Construction method of gift box packaging design three-dimensional model

Abstract

The invention relates to the technical field of gift packaging three-dimensional modeling, and discloses a construction method of a gift box packaging design three-dimensional model. The method comprises the following steps: collecting natural language description, a freehand sketch image and a physical material attribute parameter; analyzing a natural language to construct a design intention map, and analyzing a sketch to generate a geometric feature code; performing knowledge alignment fusion on the two to form enhanced design semantic representation; generating an initial three-dimensional design framework by using a generative design algorithm; material parameters are integrated, compatibility simulation test is carried out, parameter self-adaptive adjustment circulation is started according to a feasibility evaluation report, and component parameters are dynamically optimized; and rendering and outputting a final model. Through deep fusion of multi-modal design information and preposed dynamic optimization of physical attributes, the accuracy of design intention expression and the producibility of a three-dimensional model are improved, and automation and intelligentization of a design process are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of three-dimensional modeling of gift packaging, in particular to a method for constructing a three-dimensional model of a gift box packaging design. BACKGROUND

[0002] In the existing gift box packaging design process, the creative input method of the designer is relatively single. When the designer tries to combine natural language description and hand-drawn sketch to express the concept, the existing technology usually adopts independent data processing channels. The natural language processing module is responsible for analyzing the text keywords, while the computer vision module focuses on identifying the geometric contour of the sketch. This parallel processing method causes a split between semantic information and geometric information, and the style intention conveyed by the text is difficult to effectively associate and verify with the specific morphological features of the sketch, making the generated three-dimensional concept model often deviate from the comprehensive intention of the designer.

[0003] After the preliminary construction of the three-dimensional model, the integration of material properties is usually in the back-end of the design. The traditional method is to first complete the model construction, then assign physical material parameters to it, and test the structural strength or printing effect through limited simulation. This method is a passive test, and once it is found that the material does not match the design structure, the designer needs to manually return to the initial design step for substantial modification. This repetition not only is inefficient, but also heavily relies on the personal experience of the designer, making it difficult to ensure the feasibility and optimization of the design scheme.

[0004] The current technology lacks a closed-loop system that can pass through from multi-modal creative input to physical implementation verification. The deep semantics of design intention and the geometric expression of morphological features cannot be uniformly represented, and the constraints of material physical properties on design structure cannot be actively and dynamically fed back and optimized in the early and process of design. This leads to long design iteration cycle, high risk of design result manufacturability, and inaccurate creative expression. SUMMARY

[0005] The purpose of the present application is to provide a method for constructing a three-dimensional model of a gift box packaging design to solve the problems raised in the background art.

[0006] To achieve the above purpose, the present application provides a method for constructing a three-dimensional model of a gift box packaging design, which comprises: collecting diversified design input data submitted by users, the design input data including natural language description, hand-drawn sketch image and physical material attribute parameter; performing intention analysis processing on the natural language description, extracting key design terms and constructing a design intention graph; performing structured feature analysis on the hand-drawn sketch image to identify contour elements and decorative patterns within the sketch and generate geometric feature encoding; aligning and fusing the design intent graph with the geometric feature encoding to form an enhanced design semantic representation; generating an initial three-dimensional design framework using generative design algorithms on the enhanced design semantic representation; integrating physical material attribute parameters into the initial three-dimensional design framework to perform material compatibility simulation verification and obtain a feasibility evaluation report; starting a parameter self-adaptive adjustment cycle according to the feasibility evaluation report to dynamically optimize the component parameters of the three-dimensional design framework; visualizing the optimized three-dimensional model through a real-time rendering engine and delivering the final design file.

[0007] Preferably, the intent parsing process is performed on the natural language description to extract key design terms and construct a design intent graph, including: using a syntax analysis tool to decompose the sentence structure in the natural language description to identify subject objects, attribute modifiers, and action instructions; filtering professional vocabulary related to gift box design from the decomposition results to form a candidate term set; querying a domain knowledge base to verify the validity of the candidate terms, removing ambiguous terms and supplementing synonym extensions; establishing a node relationship network based on the verified term set, where nodes represent design elements and edges represent spatial or logical relationships between elements; applying a graph embedding algorithm to encode the node relationship network into a low-dimensional vector to generate a design intent graph.

[0008] Preferably, the structured feature analysis is performed on the hand-drawn sketch image to identify contour elements and decorative patterns within the sketch and generate geometric feature encoding, including: performing preprocessing operations on the hand-drawn sketch image, including line enhancement and noise removal, to obtain a clear sketch; using an edge detection algorithm to extract continuous contour lines in the sketch and segment the contour lines into independent geometric units; calculating shape descriptors for each geometric unit, including curvature distribution and symmetry indicators; identifying repeated patterns and texture regions within the geometric unit, extracting local feature points, and generating a feature description matrix; aggregating the shape descriptors and feature description matrices of all geometric units into a hierarchical feature structure to output the geometric feature encoding.

[0009] Preferably, the knowledge alignment and fusion of the design intent graph with the geometric feature encoding to form an enhanced design semantic representation includes: Project the vector representation of the design intent graph into the vector space of the geometry feature encoding, and calculate the semantic similarity matrix between the two; Establish a mapping relationship table between the intent and the geometric elements according to the semantic similarity matrix; Adjust the weight of the geometry feature encoding through the cross-modal attention mechanism, and highlight the feature dimensions highly related to the design intent; Concatenate the weighted geometry feature encoding and the design intent graph, and reduce the dimension through a nonlinear transformation layer to generate an enhanced design semantic representation.

[0010] Preferably, the generated design algorithm is used to generate three-dimensional concept generation for the enhanced design semantic representation, and an initial three-dimensional design framework is output, including: Input the enhanced design semantic representation into the pre-trained three-dimensional generation neural network, which consists of an encoder and a decoder; The encoder converts the enhanced design semantic representation into a hidden space vector, and the decoder generates a voxel grid according to the hidden space vector, and performs surface reconstruction processing on the voxel grid to generate a smooth three-dimensional mesh model; Extract key structure points from the three-dimensional mesh model, construct a parameterized skeleton system, bind the parameterized skeleton system with the three-dimensional mesh model, and form an editable initial three-dimensional design framework.

[0011] Preferably, the physical material attribute parameters are integrated into the initial three-dimensional design framework, material compatibility simulation inspection is performed, and a feasibility evaluation report is obtained, including: Analyze the elastic modulus, density and surface friction coefficient in the physical material attribute parameters; Set the material parameters in the physics engine and import the initial three-dimensional design framework into the simulation environment; Apply virtual external force to simulate the bearing and deformation behavior of the gift box, and record the stress distribution data; Detect whether the stress concentration area exceeds the material strength threshold, and check whether interference occurs between components; Generate a detailed report containing stress exceeding points and interference problems as a feasibility evaluation report.

[0012] Preferably, the parameter adaptive adjustment cycle is started according to the feasibility evaluation report, and the component parameters of the three-dimensional design framework are dynamically optimized, including: Extract the location and severity indicators of the problem area from the feasibility evaluation report, locate the corresponding component parameters including thickness, angle and connection method according to the problem area, establish a parameter optimization objective function with the constraints of minimizing stress and avoiding interference, and use the gradient descent algorithm to iteratively adjust the component parameters. After each iteration, re-run the material compatibility simulation inspection, terminate the cycle when the improvement rate of three consecutive iterations is less than the threshold, and output the optimized three-dimensional design framework.

[0013] Preferably, the visualized optimized three-dimensional model through the real-time rendering engine comprises: Converting the optimized three-dimensional design framework into a rendering-friendly mesh format, setting a virtual lighting environment and material map, and making the model present a realistic visual effect; Allowing the user to interactively rotate and scale the model and view details from different perspectives, generating a dynamic preview animation to show the opening and closing process of the gift box, and outputting high-resolution rendered images and video files as final design files.

[0014] Preferably, the method further comprises a design iteration learning mechanism: Recording the parameter modification history and simulation results in each parameter adaptive adjustment cycle, building a design decision database, storing successful and failed design cases, using machine learning algorithms to analyze the design decision database, and discovering the implicit rules between parameter adjustment and simulation results; When processing a new design task, the learned rules are used to recommend initial parameter settings, accelerating convergence.

[0015] Preferably, the method further comprises a multi-user collaborative optimization process: Allowing multiple designers to submit design input data simultaneously and merging them into collective design intent, using a voting mechanism to filter the most popular geometric feature codes, forming a consensus design semantic representation, introducing a conflict detection algorithm in the parameter adaptive adjustment cycle to solve parameter preference conflicts between different designers, and finally collecting feedback scores from each designer to update the collaborative optimization strategy.

[0016] Compared with the prior art, the present application has the following advantages: By analyzing the design intent graph of natural language and generating geometric feature codes through structured feature analysis of sketch images, and aligning and fusing the two, an enhanced design semantic representation is formed. This process realizes the mapping and unification of text semantic space and sketch geometric space, enabling style terms to establish precise correspondence with specific contour lines and decorative patterns recognized by sketches. Multi-modal design input information is no longer isolated data fragments, but an organic whole with internal connections and mutual interpretation, improving the understanding depth and representation accuracy of designers' complex and ambiguous creative intent, and ensuring that the generated three-dimensional concept framework can faithfully reflect the original creativity.

[0017] The physical material property parameters are integrated into the initial three-dimensional design framework and material compatibility simulation tests are performed, and the parameter adaptive adjustment cycle is started according to the generated feasibility evaluation report. The material constraints are preposed and integrated into the design optimization process, and the system automatically adjusts the component parameters such as box wall thickness, buckle structure gap, etc. according to the simulation results. This mechanism changes the originally post-positioned and passive material verification into an active optimization drive throughout the design process, so that the three-dimensional model continuously meets the manufacturability requirements of the physical world during the evolution process. The design iteration is changed from an external cycle relying on manual intervention to an endogenous adaptive cycle of the system, reducing the risk of rework caused by material and structure mismatch, and shortening the period from concept to producible scheme. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The working principle diagram of the construction method of the three-dimensional model of the gift box packaging design described in the present application; Figure 2 The flow chart for constructing the design intention map; Figure 3 The flow chart for generating enhanced design semantic representation; Figure 4 The relationship diagram of the objective function and the constraint condition; Figure 5 The bar chart of the importance weight distribution of the gift box packaging design parameters. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0020] Please refer to Figure 1The application provides a method for constructing a three-dimensional model of a gift box packaging design, which comprises: collecting diversified design input data submitted by a user, the design input data including natural language descriptions, hand-drawn sketch images and physical material attribute parameters. Intention analysis processing is performed on the natural language descriptions to extract key design terms and construct a design intention graph. Structured feature analysis is performed on the hand-drawn sketch images to identify contour elements and decorative patterns in the sketch and generate geometric feature codes. The design intention graph and the geometric feature codes are knowledge-aligned and fused to form an enhanced design semantic representation. A generative design algorithm is used to generate a three-dimensional concept from the enhanced design semantic representation to output an initial three-dimensional design framework. The physical material attribute parameters are integrated into the initial three-dimensional design framework, material compatibility simulation inspection is performed, and a feasibility evaluation report is obtained. According to the feasibility evaluation report, a parameter adaptive adjustment cycle is started to dynamically optimize the component parameters of the three-dimensional design framework. The optimized three-dimensional model is visualized through a real-time rendering engine, and the final design file is delivered.

[0021] Embodiment 1: see Figure 2 In the specific embodiment, the intention analysis processing performed on the natural language description involves multiple steps, a syntax analysis tool is used to decompose the sentence structure in the natural language description, the subject object, attribute modification and action instruction are identified, the professional vocabulary related to the gift box design is screened from the decomposition result to form a candidate term set. The validity of the candidate terms is verified by querying the domain knowledge base, the ambiguous terms are removed and the synonym expansion is supplemented, and a node relationship network is established based on the verified term set, wherein the node represents the design element and the edge represents the spatial or logical relationship between the elements. The node relationship network is coded into a low-dimensional vector by applying a graph embedding algorithm to generate a design intention graph. In the specific implementation, the syntax analysis tool uses a dependency parsing method to identify the subject-predicate-object structure and the modification components in the sentence, so as to extract the entities and attributes related to the gift box design. In some embodiments, the domain knowledge base contains a professional vocabulary table and historical design case data in the field of gift box design, which is used to verify the accuracy and consistency of the terms. Optionally, the synonym expansion is based on semantic similarity calculation, and similar terms are retrieved from the knowledge base to enrich the candidate set. It can be understood that the construction of the node relationship network depends on the graph data structure, and the edge weight between the nodes reflects the element association strength. In the specific implementation, the graph embedding algorithm uses the following formula to calculate the node embedding vector: ; Wherein: represents the embedding vector of node , is the neighbor node set of node , is a trainable weight matrix, is the feature vector of node , is a bias term. In some embodiments, the feature vector is derived from semantic encoding of the term, obtained by pre-trained language models. Optionally, the set of neighbor nodes is defined based on spatial or logical relationship between nodes, such as containment or adjacency. It is appreciated that the dimension of the embedding vector is set by hyperparameters to balance expressiveness and computational efficiency. In a specific implementation, the generated design intent graph is represented as a low-dimensional vector for subsequent knowledge alignment and fusion process.

[0022] Embodiment 2: refer to Figure 3 In a specific implementation, performing structured feature analysis on the hand-drawn sketch image involves preprocessing operations including line enhancement and noise removal to obtain a clear sketch, using an edge detection algorithm to extract continuous contour lines in the sketch and segment the contour lines into independent geometric units, calculating shape descriptors including curvature distribution and symmetry indicators for each geometric unit, identifying repetitive patterns and texture regions within the geometric unit and extracting local feature points to generate a feature description matrix, aggregating shape descriptors and feature description matrices of all geometric units into a hierarchical feature structure and outputting a geometric feature code. In a specific implementation, line enhancement uses morphological operations to enhance the continuity of sketch lines, noise removal uses filtering algorithms to reduce image impurities, and the edge detection algorithm identifies contour boundaries based on the Canny operator and segments the contour lines into independent geometric units according to connection points. In some embodiments, shape descriptor calculation includes analyzing the curvature variation of geometric unit boundary points using a curvature scale space method, and symmetry indicators are evaluated by a mirror symmetry detection algorithm to assess the symmetry degree of the geometric unit. Optionally, local feature point extraction uses the SIFT algorithm to generate a feature description matrix, and hierarchical feature structure integrates shape descriptors and feature description matrices through a tree-like data organization method. The specific implementation of local feature point extraction using the SIFT algorithm to generate a feature description matrix includes the following processes. First, after preprocessing the sketch image to obtain a clear sketch, the system identifies repetitive patterns and texture regions within the geometric unit; then, the SIFT algorithm is applied to detect key points in these regions as local feature points, and a feature description matrix is generated based on the image information around the key points, which quantifies the unique attributes of local features. Next, the system integrates the shape descriptors (such as curvature distribution and symmetry indicators) of each geometric unit with the feature description matrix through a tree-like data organization method, where the tree structure hierarchically associates feature data of different geometric units, ultimately forming a unified geometric feature code output.

[0023] In specific implementations, when fusing the design intent graph with the geometry feature encoding in knowledge alignment, the vector representation of the design intent graph is projected into the vector space of the geometry feature encoding and a semantic similarity matrix between the two is calculated, a mapping relationship table of intent and geometric elements is established according to the semantic similarity matrix, the weight of the geometry feature encoding is adjusted through a cross-modal attention mechanism to highlight the feature dimensions highly relevant to the design intent, the weighted geometry feature encoding is spliced with the design intent graph and is reduced in dimension through a nonlinear transformation layer to generate an enhanced design semantic representation. In some embodiments, the projection operation uses a linear transformation matrix to map the design intent graph vector into the vector space of the geometry feature encoding, and the formula for calculating the semantic similarity matrix is: ; wherein: denotes the semantic similarity matrix, denotes the projected design intent vector, denotes the geometry feature encoding vector, denotes the matrix transpose. Optionally, the mapping relationship table is generated based on the threshold screening of the semantic similarity matrix to establish the correspondence between the intent and the geometric elements, and the cross-modal attention mechanism uses a query-key value attention model to calculate the weight vector.

[0024] In specific implementations, when generating a three-dimensional concept from the enhanced design semantic representation using a generative design algorithm, the enhanced design semantic representation is input into a pre-trained three-dimensional generation neural network, the neural network is composed of an encoder and a decoder, the encoder converts the enhanced design semantic representation into a latent space vector, the decoder generates a voxel grid according to the latent space vector, the voxel grid is processed for surface reconstruction to generate a smooth three-dimensional mesh model, key structure points in the three-dimensional mesh model are extracted to construct a parameterized skeleton system, and the parameterized skeleton system is bound with the three-dimensional mesh model to form an editable initial three-dimensional design framework. In specific implementations, the three-dimensional generation neural network adopts a variational autoencoder architecture, the encoder is composed of fully connected layers and convolutional layers, the decoder is composed of deconvolutional layers and up-sampling layers, and the latent space vector is calculated through forward propagation of the encoder. In some embodiments, the voxel grid generation process is represented by the following formula: ; wherein: denotes the generated voxel grid, denotes the decoder, denotes the encoder, The enhanced design semantic representation is indicative of the surface reconstruction process. Optionally, the surface reconstruction process extracts isosurfaces from the voxel grid using a marching cubes algorithm, and the key structural points are determined by a mesh curvature analysis and a feature point detection algorithm. In an embodiment, the mesh curvature analysis involves calculating the curvature values of the points on the surface of the three-dimensional mesh model, and identifying the high curvature regions by analyzing the curvature distribution, which usually correspond to edges or corners as the key structural points. In an embodiment, the feature point detection algorithm is applied on the three-dimensional mesh to detect local geometric features such as corner points or texture change points, and the feature point positions are determined by comparing the geometric attributes of the neighborhood points. The algorithm scans the mesh surface to evaluate the local geometric variations of each point, and uses a feature response function to filter the salient points, which are used to identify the important structural points and support the construction of the parametric skeleton system. It can be understood that the parametric skeleton system is composed of joint points and connecting bones, and the binding operation associates the vertices of the three-dimensional mesh model to the skeleton system through skinning weights.

[0025] In an embodiment, when integrating the physical material attribute parameters into the initial three-dimensional design framework to perform the material compatibility simulation test, the elastic modulus, density, and surface friction coefficient in the physical material attribute parameters are analyzed, the material parameters are set in the physics engine, and the initial three-dimensional design framework is imported into the simulation environment. Virtual external forces are applied to simulate the bearing and deformation behavior of the gift box, and stress distribution data are recorded. It is detected whether the stress concentration area exceeds the material strength threshold, and it is checked whether interference occurs between components. A detailed report containing stress exceeding points and interference problems is generated as a feasibility evaluation report. In some embodiments, the physics engine uses a simulation system based on the finite element method, and the material parameters are configured according to the physical material attribute parameters to set the elastic modulus, density, and surface friction coefficient. Optionally, the virtual external forces include pressure load and torque load, and the stress distribution data are obtained by solving the mechanical equilibrium equation. It can be understood that the stress concentration area detection is realized by comparing the node stress value with the material strength threshold, and the component interference check identifies the grid penetration phenomenon through the collision detection algorithm. The feasibility evaluation report records all detection results in the form of a structured document.

[0026] In some embodiments, the parameter adaptive adjustment loop is initiated based on the feasibility assessment report, the location and severity indicators of the problematic regions are extracted from the feasibility assessment report, the corresponding component parameters including thickness, angle and connection mode are located according to the problematic regions, the parameter optimization objective function is established to minimize the stress and avoid interference as constraint conditions, the gradient descent algorithm is used to iteratively adjust the component parameters, the material compatibility simulation is re-run after each iteration for inspection, and the loop is terminated and the optimized three-dimensional design framework is output when the improvement rate of three consecutive iterations is lower than the threshold value. In some embodiments, the location of the problematic region is identified by three-dimensional coordinates, the severity indicator is quantified based on the stress exceeding value and the interference distance, and the component parameters including thickness, angle and connection mode are extracted from the metadata of the three-dimensional design framework. The parameter optimization objective function is defined as: ; wherein: represents the objective function value, represents the component parameter vector including thickness, angle and connection mode, and are weight coefficients, represents the stress value of the th node, represents the distance value of the th interference point, and represent the number of nodes and interference points, respectively. In some embodiments, the gradient descent algorithm calculates the partial derivative of the objective function with respect to the parameter vector to update the parameters, and the material compatibility simulation is re-run after each iteration to obtain new stress distribution data and interference detection results. Optionally, the improvement rate is calculated by comparing the changes in the objective function values of consecutive iterations, and the threshold value is set to a predetermined percentage. It can be understood that the termination condition of the loop ensures that the optimization process stops when it converges stably. In some embodiments, referring to Table 1, the parameter adaptive adjustment loop involves the mapping relationship between the problematic regions and the component parameters.

[0027] Table 1: Mapping table of problematic regions and component parameters

[0028] In some embodiments, the component parameter thickness is adjusted by linear transformation, the angle is adjusted by rotation transformation, and the connection mode is adjusted by enumeration selection. Optionally, the material compatibility simulation uses the same physical engine settings in each iteration to ensure consistency. It can be understood that the optimized three-dimensional design framework is output in a standard mesh format.

[0029] In specific implementations, the optimized three-dimensional model is visualized by a real-time rendering engine, the optimized three-dimensional design framework is converted into a rendering-friendly mesh format, a virtual lighting environment and material maps are set to make the model present a realistic visual effect, user interaction is allowed to rotate and scale the model and view details from different perspectives, a dynamic preview animation is generated to demonstrate the opening and closing process of the gift box, and high-resolution rendered images and video files are output as the final design files. In specific implementations, the rendering-friendly mesh format includes OBJ or FBX format, the virtual lighting environment uses physically-based rendering techniques to set point lights and directional lights. In some embodiments, the material maps apply diffuse and normal maps according to physical material property parameters, and user interaction is achieved through rotation and scaling operations by a graphical user interface. Optionally, the dynamic preview animation simulates the hinge movement of the gift box through keyframe animation techniques, and the high-resolution rendered images and video files are output in PNG and MP4 formats. It can be understood that the final design files contain metadata describing design parameters and material properties.

[0030] Referring to Figure 4 , the relationship between the objective function value and the average stress value, the average interference distance is intuitively presented to quantitatively guide the parameter optimization process. In the relationship between the objective function value and the average stress value in Figure a, the horizontal axis is the average stress value (MPa), the vertical axis is the objective function value, the dashed line is the stress threshold (60 MPa), and the right color bar represents the number of iterations. As the average stress value gradually increases from 40 MPa to 120 MPa, the objective function value shows a downward trend as a whole, and when the average stress value exceeds the stress threshold of 60 MPa, the downward trend of the objective function value remains continuous, and the number of iterations decreases as the average stress value increases. In the relationship between the objective function value and the average interference distance in Figure b, the horizontal axis is the average interference distance (mm), the vertical axis is the objective function value, the dashed line is the minimum interference distance (2 mm), and the right color bar represents the number of iterations. As the average interference distance gradually increases from 0 mm to 4.0 mm, the objective function value shows an overall upward trend, and when the average interference distance exceeds the minimum interference distance of 2 mm, the upward trend of the objective function value continues, and the number of iterations increases as the average interference distance increases. These relationships provide quantitative basis for iterative adjustment of component parameters (thickness, angle, connection mode, etc.) based on gradient descent algorithm, and by monitoring the changes of the objective function value, the average stress value and the average interference distance in real time, it can be judged whether the parameter optimization converges, and it is ensured that the stress and interference problems of the final three-dimensional design framework meet the design requirements.

[0031] In some embodiments, the design iteration learning mechanism records the parameter modification history and simulation results in each parameter self-adaptive adjustment cycle, builds a design decision database to store successful and failed design cases, uses a machine learning algorithm to analyze the design decision database to find the implicit rules between parameter adjustment and simulation results, and recommends initial parameter settings using the learned rules to accelerate convergence when processing new design tasks. In some embodiments, the parameter modification history includes adjustment records of thickness, angle, and connection mode, the simulation results include stress distribution data and interference detection results, and the design decision database uses a relational database structure to store the mapping relationship between parameter vectors and simulation results. In some embodiments, the machine learning algorithm uses a random forest regression model to analyze the relationship between parameter adjustment and simulation results, and the implicit rules are obtained by feature importance analysis to get the influence weight of parameters on simulation results. Optionally, the recommended initial parameter settings use the following optimization formula: ; wherein: represents the recommended initial parameter vector, represents the parameter vector, represents the number of similar cases, represents the weight coefficient of the th case, represents the predicted result corresponding to the parameter vector, represents the actual simulation result of the th case. It can be understood that accelerating convergence is achieved by reducing the number of iterations of the parameter self-adaptive adjustment cycle.

[0032] In some embodiments, the multi-user collaborative optimization process allows multiple designers to submit design input data simultaneously and merge them into collective design intent, uses a voting mechanism to filter the most popular geometric feature codes to form a consensus design semantic representation, introduces a conflict detection algorithm in the parameter self-adaptive adjustment cycle to solve parameter preference conflicts between different designers, and finally collects feedback scores from each designer for updating the collaborative optimization strategy after the three-dimensional model is generated. In some embodiments, the collective design intent is generated by weighted fusion of design input data from multiple designers, and the voting mechanism sorts and filters based on the scores of the designers on the geometric feature codes. In some embodiments, the conflict detection algorithm identifies mutually exclusive options in parameter settings and uses a negotiation strategy to solve conflicts, and the parameter preference conflicts include thickness setting differences and connection mode selection differences. Optionally, the feedback score uses a five-point scale to collect the designers' evaluation of each dimension of the final three-dimensional model, and the collaborative optimization strategy adjusts the weight distribution of the voting mechanism according to the feedback score. It can be understood that updating the collaborative optimization strategy optimizes the weight distribution parameters through reinforcement learning algorithm to improve the efficiency of subsequent multi-user collaborative design.

[0033] Referring to Figure 5The importance weight distribution of each physical material attribute parameter and structure parameter is shown. The importance weight of thickness (mm) is 0.32, which is the key parameter affecting the design; the weight of angle (degree) is 0.25, the weight of connection mode is 0.18, the weight of material density is 0.10, the weight of elastic modulus is 0.08, and the weight of surface friction coefficient is 0.07. These weights reflect the influence degree of each parameter on the simulation results (such as stress distribution, interference detection) in the design iteration learning mechanism, provide an intuitive basis for parameter importance analysis based on machine learning algorithm, and provide a quantitative reference for solving parameter preference conflicts and recommending initial parameters in the multi-user collaborative optimization process, helping to accelerate design convergence and improve design efficiency.

[0034] It should be noted that, in this document, the terms such as first and second are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between such entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus including a series of elements includes not only those elements, but also other elements not explicitly listed, or inherent to such a process, method, article or apparatus.

[0035] Although embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a three-dimensional model of a gift box packaging design, characterized in that, The method performs the following sequence of operations: Collect diverse design input data submitted by users, including natural language descriptions, hand-drawn sketches, and physical material property parameters; Perform intent parsing on natural language descriptions, extract key design terms, and construct a design intent graph; Perform structured feature analysis on hand-drawn sketch images to identify outline elements and decorative patterns within the sketches and generate geometric feature codes; By aligning and fusing design intent maps with geometric feature encodings to form an enhanced design semantic representation; Generative design algorithms are used to generate three-dimensional concepts from the enhanced design semantic representation, producing an initial three-dimensional design framework; Integrate physical material property parameters into the initial three-dimensional design framework, perform material compatibility simulation verification, and obtain a feasibility assessment report; Based on the feasibility assessment report, an adaptive parameter adjustment cycle is initiated to dynamically optimize the component parameters of the 3D design framework; The optimized 3D model is visualized through a real-time rendering engine, and the final design documents are delivered.

2. The method for constructing a three-dimensional model of gift box packaging design according to claim 1, characterized in that, The process of performing intent parsing on the natural language description, extracting key design terms, and constructing a design intent map includes: Use syntax analysis tools to decompose the sentence structure in natural language descriptions and identify the subject, attribute modifiers, and action instructions; From the decomposition results, professional terms related to gift box design were selected to form a candidate terminology set; The domain knowledge base is queried to verify the validity of candidate terms, remove ambiguous terms, and supplement synonym expansion; A node relationship network is established based on the validated terminology set, where nodes represent design elements and edges represent spatial or logical relationships between elements. The graph embedding algorithm is used to encode the node relationship network into a low-dimensional vector to generate a design intent graph.

3. The method for constructing a three-dimensional model of gift box packaging design according to claim 1, characterized in that, The process of performing structured feature analysis on the hand-drawn sketch image, identifying contour elements and decorative patterns within the sketch, and generating geometric feature codes includes: Preprocessing operations are performed on the hand-drawn sketch image, including line enhancement and noise removal, to obtain a clear sketch; An edge detection algorithm is used to extract continuous contour lines from the sketch and divide the contour lines into independent geometric units; Calculate a shape descriptor for each geometric unit, including curvature distribution and symmetry index; Identify repeating patterns and texture regions within geometric units, extract local feature points, and generate a feature description matrix; The shape descriptors and feature description matrices of all geometric units are aggregated into a hierarchical feature structure, and the geometric feature codes are output.

4. The method for constructing a three-dimensional model of gift box packaging design according to claim 1, characterized in that, The process of aligning and fusing design intent maps with geometric feature encodings to form an enhanced design semantic representation includes: The vector representation of the design intent map is projected onto the vector space of the geometric feature encoding, and the semantic similarity matrix between the two is calculated. A mapping table between intents and geometric elements is established based on the semantic similarity matrix; By adjusting the weights of geometric feature encoding through a cross-modal attention mechanism, feature dimensions that are highly relevant to the design intent are highlighted; The weighted geometric feature codes are concatenated with the design intent map, and then dimensionality is reduced through a nonlinear transformation layer to generate an enhanced design semantic representation.

5. The method for constructing a three-dimensional model of gift box packaging design according to claim 1, characterized in that, The process employs a generative design algorithm to generate three-dimensional concepts from the enhanced design semantic representation, producing an initial three-dimensional design framework, including: The design will enhance the semantic representation input of the pre-trained 3D generative neural network, which consists of an encoder and a decoder; The encoder converts the enhanced design semantic representation into latent space vectors, the decoder generates a voxel mesh based on the latent space vectors, performs surface reconstruction on the voxel mesh, and generates a smooth 3D mesh model. Key structural points are extracted from the 3D mesh model, a parametric skeleton system is constructed, and the parametric skeleton system is bound to the 3D mesh model to form an editable initial 3D design framework.

6. The method for constructing a three-dimensional model of gift box packaging design according to claim 1, characterized in that, The physical material property parameters are integrated into the initial three-dimensional design framework, and material compatibility simulation is performed to obtain a feasibility assessment report, including: Analyze the elastic modulus, density, and surface friction coefficient among the physical material properties; Set the material parameters in the physics engine and import the initial 3D design framework into the simulation environment; Virtual external forces are applied to simulate the load-bearing and deformation behavior of gift boxes, and stress distribution data is recorded; Check whether stress concentration areas exceed the material strength threshold and check whether interference occurs between components; Generate a detailed report, including stress exceedance points and interference issues, as a feasibility assessment report.

7. The method for constructing a three-dimensional model of gift box packaging design according to claim 6, characterized in that, The step of initiating an adaptive adjustment cycle of parameters based on the feasibility assessment report to dynamically optimize the component parameters of the 3D design framework includes: Extract the location and severity indicators of the problem area from the feasibility assessment report. Based on the problem area, locate the corresponding component parameters, including thickness, angle and connection method, and establish a parameter optimization objective function. With the constraints of minimizing stress and avoiding interference, use the gradient descent algorithm to iteratively adjust the component parameters. After each iteration, rerun the material compatibility simulation test. When the improvement rate of three consecutive iterations is lower than the threshold, terminate the loop and output the optimized three-dimensional design framework.

8. The method for constructing a three-dimensional model of gift box packaging design according to claim 1, characterized in that, The optimized 3D model, visualized through a real-time rendering engine, includes: The optimized 3D design framework is converted into a rendering-friendly mesh format, and virtual lighting environment and material maps are set to make the model present a realistic visual effect. It allows users to interactively rotate and zoom the model, view details from different perspectives, generate dynamic preview animations, demonstrate the opening and closing process of the gift box, and output high-resolution rendered images and video files as the final design files.

9. The method for constructing a three-dimensional model of gift box packaging design according to claim 1, characterized in that, The method also includes designing an iterative learning mechanism: Record the parameter modification history and simulation results in each parameter adaptive adjustment cycle, build a design decision database, store successful and unsuccessful design cases, and use machine learning algorithms to analyze the design decision database to discover the implicit patterns between parameter adjustment and simulation results; When dealing with new design tasks, prioritize applying the learned patterns to recommend initial parameter settings to accelerate convergence.

10. The method for constructing a three-dimensional model of gift box packaging design according to claim 1, characterized in that, The method also includes a multi-user collaborative optimization process: Multiple designers are allowed to submit design input data simultaneously and merge it into a collective design intent. A voting mechanism is used to select the most popular geometric feature encodings to form a consensus design semantic representation. A conflict detection algorithm is introduced in the parameter adaptive adjustment loop to resolve parameter preference conflicts between different designers. Finally, after the 3D model is generated, feedback scores from each designer are collected to update the collaborative optimization strategy.

Images