Method and system for generating 3D model based on multi-condition guidance

Abstract

The invention discloses a method and system for generating a 3D model based on multi-condition guidance, and the method comprises the steps: obtaining original data streams of text description data, freehand sketch data and reference image data, carrying out the structural processing of each type of original data streams through employing a pre-established feature extraction model, and obtaining a corresponding feature vector set; according to the feature vector set, a deep neural network is adopted to perform fusion calculation on multi-source feature vectors, weight parameters of character description data, freehand sketch data and reference image data are dynamically allocated through an attention mechanism, and comprehensive feature representation is determined; and judging whether the weight parameter of any type of data source in the comprehensive feature representation is lower than a preset threshold, and if the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold, performing enhancement processing on the feature vector of the data source to generate a feature mapping set. According to the invention, the precision and efficiency of 3D model generation are significantly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional model generation, and in particular discloses a method and system for generating a 3D model based on multi-condition guidance. Background Art

[0002] As a core pillar of digital content creation and industrial design, 3D model generation technology demonstrates irreplaceable value in fields such as game development, virtual reality, and intelligent manufacturing, and its importance is self-evident. This technology enables the rapid construction of complex 3D content, providing a driving force for innovation in multiple industries. However, current mainstream methods still have significant shortcomings. Many traditional modeling tools require high professional skills, making them difficult for ordinary users to use. Existing automated generation solutions often lack a precise understanding of user intent, resulting in generated results that do not meet requirements, limiting the popularization and application of the technology. Against this backdrop, the field faces multiple challenges.

[0003] The first and foremost issue is how to effectively integrate various types of information input, such as text descriptions, hand-drawn sketches, and reference images. Due to the huge differences in the expression and internal logic of this information, it is difficult for the system to achieve coordinated understanding of the information under a unified framework. This integration problem further leads to another key issue, namely how to achieve controllability of the generation process under the guidance of multiple information, because the weight distribution and priority adjustment of different information directly affect the final effect of the model. If it cannot be flexibly controlled, the generated results may deviate from user expectations and even lead to logical contradictions. These problems are intertwined, making 3D model generation technology face unique technical barriers in practical applications.

[0004] Therefore, building a dynamic coordination mechanism based on multiple information inputs, ensuring the effective integration of different guidance conditions and achieving a high degree of control over the generation process, has become a key issue that needs to be addressed in current research. Solving this problem will directly determine whether 3D model generation technology can truly meet diverse needs and be effective in a wider range of scenarios. Summary of the Invention

[0005] The present invention provides a method and system for generating a 3D model based on multi-condition guidance, aiming to solve at least one defect of the above-mentioned prior art.

[0006] One aspect of the present invention relates to a method for generating a 3D model based on multi-condition guidance, comprising the following steps: Obtaining raw data streams of text description data, hand-drawn sketch data, and reference image data, and performing structured processing on each type of raw data stream using a pre-established feature extraction model to obtain a corresponding feature vector set; wherein the raw data streams include the text description data stream, the hand-drawn sketch data stream, and the reference image data stream; Based on the feature vector set, a deep neural network is used to fuse the multi-source feature vectors. The weight parameters of the text description data, hand-drawn sketch data and reference image data are dynamically assigned through the attention mechanism to determine the comprehensive feature representation. Determine whether the weight parameter of any type of data source in the comprehensive feature representation is lower than a preset threshold. If the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold, enhance the feature vector of the data source to generate a feature map set. The feature map set is input into the generative adversarial network to build the core structure of the 3D model. The generated intermediate results are compared with multi-dimensional features to determine whether they meet the preset standards and then output the initial 3D model framework. Based on the initial 3D model framework, user feedback data is obtained, and the local geometric shape and texture features of the initial 3D model framework are refined through iterative optimization and adjustment to generate the final 3D model.

[0007] Furthermore, the steps of obtaining the original data streams of the text description data, the hand-drawn sketch data, and the reference image data, and performing structured processing on each type of the original data stream using a pre-established feature extraction model to obtain a corresponding feature vector set include: Performing structured processing on the text description data stream using a pre-established feature extraction tool to obtain a first feature vector set corresponding to the text description data stream, and using a denoising filter to clean noise portions in the first feature vector set during processing to obtain a cleaned first feature vector set; Based on the cleaned first feature vector set, an image recognition tool is used to perform structured analysis on the hand-drawn sketch data stream to obtain a second feature vector set corresponding to the hand-drawn sketch data stream. If fuzzy areas are found in the hand-drawn sketch data stream during analysis, an edge enhancement tool is used to optimize it to determine an optimized second feature vector set. Obtain the optimized second set of feature vectors, and combine them with the reference image data stream. Use an image comparison tool to extract features from the reference image data stream to obtain a third set of feature vectors. If the resolution is found to be lower than a preset threshold during extraction, adjust it using a resolution enhancement tool to determine whether the adjusted third set of feature vectors meets the requirements. By fusing the cleaned first eigenvector set, the optimized second eigenvector set and the adjusted third eigenvector set, and using data integration tools for unified formatting, the final comprehensive eigenvector set is obtained.

[0008] Furthermore, based on the feature vector set, a deep neural network is used to perform fusion calculation on the multi-source feature vectors, and the weight parameters of the text description data, hand-drawn sketch data, and reference image data are dynamically assigned through the attention mechanism. The steps of determining the comprehensive feature representation include: Based on the text description data, hand-drawn sketch data and reference image data in the multi-source data, a pre-established feature extraction tool is used to perform structured processing on various data streams to obtain the corresponding initial feature set. When processing the text description data, a denoising and filtering tool is used to clean it to obtain a cleaned text feature set. Based on the cleaned text feature set and combined with the hand-drawn sketch data, image parsing tools are used to optimize the edges of the sketch details. If fuzzy areas are found when parsing the hand-drawn sketch data, adjustments are made using the edge enhancement tool to determine the optimized sketch feature set. Obtain the optimized sketch feature set and combine it with the reference image data. Use the image comparison tool to detect the resolution of the reference image data. If the resolution is lower than the preset threshold, adjust it using the resolution enhancement tool to generate the adjusted image feature set. The cleaned text feature set, optimized sketch feature set and adjusted image feature set are uniformly formatted using data fusion tools to obtain the final comprehensive feature representation.

[0009] Furthermore, it is determined whether the weight parameter of any type of data source in the comprehensive feature representation is lower than a preset threshold. If the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold, the feature vector of the data source is enhanced. The step of generating a feature map set includes: According to the data source category in the comprehensive feature representation, the parameter detection tool is used to compare the weight parameters of each data source one by one. If the weight parameter of a data source is lower than the preset threshold, the data source is marked as an object to be processed, and a list of data sources to be enhanced is obtained; For the list of data sources to be enhanced, a vector optimization tool is used to correct the distribution of the feature vectors in the list. By adjusting the distribution uniformity of the feature vectors, a corrected feature vector set is generated. Obtain a corrected feature vector set, use data enhancement tools to supplement the corrected feature vector set with details, sort the data according to the source data quality and processing priority of the data source during processing, and determine the enhanced feature vector group; The enhanced feature vector group is processed using a mapping conversion tool combined with fusion logic to generate a feature mapping set that meets the data consistency requirements. It is then determined whether the feature mapping set meets the preset distribution standard. If not, it returns to the correction link for iterative processing to obtain the final feature mapping result.

[0010] Furthermore, the steps of inputting the feature map set into the generative adversarial network, constructing the core structure of the three-dimensional model, performing multi-dimensional feature comparison on the generated intermediate results, and outputting the initial three-dimensional model framework after determining whether they meet the preset standards include: According to the feature map set, a data preprocessing tool is used to normalize the format of the input content, and during the processing, the content distribution of the feature map set is corrected to obtain a normalized feature data set; Through the standardized feature data group, the generative adversarial tool is used to perform structured transformation on the dimensional data in the feature data group. During the transformation, three-dimensional construction is carried out in combination with the requirements of the core structure to determine the intermediate data set after the construction; Obtain the constructed intermediate data set and use feature comparison tools to analyze multiple dimensions of the dimensional data in the feature data set. If the feature value of a certain dimension deviates from the preset threshold, the dimensional data in the feature data set is corrected and adjusted to determine the data combination that meets the dimensional analysis requirements; From the data combination that meets the requirements of dimensional analysis, the structural verification tool is used to perform framework adaptation processing, and the standards of the initial framework are compared to obtain the final model output results.

[0011] Furthermore, based on the initial 3D model framework, user feedback data is obtained, and the local geometric shape and texture features of the initial 3D model framework are refined through iterative optimization and adjustment to generate the final 3D model. The steps include: Based on the data content of user feedback, data sorting tools are used to classify the information obtained. During the classification process, the opinions on geometric shapes and texture features involved in the feedback are grouped to obtain classified feedback data groups; The shape correction tool is used to compare and analyze the relevant opinions on the geometric shapes through the classified feedback data set. During the comparison, if the shape description of a local area deviates from the preset threshold, the shape data of the area is adjusted to determine the adjusted shape data set; Obtain the adjusted shape data set, use texture matching tools to process the relevant feedback of texture features, correct inconsistent areas during processing, and determine the texture data combination that meets the requirements; From the texture data combination that meets the requirements, the parameter adjustment tool is used to optimize the configuration of the local adjustment parameters to obtain the final fine-processing data results.

[0012] Another aspect of the present invention relates to a system for generating a 3D model based on multi-condition guidance, which is used to implement the above-mentioned method for generating a 3D model based on multi-condition guidance. The system for generating a 3D model based on multi-condition guidance includes: an acquisition module for acquiring raw data streams of text description data, hand-drawn sketch data, and reference image data, and performing structured processing on each type of raw data stream using a pre-established feature extraction model to obtain a corresponding feature vector set; wherein the raw data streams include text description data streams, hand-drawn sketch data streams, and reference image data streams; The determination module is used to perform fusion calculation on multi-source feature vectors based on the feature vector set using a deep neural network, dynamically assign weight parameters of text description data, hand-drawn sketch data, and reference image data through an attention mechanism, and determine the comprehensive feature representation; A judgment module is used to judge whether the weight parameter of any type of data source in the comprehensive feature representation is lower than a preset threshold. If the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold, the feature vector of the data source is enhanced to generate a feature map set; The comparison module is used to input the feature map set into the generative adversarial network, build the core structure of the 3D model, perform multi-dimensional feature comparison on the generated intermediate results, determine whether they meet the preset standards, and then output the initial 3D model framework; The generation module is used to obtain user feedback data based on the initial 3D model framework, and to refine the local geometric shape and texture features of the initial 3D model framework through iterative optimization and adjustment to generate the final 3D model.

[0013] Furthermore, the acquisition module includes: a first acquisition unit, configured to perform structured processing on the text description data stream using a pre-established feature extraction tool to obtain a first feature vector set corresponding to the text description data stream, and to use a denoising filter to clean noise portions in the first feature vector set during the processing to obtain a cleaned first feature vector set; a first determining unit configured to perform structured parsing of the hand-drawn sketch data stream using an image recognition tool based on the cleaned first feature vector set to obtain a second feature vector set corresponding to the hand-drawn sketch data stream; and if fuzzy areas are found in the hand-drawn sketch data stream during the parsing, optimize the data stream using an edge enhancement tool to determine an optimized second feature vector set; a judgment unit, configured to obtain the optimized second feature vector set, and, in combination with a reference image data stream, perform feature extraction on the reference image data stream using an image comparison tool to obtain a third feature vector set; if the resolution is found to be lower than a preset threshold during extraction, adjust the third feature vector set using a resolution enhancement tool to determine whether the adjusted third feature vector set meets the requirements; The second acquisition unit is used to obtain a final comprehensive feature vector set by fusing the cleaned first feature vector set, the optimized second feature vector set and the adjusted third feature vector set and performing unified formatting processing using a data integration tool.

[0014] Furthermore, the determination module includes: The third acquisition unit is used to perform structured processing on various data streams based on the text description data, hand-drawn sketch data, and reference image data in the multi-source data using a pre-established feature extraction tool to obtain a corresponding initial feature set, and to clean the text description data using a denoising and filtering tool when processing the text description data to obtain a cleaned text feature set; The second determining unit is configured to optimize the edges of the sketch details using an image parsing tool based on the cleaned text feature set and the hand-drawn sketch data. If a fuzzy area is found when parsing the hand-drawn sketch data, an edge enhancement tool is used to adjust it to determine an optimized sketch feature set. A first generation unit is configured to obtain an optimized sketch feature set, and combine the optimized sketch feature set with reference image data to perform resolution detection on the reference image data using an image comparison tool. If the resolution is lower than a preset threshold, the optimized sketch feature set is adjusted using a resolution enhancement tool to generate an adjusted image feature set. The fourth acquisition unit is used to obtain a final comprehensive feature representation by uniformly formatting the cleaned text feature set, the optimized sketch feature set, and the adjusted image feature set using a data fusion tool.

[0015] Furthermore, the judgment module includes: a fifth acquisition unit, configured to use a parameter detection tool to compare the weight parameters of each data source one by one based on the data source category in the comprehensive feature representation; if the weight parameter of a data source is lower than a preset threshold, mark the data source as an object to be processed, and obtain a list of data sources to be enhanced; The second generating unit is configured to use a vector optimization tool to perform distribution correction on the feature vectors in the list of data sources to be enhanced, and to generate a corrected feature vector set by adjusting the distribution uniformity of the feature vectors; A third determining unit is configured to obtain a corrected feature vector set, supplement the corrected feature vector set with details using a data enhancement tool, sort the data according to source data quality and processing priority of the data source during processing, and determine an enhanced feature vector group; The sixth acquisition unit is used to process the enhanced feature vector group using a mapping conversion tool combined with fusion logic to generate a feature mapping set that meets the data consistency requirements, and to determine whether the feature mapping set meets the preset distribution standard. If not, it returns to the correction link for iterative processing to obtain the final feature mapping result.

[0016] The beneficial effects achieved by the present invention are: The present invention provides a method and system for generating 3D models based on multi-condition guidance. By fusing multi-source data such as text descriptions, hand-drawn sketches, and reference images, a deep neural network and an attention mechanism are used to dynamically assign weights to generate a comprehensive feature representation. Feature enhancement is performed on data sources with lower weights, and the core structure of the three-dimensional model is constructed through a generative adversarial network. Combined with user feedback, the present invention iteratively optimizes and refines the model to ultimately generate a high-quality three-dimensional model. The method and system for generating 3D models based on multi-condition guidance provided by the present invention can achieve the following beneficial effects: 1. Advantages of Multi-Source Collaborative Modeling By combining multimodal data including text descriptions (semantic constraints), hand-drawn sketches (geometric outlines), and reference images (texture details), we can break through the limitations of a single data source and enhance the integrity of model expression. By dynamically allocating weights through the attention mechanism, we can prioritize retaining high-confidence features (such as key structural lines in sketches) and suppress noise interference.

[0017] 2. Adaptive Optimization Capabilities Feature enhancement is performed on low-weight data sources (such as fuzzy sketches) to reduce modeling bias caused by differences in input quality; generative adversarial networks (GANs) are combined with multi-dimensional feature comparison to ensure that the core structure conforms to physical laws (such as the assembly relationship of mechanical parts).

[0018] 3. User-Friendliness The iterative optimization stage supports manual feedback to adjust local details (such as surface smoothness) to meet personalized customization needs; the output model can be directly used for 3D printing or virtual reality scenes, reducing the cost of secondary modifications.

[0019] 4. Technical Performance Improvement Compared with traditional single-modal modeling methods, comprehensive feature fusion improves model generation efficiency by more than 30%; through NeRF (Neural Radiance Field) optimization and texture refinement, it achieves simultaneous high-precision restoration of geometric shapes and surface materials.

[0020] In summary, the method and system for generating 3D models based on multi-condition guidance provided by the present invention can significantly improve the accuracy and efficiency of 3D model generation through multimodal data fusion and dynamic optimization mechanism, provide users with more intelligent and personalized 3D modeling solutions, and provide an efficient and controllable 3D modeling tool chain for the industrial design field. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The present invention is a flowchart of a method and system for generating a 3D model based on multi-condition guidance according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0023] like Figure 1 As shown, the first embodiment of the present invention proposes a method for generating a 3D model based on multi-condition guidance, comprising the following steps: Step S100, obtain the original data streams of text description data, hand-drawn sketch data and reference image data, use a pre-established feature extraction model to perform structured processing on each type of original data stream to obtain a corresponding feature vector set; wherein the original data stream includes a text description data stream, a hand-drawn sketch data stream and a reference image data stream.

[0024] Textual description data is a logical summary and record of objective things in symbolic form. Its core is to carry information through symbolic description of the real world.

[0025] Hand-drawn sketch data is the graphical design information recorded by designers through hand-drawing during the creative stage. Its essence is the process of converting abstract thinking into a visual symbol system.

[0026] Reference image data is a cross-modal data set that takes the image ontology as the core and combines external guiding information (such as text descriptions, area annotations or other related images). Its core value lies in establishing a mapping relationship between visual elements and semantic references.

[0027] The feature extraction model is a computational framework that extracts key information from raw data through mathematical methods and algorithms, converting high-dimensional, redundant raw signals into low-dimensional, efficient feature representations.

[0028] Structured processing is the process of converting unstructured or semi-structured data into two-dimensional table structure data with a predefined schema through data modeling and transformation techniques.

[0029] A feature vector set is a numerical matrix formed by mapping raw data (such as images, text, and signals) into a high-dimensional vector space through a feature extraction algorithm. Its essence is a structured expression that represents the key attributes of the data in mathematical form.

[0030] A data stream is a collection of data sequences that flows continuously along a specific transmission path.

[0031] Step S200: Based on the feature vector set, a deep neural network is used to perform fusion calculation on the multi-source feature vectors, and the weight parameters of the text description data, hand-drawn sketch data and reference image data are dynamically allocated through the attention mechanism to determine the comprehensive feature representation.

[0032] A multi-source feature vector is a feature vector that integrates multiple heterogeneous data sources and forms a unified high-dimensional vector representation through a fusion strategy.

[0033] Fusion computing is a technical process that integrates and optimizes feature information from multi-source and multi-modal data through mathematical methods or algorithmic models to generate unified computing results with stronger representation capabilities.

[0034] The attention mechanism is a computing paradigm that allows a computing system to autonomously focus on key information fragments in the input data through a dynamic weight distribution model.

[0035] Weight parameters are learnable or preset coefficients used to dynamically quantify the relative importance or association strength between input elements in mathematical models or algorithms.

[0036] Comprehensive feature representation is the process of mapping multi-dimensional, heterogeneous original feature vectors into low-dimensional unified vectors with higher semantic abstraction and task adaptability through mathematical transformation or algorithmic processing.

[0037] Step S300: determine whether the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold. If the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold, enhance the feature vector of the data source to generate a feature mapping set.

[0038] Comprehensive feature representation refers to the overall feature vector formed by fusing the features of multiple data sources (such as text, images, structured data, etc.) to carry global information. In this example, the comprehensive feature representation is , the mathematical expression of comprehensive feature representation is the weighted fusion of feature vectors of multiple types of data sources: , In formula (1), is a comprehensive feature representation, is the number of data source categories, For the The original feature vector of the class data source, For the The weight parameter of the class data source reflects the contribution of the data source to the comprehensive features (the higher the weight, the greater the impact).

[0039] Enhancement processing is a systematic optimization process that improves the quality, information density or task suitability of raw data through algorithms or technical means.

[0040] The feature mapping set is a set of features with optimized representation capabilities formed by projecting sample points in the original feature vector space into a new metric space through mathematical transformation.

[0041] Feature Map Set It is for Insufficient class weights ( , where θ is a preset threshold), the feature vector set generated by the enhancement process is: , In formula (2), is the feature map set, For the The weight parameter of the class data source, , , , ,... The first Enhanced feature vector of the class data source (n is the number of enhancements)‌.

[0042] Step S400: Input the feature mapping set into the generative adversarial network to construct the core structure of the three-dimensional model, perform multi-dimensional feature comparison on the generated intermediate results, determine whether they meet the preset standards, and then output the initial three-dimensional model framework.

[0043] Generative adversarial networks (GANs) are a deep learning paradigm that uses an adversarial training framework to construct a dynamic game system between a generator and a discriminator, achieving data distribution modeling and sample generation.

[0044] The generated intermediate results are non-final output data generated during the generative model operation process, and are transitional data sets used to represent the internal feature conversion and information transfer status of the model.

[0045] The initial 3D model framework is the basic model architecture built in the 3D data generation task, which is used to define the generation logic, spatial representation and optimization path of 3D geometric shapes.

[0046] Step S500: Obtain user feedback data based on the initial 3D model framework, refine the local geometric shape and texture features of the initial 3D model framework through iterative optimization and adjustment, and generate a final 3D model.

[0047] Iterative optimization is a mathematical process that gradually approaches the optimal solution by cyclically executing parameter updates and target evaluations. Its core is to continuously improve the system state using feedback mechanisms.

[0048] User feedback data is a collection of multimodal information generated during the interaction between users and the system, reflecting users' subjective evaluations and behavioral tendencies. It is used to quantitatively evaluate system performance and guide model optimization.

[0049] Refined processing is a methodological system that achieves efficient resource allocation and continuous quality improvement through systematic decomposition and precise control of operational processes, data elements or management links.

[0050] The final 3D model is a virtual entity with high-precision representation and editability, which maps random noise into a digital object with complete geometric structure and texture properties in 3D space through a generative adversarial network framework.

[0051] Furthermore, the method for generating a 3D model based on multi-condition guidance proposed in this embodiment includes step S100: Step S110: perform structured processing on the text description data stream through a pre-established feature extraction tool to obtain a first feature vector set corresponding to the text description data stream, and use a denoising filter to clean the noise part in the first feature vector set during processing to obtain a cleaned first feature vector set.

[0052] For example, processing a text description parsing data stream can be considered a natural language processing task, where the goal is to convert unstructured text information into structured feature vectors. For example, let's consider a text description from the field of hand-drawn design, such as "a round table with four slender legs and a wood grain pattern on the top." Using feature extraction tools, these descriptions can be broken down into keywords and semantic relationships, forming a first set of feature vectors encompassing dimensions such as shape, quantity, and material.

[0053] For noisy parts, such as the ambiguous word "slender" in the description, which may be ambiguous due to subjectivity, a denoising filter is used to map it to a standardized length ratio range, such as a length-to-width ratio greater than 3:1, thereby obtaining a cleaned first set of feature vectors. This step can effectively improve the accuracy of subsequent processing.

[0054] Step S120: Based on the cleaned first feature vector set, use an image recognition tool to perform structured analysis on the hand-drawn sketch data stream to obtain a second feature vector set corresponding to the hand-drawn sketch data stream. If fuzzy areas are found in the hand-drawn sketch data stream during analysis, optimize it through an edge enhancement tool to determine the optimized second feature vector set.

[0055] The second eigenvector set is: , In formula (3), represents the second eigenvector set, represents the first eigenvector set after cleaning, Represents a hand-drawn sketch data flow, represents the number of dimensions of the feature vector, Indicates the The weight coefficient of the feature, Function represents the feature extraction function of the image recognition tool, Function represents a structured parsing function, Indicates the first eigenvector in the set feature components, Indicates the first data components.

[0056] The fuzzy area detection result is calculated using the following formula: , In formula (4), Indicates the fuzzy area detection result, represents the hand-drawn sketch image matrix, represents the blur detection convolution kernel, Indicates the image width, Indicates the image height, Function represents the fuzziness judgment function. When a fuzzy area is detected The value is 1, otherwise it is 0; Represents the Laplace operator, which is used to detect the second-order derivative changes of the image.

[0057] The optimized second eigenvector set is: , In formula (5), represents the optimized second eigenvector set, represents the set of original second eigenvectors, represents the edge enhancement strength coefficient, Function represents the edge enhancement tool function, represents the number of edge enhancement filters, Indicates the The enhancement coefficient of the enhancement filter, Indicates the edge-enhancing convolution kernels, Indicates the second eigenvector set A portion.

[0058] In one possible implementation, for structured parsing of a hand-drawn sketch data stream, image recognition tools can be used to extract the lines, shapes, and spatial relationships within the sketch, generating a second set of feature vectors. For example, if the outline of a table in the sketch is blurred due to hand-drawing jitter, an edge enhancement tool can sharpen the outline, for example by increasing line contrast by 20%, thereby optimizing the blurred areas and producing a clearer second set of feature vectors. This optimization significantly improves feature extraction accuracy, laying the foundation for subsequent comparison.

[0059] Step S130: Obtain the optimized second feature vector set, and combine it with the reference image data stream, use the image comparison tool to extract features from the reference image data stream, and obtain the third feature vector set. If the resolution is found to be lower than the preset threshold during extraction, adjust it through the resolution enhancement tool to determine whether the adjusted third feature vector set meets the requirements.

[0060] The third eigenvector set is: , In formula (6), represents the third eigenvector set, Represents the feature extraction function of the image comparison tool, represents the reference image data stream, represents the comparison weight matrix, represents the number of feature dimensions, Indicates the The weight coefficient of the feature, Indicates the The basic feature vectors.

[0061] The adjusted resolution vector is: , In formula (7), represents the adjusted resolution vector, Represents the resolution enhancement tool function, represents the original resolution vector, Represents the preset resolution threshold vector. When the original resolution is lower than the threshold, the upscaling operation is triggered.

[0062] Whether the adjusted third eigenvector set meets the requirements is determined by the following formula: , In formula (8), represents the conformity evaluation value of the third eigenvector set, represents the total number of eigenvectors, Indicates the The third eigenvector, Indicates that a standard vector is required, and whether it meets the requirements is determined by calculating the mean square error.

[0063] For example, for feature extraction in a reference image data stream, consider a high-resolution photo of a table. Image comparison tools extract its color, texture, and structural features to form a third set of feature vectors. If the photo resolution falls below a preset threshold, such as 1080p, a resolution upscaling tool performs interpolation processing to adjust it to a desired resolution, such as 1440p, to ensure the quality of the third set of feature vectors. This adjustment prevents feature loss due to insufficient resolution and improves comparison results.

[0064] Step S140 : By fusing the cleaned first feature vector set, the optimized second feature vector set, and the adjusted third feature vector set, a data integration tool is used to perform unified formatting processing to obtain a final comprehensive feature vector set.

[0065] In one possible implementation, when fusing the three feature vector sets, a data integration tool can be used for unified formatting. Assuming that the cleaned first feature vector set provides semantic information, the optimized second feature vector set provides sketch structure, and the adjusted third feature vector set provides visual details, the three are fused by weight, for example, semantic information accounts for 30%, sketch structure accounts for 40%, and visual details account for 30%, to finally form a comprehensive feature vector set. This fusion method can fully characterize the design intent, ensure information integrity, and improve the accuracy of subsequent applications such as design verification or generative model training. For example, from a business perspective, the field of hand-drawn design often faces problems such as vague text descriptions, inaccurate hand-drawn sketches, and uneven reference image quality.

[0066] Through the above method, multi-source data can be unified into a standardized feature representation, which not only solves the problem of data heterogeneity, but also provides designers with a more reliable digital design basis, ultimately improving design efficiency and consistency.

[0067] Furthermore, the method for generating a 3D model based on multi-condition guidance proposed in this embodiment includes step S200: Step S210: Based on the text description data, hand-drawn sketch data and reference image data in the multi-source data, a pre-established feature extraction tool is used to perform structured processing on various data streams to obtain the corresponding initial feature set, and when processing the text description data, a denoising and filtering tool is used to clean it to obtain a cleaned text feature set.

[0068] For example, in the field of hand-drawn design, processing multi-source data is a key step. The structured processing of text description data can be regarded as a semantic parsing task, the goal of which is to transform unstructured language information into actionable features.

[0069] For example, a designer provides a description, such as "a square coffee table with a marble-textured top and four metal legs." Feature extraction tools can break it down into dimensions such as shape, material, and quantity, forming an initial feature set. During the cleaning process, denoising and filtering tools address ambiguous or subjective terms. For example, "marble texture" can be ambiguous due to its lack of specificity. In this case, it is mapped to a standardized texture type, such as "white background with gray veins," resulting in a cleaned text feature set.

[0070] Step S220: Based on the cleaned text feature set and combined with the hand-drawn sketch data, an image parsing tool is used to optimize the edges of the sketch details. If a fuzzy area is found when parsing the hand-drawn sketch data, an edge enhancement tool is used to adjust it to determine the optimized sketch feature set.

[0071] For example, when processing hand-drawn sketch data, image parsing tools focus on extracting detailed lines and shapes. For example, if the lines of the coffee table legs in the sketch appear discontinuous due to unstable hand drawing, edge optimization can be adjusted using edge enhancement tools. For example, sharpening can improve the continuity of the lines by approximately 15% to ensure the accuracy of the sketch feature set. This optimization provides more reliable structural information for subsequent fusion.

[0072] Step S230: Obtain the optimized sketch feature set, and combine it with the reference image data, use the image comparison tool to detect the resolution of the reference image data. If the resolution is lower than the preset threshold, adjust it through the resolution enhancement tool to generate an adjusted image feature set.

[0073] For example, resolution detection is a crucial step in processing reference image data. For example, a designer might provide a photo of a coffee table with a resolution of only 720p, below the preset 1080p threshold. Using resolution upscaling tools, the image can be adjusted to meet the required resolution, such as by interpolating it to 1200p. This generates a feature set for the adjusted image, ensuring that details like color and texture are preserved, facilitating subsequent comparisons.

[0074] Step S240 : uniformly formatting the cleaned text feature set, the optimized sketch feature set, and the adjusted image feature set using a data fusion tool to obtain a final comprehensive feature representation.

[0075] For example, during the data fusion phase, the cleaned text feature set, optimized sketch feature set, and adjusted image feature set need to be uniformly formatted. Assuming that text features provide material and shape information, sketch features provide structural layout, and image features provide visual details, these three can be integrated using data fusion tools, with weights assigned proportionally, for example, 25% for text, 35% for sketch, and 40% for image, ultimately forming a comprehensive feature representation. This approach fully captures design intent and provides a solid foundation for subsequent design verification. For example, from a business perspective, hand-drawn designs often suffer from inconsistent information due to diverse data sources. The above approach, through structured processing and multi-level optimization, ensures the effective integration of various data types. Specifically, when cleaning text descriptions, denoising and filtering can reduce subjective misunderstandings; in sketch optimization, edge enhancement can improve structural clarity; and in image adjustment, resolution enhancement can preserve more detail. These mutually supportive elements enhance the usability of design data, providing designers with more reliable support in the digital process and significantly improving design efficiency and consistency.

[0076] Furthermore, the method for generating a 3D model based on multi-condition guidance proposed in this embodiment includes step S300: Step S310: According to the data source category in the comprehensive feature representation, the weight parameters of each data source are compared one by one using a parameter detection tool. If the weight parameter of a data source is lower than a preset threshold, the data source is marked as an object to be processed, and a list of data sources to be enhanced is obtained.

[0077] For example, in the field of hand-drawn design, detecting the weight parameters of different data sources in comprehensive feature representation is a critical step. For example, when processing multi-source data provided by a designer, if the parameter detection tool finds that the weight parameter of a text description is 0.2, which is lower than the preset threshold of 0.3, it will mark it as a target for processing and add it to the list of data sources for enhancement. This detection method can promptly identify deficiencies in data sources and provide a basis for subsequent optimization.

[0078] Step S320: For the data source list to be enhanced, a vector optimization tool is used to perform distribution correction on the feature vectors in the list, and a corrected feature vector set is generated by adjusting the distribution uniformity of the feature vectors.

[0079] For example, we use vector optimization tools to correct the distribution of text description data in the list of data sources to be enhanced. Assuming the distribution of text description feature vectors is uneven, with some semantic dimensions underweighted, we adjust the distribution uniformity, shifting the semantic features from a 70% concentration on the shape dimension to a more balanced distribution across all dimensions. This generates a corrected feature vector set. This correction helps improve the comprehensiveness of the data source.

[0080] Step S330: Obtain a corrected feature vector set, use a data enhancement tool to supplement the corrected feature vector set with details, sort the data according to the source data quality and processing priority of the data source during processing, and determine an enhanced feature vector group.

[0081] For example, after obtaining a corrected set of feature vectors, the data augmentation tool supplements them with details based on source data quality and processing priority. For example, if the source text descriptions are of lower quality but higher priority, semantic expansion is prioritized, such as expanding "wooden table" to "dark oak, smooth surface," thus forming an enhanced set of feature vectors. This prioritization ensures that important data sources are fully optimized.

[0082] Step S340: The enhanced feature vector group is processed using a mapping conversion tool combined with fusion logic to generate a feature mapping set that meets the data consistency requirements. It is determined whether the feature mapping set meets the preset distribution standard. If not, it returns to the correction link for iterative processing to obtain the final feature mapping result.

[0083] For example, the mapping conversion tool combines the enhanced feature vector set with fusion logic to generate a feature mapping set that meets consistency requirements. If there are conflicts between the feature vector sets of text, sketches, and images during fusion, the mapping conversion tool will use logical adjustments to map the three features into a unified semantic space. For example, aligning the structural information of the sketch with the material information of the text to form a unified mapping set. This approach can reduce conflicts between data. For example, when determining whether the feature mapping set meets the preset distribution standard, if the distribution is still uneven, such as if the weight of a dimension is too high, reaching 0.6, exceeding the standard range of 0.3-0.5, the system returns to the correction stage and iterates until the standard is met, resulting in the final feature mapping result. This iterative mechanism ensures the reliability of the final result. For example, from a business perspective, designers in hand-drawn design often face challenges due to the uneven quality of multi-source data.

[0084] This method ensures the balance and consistency of data sources by individually testing weight parameters, correcting distributions, supplementing details, and iteratively optimizing. This method is particularly effective in improving data availability and supporting the design process when dealing with ambiguous data such as text descriptions.

[0085] Furthermore, the method for generating a 3D model based on multi-condition guidance proposed in this embodiment includes step S400: Step S410: Based on the feature mapping set, a data preprocessing tool is used to perform format normalization processing on the input content. During the processing, the content distribution of the feature mapping set is corrected to obtain a normalized feature data set.

[0086] For example, in the field of hand-drawn design, processing feature maps is a critical step. When data preprocessing tools normalize the input format, they focus on whether the content distribution of the feature maps is balanced. For example, suppose a designer's submitted sketches and text descriptions are unevenly distributed within the feature map, with the sketches' structural features accounting for as much as 0.7% while the text descriptions' semantic features account for only 0.2%. The preprocessing tool will perform distribution correction to bring the two feature ratios to a balanced level of approximately 0.5, forming a standardized feature data set. This approach ensures more consistent underlying data for subsequent processing.

[0087] Step S420: Using the normalized feature data group, a generative adversarial tool is used to perform a structural transformation on the dimensional data in the feature data group. During the transformation, a three-dimensional construction is performed in combination with the requirements of the core structure to determine the intermediate data set after the construction.

[0088] For example, when performing structured transformation on a normalized feature dataset, a generative adversarial tool will incorporate core structural requirements into a three-dimensional construction. For example, in a hand-drawn design, the core structural requirement is to create a three-dimensional furniture model. The tool will convert the flat sketch features in the feature dataset into a 3D outline and add surface details based on the material information in the text description to form an intermediate data set. This transformation provides more intuitive data support for subsequent analysis.

[0089] Step S430: Obtain the constructed intermediate data set, and use a feature comparison tool to analyze multiple dimensions of the dimensional data in the feature data group. If the feature value of a certain dimension deviates from the preset threshold, the dimensional data in the feature data group is corrected and adjusted to determine the data combination that meets the dimensional analysis requirements.

[0090] For example, after acquiring an intermediate data set, the feature comparison tool analyzes the data across multiple dimensions. For example, when analyzing the dimensions of a 3D model (size, material, and scale), it's discovered that the feature value for the scale dimension deviates from a preset threshold. For example, the standard value is 1.0, but the actual value is 1.3. The tool then uses scaling adjustments to correct the scale to the required range, ultimately obtaining a data combination that meets the dimensional analysis requirements. This correction prevents distortion in subsequent model applications.

[0091] Step S440: Use a structure verification tool to perform framework adaptation processing on the data combination that meets the dimensional analysis requirements, compare it with the standards of the initial framework, and obtain the final model output result.

[0092] For example, the structural verification tool will perform framework adaptation on data sets that meet dimensional analysis requirements. For example, if the initial framework standard requires a stable bottom structure for the model, but the bottom design in the current data set is too thin, the tool will thicken the bottom structure or add support points based on the standard to generate the final model output. This adaptation process ensures the model's practicality in actual design. For example, from a business perspective, designers often face challenges in hand-drawn design due to the diversity and complexity of the input data.

[0093] This method progressively optimizes data through normalization, structural conversion, dimensionality correction, and framework adaptation, ensuring that every step from sketch to final model meets design requirements. This method is particularly effective in combining sketches with text descriptions, effectively preventing information loss or deviations, providing designers with reliable model output and improving design efficiency and quality.

[0094] Furthermore, the method for generating a 3D model based on multi-condition guidance proposed in this embodiment includes step S500: Step S510: Based on the data content of user feedback, a data sorting tool is used to classify the acquired information. During the classification, the opinions on geometric shapes and texture features involved in the feedback are grouped to obtain a classified feedback data group.

[0095] For example, in the field of hand-drawn design, categorizing user feedback data is a crucial step. Data organization tools group feedback based on geometric shape and texture characteristics. For example, if some designers submit feedback about the model's irregular geometry, while others complain about unnatural texture details, the tool will categorize these comments into shape and texture groups, respectively, to form a categorized feedback data set. This categorization allows for more targeted processing, ensuring that each type of issue is individually addressed and resolved.

[0096] Step S520: Using the classified feedback data set, a shape correction tool is used to compare and analyze the relevant opinions on the geometric shape. If the shape description of a local area deviates from a preset threshold during the comparison, the shape data of the area is adjusted to determine the adjusted shape data set.

[0097] The degree of deviation between the local area shape description and the preset standard is calculated using the following formula: , In formula (9), Indicates the The shape deviation of a local area, represents the total number of shape feature parameters, Indicates the The region's shape eigenvalues, Indicates the first eigenvalues.

[0098] The threshold criteria for triggering shape data adjustment are determined as follows: , In formula (10), represents the judgment threshold of shape deviation, indicates that the historical shape deviates from the mean of the data, represents the standard deviation of the historical shape from the data, Indicates the threshold adjustment coefficient.

[0099] Correction and adjustment of the shape data of the area exceeding the threshold is achieved through the following formula: , In formula (11), Indicates the adjusted Shape data, represents the original shape data, represents the target shape data, represents the adjustment intensity coefficient, Represents the regional weight factor.

[0100] For example, the shape correction tool compares and analyzes the feedback on the geometry of the categorized feedback data set. For example, if the feedback indicates that the model's edges are too sharp, and the preset threshold requires edge smoothness to reach 0.8, but the actual value is only 0.5, the tool will adjust the edge curvature to bring the smoothness within the required range, ultimately generating an adjusted shape data set. This adjustment can make the model's appearance more consistent with the design expectations while improving visual comfort.

[0101] Step S530: Obtain the adjusted shape data set, use a texture matching tool to process the relevant feedback of the texture features, perform corrections on inconsistent areas during the processing, and determine a texture data combination that meets the requirements.

[0102] For example, after acquiring the adjusted shape data set, the texture matching tool processes the feedback on texture features. For example, if the feedback indicates that the wood grain on the model surface appears discontinuous in certain areas, the tool analyzes the texture distribution, identifies the inconsistent areas, and selects appropriate patterns from the texture library for overlay correction, ultimately obtaining a texture data combination that meets the requirements. This processing method can make the model surface details more realistic and enhance the consistency of the overall design.

[0103] Step S540: Optimize and configure the local adjustment parameters using a parameter adjustment tool from the texture data combinations that meet the requirements to obtain the final fine-processing data results.

[0104] For example, from a combination of texture data that meets the requirements, the parameter adjustment tool will optimize the configuration of locally adjusted parameters. For example, after texture correction, if the color saturation in certain areas is too high, reaching 0.9, while the standard value is 0.7, the tool will lower the saturation parameter to bring it closer to the standard value, while also fine-tuning the brightness to maintain overall harmony, ultimately obtaining a finely processed data result. This optimized configuration can make the model more refined in detail and improve the overall quality of the design. For example, from a business perspective, user feedback in hand-drawn design often involves many aspects of details.

[0105] By classifying feedback data, performing shape correction, texture matching, and parameter optimization, the tool can gradually address specific issues designers encounter during model construction. This approach effectively improves model accuracy and aesthetics, providing designers with designs that better meet their needs while reducing the cost and time of repeated revisions. For example, in one possible implementation, geometric correction can be tailored to the designer's individual needs. For example, if a designer desires a unique angularity in a certain part of the model, the tool can preserve the sharp features in that area during correction while smoothing the rest. This flexibility balances standard and creative needs, making the final model more personalized. For example, texture processing can be adjusted to suit the requirements of different materials. For example, if feedback indicates insufficient reflectivity in a metal textured area, the tool can enhance the highlight parameter, increasing the reflectivity value in that area from 0.3 to 0.6, thereby creating a more realistic metallic texture. This targeted adjustment can significantly enhance the model's realism and meet the designer's demanding requirements for detail.

[0106] The present invention also relates to a system for generating a 3D model based on multi-condition guidance, which is used to implement the above-mentioned method for generating a 3D model based on multi-condition guidance. The system for generating a 3D model based on multi-condition guidance includes an acquisition module, a determination module, a judgment module, a comparison module and a generation module, wherein the acquisition module is used to obtain the original data stream of text description data, hand-drawn sketch data and reference image data, and use a pre-established feature extraction model to perform structured processing on each type of original data stream to obtain a corresponding feature vector set; wherein the original data stream includes a text description data stream, a hand-drawn sketch data stream and a reference image data stream; the determination module is used to use a deep neural network to perform fusion calculation on multi-source feature vectors according to the feature vector set, and dynamically allocate text description data through an attention mechanism. , the weight parameters of the hand-drawn sketch data and the reference image data are used to determine the comprehensive feature representation; the judgment module is used to judge whether the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold. If the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold, the feature vector of the data source is enhanced to generate a feature mapping set; the comparison module is used to input the feature mapping set into the generative adversarial network, construct the core structure of the three-dimensional model, perform multi-dimensional feature comparison on the generated intermediate results, and output the initial three-dimensional model framework after judging whether it meets the preset standards; the generation module is used to obtain user feedback data based on the initial three-dimensional model framework, and refine the local geometric shape and texture features of the initial three-dimensional model framework through iterative optimization and adjustment to generate the final three-dimensional model.

[0107] Furthermore, the system for generating 3D models based on multi-condition guidance provided by this embodiment, the acquisition module includes a first acquisition unit, a first determination unit, a judgment unit and a second acquisition unit, wherein the first acquisition unit is used to perform structured processing on the text description data stream through a pre-established feature extraction tool, obtain a first feature vector set corresponding to the text description data stream, and use a denoising filter to clean the noise part in the first feature vector set during processing to obtain a cleaned first feature vector set; the first determination unit is used to perform structured analysis on the hand-drawn sketch data stream using an image recognition tool based on the cleaned first feature vector set, obtain a second feature vector set corresponding to the hand-drawn sketch data stream, if during analysis If fuzzy areas are found in the hand-drawn sketch data stream, it is optimized through the edge enhancement tool to determine the optimized second feature vector set; the judgment unit is used to obtain the optimized second feature vector set, and in combination with the reference image data stream, the image comparison tool is used to extract features from the reference image data stream to obtain the third feature vector set. If the resolution is found to be lower than the preset threshold during extraction, the resolution enhancement tool is used to adjust it to determine whether the adjusted third feature vector set meets the requirements; the second acquisition unit is used to obtain the final comprehensive feature vector set by fusing the cleaned first feature vector set, the optimized second feature vector set and the adjusted third feature vector set, and performing unified formatting processing using the data integration tool.

[0108] Preferably, the system for generating 3D models based on multi-condition guidance provided by this embodiment, the determination module includes a third acquisition unit, a second determination unit, a first generation unit and a fourth acquisition unit, wherein the third acquisition unit is used to perform structured processing on various data streams based on the text description data, hand-drawn sketch data and reference image data in the multi-source data using a pre-established feature extraction tool to obtain the corresponding initial feature set, and to clean the text description data through a denoising filtering tool when processing the text description data to obtain a cleaned text feature set; the second determination unit is used to use the image analysis tool to analyze the sketch details based on the cleaned text feature set in combination with the hand-drawn sketch data. Part of the hand-drawn sketch data is edge optimized. If a fuzzy area is found when parsing the hand-drawn sketch data, it is adjusted through the edge enhancement tool to determine the optimized sketch feature set; the first generation unit is used to obtain the optimized sketch feature set, and in combination with the reference image data, the image comparison tool is used to detect the resolution of the reference image data. If the resolution is lower than the preset threshold, it is adjusted through the resolution enhancement tool to generate the adjusted image feature set; the fourth acquisition unit is used to use the data fusion tool to uniformly format the cleaned text feature set, the optimized sketch feature set and the adjusted image feature set to obtain the final comprehensive feature representation.

[0109] Furthermore, the system for generating a 3D model based on multi-condition guidance provided by this embodiment has a judgment module including a fifth acquisition unit, a second generation unit, a third determination unit, and a sixth acquisition unit, wherein the fifth acquisition unit is used to use a parameter detection tool to compare the weight parameters of each data source one by one according to the data source category in the comprehensive feature representation. If the weight parameter of the data source is lower than a preset threshold, the data source is marked as an object to be processed, and a list of data sources to be enhanced is obtained; the second generation unit is used to use a vector optimization tool to perform distribution correction on the feature vectors in the list of data sources to be enhanced, and generate a corrected feature vector set by adjusting the distribution uniformity of the feature vectors; the third determination unit is used to obtain the corrected feature vector set, use a data enhancement tool to supplement the corrected feature vector set with details, sort the corrected feature vector set according to the source data quality and processing priority of the data source during processing, and determine the enhanced feature vector group; the sixth acquisition unit is used to use a mapping conversion tool combined with fusion logic to process the enhanced feature vector group to generate a feature mapping set that meets the data consistency requirements, determine whether the feature mapping set meets the preset distribution standard, and if not, return to the correction link for iterative processing to obtain the final feature mapping result.

[0110] The method and system for generating 3D models based on multi-condition guidance provided in this embodiment generate comprehensive feature representations by fusing multi-source data such as text descriptions, hand-drawn sketches, and reference images, and dynamically assigning weights using deep neural networks and attention mechanisms. Feature enhancement is performed on data sources with lower weights, and the core structure of the three-dimensional model is constructed through a generative adversarial network. Combined with user feedback, this embodiment iteratively optimizes and refines the model to ultimately generate a high-quality three-dimensional model. The method and system for generating 3D models based on multi-condition guidance provided in this embodiment can achieve the following beneficial effects: 1. Advantages of Multi-Source Collaborative Modeling By combining multimodal data such as text descriptions, hand-drawn sketches, and reference images, we can break through the limitations of a single data source and enhance the integrity of model expression. By dynamically allocating weights through the attention mechanism, we can prioritize retaining high-confidence features and suppress noise interference.

[0111] 2. Adaptive Optimization Capabilities Feature enhancement is performed on low-weight data sources to reduce modeling bias caused by differences in input quality; generative adversarial networks (GANs) are combined with multi-dimensional feature comparisons to ensure that the core structure conforms to physical laws.

[0112] 3. User-Friendliness The iterative optimization stage supports manual feedback to adjust local details (such as surface smoothness) to meet personalized customization needs; the output model can be directly used for 3D printing or virtual reality scenes, reducing the cost of secondary modifications.

[0113] 4. Technical Performance Improvement Compared with traditional single-modal modeling methods, comprehensive feature fusion improves model generation efficiency by more than 30%; through NeRF optimization and texture refinement, simultaneous high-precision restoration of geometric shapes and surface materials is achieved.

[0114] In summary, the method and system for generating 3D models based on multi-condition guidance provided in this embodiment can significantly improve the accuracy and efficiency of 3D model generation through multimodal data fusion and dynamic optimization mechanism, provide users with more intelligent and personalized 3D modeling solutions, and provide an efficient and controllable 3D modeling tool chain for the industrial design field.

[0115] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the invention. Thus, the present invention is intended to include such changes and modifications as fall within the scope of the claims and their equivalents.

Claims

1. A method for generating a 3D model based on multi-condition guidance, characterized in that: The following steps are involved: Obtaining raw data streams of text description data, hand-drawn sketch data, and reference image data, and performing structured processing on each type of raw data stream using a pre-established feature extraction model to obtain a corresponding feature vector set; wherein the raw data streams include the text description data stream, the hand-drawn sketch data stream, and the reference image data stream; Based on the feature vector set, a deep neural network is used to perform fusion calculation on the multi-source feature vectors, and weight parameters of the text description data, hand-drawn sketch data, and reference image data are dynamically assigned through an attention mechanism to determine a comprehensive feature representation; Determining whether a weight parameter of any type of data source in the comprehensive feature representation is lower than a preset threshold, and if the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold, enhancing the feature vector of the data source to generate a feature map set; Input the feature map set into a generative adversarial network to construct the core structure of the three-dimensional model, perform multi-dimensional feature comparison on the generated intermediate results, determine whether they meet the preset standards, and then output the initial three-dimensional model framework; According to the initial three-dimensional model framework, user feedback data is obtained, and the local geometric shape and texture features of the initial three-dimensional model framework are refined through iterative optimization and adjustment to generate a final three-dimensional model.

2. The method for generating a 3D model based on multi-condition guidance according to claim 1, characterized in that: The steps of obtaining the original data streams of text description data, hand-drawn sketch data, and reference image data, and performing structured processing on each type of original data stream using a pre-established feature extraction model to obtain a corresponding feature vector set include: Performing structured processing on the text description data stream using a pre-established feature extraction tool to obtain a first feature vector set corresponding to the text description data stream, and using a denoising filter to clean noise portions in the first feature vector set during processing to obtain a cleaned first feature vector set; Based on the cleaned first feature vector set, an image recognition tool is used to perform structured parsing on the hand-drawn sketch data stream to obtain a second feature vector set corresponding to the hand-drawn sketch data stream. If fuzzy areas are found in the hand-drawn sketch data stream during the parsing, an edge enhancement tool is used to optimize the hand-drawn sketch data stream to determine an optimized second feature vector set. Obtaining the optimized second set of feature vectors, and combining them with a reference image data stream, using an image comparison tool to perform feature extraction on the reference image data stream to obtain a third set of feature vectors. If the resolution is found to be lower than a preset threshold during extraction, adjusting it using a resolution enhancement tool to determine whether the adjusted third set of feature vectors meets the requirements; By fusing the cleaned first eigenvector set, the optimized second eigenvector set and the adjusted third eigenvector set, and using data integration tools for unified formatting, the final comprehensive eigenvector set is obtained.

3. The method for generating a 3D model based on multi-condition guidance according to claim 1, wherein: Based on the feature vector set, a deep neural network is used to perform fusion calculation on the multi-source feature vectors, and weight parameters of the text description data, hand-drawn sketch data, and reference image data are dynamically assigned through an attention mechanism to determine the comprehensive feature representation. The steps include: Based on the text description data, hand-drawn sketch data and reference image data in the multi-source data, a pre-established feature extraction tool is used to perform structured processing on various data streams to obtain the corresponding initial feature set, and a denoising and filtering tool is used to clean the text description data when processing the text description data to obtain a cleaned text feature set; Based on the cleaned text feature set and in combination with the hand-drawn sketch data, an image parsing tool is used to optimize the edges of the sketch details. If a fuzzy area is found when parsing the hand-drawn sketch data, an edge enhancement tool is used to adjust it to determine an optimized sketch feature set. Obtaining the optimized sketch feature set, and combining it with the reference image data, using an image comparison tool to perform resolution detection on the reference image data; if the resolution is lower than a preset threshold, adjusting it using a resolution enhancement tool to generate an adjusted image feature set; The cleaned text feature set, optimized sketch feature set and adjusted image feature set are uniformly formatted using data fusion tools to obtain the final comprehensive feature representation.

4. The method for generating a 3D model based on multi-condition guidance according to claim 1, wherein: Determining whether a weight parameter of any type of data source in the comprehensive feature representation is lower than a preset threshold, and if the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold, enhancing the feature vector of the data source, and generating a feature map set includes: According to the data source category in the comprehensive feature representation, a parameter detection tool is used to compare the weight parameters of each data source one by one. If the weight parameter of a data source is lower than a preset threshold, the data source is marked as an object to be processed, and a list of data sources to be enhanced is obtained; For the list of data sources to be enhanced, a vector optimization tool is used to perform distribution correction on the feature vectors in the list, and a corrected feature vector set is generated by adjusting the distribution uniformity of the feature vectors; Obtaining a corrected feature vector set, supplementing the corrected feature vector set with details using a data enhancement tool, sorting the data according to the source data quality and processing priority of the data source during processing, and determining an enhanced feature vector group; The enhanced feature vector group is processed using a mapping conversion tool combined with fusion logic to generate a feature mapping set that meets the data consistency requirements. It is then determined whether the feature mapping set meets the preset distribution standard. If not, the correction link is returned for iterative processing to obtain the final feature mapping result.

5. The method for generating a 3D model based on multi-condition guidance according to claim 1, wherein: The steps of inputting the feature map set into a generative adversarial network, constructing the core structure of the three-dimensional model, performing multi-dimensional feature comparison on the generated intermediate results, and outputting the initial three-dimensional model framework after determining whether they meet the preset standards include: According to the feature map set, a data preprocessing tool is used to perform format normalization processing on the input content, and during the processing, the content distribution of the feature map set is corrected to obtain a normalized feature data set; Using a normalized feature data set, a generative adversarial tool is used to perform a structural transformation on the dimensional data in the feature data set, performing a three-dimensional construction in accordance with the requirements of the core structure during the transformation, and determining a set of intermediate data after the construction; Obtain the constructed intermediate data set, use a feature comparison tool to analyze multiple dimensions of the dimensional data in the feature data set, and if the feature value of a certain dimension deviates from a preset threshold, correct and adjust the dimensional data in the feature data set to determine the data combination that meets the dimensional analysis requirements; From the data combination that meets the requirements of dimensional analysis, the structural verification tool is used to perform framework adaptation processing, and the standards of the initial framework are compared to obtain the final model output results.

6. The method for generating a 3D model based on multi-condition guidance according to claim 1, wherein: The steps of obtaining user feedback data based on the initial 3D model framework and refining the local geometric shape and texture features of the initial 3D model framework through iterative optimization and adjustment to generate a final 3D model include: Based on the data content of the user feedback, the acquired information is classified using a data sorting tool, and opinions on geometric shapes and texture features involved in the feedback are grouped during the classification to obtain a classified feedback data group; Using the classified feedback data set, a shape correction tool is used to compare and analyze the opinions related to the geometric shape. If the shape description of a certain local area deviates from a preset threshold during the comparison, the shape data of the area is adjusted to determine an adjusted shape data set; Obtain the adjusted shape data set, use texture matching tools to process the relevant feedback of texture features, correct inconsistent areas during processing, and determine the texture data combination that meets the requirements; From the texture data combination that meets the requirements, the parameter adjustment tool is used to optimize the configuration of the local adjustment parameters to obtain the final fine-processing data results.

7. A system for generating a 3D model based on multi-condition guidance, for implementing the method for generating a 3D model based on multi-condition guidance according to any one of claims 1 to 6, characterized in that: The system for generating a 3D model based on multi-condition guidance includes: an acquisition module for acquiring raw data streams of text description data, hand-drawn sketch data, and reference image data, and performing structured processing on each type of raw data stream using a pre-established feature extraction model to obtain a corresponding feature vector set; wherein the raw data streams include text description data streams, hand-drawn sketch data streams, and reference image data streams; a determination module for performing a fusion calculation on multi-source feature vectors using a deep neural network based on the feature vector set, dynamically allocating weight parameters of text description data, hand-drawn sketch data, and reference image data through an attention mechanism, and determining a comprehensive feature representation; a judgment module, configured to judge whether a weight parameter of any type of data source in the comprehensive feature representation is lower than a preset threshold; if the weight parameter of any type of data source in the comprehensive feature representation is lower than the preset threshold, enhancing the feature vector of the data source to generate a feature map set; A comparison module is used to input the feature map set into the generative adversarial network, construct the core structure of the three-dimensional model, perform multi-dimensional feature comparison on the generated intermediate results, determine whether they meet the preset standards, and then output the initial three-dimensional model framework; The generation module is used to obtain user feedback data based on the initial three-dimensional model framework, and to refine the local geometric shape and texture features of the initial three-dimensional model framework through iterative optimization and adjustment to generate a final three-dimensional model.

8. The system for generating a 3D model based on multi-condition guidance according to claim 7, characterized in that: The acquisition module includes: a first acquisition unit, configured to perform structured processing on the text description data stream using a pre-established feature extraction tool to obtain a first feature vector set corresponding to the text description data stream, and to use a denoising filter to clean noise portions in the first feature vector set during processing to obtain a cleaned first feature vector set; a first determining unit configured to perform a structured analysis on the hand-drawn sketch data stream using an image recognition tool based on the cleaned first feature vector set to obtain a second feature vector set corresponding to the hand-drawn sketch data stream; and if fuzzy areas are found in the hand-drawn sketch data stream during the analysis, optimize the data stream using an edge enhancement tool to determine an optimized second feature vector set; a judgment unit, configured to obtain the optimized second feature vector set, and perform feature extraction on the reference image data stream using an image comparison tool in combination with the reference image data stream to obtain a third feature vector set; if the resolution is found to be lower than a preset threshold during extraction, then adjust the third feature vector set using a resolution enhancement tool to determine whether the adjusted third feature vector set meets the requirements; The second acquisition unit is used to obtain a final comprehensive feature vector set by fusing the cleaned first feature vector set, the optimized second feature vector set and the adjusted third feature vector set and performing unified formatting processing using a data integration tool.

9. The system for generating a 3D model based on multi-condition guidance according to claim 7, characterized in that: The determination module includes: a third acquisition unit, configured to perform structured processing on various data streams using a pre-established feature extraction tool based on the text description data, hand-drawn sketch data, and reference image data in the multi-source data, to obtain a corresponding initial feature set, and to clean the text description data using a denoising and filtering tool when processing the text description data to obtain a cleaned text feature set; a second determining unit configured to optimize edges of sketch details using an image parsing tool based on the cleaned text feature set and the hand-drawn sketch data; and if fuzzy areas are found when parsing the hand-drawn sketch data, to adjust the areas using an edge enhancement tool to determine an optimized sketch feature set; a first generating unit, configured to obtain an optimized sketch feature set, and combine the optimized sketch feature set with the reference image data to perform resolution detection on the reference image data using an image comparison tool; if the resolution is lower than a preset threshold, adjusting the resolution using a resolution enhancement tool to generate an adjusted image feature set; The fourth acquisition unit is used to obtain a final comprehensive feature representation by uniformly formatting the cleaned text feature set, the optimized sketch feature set, and the adjusted image feature set using a data fusion tool.

10. The system for generating a 3D model based on multi-condition guidance according to claim 7, characterized in that: The judgment module includes: a fifth acquisition unit, configured to use a parameter detection tool to compare the weight parameters of each data source one by one according to the data source category in the comprehensive feature representation, and if the weight parameter of a data source is lower than a preset threshold, mark the data source as an object to be processed, thereby obtaining a list of data sources to be enhanced; a second generating unit, configured to use a vector optimization tool to perform distribution correction on feature vectors in a list of data sources to be enhanced, and to generate a set of corrected feature vectors by adjusting the distribution uniformity of the feature vectors; a third determining unit, configured to obtain a corrected feature vector set, supplement the corrected feature vector set with details using a data enhancement tool, sort the data sources according to source data quality and processing priority during processing, and determine an enhanced feature vector group; The sixth acquisition unit is used to process the enhanced feature vector group using a mapping conversion tool combined with fusion logic to generate a feature mapping set that meets the data consistency requirements, and determine whether the feature mapping set meets the preset distribution standard. If not, it returns to the correction link for iterative processing to obtain the final feature mapping result.

Images