Material map generation method, electronic device, storage medium and program product

Abstract

The application provides a material map generation method, an electronic device, a storage medium and a program product. The method comprises the following steps: in response to a material map generation operation triggered by a terminal device for a target three-dimensional model, rendering the target three-dimensional model to obtain a rendering atlas of the target three-dimensional model; using a plurality of preset lighting conditions configured in a large model, respectively performing re-lighting processing on the rendering atlas to obtain a plurality of re-lighting effect pictures; performing material property analysis on the target three-dimensional model according to the plurality of re-lighting effect pictures to obtain a target material map of the target three-dimensional model, and sending the target material map to the terminal device. The method can accurately generate a material map corresponding to a three-dimensional model.

Description

Technical Field

[0001] This application relates to physically based rendering technology, and more particularly to a method for generating material textures, electronic devices, storage media, and program products. Background Technology

[0002] In the PBR (Physically Based Rendering) rendering process, the following three types of material maps are typically generated to represent the physical material properties of a 3D model: 1. Reflectance map (Albedo Texture or Base Color map), used to describe the basic color and reflective properties of the material, excluding lighting and reflection information; 2. Normal map, used to simulate microscopic bump details, by changing the direction of the normals to produce subtle changes in light reflection; 3. Metallic and Roughness map (a hybrid map of metallic and roughness), used to describe the metallic smoothness of the material, reflecting the physical properties of the object under lighting.

[0003] In related technologies, the generation process of the above three material textures generally involves generating the corresponding material texture from a single image. Specifically: 3D modelers create a medium-precision model corresponding to the original artwork image. The medium-precision model is then placed into sculpting software, which adds details to the model to generate a high-precision model. The high-precision model is then topologically reconstructed to generate a low-precision model that can be used in the rendering engine. Finally, the three material textures are drawn based on the texture information of the low-precision model. The Normal Texture is baked onto the low-precision model from the high-precision model to form the texture. The Albedo Texture, Metallic, and Roughness blending maps are drawn step by step by artists based on the texture information of the low-precision model.

[0004] However, the aforementioned material mapping generation methods have low accuracy in generating material maps corresponding to 3D models. Summary of the Invention

[0005] This application provides a material texture generation method, an electronic device, a computer-readable storage medium, and a computer program product, which can accurately generate material textures corresponding to three-dimensional models.

[0006] This application provides a material texture generation method, applied to a server, the method including: In response to the material texture generation operation triggered by the terminal device for the target 3D model, the target 3D model is rendered to obtain the rendering atlas of the target 3D model; Using the various lighting conditions configured in the preset large model, the rendered atlas is relit to obtain multiple relit effect images; Based on multiple re-lighting effect images, the material properties of the target 3D model are analyzed to obtain the target material texture map of the target 3D model, and then the target material texture map is sent to the terminal device.

[0007] In the above scheme, the rendering atlas includes first renderings of the target 3D model from multiple viewpoints; using various lighting conditions configured in the preset large model, the rendering atlas is relit to obtain multiple relit effect images, including: For each viewpoint, the first rendered image is relit using various lighting conditions configured in the preset large model, resulting in multiple relit effect images. Based on multiple re-lighting effect images, material property analysis was performed on the target 3D model to obtain the target material texture map of the target 3D model, including: Based on multiple relighting effect images corresponding to each viewpoint, a pre-trained material analysis model is used to analyze the material properties of the target 3D model to obtain the local material texture map corresponding to each viewpoint. The pre-trained material analysis model includes multiple branch decoders. The branch decoders are used to output the target material properties corresponding to multiple relighting effect images when multiple relighting effect images are input. The material analysis model is used to output different target material properties through each branch decoder and generate a local material texture map based on multiple target material properties. Based on the local material textures corresponding to all viewpoints, determine the target material texture of the target 3D model.

[0008] In the above scheme, the target 3D model is rendered to obtain a rendering atlas of the target 3D model, including: Determine the texture map of the target 3D model; Based on texture mapping, a preset rendering engine is invoked to render the target 3D model from multiple perspectives, resulting in a rendering atlas of the target 3D model.

[0009] In the above scheme, determining the texture map of the target 3D model includes: Determine the first texture map of the target 3D model under the preset first-view perspective; Rotate the preset rendering camera around the central axis of the target 3D model multiple times by a preset angle until the number of rotations reaches a preset threshold. After each rotation operation is completed, based on the acquired historical texture map set, the second texture map of the target 3D model in the second view is predicted using the rotated rendering camera and the preset large model. The historical texture map set includes the first texture map and the second texture map of the target 3D model in multiple historical rotation processes. The texture map of the target 3D model is determined based on the first texture map and multiple second texture maps.

[0010] In the above scheme, determining the first texture map of the target 3D model under a preset first viewpoint includes: The target 3D model is rendered under a preset first viewpoint to obtain the first geometric feature map of the target 3D model. The first geometric feature map represents the edge features, normal features and depth features of the target 3D model under the preset first viewpoint. Based on the first geometric feature map, the second rendered image of the target 3D model under the preset first view is predicted using a preset large model. By mapping the color information of each pixel in the second rendered image to the surface of the target 3D model, the first texture map of the target 3D model is obtained.

[0011] In the above scheme, based on the acquired historical texture map set, the second texture map of the target 3D model in the second view is predicted using a rotated rendering camera and a preset large model, including: For the Nth rotation process, obtain a set of historical texture maps that have been determined in multiple historical rotation processes; N is a positive integer; The target 3D model is rendered using a rotated rendering camera from a second perspective to obtain the second geometric feature map of the target 3D model. The second geometric feature map represents the edge features, normal features and depth features of the target 3D model from the second perspective. Based on the historical texture map set, the local texture map and mask map of the target 3D model in the second view are determined. The mask map is used to locate local areas of the target 3D model that do not have texture in the second view. Based on local texture mapping and second geometric feature mapping, a third rendered image corresponding to the mask image is predicted using a pre-set large model. By mapping the color information of each pixel in the third rendering image to the surface of the target 3D model, a second texture map of the target 3D model is obtained.

[0012] This application embodiment also provides another method for generating material texture maps, applied to terminal devices, the method including: In response to the model determination operation for the material map generation interface, the target 3D model of the object to be processed is displayed in the material map generation interface. In response to the material texture generation operation for the target 3D model, a material texture generation command is sent to the server, which carries the target 3D model. The server renders the target 3D model based on the material texture generation command to obtain a rendering atlas of the target 3D model. Using multiple lighting conditions configured in the preset large model, the rendering atlas is relit to obtain multiple relit effect images. Based on the multiple relit effect images, the material properties of the target 3D model are analyzed to obtain the target material texture of the target 3D model, and the target material texture is fed back to the terminal device. The target material map is displayed in the material map generation interface.

[0013] This application provides an electronic device, which includes: Memory is used to store executable instructions or computer programs. When a processor executes computer-executable instructions or computer programs stored in memory, it implements the material texture generation method provided in the embodiments of this application.

[0014] This application provides a computer-readable storage medium storing a computer program or computer-executable instructions, which, when executed by a processor, implements the material texture generation method provided in this application.

[0015] This application provides a computer program product, including a computer program or computer executable instructions. When the computer program or computer executable instructions are executed by a processor, the above-mentioned material texture generation method provided in this application is implemented.

[0016] In this embodiment, the server responds to a material texture generation operation triggered by a terminal device for a target 3D model, renders the target 3D model, and obtains a rendering atlas of the target 3D model. Using multiple lighting conditions configured in a pre-defined large model, the rendering atlas is relit to obtain multiple relit effect images. Based on these multiple relit effect images, material property analysis is performed on the target 3D model to obtain the target material texture, which is then sent to the terminal device. Throughout this process, multi-dimensional lighting conditions are introduced into the rendering atlas of the target 3D model through large model relighting. This ensures that subsequent material property analysis is based on relit effect images under different lighting conditions, eliminating the influence of lighting conditions on material properties and improving the accuracy of material texture generation for the target 3D model. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of the material texture generation system provided in the embodiments of this application; Figure 2 This is a schematic flowchart of the server-side material texture generation method provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the material analysis model provided in the embodiments of this application; Figure 4 This is a schematic diagram of the rotation of the rendering camera around the central axis of the target 3D model provided in the embodiments of this application; Figure 5 This is a schematic diagram of the second rendering of a target 3D model provided in an embodiment of this application; Figure 6 This is a schematic diagram of the first texture map of a target 3D model provided in an embodiment of this application; Figure 7 This is a schematic diagram of the first geometric feature map of a target 3D model provided in an embodiment of this application; Figure 8 This is a schematic flowchart illustrating the method for generating material textures using a terminal device, as provided in an embodiment of this application. Figure 9 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] During PBR rendering, the following three base material maps are typically generated to represent the physical material properties of the 3D model (sometimes additional maps are added to enhance details): 1. Albedo map (or Base Color map): used to describe the basic color and reflective properties of the material, excluding lighting and reflection information; 2. Normal map: used to simulate microscopic bump details, changing the direction of the normals to produce subtle changes in light reflection; 3. Metallic and Roughness blend map: used to describe the metallic smoothness of the material, reflecting the physical properties of the object under lighting.

[0020] In related technologies, 3D modelers primarily use various commercially available tools to generate PBR textures from a photograph. Specifically, 3D modelers work through a series of steps to create a medium-precision model corresponding to the concept art provided by the artists. This model is then sculpted in sculpting software to add detail, generating a very high-precision model. Next, the high-precision model is topologically analyzed to generate a low-precision model that can be used in the rendering engine. The three types of textures mentioned above are then drawn by unfolding the UV (UV Coordinates) of the low-precision model. The Normal texture is baked onto the low-precision model from the high-precision model, while the Albedo texture, Metallic, and Roughness blending maps are drawn step-by-step by the artists. The UV values ​​are texture coordinates, often called UV coordinates or texture coordinates, a two-dimensional coordinate system used to locate pixels on the texture image.

[0021] The problems with the above solutions are mainly reflected in the following aspects: 1. Insufficient information: A concept art piece usually only provides surface color information (albedo or diffuse texture), lacking the multi-dimensional information required for key PBR parameters such as reflectivity, roughness, and metallicity. This results in potentially inaccurate or incomplete PBR textures. 2. Lighting and environment dependence: Lighting conditions and environmental reflections in a photograph affect color and reflection effects, making it difficult to distinguish between the material's inherent properties and the influence of external lighting, thus leading to texture deviations. 3. Insufficient reflection and detail reproduction: The reflective characteristics and micro-details (such as scratches and bumps) of realistic materials are difficult to accurately extract from a single photograph. 4. Consistency of viewpoint and lighting: The angle and lighting conditions of a photograph may be unsatisfactory, causing inconsistent texture performance in different scenes, affecting realism.

[0022] Therefore, current material mapping generation methods have low accuracy in generating material maps corresponding to 3D models.

[0023] To address the shortcomings of the aforementioned related technologies, this application provides a material texture generation method that can improve the accuracy of material texture generation. To illustrate the technical solution of this invention, specific embodiments are described below.

[0024] This application provides a method for generating material textures, such as... Figure 1 As shown, the method is applied to a material texture generation system 101, which includes multiple terminal devices 102 and a server 103. The multiple terminal devices 102 and the server 103 communicate bidirectionally through a network. Each terminal device 102 is equipped with material texture generation software 104, and the server 103 communicates bidirectionally with a database 105.

[0025] Taking the interaction between one of the terminal devices 102 and the server 103 as an example, the user opens the material mapping generation software 104 on the terminal device 102. The terminal device 102 displays the material mapping generation interface to the user. The user selects the target 3D model from multiple candidate 3D models on the material mapping generation interface and triggers the material mapping generation control. The terminal device 102 responds to the triggering operation of the material mapping generation control and sends a material mapping generation request for the target 3D model to the server 103. The server 103 responds to the material mapping generation operation for the target 3D model triggered by the terminal device 102, renders the target 3D model, obtains the rendering atlas of the target 3D model, loads a preset large model from the database 105, and uses the various lighting conditions configured in the preset large model to relight the rendering atlas, obtaining multiple relighting effect images. Based on the multiple relighting effect images, the server analyzes the material properties of the target 3D model, obtains the target material map of the target 3D model, and sends the target material map to the terminal device 102. The terminal device 102 displays the target material texture to the user in the material texture generation interface.

[0026] Figure 2 A flowchart of a material texture generation method provided in this application embodiment, taking the method applied to server 103 as an example, the material texture generation method provided in this application includes the following steps: S201, responding to the material texture generation operation for the target 3D model triggered by the terminal device, renders the target 3D model to obtain the rendering atlas of the target 3D model.

[0027] The material textures used in this application are PBR textures, a standardized combination of material textures that follows realistic physical optics principles. They define materials through physical properties (such as roughness and refractive index), ensuring realistic and consistent material appearance under different lighting and environments. A standard PBR texture typically includes at least three basic material textures: 1. An Albedo map (or Base Color map) describing the material's basic color and reflectivity, excluding lighting and reflection information; 2. A Normal map simulating microscopic bumps and depressions, changing the normal direction to create subtle variations in light reflection; 3. A Metallic and Roughness blend map describing the material's metallic smoothness, reflecting the object's physical properties under lighting. Additionally, other material textures can be added to enhance the details of the 3D model.

[0028] Rendering is a core technology in computer graphics, used to convert 3D models or scene data into 2D visual images or videos that humans can understand.

[0029] A 3D model is a digital object form constructed in computer 3D space using points, lines, and surfaces, possessing three dimensions: length, width, and height. This application not only supports the generation of material textures for single, continuous, tiled models, but also supports the generation of material textures for any model with a complex structure, such as everyday objects and biological objects. Among them, a single, continuous, tiled model refers to a model that is infinitely tiled up, down, left, and right, without a model outline or its own concave and convex structure, such as a wooden floor. Everyday object models include, but are not limited to, table models and washing machine models, while biological models include, but are not limited to, animal and plant models.

[0030] In this application embodiment, when generating material maps from a single photograph in related technologies, a photograph typically only provides surface color information (albedo or diffuse map), lacking the multi-dimensional information required for key PBR parameters such as reflectivity, roughness, and metallicity. Therefore, this application improves the accuracy of material map generation by performing material property analysis on the target 3D model, analyzing material properties such as reflectivity, roughness, and metallicity in the photograph, and obtaining a material map based on all analyzed material properties. However, since the lighting conditions in a photograph are fixed, it is difficult to distinguish whether the color and reflection effect of the photograph are due to the model's own material properties or the influence of external lighting, resulting in deviations in the material maps generated when performing material property analysis on the photograph. Therefore, this application proposes to introduce multi-dimensional lighting conditions of the target 3D model through relighting, such as light source color, light source direction, changes in shadows on the surface of the target 3D model, and physical properties of the object surface, such as specular reflection and changes in specular spots. This allows subsequent material analysis to separate the influence of the model's own material properties from external lighting, improving the accuracy of material map generation.

[0031] To facilitate relighting of the target 3D model, the target 3D model must first be rendered before relighting. This involves converting the 3D target model into a 2D rendering atlas, and then relighting the rendering atlas.

[0032] Specifically, the target 3D model can be a 3D model pre-stored in a database or a 3D model input by the user in real time.

[0033] Taking the example that the target 3D model can be a 3D model pre-stored in the database, the user selects the target 3D model from multiple candidate 3D models on the material map generation interface of the terminal device and triggers the material map generation control. The terminal device responds to the trigger operation of the material map generation control and sends the model identifier of the target 3D model to the server. The server determines the target 3D model that matches the model identifier from the database.

[0034] Taking a target 3D model that is input by the user in real time as an example, the user draws the target 3D model on the material map generation interface, and after the target 3D model is drawn, the user triggers the material map generation control. The terminal device responds to the trigger operation of the material map generation control and sends the drawn target 3D model to the server. The server obtains the drawn target 3D model.

[0035] After the server determines or obtains the target 3D model, it renders the target 3D model to obtain a realistic image set of the target 3D model, which is the rendering atlas of the target 3D model. In the embodiments of this application, rendering the target 3D model refers to calling the rendering engine to render the target 3D model. The rendering engine is a software system specifically responsible for generating the final 2D image / video from 3D scene data (model, texture, lighting, materials, animation, camera, etc.) in real time or offline through mathematical algorithms and hardware acceleration. In other words, the rendering engine can convert 3D data into images that people can see.

[0036] The methods for rendering a target 3D model using a rendering engine include, but are not limited to: using the rendering camera within the rendering engine or using rasterization technology within the rendering engine to render the target 3D model and obtain a rendering atlas of the target 3D model. The rendering camera is a virtual object within the rendering engine used to define the viewing angle, imaging parameters, and visible range. It is responsible for projecting the 3D scene onto a 2D image plane through a series of transformations, determining the composition, perspective, and visual characteristics of the final rendering result. When using the rendering camera within the rendering engine to render the target 3D model, the rendering camera can be adjusted to orthogonal mode. In orthogonal mode, the light rays from the rendering camera are a set of parallel beams perpendicular to the imaging plane, which can eliminate perspective effects when rendering the target 3D model and improve the accuracy of the generated rendering atlas.

[0037] In the embodiments of this application, the rendering atlas of the target 3D model can be the overall rendering atlas of the target 3D model, that is, the surface of the target 3D model is unfolded, and the unfolded target 3D model is rendered as a whole to obtain the rendering atlas of the target 3D model.

[0038] In this embodiment of the application, the rendering atlas of the target 3D model can also be the first rendering images of the target 3D model from multiple perspectives, where a perspective refers to the viewpoint from which the target 3D model can be observed. That is, by rendering the target 3D model from multiple perspectives, the first rendering images of the target 3D model from multiple perspectives are obtained, and based on the multiple first rendering images, the rendering atlas of the target 3D model is obtained.

[0039] In this embodiment, the target 3D model itself has relatively little texture information, making it impossible to accurately analyze the material properties of the target 3D model during subsequent material property analysis. Therefore, it is also possible to obtain the texture map of the target 3D model and then render the target 3D model based on the texture map to obtain a rendering atlas of the target 3D model. At this time, the rendering atlas contains rich texture information. Therefore, after eliminating the influence of lighting on material properties by relighting, the result of material property analysis of the target 3D model based on multiple relighting effect maps containing rich texture information is more accurate.

[0040] Determining the texture map of the target 3D model includes, but is not limited to: tiling the surface of the target 3D model to determine the complete texture map; or determining the local texture maps of the target 3D model from different viewpoints and combining the local texture maps from all viewpoints into a complete texture map. Further, determining the local texture maps of the target 3D model from different viewpoints can be achieved by configuring multiple texture map generation tools corresponding to different viewpoints to generate local texture maps for each viewpoint of the target 3D model; or by configuring a single texture map generation tool, controlling the texture map generation tool or the target 3D model itself to rotate multiple times by a certain angle, and after each rotation, using the texture map generation tool to generate the local texture map of the target 3D model from the current viewpoint.

[0041] S202 utilizes various lighting conditions configured in the preset large model to relight the rendered image set, resulting in multiple relighting effect images.

[0042] Large models refer to artificial intelligence models with a huge number of parameters that are trained based on deep learning. Large models include, but are not limited to, AIGC models (Artificial Intelligence Generated Content) and large inference and decision-making models.

[0043] The large model includes a pre-trained relighting model with various lighting conditions configured, differing in lighting direction, intensity, and color. These conditions are obtained by training the relighting model using initial training data, which includes historical renders and relighting effects under various lighting conditions. In practical applications, to improve the relighting effect, the relighting model can be IC-LIGHTV2 or other superior models.

[0044] A re-lighting rendering is a new version of the rendering created by readjusting the lighting direction, intensity, and color of the lighting on the original rendering. The core is to change the lighting and shadows without altering the model or spatial structure.

[0045] In this embodiment of the application, the rendered atlas is passed into the large model. The large model uses the various lighting conditions configured by the relighting model to relight the rendered atlas, and obtains the relighting effect image under each lighting condition.

[0046] If the rendering atlas includes first renderings of the target 3D model from multiple perspectives, then for each first rendering, a relighting model is used to perform relighting processing under various lighting conditions to obtain the relighting effect of each first rendering under various lighting conditions.

[0047] If the rendering atlas includes a complete rendering of the target 3D model, then the relighting model is used to relight the complete rendering under various lighting conditions to obtain the relighting effect of the complete rendering under various lighting conditions.

[0048] In this embodiment, the lighting parameters in the relighting model are trained using various historical lighting conditions, enabling the trained relighting model to output relighting effect images that match the lighting conditions used during training. Specifically, a first historical rendering image and a first historical relighting effect image are obtained under various historical lighting conditions. Based on these images, the lighting conditions in the relighting model are trained, fixing them to parameters that are the same as or similar to those under the various historical lighting conditions. The trained relighting model, given an input rendering image, can output multiple relighting effect images after relighting the rendering image using various lighting conditions. Furthermore, since relighting can generate extremely realistic lighting effects, it provides an important basis for reasoning about 3D models such as metal objects. Analogous to the shortcomings of current technology in performing single-dimensional reasoning about 3D models, this application can use relighting technology to supplement multi-dimensional lighting information, and enable the subsequent material property analysis to exclude the influence of lighting factors on material properties, making the material texture generation process more accurate.

[0049] In other embodiments, without relying on a preset large model, multiple relit images can be obtained by relighting the rendered atlas using physical lighting algorithms, manual lighting, and other techniques. For example, the material map generation interface has multiple lighting conditions set. The user can manually select the desired lighting conditions from the material map generation interface and use the selected lighting conditions to relight the rendered atlas to obtain multiple relit images.

[0050] S203, based on multiple re-lighting effect images, performs material property analysis on the target 3D model, obtains the target material texture map of the target 3D model, and sends the target material texture map to the terminal device.

[0051] Material properties refer to the physical, optical, and textural parameters of an object's surface, which determine the object's color, reflection, roughness, transparency, and other effects under light illumination, including but not limited to diffuse reflection, normals, roughness, metallicity, and ambient light occlusion.

[0052] In this embodiment, multi-dimensional lighting information under various lighting conditions is introduced during the relighting process. Therefore, material property analysis of the target 3D model can be performed based on multiple relighting effect images, eliminating the influence of lighting conditions on material property analysis and obtaining an accurate target material texture map of the target 3D model. After obtaining the target material texture map, the server sends the target material texture map to the terminal device, and the terminal device displays the target material texture map to the user in the material texture map generation interface.

[0053] It needs to be explained that material properties are inherent properties of the model itself and are generally unaffected by lighting. However, in related technologies, because the lighting conditions in a photograph are fixed, it is difficult to distinguish whether the color and reflection effects of the photograph are due to the model's inherent material properties or the influence of external lighting. This leads to deviations in the material maps generated when performing material property analysis on the photograph. Therefore, by obtaining multiple relit effect images of the target 3D model, it is possible to determine the fixed material properties that are unaffected by lighting conditions under various lighting conditions during material property analysis, thereby eliminating the interference of lighting conditions. Then, the determined material properties of different categories are mapped to the corresponding PBR texture channels.

[0054] In this embodiment of the application, the method for analyzing the material properties of a target 3D model includes, but is not limited to: performing material property analysis on multiple re-lighting effect images using a pre-trained material analysis model; wherein, the pre-trained material analysis model supports outputting multiple material textures of the 3D model when multiple re-lighting effect images of the 3D model are input.

[0055] Furthermore, the pre-trained material analysis model includes multiple branch decoders. The branch decoders are used to output the target material properties corresponding to multiple relighting effect images given multiple input relighting effect images. The material analysis model is used to output different target material properties through each branch decoder and map different types of material properties to the corresponding PBR map channels to generate corresponding material maps. Alternatively, the material analysis tool can decompose the metallicity, roughness, reflection intensity, and other material properties of each relighting effect image according to standard parameters, and then map different types of material properties to the corresponding PBR map channels to generate corresponding material maps.

[0056] The aforementioned material texture generation scheme involves the server responding to a terminal device's triggering operation to generate material textures for the target 3D model. The server renders the target 3D model, obtaining a rendering atlas. Using various lighting conditions configured within a pre-defined large model, the rendering atlas is relit, resulting in multiple relit effect images. Based on these relit effect images, material property analysis is performed on the target 3D model to obtain the target material texture, which is then sent to the terminal device. Throughout this process, the relit of the large model introduces multi-dimensional lighting conditions into the target 3D model's rendering atlas. This ensures that subsequent material property analysis is based on relit effect images under different lighting conditions, eliminating the influence of lighting conditions on material properties and improving the accuracy of material texture generation for the target 3D model.

[0057] In other embodiments of this application, the rendering atlas includes a first rendering image of the target 3D model from multiple perspectives; using multiple lighting conditions configured in a preset large model, the rendering atlas is relit to obtain multiple relit effect images, including: for each perspective, using multiple lighting conditions configured in a preset large model, the first rendering image is relit to obtain multiple relit effect images.

[0058] Based on multiple re-lighting effect images, material property analysis is performed on the target 3D model to obtain the target material texture map of the target 3D model. This includes: based on multiple re-lighting effect images corresponding to each viewpoint, using a pre-trained material analysis model to perform material property analysis on the target 3D model to obtain the local material texture map corresponding to each viewpoint; and based on the local material texture maps corresponding to all viewpoints, determining the target material texture map of the target 3D model.

[0059] The pre-trained material analysis model includes multiple branch decoders. The branch decoders are used to output the target material properties corresponding to multiple relighting effect images when multiple relighting effect images are input. The material analysis model is used to output different target material properties through each branch decoder and generate local material maps based on multiple target material properties.

[0060] Among them, the perspective characterizes the direction, position, and visible range of the observed target's 3D model, which determines the content of the rendered image.

[0061] In the embodiments of this application, when rendering the target 3D model through a rendering camera, since the target 3D model is a model in a 3D coordinate system and has a spatial structure, the viewpoint captured by the rendering camera is limited, making it difficult to complete the rendering of the target 3D model in one go. Therefore, the target 3D model can be rendered from multiple viewpoints to obtain the first rendering image of the target 3D model from multiple viewpoints, and a rendering image set of the target 3D model can be obtained based on the multiple first rendering images.

[0062] Rendering the target 3D model from multiple perspectives can be achieved by setting up multiple rendering cameras around the target 3D model and rendering the target 3D model from each perspective using multiple rendering cameras to obtain a first rendering image from each perspective, and then assembling a rendering atlas from all the first rendering images from all perspectives. Alternatively, a rendering camera can be set up near the target 3D model and the target 3D model or the rendering camera can be rotated multiple times around the model's central axis by a preset angle. After each rotation, the target 3D model can be rendered using the rendering camera to obtain a first rendering image from the current perspective, and then assembling a rendering atlas from all the first rendering images from all perspectives.

[0063] In this embodiment, for each viewpoint, the first rendered image is relit using various lighting conditions configured in the pre-set large model's relit lighting model, resulting in multiple relit effect images for that viewpoint. Then, based on these multiple relit effect images, material property analysis is performed on the target 3D model. Specifically, from different lighting viewpoints, light intensities, and light source colors, the material texture of each pixel in the target 3D model is inferred to obtain the local material texture corresponding to each viewpoint. This local material texture for each viewpoint includes, but is not limited to, three material textures, such as an albedo texture, a normal texture, and a metallic roughness texture. The target material texture of the target 3D model is obtained by combining the local material textures corresponding to all viewpoints.

[0064] In the embodiments of this application, the material property analysis of the target 3D model is generally achieved by a pre-trained material analysis model. That is, based on multiple relighting effect images corresponding to each viewpoint, the material property analysis of the target 3D model is performed using the pre-trained material analysis model to obtain the local material texture map corresponding to each viewpoint.

[0065] The training process of the material analysis model generally includes: obtaining second historical relighting effect maps of the historical 3D model under various lighting conditions, and the theoretical material map of the historical 3D model; extracting the historical relighting effect map features of each second relighting effect map; predicting multiple predicted material maps of the historical 3D model based on the features of multiple historical relighting effect maps; determining the predicted relighting effect maps of the historical 3D model under various lighting conditions based on the multiple predicted material maps; and training the material analysis model based on the first loss between the predicted relighting effect map and the second historical relighting effect map under the same lighting condition, the second loss between the predicted material map and the theoretical material map, and the preset penalty loss.

[0066] More specifically, the pre-trained material analysis model includes multiple branch decoders. These branch decoders, given multiple relighting effect images as input, output the target material properties corresponding to those images. The material analysis model then uses each branch decoder to output different target material properties and generates local material maps based on these multiple properties. The structure of the material analysis model is as follows: Figure 3 As shown, it includes at least an input layer, a feature extraction network, and a multi-branch decoder; the training process of the material analysis model includes the following stages: 1. Data Input Stage: The second historical relighting effect diagram (I) of the historical 3D model under various lighting conditions. real The input dataset D={I}, which together with the theoretical material property set T of the second lighting effect image, constitutes the input layer. real (i), T}, i=1~N, where N is the total number of second historical relighting effect images. The theoretical material attribute set T includes theoretical material maps corresponding to various physical material attributes. In practical applications, the second historical relighting effect images of the historical 3D model under various lighting conditions can be obtained by dividing the PBR rendering process of the historical 3D model into ambient lighting and direct lighting parts (rendered in separate layers) through the rendering equation. The rendering engine renders the effects of these two parts and the final effect separately, resulting in a series of second historical relighting effect images. At the same time, the theoretical material maps of the historical 3D model can be obtained before rendering. Based on the second historical relighting effect images of the historical 3D model under various lighting conditions and the theoretical material maps of the historical 3D model, the material analysis model is trained.

[0067] An example of the rendering equation is as follows:

[0068] in, The emitted radiation brightness, i.e., from the scene point Along the direction of launch The brightness of the light ultimately seen by the camera; scene point It involves calculating the position of the object's surface illuminated by the light; and the direction of emission. It is the direction in which the light is directed towards the camera; It refers to self-emitting radiant light, that is, the light emitted by the object itself (such as a light bulb or the sun), not reflected light; ∫ H 2 It is the surface integral over the upper hemisphere, representing the sum of all incident light directions within the entire hemispherical space above the object's surface; It is a scattering function used to describe the surface material of an object; incident light What percentage will be reflected in the direction of emission? (For example, strong specular reflection and diffuse wall reflection); It is the incident radiation brightness, from the direction The brightness of the external light illuminating point x; It is the cosine of the angle between the incident direction and the surface normal. The more perpendicular the light rays are to the surface, the stronger their contribution to illumination; the more oblique the light rays are, the weaker their contribution. i It is the solid angle element of the incident direction. The entire rendering equation means that the total light intensity emitted from a point on the object's surface towards the eye = the light emitted from that point itself + the sum of the light emitted from all directions after reflection by the object's surface material and then towards the eye.

[0069] 2. Feature extraction stage: Use a feature extraction network to extract the features of the historical relighting effect image for each second historical relighting effect image. The feature extraction network includes, but is not limited to, convolutional neural networks, attention neural networks, etc.

[0070] Let the parameters of the feature extraction network be F. θ Then, for each second historical relighting effect image, the historical relighting effect image feature f=F θ (I real ), f∈R d , where θ is the trainable parameter of the feature extraction network and d is the feature dimension.

[0071] 3. Multi-branch decoding: Using a shared feature f, the corresponding physical material properties are generated simultaneously by multiple independent branch decoders (unaffected by lighting). Based on the physical material properties generated by all decoders, a predicted material map is obtained. The branch decoders include the following: Albedo branch of diffuse reflection: A=D A (f); Normal branch: N=D N (f); Roughness branch: R=D R (f); Metallic branch: M=D M(f); Ambient shading (AO) branch: O=D O (f); Each decoder All of them are lightweight neural networks.

[0072] 4. Compositing and Rendering Stage: The generated predicted material textures are input into the differentiable rendering engine R. The differentiable rendering engine R outputs predicted relighting effect maps (I) of the historical 3D model under each lighting condition. pred =R(A, N, R, M, O), allows gradients to backpropagate from pixel-level errors to material parameters.

[0073] 5. Physical Verification and Supervision Signal Stage: Calculate the first loss between the predicted material map corresponding to each physical material attribute and the theoretical material map corresponding to the theoretical material attribute set T. The first loss can be determined based on the L1 loss function or mean squared error loss function, etc. Taking L1 loss as an example, the first loss L... attr The calculation expression is:

[0074] in, For any physical material property, For any physical material property, it is the predicted material texture.

[0075] Physical rendering can be achieved through rendering consistency (such as rendering errors under multiple lighting conditions) or physical constraints (such as energy conservation), and the rationality of generative model training can be improved through explicit attribute supervision or implicit physical constraints.

[0076] It can also calculate the second loss between the second historical relighting effect map and the predicted relighting effect map corresponding to each lighting condition, as well as the preset penalty loss, which is the physical constraint that needs to be met during the physical rendering process.

[0077] Loss function combination: based on the first loss, the second loss Calculate the total loss by comparing it with the preset penalty loss. Total loss The expressions include:

[0078] in, , These are the weighting coefficients corresponding to the second loss and the preset penalty loss, respectively.

[0079] 6. Parameter Optimization: Based on the total loss calculated in the physical verification and supervision signal stages, all trainable parameters of the material analysis model are updated through backpropagation, such as the training parameters Θ of the feature extraction network and the branch decoder. Differentiable rendering engines like R update trainable parameters by subtracting the learning rate multiplied by the gradient η of the total loss with respect to all trainable parameters Θ. ΘL total , where Θ←Θ η ΘL total Θ={θ,D A D N Iterate until convergence or the preset epoch is reached to obtain the trained material analysis model.

[0080] In this embodiment, after determining the target material texture of the target 3D model, the target material texture can be verified. The verification method includes: based on the target material texture, calling the rendering engine to render the target 3D model from a specific angle to obtain a third lighting effect under a certain lighting condition; comparing the third lighting effect with a fourth lighting effect using similar lighting conditions when the material analysis model was pre-trained to determine whether the current target material texture meets the expected effect. If it does not meet the expectation, it is continuously corrected by training the model. If it meets the expectation, the process is terminated, and the generation of the PBR material texture is completed. This verification process makes the quality of the generated texture controllable.

[0081] In the above embodiments, first rendering images of the target 3D model from multiple perspectives are obtained, making the acquisition of the rendering atlas more comprehensive and complete. Combined with a relighting model under multiple lighting conditions, the first rendering images are relit respectively. Based on multiple relighting effect images corresponding to each perspective, a pre-trained material analysis model is used to perform material analysis on the target 3D model. This allows the influence of lighting on material properties to be eliminated during material property analysis. The material analysis model focuses more on the analysis of physical material properties such as metallicity, roughness, and reflectivity, thereby improving the accuracy of target material texture generation for the target 3D model.

[0082] In other embodiments of this application, the target 3D model is rendered to obtain a rendering atlas of the target 3D model, including: Determine the texture map of the target 3D model; based on the texture map, call the preset rendering engine to render the target 3D model from multiple perspectives to obtain the rendering atlas of the target 3D model.

[0083] Texture mapping is the process of mapping the UV (Texture UV Coordinates) coordinates of a two-dimensional texture space to the surface space of a three-dimensional model. By sampling texture image data, it achieves detailed representation of the object's surface. Specifically, texture mapping stores the color information of the object's surface, and through mapping relationships, it endows the three-dimensional model with rich color details, replacing a single solid color and determining the final color effect of the model.

[0084] In this embodiment, the target 3D model is generally a low-precision model, lacking effective information, especially texture information. Therefore, when performing material property analysis on the target 3D model, due to the limited effective texture information, issues such as mistaking shadows for black stains or specular highlights for white materials may occur. Therefore, it is necessary to first determine the texture map of the target 3D model, and then render the target 3D model from multiple perspectives based on the texture map to obtain a rendering atlas of the target 3D model.

[0085] The texture map contains the color and geometric feature information of the target 3D model. The geometric feature information represents the edge features, normal features, and depth features of the target 3D model. Therefore, determining the texture map of the target 3D model is equivalent to determining its color, edge features, normal features, and depth features. After determining this information, the target 3D model can be rendered from multiple perspectives based on the determined texture map to obtain a rendering atlas of the target 3D model. This rendering atlas also contains the color and geometric feature information of the target 3D model, providing a more effective and informative reference benchmark for subsequent material property analysis. This informs the material analysis model of the color and geometric feature information of the target 3D model, allowing it to focus more on analyzing features such as roughness and metallicity, rather than guessing the basic color and geometric features.

[0086] In this embodiment, the texture map of the target 3D model can be generated directly from the large model. For example, the texture map of the target 3D model can be generated using the AIGC model. When generating the texture map of the target 3D model by relying on the large model, an accurate and appropriate prompt word needs to be designed so that the large model can generate an accurate texture map; or it can be generated using geometric projection method, UV coordinate projection method, etc., which is not limited here.

[0087] In this embodiment, the texture map of the target 3D model is the complete texture map of the target 3D model, rather than the texture map from a single viewpoint. The method for determining the texture map may be to first unfold the surface of the target 3D model into a two-dimensional plane and determine the texture map of this two-dimensional plane as the texture map of the target 3D model; or to determine the local texture maps of the target 3D model from different viewpoints and combine the local texture maps from all viewpoints into a complete texture map.

[0088] When the local texture maps of the target 3D model are determined from multiple perspectives, multiple texture map generation tools corresponding to different perspectives can be configured to generate local texture maps of the target 3D model from each perspective; or a single texture map generation tool can be configured to control the rotation of the texture map generation tool or the target 3D model itself to generate local texture maps of the target 3D model from each perspective.

[0089] In this embodiment, rendering the target 3D model from multiple perspectives is generally achieved by calling a preset rendering engine. The rendering engine can be a real-time rendering tool such as a rendering camera or rendering software. When the rendering engine is a rendering camera, the rendering camera needs to be an orthographic camera to eliminate perspective distortion during the rendering process. Furthermore, rendering the target 3D model from multiple perspectives based on texture mapping by calling the preset rendering engine means drawing the texture map of the target 3D model onto the rendering camera of each perspective to obtain a rendering atlas.

[0090] In the above embodiments, the texture map of the target 3D model can provide a basic texture distribution, such as color information and geometric feature information, for material property analysis. This allows the material property analysis to focus on analyzing other more important material properties, such as roughness and metallicity, without having to guess the color and geometric features of the target 3D model. Therefore, the rendering atlas of the target 3D model is obtained based on the texture map of the target 3D model, making the subsequent material analysis process more accurate.

[0091] In this embodiment of the application, determining the texture map of the target 3D model includes: Determine the first texture map of the target 3D model in a preset first viewpoint; rotate the preset rendering camera around the central axis of the target 3D model multiple times by a preset angle until the number of rotations reaches a preset threshold; after each rotation operation, based on the acquired historical texture map set, predict the second texture map of the target 3D model in a second viewpoint using the rotated rendering camera and the preset large model, wherein the historical texture map set includes the first texture map and multiple second texture maps of the target 3D model during historical rotations; determine the texture map of the target 3D model based on the first texture map and multiple second texture maps.

[0092] The historical texture map set includes a first texture map and second texture maps of the target 3D model during multiple historical rotations.

[0093] In this embodiment of the application, taking the texture mapping generation tool as an example that includes at least a rendering camera in the rendering engine, after the target 3D model is placed, the face of the target 3D model or the angle of the model surface that can cover the target 3D model to the greatest extent is used as a preset first viewpoint, and the rendering camera is controlled to determine the first texture map of the target 3D model under this viewpoint.

[0094] Next, rotate the rendering camera around the central axis of the target 3D model by a preset angle to determine the second viewpoint of the rendering camera after rotation, and determine the second texture map of the target 3D model in the second viewpoint. Repeat the above operations, rotating the rendering camera around the central axis of the model multiple times by a preset angle until the number of rotations reaches a preset threshold, to obtain the second texture map of the target 3D model in the corresponding second viewpoint. Finally, combine the first texture map with multiple second texture maps to form the texture map of the target 3D model. The preset threshold for rotation is determined by the preset angle of rotation. When the total angle of multiple rotations is greater than 360°, the number of rotations can be considered to have reached the preset threshold. In practical applications, the preset angle for each rotation of the rendering camera around the central axis of the target 3D model can be set to 20°-45°, and the rotation process is as follows: Figure 4 As shown, the dashed line represents the original viewpoint of the rendering camera during the previous rotation, while the solid line represents the new viewpoint of the rendering camera during the current rotation.

[0095] In this embodiment, to rotate the rendering camera around the central axis of the target 3D model by a preset angle, it is necessary to determine parameters such as the direction and angle of rotation of the rendering camera in advance. Based on these parameters, the rendering camera is controlled to rotate multiple times around the central axis of the model in the preset direction by a preset angle. It should be noted that the rotation direction of the rendering camera should be consistent during multiple rotations, for example, a consistent clockwise rotation or a consistent counterclockwise rotation, to avoid inefficiency caused by repeated rendering.

[0096] In the embodiments of this application, determining the first texture map of the target 3D model under a preset first viewpoint means mapping the color information of the target 3D model to the surface of the target 3D model corresponding to the viewpoint, including but not limited to: mapping the color information of the target 3D model to the surface of the target 3D model corresponding to the viewpoint based on the UV coordinates of the target 3D model under the preset first viewpoint; or mapping the color information of the target 3D model to the surface of the target 3D model corresponding to the viewpoint through a geometric projection method, without relying on the UV coordinates of the target 3D model under the preset first viewpoint; or mapping the color information of the target 3D model to the surface of the target 3D model corresponding to the viewpoint using a mathematical algorithm based on texture map generation, which is not limited here.

[0097] In the embodiments of this application, the second texture map determined after each rotation may be different from the previously acquired historical texture map set, or it may overlap with the previously acquired historical texture map set. This can be determined by the preset rotation angle and the rendering field of view corresponding to the rendering camera. For example, when the rendering field of view of the rendering camera is 10°, if the preset angle of each rotation is 8°, then the second texture map determined in two adjacent rotation processes will have a 2° overlap.

[0098] If the second texture map determined after each rotation differs from the previously acquired set of historical texture maps, the second texture map can be determined using the method described in the above embodiment for determining the first texture map of the target 3D model under a preset first viewpoint. However, this may lead to inconsistencies in the texture maps generated for the target 3D model across different viewpoints. Therefore, a progressive generation scheme can be adopted, meaning that the second texture map determined after each rotation overlaps with the previously acquired set of historical texture maps.

[0099] When there is overlap between the second texture map determined after each rotation and the previously acquired set of historical texture maps, the second texture map of the target 3D model in the second viewpoint can be predicted based on the previously acquired set of historical texture maps, using the rotated rendering camera and a preset large model. In this process, texture maps in all views are supplemented from the texture maps of the previous viewpoint, ensuring continuity between different views. This prediction process can be implemented using a large model or a texture map generation tool; no specific limitation is made here. Taking a preset large model as an example, the preset large model supports predicting the texture map of the undetermined texture portion of the target 3D model in a certain viewpoint when the previously acquired set of historical texture maps of the target 3D model is input. Therefore, the second texture map of the target 3D model in the second viewpoint can be predicted based on the preset large model.

[0100] In this embodiment of the application, when both the first texture map and the second texture map are generated based on a preset large model, the same prompt can be reused to generate the texture map. For example, the same prompt "Generate a texture map of model A from this viewpoint, and the generated texture map is accurate, realistic, rich in detail, without distortion, without redundant objects, without borders, and the edges, lighting effects and foreground and background relationships are completely matched with model A, with a resolution of 8K" can be reused to generate the first texture map and the second texture map.

[0101] In this embodiment, the obtained texture map of the target 3D model is a texture map with fixed-direction lighting effects. If the conventional method of generating three PBR texture maps from a single image is used, there will be defects due to insufficient inference factors. Therefore, this application can further solve this problem by fully linking the rendering engine with the large model. That is, based on the texture map of the target 3D model, the rendering engine is called to render the target 3D model to obtain the rendering atlas of the target 3D model. Using the multiple lighting conditions configured in the preset large model, the rendering atlas is relit to obtain multiple relit effect maps. Based on the multiple relit effect maps, the material properties of the target 3D model are analyzed to obtain the accurate target material map of the target 3D model.

[0102] In the above embodiments, firstly, the first texture map of the target 3D model under a preset first viewpoint is determined. Then, after each rotation operation, based on the acquired historical texture map set, the second texture map of the target 3D model under the corresponding second viewpoint is predicted using a preset large model. Through this progressive texture map generation method, the texture maps under all views are supplemented by the texture maps of the previous viewpoint and reuse the same prompt word. Therefore, the 360-degree rendering atlas of the model obtained later is very stable, avoiding the problem of inconsistent texture maps generated for different views when the model is simultaneously generated for the large model from all angles. This provides the necessary conditions for subsequent material property analysis.

[0103] In other embodiments of this application, determining the first texture map of the target 3D model under a preset first viewpoint includes: The target 3D model is rendered under a preset first viewpoint to obtain the first geometric feature map of the target 3D model; based on the first geometric feature map, a second rendering map of the target 3D model under the preset first viewpoint is predicted using a preset large model; by mapping the color information of each pixel in the second rendering map to the surface of the target 3D model, the first texture map of the target 3D model is obtained.

[0104] Among them, the first geometric feature map represents the edge features, normal features and depth features of the target 3D model under the preset first view.

[0105] Edge features are regions where spatial grayscale, geometric curvature, normal, and depth values ​​change abruptly, representing the geometric features of the model's surface structure contours, object boundaries, geometric transitions, and the intersection of edges and textures. Normal features are geometric features composed of the unit normal vectors of each vertex on the 3D model's surface, describing the spatial pose of the model's local surface, such as orientation, tilt direction, and concavity / convexity trends. Depth features are spatial features based on the depth distance values ​​of each point on the 3D model's surface relative to the camera reference plane, describing the 3D spatial position information of points on the model's surface, such as distance, spatial location, and relative elevation.

[0106] In this embodiment, the first texture map of the target 3D model under a preset first viewpoint is obtained. This first texture map contains color and geometric feature information of the target 3D model under the preset first viewpoint. Therefore, the target 3D model is first rendered under the preset first viewpoint to obtain a second rendered image of the target 3D model. This converts the target 3D model from 3D data into a 2D image. The second rendered image contains geometric feature information of the target 3D model under the preset first viewpoint. For example, a certain second rendered image... Figure 5 As shown, the color information of each pixel in the second rendered image is then mapped onto the surface of the target 3D model to obtain the first texture map of the target 3D model. Figure 5 The first texture map corresponding to the second rendered image in the image is as follows: Figure 6 As shown.

[0107] In this embodiment of the application, when rendering the target 3D model under a preset first viewpoint, it is necessary to introduce the geometric feature information of the target 3D model under the preset first viewpoint. Therefore, rendering the target 3D model under the preset first viewpoint results in a first geometric feature map of the target 3D model, including an edge map, a model normal map, and a model depth map, such as... Figure 7 As shown. The reason for outputting these three images is to match the corresponding Canny, Normal, and Depth nodes in the preset large model. This allows the large model to determine the edges, lighting effects, and foreground / background relationships of the target 3D model based on its edge map, normal map, and depth map. It then generates a second rendering image based on these elements, ensuring a perfect match between the target 3D model and its edges, lighting effects, and foreground / background relationships. Furthermore, during the application of the large model, detailed descriptive prompts can be added to input the large model and generate a realistic rendering that matches our expectations. This realistic rendering is the second rendering image.

[0108] In this embodiment of the application, rendering the target 3D model includes: placing the target 3D model into a preset rendering engine, calling the preset rendering engine to render the target 3D model, for example, calling the rendering camera in the rendering engine or other real-time rendering tools to render the target 3D model. When calling the rendering camera to render the target 3D model, the rendering camera in the rendering engine needs to be adjusted to an orthogonal mode to eliminate the perspective effect during the rendering process.

[0109] In this embodiment of the application, after mapping the color information of each pixel in the second rendering image to the surface of the target 3D model, further processing such as adding details and enlarging the image can be performed through the workflow of the large model to obtain a realistic second rendering image whose edges, lighting effects and foreground and background relationships are completely matched with the target 3D model. Moreover, the large model can automatically fill the second rendering image with lighting information. In other words, the model-covered part of the second rendering image can be fully filled with various exquisite images with lighting information.

[0110] In this embodiment, mapping the color information of each pixel in the second rendering image to the surface of the target 3D model to obtain a first texture map of the target 3D model includes: mapping the color information of each pixel in the second rendering image to the surface of the target 3D model corresponding to the viewpoint based on the UV coordinates of the target 3D model under a preset first viewpoint; or mapping the color information of each pixel in the second rendering image to the surface of the target 3D model corresponding to the viewpoint through a geometric projection method; or using a mathematical algorithm based on texture map generation to map the color information of each pixel in the second rendering image to the surface of the target 3D model corresponding to the viewpoint, which is not limited here.

[0111] Taking the mapping of color information of each pixel in the second rendered image to the surface of the target 3D model from the corresponding viewpoint based on the UV coordinates of the target 3D model in a preset first viewpoint as an example, each pixel in the model-covered part of the second rendered image can correspond to the UV values ​​of the target 3D model displayed in the rendering engine. At this time, the UV values ​​of the target 3D model can be drawn onto an image of the same size as the second rendered image. Then, through the correspondence between the pixels in the second rendered image and the model's UV values, the color information of each pixel in the second rendered image is mapped to the color map of the target 3D model's surface, resulting in the first texture map of the target 3D model. This process is actually the reprojection process.

[0112] In the above embodiments, by rendering the target 3D model under a preset first viewpoint, a first geometric feature map of the target 3D model is obtained. Then, based on the first geometric feature map, a second rendering map of the target 3D model under the preset first viewpoint is predicted. The generated second rendering map can completely match the edges, lighting effects, and front-back relationships of the target 3D model, making the subsequent texture map generation process more accurate. Furthermore, the color information of each pixel in the second rendering map can be mapped to the surface of the target 3D model. This not only adds accurate color information to the target 3D model but is also simple to operate and improves the efficiency of texture map generation.

[0113] In other embodiments of this application, based on an acquired set of historical texture maps, a second texture map of the target 3D model is predicted from a second perspective using a rotated rendering camera and a preset large model, including: For the Nth rotation process, a set of historical texture maps determined during multiple historical rotations is obtained; N is a positive integer. The target 3D model is rendered using the rotated rendering camera in a second viewpoint to obtain the second geometric feature map of the target 3D model. The second geometric feature map represents the edge features, normal features, and depth features of the target 3D model in the second viewpoint. Based on the historical texture map set, the local texture map and mask map of the target 3D model in the second viewpoint are determined. Based on the local texture map and the second geometric feature map, the third rendering map corresponding to the mask map is predicted using a large model. By mapping the color information of each pixel in the third rendering map to the surface of the target 3D model, the second texture map of the target 3D model is obtained.

[0114] Where N is a positive integer; the second geometric feature map represents the edge features, normal features and depth features of the target 3D model in the second view; the mask map is used to locate local areas of the target 3D model that do not have texture in the second view.

[0115] In this embodiment of the application, during the Nth rotation, the target 3D model is rendered again from the second perspective using the rotated rendering camera, including the edge map, normal map, depth map, and other geometric features, to obtain the second geometric feature map of the target 3D model.

[0116] Since the target 3D model corresponding to two adjacent rotation processes may have overlapping parts during multiple rotations, and the texture map of the overlapping part may have been obtained during the historical rotation process, it is necessary to obtain a set of historical texture maps that have been determined during multiple historical rotation processes. Based on the set of historical texture maps that have been determined during multiple historical rotation processes, the local texture map of the target 3D model that has been textured in the corresponding second view and the mask map part that has not yet been textured are determined. In other words, the target 3D model is divided into two parts in the second view: one part is the local texture map that has been textured, and the other part is the mask map part that has not yet been textured.

[0117] At this point, the edge map, normal map, and depth map in the second geometric feature map are completely continuous and controllable. Simultaneously, the local texture map of the rendered model perfectly conforms to the imaging principles of the rendering camera, and the mask map of the blank areas correctly represents the textureless parts of the model region. Therefore, the local texture map, mask map, and second geometric feature map can be input into a preset large-scale model rendering prediction model. This rendering prediction model supports outputting the rendering map corresponding to the mask map when the local texture map, mask map, and geometric feature map of the 3D model are input from a certain viewpoint. Therefore, the rendering prediction model can predict the third rendering map corresponding to the mask map based on the local texture map and the second geometric feature map. This third rendering map is a high-precision, realistic image that is continuous with the original image.

[0118] After supplementing the target 3D model with the third rendered image corresponding to the mask image that lacks texture in the target 3D model, a reprojection is performed based on the third rendered image to supplement the missing texture parts corresponding to the mask image. That is, the color information of each pixel in the third rendered image corresponding to the mask image is mapped to the surface of the target 3D model, resulting in the second texture map of the target 3D model. The generated second texture map can be filled into the previously obtained historical texture map set, updating the historical texture map set.

[0119] In this embodiment, the second geometric feature map and the local texture map are used as reference maps for the mask map, and various redrawing methods or variable control methods such as adding details are used to predict the third rendering map corresponding to the mask map. For example, based on the geometric feature information in the second geometric feature map and the local texture information in the local texture map, an initial rendering map corresponding to the mask map is generated, and details are added to the initial rendering map to obtain the third rendering map corresponding to the mask map; or a cartoon-style redrawing method is adopted for the initial rendering map to obtain the third rendering map corresponding to the mask map; no limitation is made here.

[0120] In this embodiment of the application, the training process of the rendering image prediction model in the large model includes: acquiring historical masking images, historical local texture maps, historical geometric feature maps, and second historical rendering images corresponding to the historical masking images of multiple historical 3D models from various perspectives; training the rendering image prediction model based on the historical masking images, historical local texture maps, historical geometric feature maps, and second historical rendering images of multiple historical 3D models from various perspectives; and outputting the rendering image corresponding to the masking image when the input is a masking image, a local texture map, or a geometric feature map of a 3D model from a certain perspective.

[0121] In the embodiments of this application, the methods for mapping the color information of each pixel in the third rendering image to the surface of the target 3D model include, but are not limited to: mapping the color information of each pixel in the third rendering image to the surface of the target 3D model from the corresponding viewpoint according to the UV coordinates of the target 3D model in the second viewpoint; or mapping the color information of each pixel in the third rendering image to the surface of the target 3D model from the corresponding viewpoint through geometric projection; or using a mathematical algorithm based on texture mapping to map the color information of each pixel in the third rendering image to the surface of the target 3D model from the corresponding viewpoint, which are not limited here.

[0122] Taking the mapping of color information of each pixel in the third rendering image to the surface of the target 3D model from the corresponding viewpoint based on the UV coordinates of the target 3D model in the second viewpoint as an example, each pixel in the model-covered part of the third rendering image can correspond to the UV value of the target 3D model displayed in the rendering engine from a specific viewpoint. At this time, the UV value of the target 3D model can be drawn onto an image of the same size as the third rendering image. Then, through the correspondence between the pixels in the third rendering image and the UV value of the model, the color information of each pixel in the third rendering image is mapped to the color map of the surface of the target 3D model to obtain the second texture map of the target 3D model.

[0123] In the above embodiments, since the historical texture map set contains images of most other perspectives of the target 3D model, based on the historical texture map set, a continuous and controllable new third rendering map can be generated by progressively supplementing the masking part. Then, the color information of each pixel in the third rendering map is mapped to the surface of the target 3D model to obtain the second texture map of the target 3D model. This can solve the problem of texture breakage in multiple perspectives, ensure the continuity of texture maps, and avoid the problem of generating texture maps in the ComfyUI workflow where large models are placed in front, behind, left, right, top, and bottom, and the texture maps under each perspective are different.

[0124] Figure 8 A flowchart illustrating a material texture generation method provided in this application embodiment, taking the method applied to terminal device 102 as an example, includes the following steps: S801, in response to the model determination operation for the material map generation interface, displays the target 3D model in the material map generation interface.

[0125] S802, in response to the material map generation operation for the target 3D model, sends a material map generation command to the server.

[0126] S803 displays the target material map in the material map generation interface.

[0127] The process involves several steps: the material texture generation command carries the target 3D model; the server renders the target 3D model based on the material texture generation command to obtain a rendering atlas of the target 3D model; using various lighting conditions configured in the preset large model, the rendering atlas is relit to obtain multiple relit effect images; based on the multiple relit effect images, the material properties of the target 3D model are analyzed to obtain the target material texture of the target 3D model, and the target material texture is fed back to the terminal device.

[0128] In this embodiment, the target 3D model can be a pre-stored 3D model in a database or a 3D model input by the user in real time. Taking a pre-stored 3D model as an example, the user selects the target 3D model from multiple candidate 3D models on the material mapping generation interface of the terminal device and triggers the material mapping generation control. The terminal device responds to the trigger operation of the material mapping generation control by sending the model identifier of the target 3D model to the server. The server then determines the target 3D model matching the model identifier from the database. Alternatively, taking a user-input 3D model as an example, the user draws the target 3D model on the material mapping generation interface and, after completing the drawing, triggers the material mapping generation control. The terminal device responds to the trigger operation of the material mapping generation control by sending the completed target 3D model to the server. The server then retrieves the completed target 3D model. The server renders the target 3D model, obtaining a rendering atlas of the target 3D model. Then, using various lighting conditions configured in the preset large model, it relights the rendering atlas, obtaining multiple relighting effect images. Based on these multiple relighting effect images, it analyzes the material properties of the target 3D model, obtaining the target material texture map, and sends the target material texture map back to the terminal device. The terminal displays the target material texture map on the material texture map generation interface. The method for the server to generate the target material texture map is described in detail in the above embodiments and will not be repeated here.

[0129] The aforementioned material texture generation method involves the server responding to a terminal device's triggering operation to generate material textures for a target 3D model. The server renders the target 3D model, obtaining a rendering atlas. Using various lighting conditions configured in a pre-defined large model, the rendering atlas is relit, resulting in multiple relit effect images. Based on these relit effect images, material property analysis is performed on the target 3D model to obtain the target material texture, which is then sent to the terminal device. Throughout this process, the relit of the large model introduces multi-dimensional lighting conditions into the target 3D model's rendering atlas. This ensures that subsequent material property analysis is based on relit effect images under different lighting conditions, eliminating the influence of lighting conditions on material properties and improving the accuracy of material texture generation for the target 3D model.

[0130] In existing technical solutions, 3D modeling artists often face challenges such as long production cycles for PBR textures, inconsistent skill levels among artists, and limited functionality in image-based PBR texture generation tools. These tools cannot generate usable PBR textures directly from a single image, requiring repeated manual adjustments before being used in rendering engines. These are all problems that current technology urgently needs to address. Therefore, this application proposes a more accurate material texture generation method. The detailed process of material texture generation is as follows: S1. Place the low-precision target 3D model created by the artists into the rendering engine, adjust the rendering camera in the rendering engine to orthogonal mode, and select the front of the target 3D model or the rendering angle that can cover the surface of the target 3D model to the greatest extent. Render the target 3D model and output the geometric feature maps of the target 3D model, namely edge line map, model normal map and model depth map.

[0131] S2. Input the output edge line map, model normal map, and model depth map into the AIGC model, and add detailed descriptive prompts so that the second rendering map generated by the AIGC model can match the edges, lighting effects, and foreground / background relationships of the target 3D model.

[0132] S3. After processing by the AIGC workflow (including model selection, detail enhancement, image magnification, etc.), the edges, lighting effects, and foreground / background relationships of the rendered image perfectly match the target 3D model in the rendering engine. Furthermore, the covered portion of the target 3D model in the second rendered image is fully filled with the AIGC-generated image containing lighting information. Each pixel of the covered portion of the target 3D model in the second rendered image corresponds to the UV value of the model displayed in the camera's view in the rendering engine. By mapping each pixel in the second rendered image to the color map of the target 3D model itself, the first texture map of the target 3D model is obtained.

[0133] S4. Rotate the rendering camera around the center point of the model by a fixed angle (e.g., 20-45 degrees), and render the edge lines, normals, and depth maps of the target 3D model from the current viewpoint again. At the same time, use the historical texture map A of the target 3D model that has been determined from this viewpoint as the fourth image input to AIGC. The parts of the target 3D model without textures from this viewpoint are drawn as white parts and passed into AIGC as the fifth image of the supplementary generation part, which is the mask map.

[0134] S5. All control maps from this perspective—edge map, normal map, depth map, mask map, and part of the historical texture map A—are placed into AIGC. Through various variable controls, a high-precision, partially realistic image corresponding to the masked part in the mask map and continuous with the original image is generated. This realistic image is the third rendering map corresponding to the mask map.

[0135] S6. The third rendering image is sent back to the rendering engine again, and the color information of each pixel in the third rendering image is mapped to the surface of the target 3D model to obtain the second texture map of the target 3D model from this viewpoint.

[0136] S7. Repeat S4 to S6 until the local texture maps of the target 3D model are obtained from all perspectives. Based on the local texture maps of the target 3D model from all perspectives, the texture map of the target 3D model is obtained. At this time, the texture map is a map with fixed-direction lighting effects.

[0137] S8. Draw the texture map of the target 3D model onto an orthographic camera at a certain angle to form a first rendering image containing the model. Input the first rendering image into AIGC. AIGC uses the latest relighting model IC-LIGHTV2 or other superior modules to relight the first rendering image to obtain relighting effect images under various lighting conditions.

[0138] The training data for the AIGC relighting parameters of the relighting model consists of the first historical rendering of the historical 3D model and the first historical relighting effect image drawn after rendering the historical 3D model under various lighting conditions such as different lighting angles, different light intensities, and different light colors. The AIGC relighting parameters of the pre-trained relighting model are fixed, and the relighting model can output relighting effect images that match the various lighting conditions used during the training of the model.

[0139] S9. Using the trained material analysis model, perform material property analysis and inference on the relighting effect images under various lighting conditions obtained in S8 to generate three texture maps: albedo, normal, and metallic roughness.

[0140] The training process of the material analysis model includes: dividing the PBR rendering of the historical 3D model into ambient lighting and direct lighting parts through the rendering equation; rendering the effects and final effects of these two parts through the rendering engine to obtain a series of second historical relighting effect maps; at the same time, determining the values ​​of three theoretical material maps of the historical 3D model before rendering, and using them as training data for the material analysis model to start training the inference model; and finally completing the training of the material analysis model through a series of convergences.

[0141] S10. Repeat S8 to S9, and put the re-lighting effect images from different perspectives into the material analysis model. Iterate and reason about the required material texture of the target 3D model until the material texture is filled.

[0142] S11. Put the three target material textures that have been inferred into the rendering engine for rendering at a specific angle to obtain a relighting effect image to match the relighting effect image with specific lighting conditions generated in AIGC. Compare the two effect images to evaluate whether the target 3D model meets the expected effect. If it meets the expectation, terminate this process and complete the generation of the target material texture.

[0143] Based on the above analysis, this technical solution provides an automated method for generating PBR material maps using AI algorithms (deep learning, image analysis technology), with broad application potential. This solution is used to automatically generate multi-dimensional PBR maps (including Albedo, Normal, Metallic, etc.) from a single input (such as a photo or simple material). This technology differs from traditional manual drawing or single-image, single-dimensional methods, featuring high automation and adaptability. This technology efficiently generates high-quality PBR maps without the need for tedious manual production or multi-angle data acquisition, significantly reducing industry barriers and costs. Therefore, it can be used for rapid iteration and reduced labor costs in fields requiring large numbers of material maps, such as game development, film and television special effects, and virtual reality, significantly improving the production efficiency and quality standards of next-generation 3D models.

[0144] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0145] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0146] It should be noted that the technical solutions described in the embodiments of the present invention can be combined arbitrarily without conflict.

[0147] In addition, in the embodiments of the present invention, "first," "second," etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0148] The material texture generation method in this application embodiment can be implemented by a processor in an electronic device, such as a central processing unit (CPU), a digital signal processor (DSP), a microcontroller unit (MCU), or a field-programmable gate array (FPGA).

[0149] To implement the method of the embodiments of this application, the embodiments of this application also provide an electronic device. Figure 6 This is a schematic diagram of the hardware composition structure of an electronic device provided in an embodiment of this application, such as... Figure 9 As shown, an electronic device is equipped with a material mapping generation system. The electronic device includes: The communication interface 901 enables information exchange with other devices, such as electronic devices. The processor 902 is connected to the communication interface 901 to enable information exchange with other devices and to execute the methods provided by one or more of the above-mentioned technical solutions when running a computer program. The computer program is stored in the memory 903.

[0150] Of course, in practical applications, the various components in an electronic device are coupled together through a bus system 604. It can be understood that the bus system 604 is used to implement communication between these components. In addition to the data bus, the bus system also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 9 The general designated all buses as Bus System 904.

[0151] The memory 903 in this embodiment is used to store various types of data to support the operation of the computer device. Examples of such data include any computer program used to operate on the electronic device.

[0152] It is understood that memory can be volatile or non-volatile, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of memories.

[0153] The methods disclosed in the embodiments of this application can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor may be a general-purpose processor, a DSP, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. A general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in memory. The processor reads the program from the memory and, in conjunction with its hardware, completes the steps of the aforementioned method.

[0154] Optionally, when the processor executes the program, it implements the corresponding processes implemented by the electronic device in the various methods of the embodiments of this application. For the sake of brevity, these will not be described in detail here.

[0155] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a first memory storing a computer program, which can be executed by a processor of a computer device to complete the aforementioned method steps. The computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM.

[0156] In the several embodiments provided in this application, it should be understood that the disclosed apparatus, computer devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components may be combined, or integrated into another system, or some features may be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0157] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0158] In addition, each functional unit in the various embodiments of this application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0159] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0160] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, etc.) to execute all or part of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0161] In an exemplary embodiment, this application also provides a computer program product, including a computer program that can be executed by a processor 902 of an electronic device to complete the steps of the material texture generation method in this application embodiment.

[0162] It should be noted that terms such as "first" and "second" are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0163] Furthermore, the technical solutions described in the embodiments of this application can be combined arbitrarily without conflict.

[0164] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for generating material texture maps, characterized in that, Applied to the server side, the method includes: In response to the material texture generation operation triggered by the terminal device for the target 3D model, the target 3D model is rendered to obtain the rendering atlas of the target 3D model; Using the various lighting conditions configured in the preset large model, the rendered image set is relit to obtain multiple relit effect images. Based on multiple re-lighting effect images, the material properties of the target 3D model are analyzed to obtain the target material texture map of the target 3D model, and the target material texture map is sent to the terminal device.

2. The method according to claim 1, characterized in that, The rendering atlas includes first renderings of the target 3D model from multiple viewpoints; the rendering atlas is then relit using various lighting conditions configured in a pre-set large model to obtain multiple relit effect images, including: For each of the aforementioned viewpoints, the first rendered image is relit using various lighting conditions configured in the preset large model, resulting in multiple relit effect images. The step of analyzing the material properties of the target 3D model based on multiple re-lighting effect images to obtain the target material texture map of the target 3D model includes: Based on multiple relighting effect images corresponding to each viewpoint, a pre-trained material analysis model is used to analyze the material properties of the target 3D model to obtain a local material texture map corresponding to each viewpoint. The pre-trained material analysis model includes multiple branch decoders, which, given the input of the multiple relighting effect images, output the target material properties corresponding to the multiple relighting effect images. The material analysis model is used to output different target material properties through each branch decoder and generate a local material texture map based on the multiple target material properties. The target material texture of the target 3D model is determined based on the local material textures corresponding to all the aforementioned viewpoints.

3. The method according to claim 1, characterized in that, The process of rendering the target 3D model to obtain a rendering atlas of the target 3D model includes: Determine the texture map of the target 3D model; Based on the texture map, a preset rendering engine is invoked to render the target 3D model from multiple perspectives, resulting in a rendering atlas of the target 3D model.

4. The method according to claim 3, characterized in that, Determining the texture map of the target 3D model includes: Determine the first texture map of the target 3D model under a preset first viewpoint; The preset rendering camera is rotated multiple times around the central axis of the target 3D model by a preset angle until the number of rotations reaches a preset threshold. After each rotation operation is completed, based on the acquired historical texture map set, the second texture map of the target 3D model in the second view is predicted using the rotated rendering camera and the preset large model. The historical texture map set includes the first texture map and the second texture map of the target 3D model in multiple historical rotation processes. The texture map of the target 3D model is determined based on the first texture map and multiple second texture maps.

5. The method according to claim 4, characterized in that, Determining the first texture map of the target 3D model under a preset first viewpoint includes: The target 3D model is rendered under a preset first viewpoint to obtain a first geometric feature map of the target 3D model, wherein the first geometric feature map represents the edge features, normal features and depth features of the target 3D model under the preset first viewpoint. Based on the first geometric feature map, the second rendered image of the target 3D model under the preset first viewpoint is predicted using the preset large model. By mapping the color information of each pixel in the second rendered image to the surface of the target 3D model, a first texture map of the target 3D model is obtained.

6. The method according to claim 4, characterized in that, The process of predicting the second texture map of the target 3D model from a second perspective using a rotated rendering camera and a preset large model, based on the acquired historical texture map set, includes: For the Nth rotation process, obtain a set of historical texture maps that have been determined in multiple historical rotation processes; N is a positive integer; The target 3D model is rendered using a rotated rendering camera from a second perspective to obtain a second geometric feature map of the target 3D model. The second geometric feature map represents the edge features, normal features, and depth features of the target 3D model from the second perspective. Based on the historical texture map set, the local texture map and mask map of the target 3D model under the second view are determined, wherein the mask map is used to locate local areas of the target 3D model where no texture exists under the second view. Based on the local texture map and the second geometric feature map, the third rendered image corresponding to the mask map is predicted using the preset large model; By mapping the color information of each pixel in the third rendering image to the surface of the target 3D model, a second texture map of the target 3D model is obtained.

7. A method for generating material texture maps, characterized in that, Applied to a terminal device, the method includes: In response to a model determination operation for the material mapping generation interface, the target 3D model of the object to be processed is displayed in the material mapping generation interface. In response to the material texture generation operation for the target 3D model, a material texture generation instruction is sent to the server, wherein the material texture generation instruction carries the target 3D model; the server renders the target 3D model based on the material texture generation instruction to obtain a rendering atlas of the target 3D model; using multiple lighting conditions configured in a preset large model, the rendering atlas is relit to obtain multiple relit effect images; based on the multiple relit effect images, the material properties of the target 3D model are analyzed to obtain the target material texture of the target 3D model, and the target material texture is fed back to the terminal device; The target material texture is displayed in the material texture generation interface.

8. An electronic device, characterized in that, The electronic device includes: Memory is used to store executable instructions or computer programs. A processor, when executing computer-executable instructions or computer programs stored in the memory, implements the method according to any one of claims 1 to 7.

9. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images