Three-dimensional model mapping construction method and device

Abstract

The invention discloses a three-dimensional model chartlet construction method and device, and relates to the technical field of three-dimensional models.The method comprises the following steps that RGB images of four preset fixed view angles are constructed based on text input information and a three-dimensional white model of a target three-dimensional model; obtaining four corresponding depth images based on the geometric structure of the target three-dimensional model and the four RGB images; and constructing a three-dimensional model map corresponding to the target three-dimensional model based on the RGB images of the four preset fixed visual angles and the corresponding depth images. According to the method, the RGB images of the multiple preset visual angles and the corresponding depth images are constructed, then three-dimensional model mapping construction is carried out, the operation process is simple and efficient, the refinement degree is high, and certain reliability is achieved.

Description

Technical Field

[0001] The present application relates to the technical field of three-dimensional models, and in particular to a method and device for constructing a three-dimensional model map. Background Art

[0002] The current technical background of Consistent-Texture mainly involves 3D model texture generation and the application of Neural Radiance Field (NeRF). Traditional 3D model generation methods usually rely on complex geometric construction processes or optimized NeRF technology. However, these methods have disadvantages such as low efficiency, high cost, and strong dependence on professional knowledge. Especially in the process of generating 3D model textures from text, these methods face huge challenges, including how to accurately understand and capture complex 3D information from limited text descriptions for consistent mapping.

[0003] In the prior art, the closest implementation scheme to Consistent-Texture includes 3D model mapping methods based on deep learning. These methods usually use neural networks to directly generate corresponding maps from 3D models, but they are not efficient and the quality of the generated 3D model maps is unstable. In addition, most existing methods require image inputs from multiple perspectives when generating 3D model maps, which increases the complexity and cost of data acquisition.

[0004] Therefore, in order to meet actual needs, a three-dimensional model mapping construction technology is now provided. Summary of the invention

[0005] In view of the defects existing in the prior art, the purpose of this application is to provide a three-dimensional model mapping construction method and device, which constructs RGB images of multiple preset perspectives and corresponding depth images to further construct three-dimensional model mapping. The operation process is simple and efficient, the degree of refinement is high, and it has a certain degree of reliability.

[0006] In order to achieve the above objectives, the technical solution adopted by this application is:

[0007] In a first aspect, the present application provides a method for constructing a three-dimensional model map, the method comprising the following steps:

[0008] Based on the text input information and the 3D white model of the target 3D model, four RGB images with preset fixed viewing angles are constructed;

[0009] Based on the geometric structure of the target three-dimensional model and the four RGB images, four corresponding depth images are obtained;

[0010] Based on the four RGB images of the preset fixed viewing angles and the corresponding depth images, a three-dimensional model map corresponding to the target three-dimensional model is constructed.

[0011] On the basis of the above technical solution, based on the text input information and the 3D white model of the target 3D model, four preset fixed-view RGB images are constructed, including the following steps:

[0012] Based on the three-dimensional white model of the target three-dimensional model, four RGB images with preset fixed viewing angles are constructed according to preset elevation angles; wherein,

[0013] The preset elevation angle is 10°.

[0014] On the basis of the above technical solution, based on the geometric structure of the target three-dimensional model and the four RGB images, four corresponding depth images are obtained, including the following steps:

[0015] Based on the geometric structure of the target three-dimensional model and the four RGB images, four corresponding depth images are obtained through a preset DINOv2 model.

[0016] On the basis of the above technical solution, based on the four RGB images of the preset fixed perspectives and the corresponding depth images, constructing a three-dimensional model map corresponding to the target three-dimensional model includes the following steps:

[0017] Based on the four depth images, construct three-plane feature information;

[0018] Based on the three-plane feature information and the four RGB images of the preset fixed viewing angles, a three-dimensional model map corresponding to the target three-dimensional model is constructed.

[0019] Based on the above technical solution, the four preset fixed viewing angles are 0°, 90°, 180° and 270° respectively.

[0020] In a second aspect, the present application provides a three-dimensional model mapping construction device, the device comprising:

[0021] A first image construction module, which is used to construct four RGB images of preset fixed viewing angles based on text input information and a three-dimensional white model of a target three-dimensional model;

[0022] A second image construction module, which is used to obtain four corresponding depth images based on the geometric structure of the target three-dimensional model and the four RGB images;

[0023] A model map construction module is used to construct a three-dimensional model map corresponding to the target three-dimensional model based on the four RGB images of the preset fixed viewing angles and the corresponding depth images.

[0024] On the basis of the above technical solution, the first image construction module is also used to construct four RGB images of preset fixed viewing angles based on the three-dimensional white model of the target three-dimensional model according to preset elevation angles; wherein,

[0025] The preset elevation angle is 10°.

[0026] On the basis of the above technical solution, the second image construction module is also used to obtain four corresponding depth images through a preset DINOv2 model based on the geometric structure of the target three-dimensional model and the four RGB images.

[0027] On the basis of the above technical solution, the model map construction module is also used to construct three-plane feature information based on the four depth images;

[0028] Based on the three-plane feature information and the four RGB images of the preset fixed viewing angles, a three-dimensional model map corresponding to the target three-dimensional model is constructed.

[0029] Based on the above technical solution, the four preset fixed viewing angles are 0°, 90°, 180° and 270° respectively.

[0030] Compared with the prior art, the advantages of this application are:

[0031] This application constructs RGB images of multiple preset perspectives and corresponding depth images to construct a three-dimensional model map. The operation process is simple and efficient, with a high degree of refinement and a certain degree of reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 A flowchart of the steps of the three-dimensional model mapping construction method according to an embodiment of the present application;

[0034] Figure 2 This is a structural block diagram of the three-dimensional model mapping construction method of an embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0036] The embodiments of the present application are further described in detail below in conjunction with the accompanying drawings.

[0037] The embodiment of the present application provides a method and device for constructing a three-dimensional model map, which constructs a three-dimensional model map by constructing RGB images of multiple preset perspectives and corresponding depth images. The operation process is simple and efficient, the degree of refinement is high, and it has a certain degree of reliability.

[0038] In order to achieve the above technical effects, the overall idea of ​​this application is as follows:

[0039] A three-dimensional model mapping construction method, the method comprising the following steps:

[0040] S1, constructing four RGB images with preset fixed viewing angles based on the text input information and the 3D white model of the target 3D model;

[0041] S2, obtaining four corresponding depth images based on the geometric structure of the target 3D model and the four RGB images;

[0042] S3. Based on four preset fixed-viewing angle RGB images and corresponding depth images, a three-dimensional model map corresponding to the target three-dimensional model is constructed.

[0043] The embodiments of the present application are further described in detail below in conjunction with the accompanying drawings.

[0044] First, see Figure 1 As shown, the embodiment of the present application provides a method for constructing a three-dimensional model map, the method comprising the following steps:

[0045] S1, constructing four RGB images with preset fixed viewing angles based on the text input information and the 3D white model of the target 3D model;

[0046] S2, obtaining four corresponding depth images based on the geometric structure of the target 3D model and the four RGB images;

[0047] S3. Based on four preset fixed-viewing angle RGB images and corresponding depth images, a three-dimensional model map corresponding to the target three-dimensional model is constructed.

[0048] It should be noted that Consistent-Texture is an innovative 3D model texture generation technology that aims to quickly generate high-quality 3D model textures from simple text prompts. This technology achieves efficient conversion from text descriptions to 3D model textures by combining the latest text-to-image conversion algorithm and Neural Radiance Field (NeRF) technology. The core advantage of Consistent-Texture lies in its fast processing speed and low requirements for input data. It only needs four structured views to generate complex 3D model textures, which greatly simplifies the complexity and time cost of traditional 3D modeling.

[0049] In addition, the technical breakthrough of Consistent-Texture is that it can generate high-quality 3D model maps from only a small number of viewpoints (only four views). This method uses an advanced text-to-image diffusion model to generate consistent and structured view maps. Specifically, a consistent view map is first generated by training a depth map-guided multi-view image diffusion model, and then the color field of NeRF is directly regressed through a transformer-based sparse view texture reconstructor. This method significantly improves the generation speed while maintaining visual quality.

[0050] In general, the main differences between the technical background of Consistent-Texture and the closest existing implementations are processing speed, input data requirements, and the quality of the final generated 3D model map. Consistent-Texture has shown significant advantages in all these aspects and represents an important advancement in the field of 3D model generation.

[0051] Therefore, the main disadvantages faced by the existing existing 3D model texture generation technology include:

[0052] 1. Slow processing speed: Traditional optimization-based NeRF technology and other 3D reconstruction methods usually have slow processing speeds and are not suitable for the needs of quickly generating 3D models.

[0053] 2. High input data requirements: Most existing technologies require a large number of perspective images to generate 3D model maps, which not only increases the difficulty of data collection, but also increases the overall computational burden.

[0054] 3. Unstable generation quality: The conversion process from text to 3D model texture is complex, and existing methods often find it difficult to ensure the consistency and high quality of the generated model.

[0055] 4. Strong reliance on professional knowledge: Existing methods usually require complex geometric construction processes and high requirements on users' professional knowledge.

[0056] In view of the above defects, the technical core of this application, namely, the main advantages of Consistent-Texture are:

[0057] 1. Improve processing speed: Consistent-Texture uses advanced diffusion models and transformer-based sparse view texture reconstructor to significantly improve the speed of generating text to 3D model maps.

[0058] 2. Reduce input data requirements: Consistent-Texture only needs a small number (four) of structured views to generate high-quality 3D model maps, greatly reducing the demand for input data.

[0059] 3. Improve generation quality: Through optimized generation process and algorithm, Consistent-Texture can generate more stable and high-quality 3D model maps.

[0060] 4. Reduce dependence on professional knowledge: Consistent-Texture simplifies the process of generating 3D model textures, making it easy for non-professionals to generate 3D model textures.

[0061] The core steps of the technical solution of the embodiment of the present application are as follows:

[0062] In the first step, the text is input into the multi-view generation network of the first stage to generate four fixed-view rgbimageRGB, that is, RGB images, elevation = 10 degrees, azimuth = 0, 90, 180, 270 degrees.

[0063] In the second step, according to the geometric structure of the 3D model itself, the depth image of the corresponding view is obtained, and the four RGB images generated in the first stage are merged and sent to the second stage network for reconstruction.

[0064] In the embodiment of the present application, RGB images of multiple preset perspectives and corresponding depth images are constructed to further construct a three-dimensional model map. The operation process is simple and efficient, the degree of refinement is high, and it has a certain degree of reliability.

[0065] Further, based on the text input information and the 3D white model of the target 3D model, four RGB images with preset fixed viewing angles are constructed, including the following steps:

[0066] Based on the three-dimensional white model of the target three-dimensional model, four RGB images with preset fixed viewing angles are constructed according to preset elevation angles; wherein,

[0067] The preset elevation angle is 10°.

[0068] For this step, the technical solution of the embodiment of the present application will be based on a diffusion model from text to image during specific implementation. The Consistent-Texture technology uses an advanced depth map-guided multi-view image diffusion model to generate images from text. This model is based on a deep learning algorithm and can understand and convert text descriptions to generate images that match the text content.

[0069] The testing process of this model is as follows:

[0070] (a) The user enters a text description of the model and provides a 3D white model (without texture).

[0071] (b) Based on the white model, four uniform viewing angles with an elevation of 10 degrees and azimuths of 0, 90, 180, and 270 degrees are selected to render the depth map of the model with a resolution of 512x512x3.

[0072] (c) Since the latent diffusion model is used for generation, four latent codes are generated first, each with a size of 64x64x4, and then the code is decoded by the pre-trained VAE Decoder to obtain the image. The specific process of generating latent codes is as shown in step (d).

[0073] (d) First, initialize 4 pure noise latent codes, and then concatenate them with the latent code of the depth map in the channel dimension to form a Tensor. This Tensor is used as the input of the diffusion model UNet, and the step-by-step denoising process of the diffusion model is used to generate a clean latent code (that is, what is needed in step (c)).

[0074] (e) In the process of (d), the user's text will be extracted from the pre-trained CLIP model and added to the forward process of UNet through cross attention.

[0075] In addition, the UNet training process is as follows:

[0076] In the testing process of this model, the trained UNet is needed to generate clean multi-view latent codes from text and depth map as input, and is trained using an internal dataset.

[0077] Further, based on the geometric structure of the target three-dimensional model and the four RGB images, obtaining four corresponding depth images includes the following steps:

[0078] Based on the geometric structure of the target three-dimensional model and the four RGB images, four corresponding depth images are obtained through a preset DINOv2 model.

[0079] Further, based on the four RGB images of the preset fixed viewing angles and the corresponding depth images, constructing a three-dimensional model map corresponding to the target three-dimensional model includes the following steps:

[0080] constructing three-plane feature information based on the four depth images;

[0081] Based on the three-plane feature information and the four RGB images of the preset fixed viewing angles, a three-dimensional model map corresponding to the target three-dimensional model is constructed.

[0082] Furthermore, the four preset fixed viewing angles are 0°, 90°, 180° and 270° respectively.

[0083] In the above process, a sparse view texture reconstructor is specifically used:

[0084] After the text-to-image diffusion model generates views, Consistent-Texture uses a sparse view texture reconstructor to reconstruct the 3D model map from these view textures. The main contents of this part include:

[0085] (1) Transformer-based architecture: The sparse view texture reconstructor adopts the transformer model as its core structure and uses its powerful processing capability to analyze and process the sparse views.

[0086] (2) Mapping process from sparse views to NeRF: The reconstructor analyzes four views and uses NeRF technology to map these two-dimensional information into three-dimensional model maps.

[0087] The test process of the sparse view texture reconstructor is as follows

[0088] (a) The input is a 3D white model (without texture) and four multi-view images generated in the first stage.

[0089] (b) Four multi-view images are passed through the pre-trained DINOv2 model to extract four image embeddings.

[0090] (c) The four image embeddings act on the forward process of the Transformer through cross attention, and the final output of the Transformer is a triplane feature.

[0091] (d) Three-plane features, i.e. three 32x32x84 feature maps;

[0092] 32x32 is the resolution, and 84 is the number of feature channels, which represents the projection features of the texture to be generated in the X, Y, and Z directions.

[0093] (e) In order to obtain the texture of the final 3D model, the 3D coordinates of the texture information are projected onto three planes to obtain the corresponding three features, and the three features are concatenated to obtain a feature with a total length of 252 dimensions. This feature is decoded through an 8-layer MLP to obtain the final three-channel color RGB value.

[0094] (f) Traverse all the coordinates of the texture that needs to be queried to obtain the entire texture information.

[0095] It should be noted that the technical solution of the embodiment of the present application is based on Consistent-Texture. Another notable feature of this technology is its efficient algorithm performance, that is, the improvement in processing speed: Consistent-Texture greatly improves the processing speed by optimizing the algorithm and model structure, and can complete complex three-dimensional models in a short time.

[0096] Furthermore, the above operations also include the Transformer training process:

[0097] (a) During the testing phase, the trained Transformer and MLP are used. In actual operation, internal datasets are used for training.

[0098] (b) During training, a view is randomly sampled to obtain the three-dimensional coordinates of the pixels in the view, and they are used to supervise the training of the transformer and MLP end-to-end.

[0099] It should be noted that the embodiments of the present application also involve the optimization and application of NeRF:

[0100] The NeRF technology in Consistent-Texture has been specially optimized to make it more suitable for the task of generating 3D model maps from sparse views. The specific contents are as follows:

[0101] First, improvement of the NeRF model: Consistent-Texture optimizes the traditional NeRF model and improves its efficiency and accuracy in processing sparse data.

[0102] The key improvement is that instead of using volume rendering, a mesh-based rasterization method is adopted.

[0103] Here are some key features of this technology: Unlike traditional NeRF, which uses volume rendering, the grid-based method relies more on geometric information to render the scene. This method first builds a three-dimensional mesh to represent the geometric structure of the scene, and then converts the mesh into a pixelated image in the rasterization stage. Its main advantages include:

[0104] Efficiency: Since the geometric mesh is used directly, this method can effectively reduce the amount of calculation, especially when processing sparse view data. Rasterization is a highly optimized process that can be accelerated by modern graphics processing hardware (such as GPU). At the same time, by avoiding complex volume rendering calculations, the improved NeRF model reduces the computational complexity of the rendering process while maintaining or improving the quality of the output image.

[0105] Accuracy: Mesh-based methods are able to produce very accurate geometric and textural details, which are crucial for texture reconstruction tasks.

[0106] Second, performance in different application scenarios: The optimized NeRF model shows good adaptability and generation effects in various application scenarios.

[0107] Of course, based on the main idea of ​​the above technical solution, adaptive adjustments can be made during actual implementation, such as:

[0108] Multi-view image synthesis: In addition to using diffusion models to generate sparse views from text, multi-view stereo vision techniques can be employed to synthesize 3D models from images captured from different angles.

[0109] Rule-based modeling: Instead of adopting a transformer-based sparse view reconstructor, a set of rules or scripts can be designed to extract model features from structured views and construct a 3D model.

[0110] Improved datasets and training strategies: The NeRF model’s ability to handle sparse views can be improved by training it with more diverse and challenging datasets.

[0111] In summary, the technical solution of the embodiment of the present application has the following technical advantages compared with the existing technical means:

[0112] First, efficient text-to-image conversion:

[0113] A core innovation of Consistent-Texture is that it can quickly generate structured image views from text descriptions. This process uses an advanced diffusion model that can accurately parse text content and convert it into a visual image that matches the description. The key to this technical point is:

[0114] Application of deep learning algorithms: Using deep learning technology for text understanding and image generation improves the accuracy and efficiency of conversion.

[0115] Generation of structured views: Four structured views are innovatively generated from text, providing complete and consistent perspective information for 3D model mapping.

[0116] Second, innovation of sparse view texture reconstruction:

[0117] Consistent-Texture's ability to reconstruct high-quality 3D model textures based on a small number of views is another key innovation. The key technologies for this include:

[0118] Transformer-based reconstructor design: The transformer architecture is used to process view information, which improves the efficiency and accuracy of conversion from sparse views to 3D model maps.

[0119] Efficient 3D mapping process: Through innovative algorithms, sparse views are efficiently mapped into 3D space to generate accurate 3D model maps.

[0120] Third, optimization of NeRF model:

[0121] Consistent-Texture's optimization of the NeRF model is another core of this technical solution. The specific optimization includes:

[0122] Performance Improvement: The NeRF model has been specially optimized to improve its efficiency and accuracy in handling sparse data.

[0123] Adaptability to application scenarios: The optimized model shows better adaptability and generation effect in different application scenarios.

[0124] Fourth: Improvement of algorithm efficiency and performance:

[0125] Consistent-Texture's improvement in algorithm efficiency and performance is one of its key selling points. The specific improvements are mainly reflected in:

[0126] Fast processing: Algorithm optimization greatly accelerates the conversion process from text to 3D model textures to meet the needs of fast generation.

[0127] High-quality output: Through comprehensive algorithm optimization, the high quality and consistency of the generated model are guaranteed.

[0128] Fifth, expansion of application scope:

[0129] Consistent-Texture has a wide range of applications, including:

[0130] Diverse application scenarios: Applicable to a variety of different fields, such as game design, virtual reality, industrial design, etc.

[0131] Easy-to-use user interface: The user interface is designed to be simple and intuitive, making it easy for non-professional users to use.

[0132] Sixth, simplification of user interface and operation:

[0133] Consistent-Texture's simplified user interface and operation process lower the user threshold.

[0134] In short, the technical solution of the embodiment of the present application has the following technical advantages:

[0135] 1. Fast generation: Consistent-Texture significantly improves the processing speed from text to 3D model maps. Compared with existing technologies, it can generate high-quality 3D model maps more quickly.

[0136] 2. Low input data requirements: Compared with existing technologies that require a large number of perspective images, Consistent-Texture only requires a small number (four) views to generate 3D model maps, greatly reducing the input data requirements.

[0137] 3. High generation quality: The optimization algorithm used by Consistent-Texture can ensure the high quality and consistency of the generated model, and improves the visual effect and detail performance of the model compared to existing technologies.

[0138] 4. Easy to operate: Consistent-Texture's user interface is simple and intuitive, and the operation process is simplified, making it easy for non-professional users to use, lowering the technical threshold.

[0139] 5. Wide range of applications: Consistent-Texture is suitable for a variety of fields, such as game design, virtual reality, etc., providing a wider range of application possibilities.

[0140] Second, see Figure 2 As shown, an embodiment of the present application provides a three-dimensional model mapping construction device, the device comprising:

[0141] A first image construction module, which is used to construct four RGB images of preset fixed viewing angles based on text input information and a three-dimensional white model of a target three-dimensional model;

[0142] A second image construction module, which is used to obtain four corresponding depth images based on the geometric structure of the target three-dimensional model and the four RGB images;

[0143] A model map construction module is used to construct a three-dimensional model map corresponding to the target three-dimensional model based on the four RGB images of the preset fixed viewing angles and the corresponding depth images.

[0144] In the embodiment of the present application, RGB images of multiple preset perspectives and corresponding depth images are constructed to further construct a three-dimensional model map. The operation process is simple and efficient, the degree of refinement is high, and it has a certain degree of reliability.

[0145] Furthermore, the first image construction module is also used to construct four RGB images of preset fixed viewing angles based on the three-dimensional white model of the target three-dimensional model according to preset elevation angles; wherein,

[0146] The preset elevation angle is 10°.

[0147] Furthermore, the second image construction module is also used to obtain four corresponding depth images through a preset DINOv2 model based on the geometric structure of the target three-dimensional model and the four RGB images.

[0148] Furthermore, the model map construction module is also used to construct three-plane feature information based on the four depth images;

[0149] Based on the three-plane feature information and the four RGB images of the preset fixed viewing angles, a three-dimensional model map corresponding to the target three-dimensional model is constructed.

[0150] Furthermore, the four preset fixed viewing angles are 0°, 90°, 180° and 270° respectively.

[0151] It should be noted that the three-dimensional model mapping construction device of the embodiment of the present application and the three-dimensional model mapping construction method mentioned in the first aspect have similar technical principles in terms of technical solutions, technical problems and technical effects, so they will not be elaborated here.

[0152] In the description of the present application, it should be noted that the terms "upper", "lower", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. Unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0153] It should be noted that, in this application, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0154] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest range consistent with the principles and novel features applied for herein.

Claims

1. A method for constructing a three-dimensional model map, characterized in that: The method comprises the following steps: Based on the text input information and the 3D white model of the target 3D model, four RGB images with preset fixed viewing angles are constructed; Based on the geometric structure of the target three-dimensional model and the four RGB images, four corresponding depth images are obtained; Based on the four RGB images of the preset fixed viewing angles and the corresponding depth images, a three-dimensional model map corresponding to the target three-dimensional model is constructed.

2. The three-dimensional model mapping construction method according to claim 1, characterized in that: Based on the text input information and the 3D white model of the target 3D model, four preset fixed-view RGB images are constructed, including the following steps: Based on the three-dimensional white model of the target three-dimensional model, four RGB images with preset fixed viewing angles are constructed according to preset elevation angles; wherein, The preset elevation angle is 10°.

3. The three-dimensional model mapping construction method according to claim 1, characterized in that: Based on the geometric structure of the target three-dimensional model and the four RGB images, four corresponding depth images are obtained, including the following steps: Based on the geometric structure of the target three-dimensional model and the four RGB images, four corresponding depth images are obtained through a preset DINOv2 model.

4. The three-dimensional model mapping construction method according to claim 1, characterized in that: Constructing a three-dimensional model map corresponding to the target three-dimensional model based on the four RGB images of the preset fixed perspectives and the corresponding depth images, comprising the following steps: constructing three-plane feature information based on the four depth images; Based on the three-plane feature information and the four RGB images of the preset fixed viewing angles, a three-dimensional model map corresponding to the target three-dimensional model is constructed.

5. The three-dimensional model mapping construction method according to claim 1, characterized in that: The four preset fixed viewing angles are 0°, 90°, 180° and 270° respectively.

6. A three-dimensional model mapping construction device, characterized in that: The device comprises: A first image construction module, which is used to construct four RGB images of preset fixed viewing angles based on text input information and a three-dimensional white model of a target three-dimensional model; A second image construction module, which is used to obtain four corresponding depth images based on the geometric structure of the target three-dimensional model and the four RGB images; A model map construction module is used to construct a three-dimensional model map corresponding to the target three-dimensional model based on the four RGB images of the preset fixed viewing angles and the corresponding depth images.

7. The three-dimensional model mapping construction device according to claim 6, characterized in that: The first image construction module is also used to construct four RGB images of preset fixed viewing angles based on the three-dimensional white model of the target three-dimensional model according to preset elevation angles; wherein, The preset elevation angle is 10°.

8. The three-dimensional model mapping construction device according to claim 6, characterized in that: The second image construction module is also used to obtain four corresponding depth images through a preset DINOv2 model based on the geometric structure of the target three-dimensional model and the four RGB images.

9. The three-dimensional model mapping construction device according to claim 6, characterized in that: The model map construction module is also used to construct three-plane feature information based on the four depth images; Based on the three-plane feature information and the four RGB images of the preset fixed viewing angles, a three-dimensional model map corresponding to the target three-dimensional model is constructed.

10. The three-dimensional model mapping construction device according to claim 6, characterized in that: The four preset fixed viewing angles are 0°, 90°, 180° and 270° respectively.

Images