the first 3D model to the server 102; the server 102 is used to receive the first 3D model, optimize the first 3D model according to the method provided in the embodiments of the present application to obtain a second 3D model, and send the second 3D model to the terminal 101; the terminal 101 receives the second 3D model and displays the second 3D model in the model display interface.

[0054] Those skilled in the art can know that the number of the above terminals can be more or less. For example, there can be only one of the above terminals, or there can be dozens or hundreds of the above terminals, or even more. The embodiments of the present application do not limit the number and device type of the terminals.

[0055] Figure 2 is a flowchart of a method for obtaining a 3D model provided by the embodiments of the present application. This method is executed by a computer device, such as Figure 2 shown, and this method includes:

[0056] 201. The computer device identifies the fineness of the first 3D model to obtain a first fineness, and the first fineness is used to represent the fineness degree of the first 3D model.

[0057] In the embodiments of the present application, in order to improve the quality of the 3D model, the first third model is optimized according to the fineness of the first 3D model, so as to improve the fineness of the 3D model, ensure that the fineness of the obtained second 3D model is higher than that of the first 3D model, ensure that the quality of the 3D model can be improved, and make the second 3D model more refined, that is, the details of the second 3D model are clearer and more accurate.

[0058] Among them, the first 3D model is a 3D model of any object. For example, the first 3D model is a 3D model of a virtual building, a 3D model of a virtual character, or a 3D model of a virtual animal, etc. The first 3D model can be represented in any form. For example, the first 3D model is a 3D mesh model composed of polygons, or a 3D model represented by point cloud data. The first fineness is any fineness. For example, there are 10 fineness levels in total, and the larger the number, the higher the accuracy. The first fineness is fineness 3 or fineness 4.

[0059] 202. The computer device encodes the first 3D model based on the first fineness to obtain a model feature corresponding to the first fineness.

[0060] In the embodiments of the present application, the fineness of the three-dimensional model can reflect the degree of fineness of the three-dimensional model. When the fineness of the first three-dimensional model is recognized, the first three-dimensional model is encoded according to the recognized fineness, so that the obtained model features can accurately represent the specific situation of the first three-dimensional model at this fineness, thereby ensuring the accuracy of the model features.

[0061] Among them, the model features corresponding to the first fineness are used to represent the first three-dimensional model, and the model features can be represented in any form. For example, the model features are feature vectors or feature matrices, such as the model features are represented in the form of three-dimensional feature matrices.

[0062] 203. The computer device predicts the model features corresponding to the second fineness based on the model features corresponding to the first fineness, and the second fineness is higher than the first fineness.

[0063] In the embodiments of the present application, when the model features of the first three-dimensional model are obtained, that is, when the model features corresponding to the first fineness are obtained, the model features corresponding to the first fineness can be refined to predict the model features of a higher fineness, so that the model features corresponding to the second fineness can represent the three-dimensional model of a higher fineness.

[0064] Among them, the second fineness is any fineness higher than the first fineness. For example, there are 10 fineness levels in total, the first fineness is fineness level 3, and the second fineness is fineness level 4 or fineness level 7, etc.

[0065] 204. The computer device decodes the model features corresponding to the second fineness to obtain the second three-dimensional model.

[0066] In the embodiments of the present application, compared with the first three-dimensional model, the model features corresponding to the second fineness can represent the three-dimensional model of a higher fineness. Therefore, by decoding the model features corresponding to the second fineness, the second three-dimensional model is obtained, so that the second three-dimensional model has a higher fineness than the first three-dimensional model.

[0067] Among them, the fineness of the second three-dimensional model is the second fineness. Compared with the first three-dimensional model, the content represented by the second three-dimensional model is more detailed. The second three-dimensional model can be represented in any form. For example, the second three-dimensional model is a three-dimensional mesh model composed of polygons or a three-dimensional model represented by point cloud data.

[0068] For example, taking the three-dimensional model of a virtual building as the first three-dimensional model, the first three-dimensional model can only present the outline of the virtual building, while the second three-dimensional model can not only present the outline of the virtual building, but also present details such as virtual doors and windows and virtual bricks and tiles of the virtual building.

[0069] In the solution provided by the embodiments of the present application, by identifying the fineness of the first three-dimensional model to be optimized, encoding the model features corresponding to the first fineness according to the first fineness of the first three-dimensional model, using the model features corresponding to the first fineness to predict the model features of a higher fineness, and then decoding to obtain the second three-dimensional model of a higher fineness. In this way, a method for optimizing the three-dimensional model is realized, which can improve the fineness of the three-dimensional model, eliminate the need for manual editing of the three-dimensional model, save the time required to obtain a three-dimensional model with high fineness, and can improve the generation efficiency of the three-dimensional model while improving the fineness of the three-dimensional model. Moreover, this three-dimensional model optimization method can be applied to any fineness, and can optimize the three-dimensional model corresponding to any fineness into a three-dimensional model corresponding to a higher fineness, improving the applicability of optimizing the three-dimensional model.

[0070] Based on the Figure 2 embodiment shown, in the embodiments of the present application, the process of identifying the fineness of the first three-dimensional model, the encoding process of the first three-dimensional model, predicting the model features corresponding to the second fineness, and the decoding process of the model features corresponding to the second fineness are described in detail. For the specific process, please refer to the following embodiments.

[0071] Figure 3 is a flowchart of a method for obtaining a three-dimensional model provided by the embodiments of the present application. This method is executed by a computer device, as Figure 3 shown. The method includes:

[0072] 301. The computer device obtains the perspective image of the first three-dimensional model.

[0073] In the embodiments of the present application, the perspective image is an image of the first three-dimensional model at any perspective. The content of the perspective image can reflect the shape or color of the first three-dimensional model. The perspective image is a two-dimensional image.

[0074] In a possible implementation manner, the perspective image includes images of the first three-dimensional model at multiple perspectives.

[0075] In the embodiments of the present application, the perspective image is equivalent to an image obtained by photographing the first three-dimensional model from multiple perspectives. Considering that the content presented by the first three-dimensional model at different perspectives is different, therefore, images of the first three-dimensional model at multiple perspectives are obtained, so as to subsequently identify the fineness of the first three-dimensional model using the images at multiple perspectives, enriching the basis for identifying the fineness of the first three-dimensional model subsequently and ensuring the accuracy of the fineness identified subsequently.

[0076] Among them, the multiple perspectives are arbitrary perspectives. For example, the multiple perspectives are the front perspective, rear perspective, left perspective, and right perspective of the first three-dimensional model.

[0077] In a possible implementation, the perspective image includes at least one of the depth map of the first 3D model, the normal map of the first 3D model, or the coordinate map of the first 3D model.

[0078] Among them, the perspective image is obtained by a virtual camera in the virtual space photographing a 3D model in the virtual space. The depth image is a 2D image representing the distances from each point on the contour of the 3D model to the virtual camera. The normal map is an image used to represent the normal directions of each point on the surface of the 3D model. The normal map can be represented in any form. For example, the normal map is represented in the form of a 3D matrix, and the 3D matrix can reflect the angles of each normal direction with the X-axis, Y-axis, and Z-axis in the virtual space. The coordinate map is an image used to represent the coordinates of each vertex in the 3D model. The coordinate map can be represented in any form. For example, the coordinate map is represented in the form of a 3D matrix, and the 3D matrix can reflect the coordinates of each vertex.

[0079] In the embodiments of the present application, the perspective image includes at least one of the depth map, the normal map, or the coordinate map, which can enrich the types of perspective images, so as to enrich the basis for subsequent recognition of the fineness of the first 3D model and ensure the accuracy of the recognized fineness subsequently.

[0080] In a possible implementation, the perspective image includes images of the first 3D model from multiple perspectives, and each image from a perspective includes at least one of the depth map, the normal map, or the coordinate map.

[0081] In a possible implementation, the process of obtaining the perspective image includes: rendering the 3D model through a rendering engine to obtain the perspective image.

[0082] Among them, the rendering engine is used to process the 3D model. The rendering engine is any type of engine, and the rendering engine is Unreal Engine (a game engine).

[0083] 302. The computer device encodes the perspective image to obtain image features.

[0084] In the embodiments of the present application, the image features can characterize the perspective image, and the image features can be represented in any form. For example, the image features are represented in the form of feature vectors or feature matrices.

[0085] In a possible implementation, if the perspective image includes multiple images, then step 302 includes: encoding each of the multiple images to obtain the features of each image, and fusing the features of the multiple images to obtain the image features.

[0086] In the embodiments of the present application, since the perspective image includes multiple images, each image is first encoded separately and then the features of the multiple images are fused to ensure that the obtained image features can accurately describe the multiple images included in the perspective image and ensure the accuracy of the image features.

[0087] 303. The computer device classifies the image features to obtain a first fineness, and the first fineness is used to represent the fineness degree of the first 3D model.

[0088] In the embodiments of the present application, since the image features can accurately describe the perspective image, the image features are classified to obtain a first accuracy, and the first accuracy obtained by classification is the fineness degree of the first 3D model.

[0089] In the embodiments of the present application, the perspective image is a 2D image of the first 3D model from any perspective, and the content of the perspective image can reflect the shape or color of the first 3D model. By encoding the perspective image and classifying the encoded image features, the first accuracy obtained by classification is the fineness degree of the first 3D model, which ensures the accuracy of the first fineness obtained by classification and also does not require directly identifying the fineness degree of the 3D model, which can reduce the computational amount of identifying the fineness degree and ensure the efficiency of identifying the fineness degree.

[0090] In a possible implementation manner, the above steps 302-303 are implemented by a fineness recognition model. That is, the process of obtaining the first fineness includes: encoding the perspective image through the fineness recognition model to obtain image features, and classifying the image features to obtain the first fineness.

[0091] Among them, the fineness recognition model is any type of neural network model, and the fineness recognition model is a geometric fineness discriminator for identifying the geometric fineness of the 3D model.

[0092] Optionally, the process of obtaining the first fineness includes: performing feature transformation on the image features based on the fineness recognition model to obtain a probability vector, where the probability vector includes probabilities corresponding to multiple fineness degrees, and determining the fineness degree corresponding to the maximum probability in the probability vector as the first fineness.

[0093] Among them, the multiple fineness degrees are preset fineness degrees. For example, the multiple fineness degrees are 10 fineness degrees.

[0094] It should be noted that the embodiments of the present application take the identification of the fineness degree based on the perspective image of the first 3D model as an example for description. In another embodiment, the above steps 301-303 do not need to be executed, but other methods are adopted to identify the fineness degree of the first 3D model to obtain the first fineness.

[0095] In a possible implementation, the process of obtaining the first fineness includes: encoding the first 3D model based on the fineness recognition model to obtain the features of the first 3D model, and classifying the features of the first 3D model to obtain the first fineness.

[0096] Among them, the features of the first 3D model can be represented in any form. For example, the features of the first 3D model are represented in the form of feature vectors or feature matrices.

[0097] 304. The computer device samples from the contour of the first 3D model based on the density corresponding to the first fineness to obtain point cloud data.

[0098] In the embodiments of the present application, different fineness levels correspond to different densities, and the density can reflect the density of points in the sampled point cloud data. Since the first fineness can reflect the fineness degree of the first 3D model, therefore, sampling points from the contour of the first 3D model according to the density corresponding to the first fineness to obtain point cloud data, so that the points in the point cloud data match the first fineness, avoiding inaccurate data caused by too few points in the point cloud data, and also avoiding large computational amounts caused by too many points in the point cloud data, ensuring the accuracy of the point cloud data and the subsequent calculation efficiency.

[0099] In a possible implementation, the fineness is positively correlated with the corresponding density.

[0100] In the embodiments of the present application, the higher the fineness, the greater the density corresponding to the fineness; the lower the fineness, the smaller the density corresponding to the fineness. Sampling the point cloud data according to the density corresponding to the fineness enables the point cloud data to accurately represent the 3D model corresponding to the fineness, ensuring the accuracy of the point cloud data.

[0101] In a possible implementation, the process of obtaining point cloud data includes the following two methods.

[0102] The first method includes: determining the spatial range occupied by the first 3D model, dividing the spatial range into equal-sized cubic grids according to the density, determining at least one sampling point located on the contour of the first 3D model from each cubic grid based on the density, and forming the determined sampling points into point cloud data.

[0103] Among them, the density can reflect the number of determined points in a single cubic grid. The greater the density, the more points are determined from the cubic grid, and the smaller the density, the fewer points are determined from the cubic grid.

[0104] In an embodiment of the present application, the spatial range occupied by the first three-dimensional model is divided into a plurality of uniform cubic grids, and then a certain number of points located on the contour of the first three-dimensional model are selected as sampling points according to the density corresponding to the fineness in each cubic grid.

[0105] Optionally, the process of determining sampling points from the cubic grids includes: for each cubic grid, randomly select at least one point located on the contour of the first three-dimensional model from the cubic grid and determine it as a sampling point; or, select at least one point with the closest distance to the center point of the cubic network from the points located on the contour of the first three-dimensional model in the cubic network and determine it as a sampling point.

[0106] The second method includes: converting the first three-dimensional model into original point cloud data and determining the distances between each point in the original point cloud data and other points; for any point, if the number of points with a distance less than the distance threshold from this point is not less than the quantity threshold, delete some points from the points with a distance less than the distance threshold from this point so that the number of points with a distance less than the distance threshold from this point is less than the quantity threshold, and determine the updated original point cloud data as the point cloud data.

[0107] Among them, the points in the original point cloud data are the points on the contour of the first three-dimensional model, and the original point cloud data can reflect the shape or color of the first three-dimensional model. The quantity threshold is an arbitrary value, and this quantity threshold matches the density corresponding to the first fineness, and the quantity threshold changes with the density.

[0108] In an embodiment of the present application, after converting the first three-dimensional model into original point cloud data, according to the density corresponding to the first fineness, delete some points from the original point cloud data so that the deleted point cloud data matches this density, ensuring the accuracy of the point cloud data.

[0109] 305. The computer device encodes the point cloud data to obtain model features corresponding to the first fineness.

[0110] In an embodiment of the present application, since the first fineness can reflect the fineness degree of the first three-dimensional model, therefore, sampling points are taken from the contour of the first three-dimensional model according to the density corresponding to the first fineness to obtain point cloud data, and then the point cloud data is encoded to obtain model features corresponding to the first fineness, so that the points in the point cloud data match the first fineness, avoiding inaccurate data caused by too few points in the point cloud data, and also being able to avoid large computational amounts caused by too many points in the point cloud data, ensuring that the point cloud data is as accurate as possible, and further ensuring the accuracy of the model features corresponding to the first fineness.

[0111] In a possible implementation, the above step 305 is implemented by an encoding model. That is, the process of obtaining the model features corresponding to the first fineness includes: encoding the point cloud data through the encoding model to obtain the model features corresponding to the first fineness.

[0112] Among them, the encoding model (Encoder) is any type of neural network model.

[0113] In the embodiments of the present application, the encoding model is used to encode the point cloud data into features. By encoding the point cloud data through the encoding model, the accuracy of the obtained model features can be ensured.

[0114] It should be noted that the embodiments of the present application take obtaining the model features corresponding to the first fineness by using the density corresponding to the first fineness as an example for illustration. In another embodiment, the above steps 304-305 do not need to be executed, but other methods are adopted to encode the first three-dimensional model based on the first fineness to obtain the model features corresponding to the first fineness.

[0115] 306. The computer device processes the model features corresponding to the first fineness through a fineness improvement model to obtain the model features corresponding to the second fineness. The fineness improvement model is used to predict the model features of the next fineness based on the input model features. The second fineness is the next fineness of the first fineness.

[0116] Among them, the fineness improvement model can be any neural network model. For example, the fineness improvement model is a Transformer (a deep learning model) structure. The fineness improvement model is used to improve the fineness of the input model features. The fineness improvement model can be applied to multiple fineness levels. That is, it can improve the model features of any input fineness level to the model features of the next fineness level. The second fineness is higher than the first fineness.

[0117] In the embodiments of the present application, the fineness improvement model is used to improve the model features of the input to the model features of the next fineness level. The model features corresponding to the first fineness are the model features input to the fineness improvement model. By processing the model features corresponding to the first fineness through the fineness improvement model, the fineness of the input model features can be improved, and the model features corresponding to the second fineness are output, avoiding the inaccurate situation of directly improving the model features corresponding to the first fineness across multiple fineness levels, and gradually improving the fineness of the model features to ensure the accuracy of the output model features.

[0118] For example, if there are 10 fineness levels in total, if the first fineness is fineness 3, then the second fineness is fineness 4; if the first fineness is fineness 4, then the second fineness is fineness 5.

[0119] In a possible implementation, step 306 includes: performing feature transformation on the model features corresponding to the first fineness through a fineness improvement model to obtain the model features corresponding to the second fineness.

[0120] In the embodiments of the present application, the fineness improvement model can perform feature transformation on the input model features to refine the input model features, so as to improve the fineness of the input model features, and further ensure the accuracy of the model features corresponding to the second fineness.

[0121] It should be noted that the embodiments of the present application are described by taking obtaining the model features corresponding to the second fineness through a fineness improvement model as an example. In another embodiment, the above step 306 does not need to be executed, but other methods are adopted to predict the model features corresponding to the second fineness based on the model features corresponding to the first fineness.

[0122] 307. The computer device decodes the model features corresponding to the second fineness to obtain distance field information, where the distance field information indicates the distance from a point in the target space to the contour of the three-dimensional model.

[0123] In the embodiments of the present application, considering that the model features corresponding to the second fineness can represent the three-dimensional model corresponding to the second fineness and can describe the shape of the three-dimensional model corresponding to the second fineness, the model features corresponding to the second fineness are decoded to determine the distance from each point in the target space to the contour of the three-dimensional model. Then, according to the distance indicated by the distance field information, the points located on the contour of the three-dimensional model are determined from the target space, and then a second three-dimensional model is generated in the target space to ensure that the second three-dimensional model matches the model features corresponding to the second fineness, that is, the fineness of the second three-dimensional model is the second fineness. The second three-dimensional model is obtained by further refining the first three-dimensional model, and the second three-dimensional model matches the first three-dimensional model, ensuring the accuracy of the second three-dimensional model and the refinement effect of the three-dimensional model.

[0124] Among them, the distance field information can be represented in any form. For example, the distance field information is represented in the form of a matrix.

[0125] In a possible implementation, step 307 is implemented through a decoding model. That is, the process of obtaining the model features corresponding to the first fineness includes: decoding the model features corresponding to the second fineness through the decoding model to obtain distance field information.

[0126] Among them, the decoding model (Decoder) is any type of neural network model.

[0127] In the embodiments of the present application, the decoding model is used to decode the model features into distance field information. Through the decoding model, the model features corresponding to the second fineness are decoded to ensure the accuracy of the obtained distance field information.

[0128] 308. The computer device generates a second three-dimensional model in the target space based on the distance field information.

[0129] In the embodiments of the present application, the distance field information includes the distances from each point in the target space to the contour of the three-dimensional model. Based on this distance field information, the points on the contour of the three-dimensional model in the target space can be determined. Furthermore, the points on the contour of the three-dimensional model in the target space are outlined to form the contour of the three-dimensional model, thereby constituting the second three-dimensional model.

[0130] In a possible implementation manner, step 308 includes: based on the distances included in the distance field information, the points with a distance of 0 in the target space are determined as contour points, the contour points are connected to adjacent contour points, and the model obtained by connecting the contour points is determined as the second three-dimensional model.

[0131] In the embodiments of the present application, the points with a distance of 0 indicated by the distance field information are the contour points of the three-dimensional model. Therefore, the contour points on the three-dimensional model can be determined from the target space, and the contour points are connected to each other according to the positional relationship between the contour points to form a three-dimensional network model. The three-dimensional network model is the second three-dimensional model corresponding to the second fineness, ensuring that the second three-dimensional model matches the model features corresponding to the second fineness, and further ensuring the accuracy of the second three-dimensional model.

[0132] It should be noted that the embodiments of the present application take decoding the model features to obtain distance field information as an example for illustration. The distance field information is SDF (Signed Distance Field, directed distance field). In another embodiment, the model features can also be decoded into other types of information, so as to obtain the second three-dimensional model based on other types of information. Other types of information can be Occupancy Map (occupancy map), UDF (User-Defined Feature, user-defined feature), etc.

[0133] It should be noted that the embodiments of the present application take decoding to obtain distance field information first and then generating the second three-dimensional model as an example for illustration. In another embodiment, it is not necessary to perform the above steps 307-308, but other methods are adopted to decode the model features corresponding to the second fineness to obtain the second three-dimensional model.

[0134] In the solution provided by the embodiments of the present application, by identifying the fineness of the first three-dimensional model to be optimized, encoding the model features corresponding to the first fineness according to the first fineness of the first three-dimensional model, and using the model features corresponding to the first fineness to predict the model features of a higher fineness, and then decoding to obtain the second three-dimensional model of a higher fineness. In this way, a method for optimizing a three-dimensional model is realized, which can improve the fineness of the three-dimensional model, and there is no need for manual editing of the three-dimensional model, saving the time required to obtain a three-dimensional model with a high fineness. It can improve the generation efficiency of the three-dimensional model while improving the fineness of the three-dimensional model. Moreover, this three-dimensional model optimization method can be applied to any fineness, and can optimize the three-dimensional model corresponding to any fineness to the three-dimensional model corresponding to a higher fineness, improving the applicability of optimizing the three-dimensional model.

[0135] Based on the above Figure 2 On the basis of the shown embodiment, the embodiments of the present application can also combine control information to predict the model features of a higher fineness. For the specific process, please refer to the following embodiments.

[0136] Figure 4 is a flowchart of a method for obtaining a three-dimensional model provided by the embodiments of the present application. This method is executed by a computer device. As Figure 4 shown, this method includes:

[0137] 401. The computer device identifies the fineness of the first three-dimensional model to obtain the first fineness, and the first fineness is used to represent the fineness degree of the first three-dimensional model.

[0138] 402. The computer device encodes the first three-dimensional model based on the first fineness to obtain the model features corresponding to the first fineness.

[0139] It should be noted that the above steps 401-402 are the same as the above steps 201-202, and will not be elaborated here.

[0140] 403. The computer device predicts the model features corresponding to the second fineness based on the model features corresponding to the first fineness and the control information. The control information is used to describe the shape or color of the three-dimensional model to be generated, and the second fineness is higher than the first fineness.

[0141] In the embodiments of the present application, in the process of refining the first three-dimensional model, in order to ensure that the refined three-dimensional model can meet the user's expectations, therefore, the control information representing the user's expectations is combined, and the control information is used as the refinement basis to refine the model features corresponding to the first fineness, so as to predict the model features corresponding to a higher fineness, so that the refined model features meet the user's expectations and ensure the controllability and accuracy of the refined model features.

[0142] Among them, the control information can be any type of information, such as text, pictures, videos, point cloud data, etc.

[0143] For example, when the control information is text, the content of the text describes the shape or color of the three-dimensional model expected by the user.

[0144] For example, when the control information is an image, the content in the image may be different from the first three-dimensional model or may be similar to the first three-dimensional model. For example, if the first three-dimensional model is a virtual building model, the content in the image can be a table and chair, or it can be another building different from the first three-dimensional model.

[0145] For another example, when the control information is point cloud data, the point cloud data is also used to represent a three-dimensional model expected by the user. The three-dimensional model represented by the point cloud data may be different from the first three-dimensional model or may be similar to the first three-dimensional model.

[0146] In a possible implementation manner, step 403 includes: encoding the control information to obtain a control feature; fusing the control feature and the model feature corresponding to the first fineness to obtain a first fusion feature; predicting the model feature corresponding to the second fineness based on the first fusion feature.

[0147] In the embodiments of the present application, by encoding the control information, the feature obtained by encoding the control information is fused with the model feature corresponding to the first fineness, so that the obtained first fusion feature can reflect the model feature corresponding to the first fineness and the control information. Furthermore, the fused feature is processed to ensure that the predicted model feature corresponding to the second fineness matches the control information, thereby ensuring the accuracy of the predicted model feature corresponding to the second fineness.

[0148] Among them, the control feature can be represented in any form. For example, the control feature is represented in the form of a feature vector or a feature matrix. The process of predicting the model feature corresponding to the second fineness based on the first fusion feature is the same as step 306 above and will not be elaborated here.

[0149] In a possible implementation manner, the control information is multi-modal information, and the control information includes at least two of text, images, or point cloud data.

[0150] In the embodiments of the present application, the refinement of the three-dimensional model can be guided by multi-modal information to ensure that the refined three-dimensional model can meet the user's expectations, thereby ensuring the accuracy of the refined three-dimensional model.

[0151] It should be noted that the embodiments of the present application are described by taking the example of predicting the model features corresponding to the second fineness in combination with control information. In another embodiment, the above step 403 does not need to be executed, but other methods are adopted to predict the model features corresponding to the second fineness based on the model features corresponding to the first fineness.

[0152] 404. The computer device decodes the model features corresponding to the second fineness to obtain a second 3D model.

[0153] It should be noted that the above step 404 is the same as step 204 above and will not be elaborated here.

[0154] In the solution provided by the embodiments of the present application, by identifying the fineness of the first 3D model to be optimized, encoding the model features corresponding to the first fineness according to the first fineness of the first 3D model, using the model features corresponding to the first fineness to predict the model features of a higher fineness, and then decoding to obtain the second 3D model of a higher fineness. In this way, a method for optimizing the 3D model is realized, which can improve the fineness of the 3D model, and there is no need for manual editing of the 3D model, saving the time required to obtain a 3D model with high fineness. It can improve the generation efficiency of the 3D model while improving the fineness of the 3D model; and this 3D model optimization method can be applied to any fineness, and can optimize the 3D model corresponding to any fineness to the 3D model corresponding to a higher fineness, improving the applicability of optimizing the 3D model.

[0155] In the above Figure 2 Based on the above-described embodiments, in the embodiments of the present application, the second fineness is higher than the first fineness, and a step-by-step refinement method is adopted. Each time, the model features of a higher fineness are obtained based on the model features corresponding to the current fineness. For the specific process, please refer to the following embodiments.

[0156] Figure 5 is a flowchart of a method for obtaining a 3D model provided by the embodiments of the present application. This method is executed by a computer device. As Figure 5 shown, this method includes:

[0157] 501. The computer device identifies the fineness of the first 3D model to obtain a first fineness, and the first fineness is used to represent the fineness degree of the first 3D model.

[0158] 502. The computer device encodes the first 3D model based on the first fineness to obtain the model features corresponding to the first fineness.

[0159] It should be noted that the above steps 501-502 are the same as steps 201-202 above and will not be elaborated here.

[0160] 503. The computer device starts from the first fineness and predicts the model features corresponding to the next fineness based on the model features corresponding to the previous fineness.

[0161] In the embodiment of the present application, the second fineness is higher than the first fineness, and there are m fineness levels between the second fineness and the first fineness, where m is a positive integer greater than 1. That is to say, the second fineness is higher than the first fineness by multiple levels. The second fineness is the fineness after the first 3D model is refined as expected. Starting from the first fineness, the fineness of the model features is improved in a step-by-step manner until the model features corresponding to the second fineness are obtained. This avoids the situation where the model features are inaccurate due to directly refining the model features corresponding to the first fineness into the model features corresponding to the second fineness. By adopting a step-by-step improvement method to refine the model features, the accuracy of the model features corresponding to the second fineness obtained is ensured.

[0162] It should be noted that the above step 503 is only illustrated by taking the prediction of the model features corresponding to the next fineness as an example. In another embodiment, according to the above step 503, the model features corresponding to higher fineness levels can be continuously predicted until the model features corresponding to the second fineness are obtained.

[0163] For example, taking the first fineness as fineness level 1, m as 2, and the second fineness as fineness level 4 as an example, the model features corresponding to the first fineness are the model features corresponding to fineness level 1; based on the model features corresponding to fineness level 1, the model features corresponding to fineness level 2 are predicted; based on the model features corresponding to fineness level 2, the model features corresponding to fineness level 3 are predicted; based on the model features corresponding to fineness level 3, the model features corresponding to fineness level 4 are predicted; based on the model features corresponding to fineness level 4, the model features corresponding to fineness level 5 are predicted.

[0164] It should be noted that the process of predicting the model features corresponding to the next fineness based on the model features corresponding to the previous fineness is the same as the above step 306 and will not be elaborated here.

[0165] It should be noted that the embodiment of the present application is illustrated by taking obtaining the model features of a higher fineness based on the model features corresponding to the current fineness each time as an example. In another embodiment, the above step 503 does not need to be executed, but other methods are adopted to predict the model features corresponding to the second fineness based on the model features corresponding to the first fineness.

[0166] 504. The computer device decodes the model features corresponding to the second fineness to obtain the second 3D model.

[0167] It should be noted that the above step 504 is the same as the above step 204 and will not be elaborated here.

[0168] In the solution provided by the embodiments of the present application, by identifying the fineness of the first three-dimensional model to be optimized, encoding the model features corresponding to the first fineness according to the first fineness of the first three-dimensional model, and using the model features corresponding to the first fineness to predict the model features of a higher fineness, and then decoding to obtain the second three-dimensional model of a higher fineness. In this way, a method for optimizing the three-dimensional model is realized, which can improve the fineness of the three-dimensional model, without the need for manual editing of the three-dimensional model, saving the time required to obtain a three-dimensional model with high fineness, and can improve the generation efficiency of the three-dimensional model while improving the fineness of the three-dimensional model; and this three-dimensional model optimization method can be applied to any fineness, and can optimize the three-dimensional model corresponding to any fineness to the three-dimensional model corresponding to a higher fineness, improving the applicability of optimizing the three-dimensional model.

[0169] Based on the above Figure 2 On the basis of the shown embodiment, in the embodiments of the present application, the second fineness is higher than the first fineness, and a step-by-step refinement method is adopted. Each time, the model features of a higher fineness are obtained based on the currently obtained model features. For the specific process, please refer to the following embodiments.

[0170] Figure 6 is a flowchart of a method for obtaining a three-dimensional model provided by the embodiments of the present application. This method is executed by a computer device. As Figure 6 shown, this method includes:

[0171] 601. The computer device identifies the fineness of the first three-dimensional model to obtain the first fineness, and the first fineness is used to represent the fineness degree of the first three-dimensional model.

[0172] 602. The computer device encodes the first three-dimensional model based on the first fineness to obtain the model features corresponding to the first fineness.

[0173] It should be noted that the above steps 501-502 are the same as the above steps 201-202, and will not be elaborated here.

[0174] 603. The computer device predicts the model features corresponding to the first third fineness based on the model features corresponding to the first fineness.

[0175] In the embodiments of the present application, the second fineness is higher than the first fineness, and there are m third fineness intervals between the second fineness and the first fineness, where m is a positive integer greater than 0. The first third fineness is the fineness interval between the first fineness and the second fineness, and the first third fineness is the next fineness higher than the first fineness. In the case of obtaining the model features corresponding to the first fineness, based on the model features corresponding to the first fineness, the model corresponding to the next fineness, that is, the model features corresponding to the first third fineness, can be predicted.

[0176] It should be noted that step 603 is the same as step 306 described above, and will not be elaborated here.

[0177] 604. When the computer device obtains the model features corresponding to the nth third fineness currently, based on the model features corresponding to the first fineness to the nth third fineness, predict the model features corresponding to the next fineness of the nth third fineness, where n is a positive integer not greater than m.

[0178] In the embodiment of the present application, the second fineness is higher than the first fineness, and there are m fineness levels between the second fineness and the first fineness. m is a positive integer greater than 1. That is to say, the second fineness is higher than the first fineness by multiple levels. The second fineness is the fineness after the first three-dimensional model is refined. Starting from the first fineness, the fineness of the model features is improved in a step-by-step manner until the model features corresponding to the second fineness are obtained. During the process of gradually refining the model features, based on the model features corresponding to the currently obtained multiple fineness levels, the model features corresponding to the next fineness can be predicted, which can enrich the basis for refining the model features, ensure the accuracy of predicting the model features, and also avoid the situation where the model features are inaccurate due to directly refining the model features corresponding to the first fineness into the model features corresponding to the second fineness. By adopting a step-by-step improvement method to refine the model features, the accuracy of the model features corresponding to the obtained second fineness is ensured.

[0179] It should be noted that the above step 604 is only described by taking the prediction of the model features corresponding to the next fineness of the nth third fineness as an example. In another embodiment, according to the above step 604, the model features corresponding to higher fineness can be continuously predicted until the model features corresponding to the second fineness are obtained.

[0180] For example, taking the first fineness as fineness 1, m as 2, and the second fineness as fineness 4 as an example, there are 2 third fineness levels between the first fineness and the second fineness; the 1st third fineness is fineness 2, and the 2nd third fineness is fineness 3; the model features corresponding to the first fineness are the model features corresponding to fineness 1; based on the model features corresponding to fineness 1, predict the model features corresponding to fineness 2; based on the model features corresponding to fineness 1 and the model features corresponding to fineness 2, predict the model features corresponding to fineness 3; based on the model features corresponding to fineness 1, the model features corresponding to fineness 2, and the model features corresponding to fineness 3, predict the model features corresponding to fineness 4.

[0181] In a possible implementation, step 604 includes: when the model features corresponding to the nth third fineness are obtained currently, fusing the model features corresponding to the first fineness to the nth third fineness to obtain a second fused feature; and predicting the model features corresponding to the next fineness of the nth third fineness based on the second fused feature.

[0182] In the embodiments of the present application, when the model features corresponding to the nth third fineness are obtained, the model features corresponding to the first fineness to the nth third fineness have been obtained at this time. First, the model features corresponding to the first fineness to the nth third fineness are fused, so as to predict the model features corresponding to the next fineness of the nth third fineness based on the fused features, so as to make full use of the currently obtained model features, enrich the basis for predicting the model features corresponding to the next fineness, and ensure the accuracy of the predicted model features.

[0183] It should be noted that the process of predicting the model features corresponding to the next fineness based on the second fused feature is the same as step 306 above, and will not be elaborated here.

[0184] It should be noted that the embodiments of the present application take obtaining the model features with higher fineness based on the currently obtained model features each time as an example for illustration. In another embodiment, it is not necessary to execute the above steps 603-604, but other methods are adopted to predict the model features corresponding to the second fineness based on the model features corresponding to the first fineness.

[0185] 605. The computer device decodes the model features corresponding to the second fineness to obtain a second three-dimensional model.

[0186] It should be noted that the above step 605 is the same as step 204 above, and will not be elaborated here.

[0187] In the solution provided by the embodiments of the present application, by identifying the fineness of the first three-dimensional model to be optimized, encoding the model features corresponding to the first fineness according to the first fineness of the first three-dimensional model, using the model features corresponding to the first fineness to predict the model features with higher fineness, and then decoding to obtain the second three-dimensional model with higher fineness. In this way, a method for optimizing the three-dimensional model is realized, which can improve the fineness of the three-dimensional model, no longer requires manual editing of the three-dimensional model, saves the time required to obtain the three-dimensional model with high fineness, and can improve the generation efficiency of the three-dimensional model while improving the fineness of the three-dimensional model; and this three-dimensional model optimization method can be applied to any fineness, and can optimize the three-dimensional model corresponding to any fineness into the three-dimensional model corresponding to a higher fineness, improving the applicability of optimizing the three-dimensional model.

[0188] It should be noted that the above-mentioned multiple optional embodiments can be combined in any way.

[0189] For example, taking the combination of the embodiments Figures 2 to 4 shown above as an example, a second 3D model can be obtained through a fineness recognition model, an encoding model, a fineness improvement model, and a decoding model. As Figure 7 shown, the method further includes the following steps 1 to 6.

[0190] Step 1: Obtain the perspective image of the first 3D model; through the fineness recognition model, encode the perspective image to obtain image features, and classify the image features to obtain the first fineness.

[0191] Among them, the perspective image includes a depth map, a normal map, and a coordinate map.

[0192] For example, the perspective image includes a depth map, a normal map, and a coordinate map, and the fineness recognition model is a geometric fineness discriminator. Then the process of recognizing the first fineness is as Figure 8 shown. By rendering the first 3D model, a depth map, a normal map, and a coordinate map are obtained, and the depth map, the normal map, and the coordinate map are input into the geometric fineness discriminator. The geometric fineness discriminator processes the depth map, the normal map, and the coordinate map and outputs the recognized first fineness.

[0193] Step 2: Sample from the contour of the first 3D model based on the density corresponding to the first fineness to obtain point cloud data.

[0194] Step 3: Encode the point cloud data through the encoding model to obtain model features corresponding to the first fineness.

[0195] Step 4: Encode the control information through the fineness improvement model to obtain control features; fuse the control features and the model features corresponding to the first fineness to obtain a first fusion feature; based on the first fusion feature, predict the model features corresponding to the second fineness.

[0196] It should be noted that in the embodiments of the present application, only the case where the second fineness is higher than the first fineness and the second fineness is the next fineness of the first fineness is taken as an example for illustration. In another embodiment, there are m third finenesses between the first fineness and the second fineness. Taking the case where there is 1 third fineness between the first fineness and the second fineness as an example, then compared with the above Figure 5When combined with the embodiments shown, the process of obtaining the model features corresponding to the second fineness includes: encoding the control information through a fineness improvement model to obtain control features, fusing the control features and the model features corresponding to the first fineness to obtain a first fusion feature; predicting the model features corresponding to the third fineness based on the first fusion feature; fusing the control features and the model features corresponding to the third fineness through a fineness improvement model to obtain a third fusion feature; predicting the model features corresponding to the second fineness based on the third fusion feature.

[0197] When combined with the above Figure 6 When combined with the embodiments shown, the process of obtaining the model features corresponding to the second fineness includes: encoding the control information through a fineness improvement model to obtain control features, fusing the control features and the model features corresponding to the first fineness to obtain a first fusion feature; predicting the model features corresponding to the third fineness based on the first fusion feature; fusing the control features, the model features corresponding to the first fineness and the model features corresponding to the third fineness through a fineness improvement model to obtain a fourth fusion feature; predicting the model features corresponding to the second fineness based on the fourth fusion feature.

[0198] Step 5: Decode the model features corresponding to the second fineness through a decoding model to obtain distance field information.

[0199] Step 6: Generate a second three-dimensional model in the target space based on the distance field information.

[0200] In the embodiment of the present application, the fineness recognition model is a geometric fineness discriminator, and both the encoding model and the decoding model are sub-models of the multi-precision variational autoencoder module. That is, the encoding model is the encoder in the multi-precision variational autoencoder module, and the decoding model is the decoder in the multi-precision variational autoencoder module. The fineness improvement model can be a multi-precision autoregressive model. Taking the first 3D model as an example of a 3D mesh model, this 3D mesh model is the geometric information of the virtual object. The multi-precision variational autoencoder module maps the geometric information of the virtual object into the latent space features of any one of multiple different fineness levels, that is, the model features corresponding to this fineness level. The multi-precision autoregressive model can predict the latent space features of the next fineness level based on the latent space features of the previous fineness level. The geometric fineness discriminator can automatically determine the corresponding fineness level based on the perspective image of the 3D model. Then, through the multi-precision autoregressive model, the latent space features of this model are gradually optimized and improved from the corresponding fineness level to a higher fineness level. Finally, the decoding module of the variational autoencoder decodes the latent space features of the corresponding high fineness level into a 3D model of the corresponding high fineness level. Based on the above geometric fineness discriminator, multi-precision variational autoencoder, and multi-precision autoregressive model, the refinement of the 3D model can be realized to refine the input rough 3D model and generate a more refined 3D model.

[0201] According to the above-described embodiment, a 3D model in any format can be input. The fineness recognition model first recognizes the corresponding fineness based on the rendered perspective image. Then, the encoding model combines the recognized fineness and encodes the 3D model into the latent space features corresponding to this fineness level. Subsequently, the fineness improvement model will, based on the latent space features corresponding to this fineness level, improve this latent space feature to a higher fineness level to predict the latent space features corresponding to the next fineness level. Finally, the decoding model decodes the latent space features corresponding to the higher fineness level into the corresponding 3D model.

[0202] On the basis of the above-described embodiment, before obtaining the second 3D model through the fineness recognition model, encoding model, fineness improvement model, and decoding model, it is also necessary to train the fineness recognition model, encoding model, fineness improvement model, and decoding model.

[0203] In a possible implementation, the encoding model and the decoding model are sub-models of a multi-precision variational autoencoder module, so that the encoding model and the decoding model can be trained simultaneously. The training process includes: obtaining sample point cloud data corresponding to any level of detail, encoding the sample point cloud data through the encoding model to obtain model features corresponding to the sample point cloud data, decoding the model features through the decoding model to obtain predicted distance field information, and determining predicted point cloud data in the target space based on the predicted distance field information; training the encoding model and the decoding model based on the sample point cloud data and the predicted point cloud data.

[0204] Among them, the predicted point cloud data is composed of points with a distance of 0 indicated by the predicted distance field information in the target space.

[0205] For example, the process of obtaining the predicted point cloud data through the encoding model and the decoding model is as Figure 9 shown.

[0206] In the embodiments of the present application, the sample point cloud data can be point cloud data corresponding to any level of detail. The number of points in the point cloud data corresponding to different levels of detail is different. The higher the level of detail, the more points there are in the point cloud data corresponding to the level of detail. For example, for the same object, the point cloud data corresponding to 4 levels of detail, the 4 levels of detail are level of detail 1, level of detail 2, level of detail 3, and level of detail 4 respectively. Level of detail 1 is the lowest, and level of detail 4 is the highest. As Figure 10 shown, the number of points included in the point cloud data corresponding to these 4 levels of detail are 256, 512, 768, and 1024 respectively.

[0207] For example, sampling a 3D model of any object with different densities to obtain sample point cloud data corresponding to multiple levels of detail. The point cloud data obtained based on low-density sampling can only describe the general outline of the object, and the point cloud data obtained based on high-density sampling can describe more details of the object. Different densities correspond to different levels of detail. According to the density used for sampling the point cloud data, the sampled sample point cloud data is divided into sample point cloud data corresponding to multiple levels of detail.

[0208] In the embodiments of the present application, the sample point cloud data is processed through the encoding model and the decoding model to obtain the predicted point cloud data. The difference between the predicted point cloud data and the sample point cloud data can reflect the accuracy of the encoding model and the decoding model. Therefore, training the encoding model and the decoding model based on the sample point cloud data and the predicted point cloud data can improve the accuracy of the encoding model and the decoding model.

[0209] Optionally, the process of training the model includes: determining a first loss value based on the sample point cloud data and the predicted point cloud data, and training the encoding model and the decoding model based on the first loss value.

[0210] Among them, the first loss value is used to represent the difference between the sample point cloud data and the predicted point cloud data. The first loss value can be determined by taking any loss function. For example, a reconstruction loss function, a KL (Kullback-Leibler divergence, relative entropy) loss function, etc. can be taken.

[0211] Optionally, the first loss value satisfies the following relationship:

[0212] L vae = ||Φ(x|z) - S|| 1 + β[D KL (q Φ (z|π)||p(z))]

[0213] Among them, L vae is used to represent the first loss value; ||Φ(x|z) - S|| 1 is equivalent to the loss value obtained based on the reconstruction loss function, [D KL (q Φ (z|π)||p(z))] is equivalent to the loss value obtained based on the KL loss function; β is equivalent to a weight, which is used to measure the importance between the reconstruction loss function and the KL loss function. Φ(x|z) is used to represent the predicted point cloud data, x is used to represent the data input to the encoding model, z is used to represent the model features output by the encoding model; S is used to represent the sample point cloud data; q Φ (z|π) is used to represent the posterior probability distribution of the model feature z obtained by the encoding model based on the input data x and other possible parameters π, which can reflect the probability distribution of the model feature z given the data x and the model parameter Φ; p(z) is used to represent the prior probability distribution of the model feature z, which is a target normal distribution, such as a Gaussian distribution; [D KL (q Φ (z|π)||p(z))] can constrain the posterior probability distribution q Φ (z|π) to be as close as possible to the prior probability distribution p(z), thereby constraining the distribution of the model feature z and avoiding the situation that the encoding model and the decoding model are overfitted or the generated data is not real.

[0214] In a possible implementation manner, the training process of the fineness improvement model includes: obtaining the sample point cloud data corresponding to the fourth fineness and the sample point cloud data corresponding to the fifth fineness, and based on the sample point cloud data corresponding to the fourth fineness through the fineness improvement model, predicting the predicted point cloud data corresponding to the fifth fineness; training the encoding model and the decoding model based on the sample point cloud data corresponding to the fifth fineness and the predicted point cloud data corresponding to the fifth fineness.

[0215] Among them, the fourth fineness is any fineness, the fifth fineness is higher than the fourth fineness, and the fifth fineness is the next fineness of the fourth fineness.

[0216] In the embodiment of the present application, the sample point cloud data corresponding to the fourth fineness and the sample point cloud data corresponding to the fifth fineness are the point cloud data of the three-dimensional model of the same object at different finenesses. The fineness improvement model can predict the point cloud data corresponding to the next fineness based on the sample point cloud data corresponding to the fourth fineness, that is, obtain the predicted point cloud data corresponding to the fifth fineness. The difference between the sample point cloud data corresponding to the fifth fineness and the predicted point cloud data corresponding to the fifth fineness can reflect the accuracy of the fineness improvement model. Therefore, based on the sample point cloud data corresponding to the fifth fineness and the predicted point cloud data corresponding to the fifth fineness, the fineness improvement model is trained to improve the accuracy of the fineness improvement model.

[0217] Optionally, based on the sample point cloud data corresponding to the fifth fineness and the predicted point cloud data corresponding to the fifth fineness, a second loss value is determined, and the encoding model and the decoding model are trained based on the second loss value.

[0218] Among them, the second loss value is used to represent the difference between the sample point cloud data corresponding to the fifth fineness and the predicted point cloud data corresponding to the fifth fineness. The second loss value can be calculated by any loss function. For example, it can be calculated by using the cross-entropy loss function.

[0219] In a possible implementation manner, the training process of the fineness recognition model includes: obtaining the perspective image of the three-dimensional model corresponding to the sample fineness, encoding the perspective image through the fineness recognition model to obtain image features, classifying the image features to obtain the predicted fineness; and training the fineness recognition model based on the predicted fineness and the sample fineness.

[0220] In the embodiment of the present application, the difference between the predicted fineness and the sample fineness can reflect the accuracy of the fineness recognition model. Training the fineness recognition model based on the predicted fineness and the sample fineness can improve the accuracy of the fineness recognition model.

[0221] Optionally, based on the predicted fineness and the sample fineness, a third loss value is determined, and the fineness recognition model is trained based on the third loss value.

[0222] Among them, the third loss value is used to represent the difference between the predicted fineness and the sample fineness, and the third loss value can be calculated by any loss function. For example, the third loss value is calculated by using the cross-entropy loss function.

[0223] It should be noted that the above description is based on the 3D model corresponding to any sample fineness as an example. In another embodiment, 3D models corresponding to multiple sample finenesses can be obtained in the above manner. Then, for the perspective images of the 3D models corresponding to each sample fineness, the fineness recognition model can be iteratively trained multiple times to improve the accuracy of the fineness recognition model. At the same time, it can also make the trained fineness recognition model applicable to 3D models corresponding to multiple finenesses and be able to recognize the fineness of any 3D model.

[0224] The embodiment of the present application provides an artificial intelligence solution for automatically converting a rough 3D model into a fine 3D model. The above fineness recognition model, encoding model, fineness improvement model, and decoding model can constitute an artificial intelligence system. Users can input any 3D model, and the format of the 3D model supports 3D model formats such as OBJ (a text file format) and GLB (a binary file format). The artificial intelligence system will automatically generate a 3D model with higher fineness in the manner shown in the above embodiment. The 3D model with higher fineness has more details on the premise of keeping the object size and semantics unchanged. Moreover, the 3D model generated by the artificial intelligence system can be directly imported into a game development engine for game production. For example, the game development engine is Unreal (Unreal Engine), Maya (Maya Engine), Unity (a game development engine), etc. Refining the 3D model based on the solution provided by the embodiment of the present application can greatly reduce the development cost of game production and shorten the development cycle.

[0225] In the embodiment of the present application, using an autoregressive model to refine the model features of the 3D model can improve the fineness and accuracy of 3D model refinement. The solution provided by the embodiment of the present application first identifies the fineness of the input 3D model according to the different fineness levels divided, and then gradually improves the fineness of the 3D model from this fineness level. The solution provided by the embodiment of the present application supports conditional input. The input condition is equivalent to control information to guide the refinement of the 3D model, which can quickly obtain a 3D model that meets the user's requirements, saving the time required to obtain a 3D model that meets the user's requirements and ensuring the convenience and efficiency of generating the 3D model. At the same time, the same autoregressive model can be used to refine the model features of models with different fineness levels, which has strong generalization.

[0226] The solution provided by the embodiments of the present application can be applied to the game production scenario, providing a complete set of game asset geometry refinement tools for game production personnel. The production personnel can input a rough 3D model and, according to the solution provided by the embodiments of the present application, can generate a refined 3D model as a game asset. The generated 3D model can be directly imported into the game engine, thus greatly reducing the game production cost, accelerating the game production speed, shortening the game production time, and making the game production more convenient.

[0227] Based on the above-described embodiments, the 3D model provided by the embodiments of the present application is as Figure 11 shown. Figure 11 Among them, the 3D model on the left is the first 3D model. Using the image as control information, the first 3D model is refined, and the obtained second 3D model is as Figure 11 shown by the 3D model on the right. The second 3D model presents more details.

[0228] Figure 12 is a schematic structural diagram of a 3D model acquisition device provided by the embodiments of the present application. As Figure 12 shown, the device includes:

[0229] An identification module 1201, configured to identify the fineness of the first 3D model to obtain a first fineness, where the first fineness is used to represent the fineness degree of the first 3D model;

[0230] An encoding module 1202, configured to encode the first 3D model based on the first fineness to obtain a model feature corresponding to the first fineness;

[0231] A prediction module 1203, configured to predict a model feature corresponding to a second fineness based on the model feature corresponding to the first fineness, where the second fineness is higher than the first fineness;

[0232] A decoding module 1204, configured to decode the model feature corresponding to the second fineness to obtain a second 3D model.

[0233] In a possible implementation manner, the prediction module 1203 is configured to process the model feature corresponding to the first fineness through a fineness improvement model to obtain a model feature corresponding to the second fineness. The fineness improvement model is used to predict the model feature of the next fineness based on the input model feature, and the second fineness is the next fineness of the first fineness.

[0234] In another possible implementation manner, the prediction module 1203 is configured to predict the model feature corresponding to the second fineness based on the model feature corresponding to the first fineness and control information, where the control information is used to describe the shape or color of the 3D model to be generated.

[0235] In another possible implementation, the prediction module 1203 is configured to encode the control information to obtain control features; fuse the control features and the model features corresponding to the first fineness to obtain first fused features; and predict the model features corresponding to the second fineness based on the first fused features.

[0236] In another possible implementation, there is an interval of m fineness levels between the second fineness level and the first fineness level, where m is a positive integer greater than zero; the prediction module 1203 is configured to start from the first fineness level and predict the model features corresponding to the next fineness level based on the model features corresponding to the previous fineness level.

[0237] In another possible implementation, there is an interval of m third fineness levels between the second fineness level and the first fineness level, where m is a positive integer greater than zero; the prediction module 1203 is configured to predict the model features corresponding to the first third fineness level based on the model features corresponding to the first fineness level; and in the case where the model features corresponding to the nth third fineness level are currently obtained, predict the model features corresponding to the next fineness level after the nth third fineness level based on the model features corresponding to the first fineness level to the nth third fineness level, where n is a positive integer not greater than m.

[0238] In another possible implementation, the prediction module 1203 is configured to, in the case where the model features corresponding to the nth third fineness level are currently obtained, fuse the model features corresponding to the first fineness level to the nth third fineness level to obtain second fused features; and predict the model features corresponding to the next fineness level after the nth third fineness level based on the second fused features.

[0239] In another possible implementation, the encoding module 1202 is configured to sample from the contour of the first three-dimensional model based on the density corresponding to the first fineness level to obtain point cloud data; and encode the point cloud data to obtain the model features corresponding to the first fineness level.

[0240] In another possible implementation, the recognition module 1201 is configured to obtain the perspective image of the first three-dimensional model; encode the perspective image to obtain image features; and classify the image features to obtain the first fineness level.

[0241] In another possible implementation, the perspective image includes images of the first three-dimensional model from multiple perspectives; or,

[0242] The perspective image includes at least one of the depth map of the first three-dimensional model, the normal map of the first three-dimensional model, or the coordinate map of the first three-dimensional model.

[0243] In another possible implementation, the decoding module 1204 is configured to decode the model features corresponding to the second fineness to obtain distance field information, where the distance field information indicates the distance from a point in the target space to the contour of the three-dimensional model; and generate a second three-dimensional model in the target space based on the distance field information.

[0244] It should be noted that: for the three-dimensional model acquisition device provided in the above embodiments, only the division of the above functional modules is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the three-dimensional model acquisition device provided in the above embodiments and the embodiments of the three-dimensional model acquisition method belong to the same concept, and the specific implementation process can be seen in the method embodiments, which will not be elaborated here.

[0245] An embodiment of the present application further provides a computer device, which includes a processor and a memory. At least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the three-dimensional model acquisition method in the above embodiments.

[0246] Optionally, the computer device is provided as a terminal. Figure 13 The structural block diagram of a terminal 1300 provided by an exemplary embodiment of the present application is shown. The terminal 1300 includes a processor 1301 and a memory 1302.

[0247] The processor 1301 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. The processor 1301 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor 1301 may also include a main processor and a coprocessor. The main processor is a processor used to process data in the wake state, also known as the CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 1301 may be integrated with a GPU (Graphics Processing Unit), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1301 may further include an AI (Artificial Intelligence) processor, and the AI processor is used to process computational operations related to machine learning.

[0248] The memory 1302 may include one or more computer-readable storage media, and the computer-readable storage media may be non-transitory. The memory 1302 may further include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash storage devices. In some embodiments, the non-transitory computer-readable storage media in the memory 1302 is used to store at least one computer program, and the at least one computer program is used to be executed by the processor 1301 to implement the three-dimensional model acquisition method provided in the method embodiments of the present application.

[0249] In some embodiments, the terminal 1300 may further optionally include: a peripheral device interface 1303 and at least one peripheral device. The processor 1301, the memory 1302, and the peripheral device interface 1303 may be connected through a bus or signal lines. Each peripheral device may be connected to the peripheral device interface 1303 through a bus, signal lines, or a circuit board. Specifically, the peripheral devices include at least one of a radio frequency circuit 1304, a display screen 1305, a camera assembly 1306, an audio circuit 1307, and a power supply 1308.

[0250] The peripheral device interface 1303 can be used to connect at least one I / O (Input / Output) related peripheral device to the processor 1301 and the memory 1302. In some embodiments, the processor 1301, the memory 1302, and the peripheral device interface 1303 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1301, the memory 1302, and the peripheral device interface 1303 can be implemented on a separate chip or circuit board, and this embodiment does not limit this.

[0251] The radio frequency circuit 1304 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 1304 communicates with a communication network and other communication devices through electromagnetic signals. The radio frequency circuit 1304 converts an electrical signal into an electromagnetic signal for transmission, or converts the received electromagnetic signal into an electrical signal. Optionally, the radio frequency circuit 1304 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and so on. The radio frequency circuit 1304 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes but is not limited to: the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G, and 5G), a wireless local area network, and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 1304 may further include a circuit related to NFC (Near Field Communication), and this application does not limit this.

[0252] The display screen 1305 is used to display the UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 1305 is a touch display screen, the display screen 1305 also has the ability to collect touch signals on or above the surface of the display screen 1305. The touch signals can be input to the processor 1301 as control signals for processing. At this time, the display screen 1305 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, there may be one display screen 1305, which is disposed on the front panel of the terminal 1300; in other embodiments, there may be at least two display screens 1305, which are respectively disposed on different surfaces of the terminal 1300 or are in a foldable design; in other embodiments, the display screen 1305 may be a flexible display screen, which is disposed on the curved surface or the folding surface of the terminal 1300. Even further, the display screen 1305 can also be set to an irregular non-rectangular shape, that is, a special-shaped screen. The display screen 1305 can be prepared using materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0253] The camera module 1306 is used to collect images or videos. Optionally, the camera module 1306 includes a front camera and a rear camera. The front camera is disposed on the front panel of the terminal, and the rear camera is disposed on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth camera, a wide-angle camera, and a telephoto camera, to implement functions such as background blurring by fusing the main camera and the depth camera, panoramic shooting by fusing the main camera and the wide-angle camera, and VR (Virtual Reality) shooting functions or other fused shooting functions. In some embodiments, the camera module 1306 may further include a flash. The flash can be a single-color temperature flash or a two-color temperature flash. A two-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation under different color temperatures.

[0254] The audio circuit 1307 may include a microphone and a speaker. The microphone is used to collect sound waves of the user and the environment, and convert the sound waves into electrical signals for input to the processor 1301 for processing, or input to the radio frequency circuit 1304 to implement voice communication. For the purpose of stereo collection or noise reduction, there may be multiple microphones, which are respectively arranged at different parts of the terminal 1300. The microphone may also be an array microphone or an omnidirectional collection microphone. The speaker is used to convert the electrical signal from the processor 1301 or the radio frequency circuit 1304 into sound waves. The speaker may be a traditional thin film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for uses such as ranging. In some embodiments, the audio circuit 1307 may further include a headphone jack.

[0255] The power supply 1308 is used to supply power to each component in the terminal 1300. The power supply 1308 may be alternating current, direct current, a disposable battery or a rechargeable battery. When the power supply 1308 includes a rechargeable battery, the rechargeable battery may be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery charged through a wired line, and a wireless rechargeable battery is a battery charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0256] Those skilled in the art can understand that Figure 13 the structure shown in does not constitute a limitation on the terminal 1300, and may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component layout.

[0257] Optionally, the computer device is provided as a server. Figure 14 is a schematic structural diagram of a server provided by an embodiment of the present application. The server 1400 may vary greatly due to different configurations or performances, and may include one or more processors (Central Processing Units, CPUs) 1401 and one or more memories 1402. Among them, at least one computer program is stored in the memory 1402, and at least one computer program is loaded and executed by the processor 1401 to implement the methods provided by the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard and an input / output interface for input and output. The server may further include other components for implementing the functions of the device, which will not be elaborated here.

[0258] The embodiment of the present application also provides a computer-readable storage medium, in which at least one computer program is stored, and the at least one computer program is loaded and executed by a processor to implement the operations performed by the three-dimensional model acquisition method in the above embodiment.

[0259] The embodiment of the present application also provides a computer program product, including a computer program, and the operations performed by the three-dimensional model acquisition method in the above embodiment are implemented when the computer program is executed by a processor.

[0260] Those of ordinary skill in the art can understand that all or part of the steps in the above embodiment can be completed by hardware, or can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium, and the above-mentioned storage medium can be a read-only memory, a disk, an optical disc, etc.

[0261] The above are only optional embodiments of the embodiments of the present application, and are not intended to limit the embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the embodiments of the present application shall be included in the protection scope of the present application.

Claims

1. A three-dimensional model acquisition method, characterized in that: The method comprises: Identifying the fineness of the first three-dimensional model to obtain a first fineness, where the first fineness is used to represent the degree of fineness of the first three-dimensional model; Based on the first fineness, encoding the first three-dimensional model to obtain model features corresponding to the first fineness; predicting, based on the model features corresponding to the first fineness, a model feature corresponding to a second fineness, the second fineness being higher than the first fineness; The model features corresponding to the second fineness are decoded to obtain a second three-dimensional model.

2. The method according to claim 1, characterized in that The predicting, based on the model features corresponding to the first fineness, the model features corresponding to the second fineness, includes: The model features corresponding to the first fineness are processed by the fineness improvement model to obtain the model features corresponding to the second fineness. The fineness improvement model is used to predict the model features of the next fineness based on the input model features. The second fineness is the next fineness of the first fineness.

3. The method according to claim 1, characterized in that The predicting, based on the model features corresponding to the first fineness, the model features corresponding to the second fineness, includes: Based on the model features and control information corresponding to the first fineness, the model features corresponding to the second fineness are predicted, and the control information is used to describe the shape or color of the three-dimensional model to be generated.

4. The method according to claim 3, characterized in that The predicting, based on the model features and control information corresponding to the first fineness, the model features corresponding to the second fineness includes: Encoding the control information to obtain a control feature; Fusing the control feature and the model feature corresponding to the first precision to obtain a first fused feature; Based on the first fusion features, predict model features corresponding to the second level of refinement.

5. The method according to claim 1, characterized in that The second fineness is separated from the first fineness by m finenesses, where m is greater than a positive integer; and predicting the model features corresponding to the second fineness based on the model features corresponding to the first fineness includes: Starting from the first fineness, the model features corresponding to the next fineness are predicted based on the model features corresponding to the previous fineness.

6. The method according to claim 1, characterized in that The second fineness is separated from the first fineness by m third finenesses, where m is greater than a positive integer; and predicting the model features corresponding to the second fineness based on the model features corresponding to the first fineness includes: Predicting a model feature corresponding to a first third level of fineness based on the model feature corresponding to the first level of fineness; When the model features corresponding to the nth third fineness are currently obtained, the model features corresponding to the next fineness of the nth third fineness are predicted based on the model features corresponding to the first fineness to the nth third fineness, where n is a positive integer not greater than m.

7. The method according to claim 6, characterized in that The method of predicting a model feature corresponding to a next fineness of the nth third fineness based on the model features corresponding to the first fineness to the nth third fineness when the model feature corresponding to the nth third fineness is currently obtained includes: When the model feature corresponding to the n-th third fineness is currently obtained, the model features corresponding to the first fineness to the n-th third fineness are fused to obtain a second fused feature; Based on the second fused features, predict the model features corresponding to the next level of refinement of the nth third level of refinement.

8. The method according to any one of claims 1 to 7, characterized in that: The encoding of the first three-dimensional model based on the first fineness to obtain a model feature corresponding to the first fineness includes: Based on the density corresponding to the first fineness, sampling is performed from the contour of the first three-dimensional model to obtain point cloud data; The point cloud data is encoded to obtain model features corresponding to the first fineness.

9. The method according to any one of claims 1 to 7, characterized in that: The step of identifying the fineness of the first three-dimensional model to obtain a first fineness includes: Acquire a perspective image of the first three-dimensional model; Encoding the perspective image to obtain image features; The image features are classified to obtain the first fineness.

10. The method according to claim 9, characterized in that The viewing angle images include images of the first three-dimensional model at multiple viewing angles; or, The perspective image includes at least one of a depth map of the first three-dimensional model, a normal map of the first three-dimensional model, or a coordinate map of the first three-dimensional model.

11. The method according to any one of claims 1 to 7, characterized in that: The decoding of the model features corresponding to the second fineness to obtain a second three-dimensional model includes: Decoding the model features corresponding to the second fineness to obtain distance field information, where the distance field information indicates the distance from a point in the target space to a contour of the three-dimensional model; The second three-dimensional model is generated in the target space based on the distance field information.

12. A three-dimensional model acquisition device, characterized in that: The device comprises: an identification module, used to identify the fineness of the first three-dimensional model to obtain a first fineness, where the first fineness is used to indicate the degree of fineness of the first three-dimensional model; an encoding module, configured to encode the first three-dimensional model based on the first fineness to obtain model features corresponding to the first fineness; a prediction module, configured to predict, based on the model features corresponding to the first fineness, model features corresponding to a second fineness, the second fineness being higher than the first fineness; A decoding module is used to decode the model features corresponding to the second fineness to obtain a second three-dimensional model.

13. A computer device, characterized in that: The computer device includes a processor and a memory, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the three-dimensional model acquisition method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by a processor to implement the operations performed by the three-dimensional model acquisition method according to any one of claims 1 to 11.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the operations performed by the three-dimensional model acquisition method according to any one of claims 1 to 11 are implemented.

Images