An artificial intelligence-based clothing design customization method and system

Abstract

The application provides a clothing design customization method and system based on artificial intelligence, which comprises the following steps: generating a two-dimensional planar pattern and a semantic segmentation map thereof; performing conformal parameterization unfolding on a three-dimensional curved surface model to obtain a pattern planar map, a bidirectional mapping relationship, and a local area deformation factor; automatically planning on the two-dimensional pattern planar map to obtain a layout scheme; fitting the two-dimensional planar pattern to a corresponding area of the two-dimensional pattern planar map according to the layout scheme, and then performing differential texture mapping on the two-dimensional planar pattern according to the local area deformation factor to obtain a pre-adapted two-dimensional pattern map; and wrapping the pre-adapted two-dimensional pattern map back onto the three-dimensional curved surface model of a target garment to generate a three-dimensional clothing design effect drawing. By adopting the scheme of the application, adaptive differential texture mapping can be realized in combination with pattern semantic structures under the deformation difference of the high-curvature area of the clothing curved surface model.

Description

Technical Field

[0001] This application relates to the field of clothing design and customization technology, and more specifically, to a method and system for clothing design and customization based on artificial intelligence. Background Technology

[0002] Customized clothing design refers to a series of creative processes, including in-depth needs communication, personalized aesthetic concepts, digital or manual pattern generation, fabric adaptation and three-dimensional cutting, and multiple rounds of sample correction, to transform a client's body characteristics, wearing scenarios, and aesthetic preferences into a wearable three-dimensional entity that is unique, well-fitting, and functional.

[0003] AI-based clothing design customization refers to the process of intelligently collecting multimodal user data, generating personalized designs driven by deep learning, and automating precise pattern adaptation and optimization through algorithms. This transforms individual physiological characteristics, aesthetic preferences, and scenario requirements into highly personalized, instantly previewable, and directly producible digital clothing solutions. Current AI-based clothing design customization processes often rely on directly wrapping two-dimensional patterns onto three-dimensional clothing surface models through a unified parametric mapping relationship. This approach struggles to effectively address the visual distortion of patterns on the three-dimensional model caused by factors such as the complex Gaussian curvature variations of clothing surfaces and differences in the degree of deformation of different body parts. Consequently, key semantic parts of the pattern (such as brand logos and main graphics) suffer unacceptable deformation and distortion, while non-key semantic parts (such as background textures) fail to fully utilize deformation for adaptive filling to conceal seams or conform to the surface, further exacerbating visual inconsistency and unprofessionalism. Therefore, how to achieve adaptive differentiated texture mapping in combination with the semantic structure of the pattern, given the deformation differences in high-curvature areas of the clothing surface model, has become a challenge for the industry. Summary of the Invention

[0004] This application provides an artificial intelligence-based method and system for clothing design and customization, which can achieve adaptive differentiated texture mapping by combining pattern semantic structure under the deformation differences in the high curvature area of ​​the clothing surface model.

[0005] In a first aspect, this application provides an artificial intelligence-based method for clothing design and customization, comprising the following steps: It receives a pattern theme description input by the user and then generates a two-dimensional planar pattern that matches the description and its corresponding semantic segmentation map; The three-dimensional surface model of the target garment is unfolded using conformal parametric method to obtain a two-dimensional pattern planar diagram, a two-way mapping relationship between the three-dimensional surface and the two-dimensional pattern, and a local area deformation factor during the mapping process. Based on the garment manufacturing rules, the pattern layout on the two-dimensional pattern is automatically planned to obtain the pattern layout scheme on the garment piece. According to the layout scheme, the two-dimensional planar pattern is attached to the corresponding area of ​​the two-dimensional plate planar image, and then the local area deformation factor is used to perform differentiated texture mapping on different semantic regions in the two-dimensional planar pattern identified by the semantic segmentation image to obtain a pre-adapted two-dimensional plate planar pattern. The pre-adapted two-dimensional pattern is wrapped back onto the three-dimensional curved surface model of the target garment through the two-way mapping relationship, generating a three-dimensional garment design rendering in which the pattern naturally fits the curved surface of the garment.

[0006] In some embodiments, conformal parametric unfolding of the three-dimensional surface model of the target garment is performed to obtain a two-dimensional pattern planar diagram, a two-way mapping relationship between the three-dimensional surface and the two-dimensional pattern, and a local area deformation factor during the mapping process, specifically including: Extract the geometric and topological data of the 3D surface model of the target garment; By solving the conformal parameterization optimization problem using the geometric and topological data, a coordinate transformation is obtained that maps a three-dimensional surface to a two-dimensional plane. Based on the coordinate transformation results, a two-dimensional plate planar diagram, a two-way mapping relationship between the three-dimensional curved surface and the two-dimensional plate, and a local area deformation factor during the mapping process are generated.

[0007] In some embodiments, the automatic planning on the two-dimensional pattern drawing according to garment manufacturing rules to obtain a pattern layout scheme on the garment piece specifically includes: Based on the garment manufacturing process rules and the properties of the two-dimensional planar pattern, geometric constraints and optimization objectives for the pattern arrangement are generated. Using the closed outline of the two-dimensional plate planar diagram as the container area and the two-dimensional planar pattern as the object to be placed, a two-dimensional irregular shape nested optimization model is constructed by combining the geometric constraints and optimization objectives. Solve the nested optimization model of the two-dimensional irregular shape to obtain the layout scheme of the pattern on the garment piece.

[0008] In some embodiments, attaching the two-dimensional planar pattern to the corresponding area of ​​the two-dimensional plate planar image according to the layout scheme specifically includes: The layout scheme is analyzed to obtain the target position, rotation angle, and scaling factor of the two-dimensional planar pattern on the two-dimensional plate planar diagram; Based on the target position, rotation angle, and scaling factor, determine the corresponding texture coordinates of each sampling point on the two-dimensional plate planar image in the two-dimensional planar pattern texture space; The two-dimensional planar pattern is sampled and transformed according to the texture coordinates, thereby fitting the two-dimensional planar pattern to the corresponding area of ​​the two-dimensional plate planar image.

[0009] In some embodiments, differentiating texture mapping is performed on different semantic regions identified by the semantic segmentation map in the two-dimensional planar pattern according to the local area deformation factor to obtain a pre-adapted two-dimensional plate pattern, specifically including: Based on the semantic segmentation map, a mapping relationship is established between different semantic regions and preset deformation resistance parameters; The differentiated texture coordinate correction amount is determined based on the deformation resistance parameter and the corresponding local area deformation factor of the semantic region to which each local region of the two-dimensional plate planar image belongs. By locally adjusting the texture coordinates of the intermediate texture image using the differentiated texture coordinate correction amount, a pre-adapted two-dimensional plate pattern is obtained.

[0010] In some embodiments, the process of wrapping the pre-adapted two-dimensional pattern back onto the three-dimensional curved surface model of the target garment through the bidirectional mapping relationship to generate a three-dimensional garment design rendering in which the pattern naturally fits the curved surface of the garment specifically includes: Based on the bidirectional mapping relationship, the texture information on the pre-adapted two-dimensional pattern is reverse-mapped to the corresponding surface of the target garment's three-dimensional curved surface model; The mapping results are processed with material and lighting settings and graphic rendering to generate a 3D clothing design rendering in which the pattern and the curved surface of the garment fit together naturally.

[0011] In some embodiments, the user-inputted pattern theme description refers to any short natural language phrase that can define the visual content of the pattern.

[0012] Secondly, this application provides an artificial intelligence-based clothing design and customization system, comprising: The generation module is used to receive the pattern theme description input by the user, and then generate a two-dimensional planar pattern that conforms to the description and its corresponding semantic segmentation map; The processing module is used to perform conformal parametric unfolding of the three-dimensional curved surface model of the target garment to obtain a two-dimensional pattern planar diagram, a two-way mapping relationship between the three-dimensional curved surface and the two-dimensional pattern, and a local area deformation factor during the mapping process. The processing module is also used to automatically plan on the two-dimensional pattern drawing according to the garment process rules to obtain the pattern layout scheme on the garment piece; The processing module is further configured to attach the two-dimensional planar pattern to the corresponding area of ​​the two-dimensional plate planar image according to the layout scheme, and then perform differentiated texture mapping on different semantic regions identified by the semantic segmentation map in the two-dimensional planar pattern according to the local area deformation factor to obtain a pre-adapted two-dimensional plate planar pattern. The execution module is used to wrap the pre-adapted two-dimensional pattern back onto the three-dimensional curved surface model of the target garment through the bidirectional mapping relationship, generating a three-dimensional garment design rendering in which the pattern and the garment surface naturally fit together.

[0013] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described artificial intelligence-based clothing design and customization method.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described artificial intelligence-based clothing design and customization method.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The AI-based clothing design and customization method and system provided in this application firstly maps the three-dimensional curved surface model of the target garment to a two-dimensional pattern planar image through conformal parametric unfolding, and simultaneously calculates the bidirectional mapping relationship and local area deformation factor between the three-dimensional curved surface and the two-dimensional pattern. This transforms the complex and difficult-to-quantify geometric deformation of the three-dimensional curved surface (especially the non-uniform stretching or compression in high curvature areas) into precisely measurable mathematical parameters, providing a quantitative data foundation for subsequent texture mapping to distinguish the degree of deformation in different areas. Subsequently, in the process of generating the pre-adapted two-dimensional pattern, based on the different semantic regions of the pattern identified by the semantic segmentation map (such as the main graphic and background texture), and combined with the local area deformation factor, differentiated texture mapping is implemented. This process can associate the semantic information of the pattern with the local geometric deformation features of the garment surface, and transform the abstract deformation parameters into specific, region-specific texture adjustment operations through a differentiated mapping strategy, thereby transforming the one-size-fits-all projection problem into a semantically protected directional adaptation problem. Furthermore, through Using a local area deformation factor as an adjustment coefficient, deformation suppression is applied to key semantic regions (such as brand logos) to maintain shape integrity. Simultaneously, non-key semantic regions (such as background textures) are allowed to adaptively fill and stretch according to the deformation factor. This process transforms external design experience into internal calculation rules driving texture mapping, translating the structural requirements of the pattern into the surface fit requirements. Differential mapping provides dual guidance for the preprocessing of patterns on 2D patterns, combining semantic fidelity and geometric adaptation. Finally, the preprocessed 2D pattern is wrapped back onto the 3D garment surface through a bidirectional mapping relationship. This process dynamically recreates a comprehensive visual effect on the 3D model based on pre-adjusted pattern data with deformation factors and semantics, maintaining the shape of key areas while achieving natural background fit. This effectively suppresses the distortion and visual distortion of the pattern subject caused by deformation differences in high-curvature areas. In summary, this solution can achieve adaptive differential texture mapping in high-curvature areas of the garment surface model, combined with the semantic structure of the pattern. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an artificial intelligence-based clothing design and customization method according to some embodiments of this application; Figure 2 This is a schematic flowchart illustrating the implementation of conformal parameterized expansion according to some embodiments of this application; Figure 3 This is a schematic diagram illustrating the process of implementing texture mapping according to some embodiments of this application; Figure 4 This is a structural schematic diagram of an artificial intelligence-based clothing design and customization system according to some embodiments of this application; Figure 5This is an internal structural diagram of a computer device that implements an artificial intelligence-based clothing design and customization method, according to some embodiments of this application. Detailed Implementation

[0017] To better understand the technical solutions in this embodiment, the technical solutions in this embodiment will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0018] refer to Figure 1 The figure is a flowchart illustrating an AI-based clothing design and customization method according to some embodiments of this application. This AI-based clothing design and customization method mainly includes the following steps: In step 101, the user inputs a pattern theme description, and then generates a two-dimensional planar pattern that matches the description and its corresponding semantic segmentation map.

[0019] In practice, users input natural language text describing the theme, style, or elements of a pattern through a graphical user interface, such as "Chinese dragon totem in ink painting style" or "simple geometric lines and polka dot background." This text description, as conditional input, is sent to a text-to-image generative AI model deployed on a server or local computing unit. This model employs a diffusion-based generative architecture, with its core being a U-Net network pre-trained on a massive dataset of clothing, fabric, and decorative patterns. During the inference phase, the model converts the text description into a high-dimensional semantic vector using a text encoder, guiding a progressive denoising process starting with random Gaussian noise. After multiple iterations, a high-resolution, diverse, and semantically consistent two-dimensional planar pattern is finally synthesized. Simultaneously, a pixel-level semantic understanding model is used to parse the generated pattern to generate the corresponding semantic segmentation map. Specifically, the generated two-dimensional planar pattern can be... The 2D planar pattern is input to a deep convolutional neural network (such as DeepLabV3+ or SegFormer architecture) pre-trained on a finely labeled pattern segmentation dataset. This network performs pixel-by-pixel classification on the input image and outputs a grayscale or color map with the same size as the original pattern, but with each pixel assigned a specific category label. The category labels are predefined based on prior design knowledge and include at least three categories: "main graphic", "decorative border", and "background texture" to distinguish semantic regions with different functions and deformation requirements in the pattern. As an alternative implementation, the text-to-image generative model and the semantic segmentation model can be designed as an integrated system using end-to-end joint training. A parallel segmentation head is added to the decoder end of the diffusion model, so that the model can directly output the semantic category probability distribution of each pixel while generating the image, thereby obtaining the 2D planar pattern and its corresponding semantic segmentation map at once.

[0020] It should be noted that the user-inputted pattern theme description mentioned in this application refers to any short natural language sentence that can define the visual content of the pattern, and is the creative starting point that drives the generation process. The two-dimensional planar pattern refers to a digital raster image created by a generative model and stored in RGB or RGBA format. Its content visually echoes the theme description and serves as a texture source for subsequent mapping and adaptation. The semantic segmentation map is a mask image aligned with the pixels of the two-dimensional planar pattern and characterized by discrete label values ​​representing the semantic composition of the image content. Its physical significance lies in deconstructing the pattern from a visual perspective into components with different "deformation properties," providing a structural basis for implementing differentiated texture mapping in subsequent steps. The text-to-image diffusion model is a type of generative artificial intelligence based on a denoising diffusion probability model. It synthesizes images from text descriptions by learning to reverse a process of gradually adding noise. This application does not limit its specific network architecture variants, noise scheduling strategies, or sampler selection. The pixel-level semantic segmentation model is a computer vision model that can assign a predefined category label to each pixel in an image. It achieves pixel-level classification by extracting and upsampling deep features of the image. This application does not limit its specific network implementation.

[0021] In step 102, the three-dimensional surface model of the target garment is unfolded using conformal parametric method to obtain a two-dimensional pattern planar diagram, a two-way mapping relationship between the three-dimensional surface and the two-dimensional pattern, and a local area deformation factor during the mapping process.

[0022] In some embodiments, reference Figure 2 As shown in the figure, this is a schematic diagram of the process of implementing conformal parametric unfolding in some embodiments of this application. The conformal parametric unfolding of the three-dimensional surface model of the target garment to obtain a two-dimensional pattern planar view, the bidirectional mapping relationship between the three-dimensional surface and the two-dimensional pattern, and the local area deformation factor during the mapping process can be achieved through the following steps: First, in step 1021, the geometric and topological data of the three-dimensional surface model of the target garment are extracted; Then, in step 1022, the conformal parameterization optimization problem is solved using the geometric and topological data to obtain the coordinate transformation that maps the three-dimensional surface to the two-dimensional plane; Finally, in step 1023, a two-dimensional plate planar view, a two-way mapping relationship between the three-dimensional curved surface and the two-dimensional plate, and a local area deformation factor in the mapping process are generated based on the coordinate transformation results.

[0023] In specific implementation, the extraction of geometric and topological data of the three-dimensional surface model of the target garment can be achieved in the following way: for example, loading the three-dimensional surface model file of the target garment, which is usually stored in the form of a triangular mesh, parsing and extracting the basic data required for its discrete surface representation as the geometric and topological data; specifically, the geometric data includes the three-dimensional spatial coordinates of each vertex constituting the surface mesh, and the topological data includes an edge list and a face list defining the mesh connection relationship, wherein each face is defined by its three vertex indices; as a preferred embodiment, the original mesh can be preprocessed before data extraction, including removing duplicate vertices, repairing non-manifold edges and holes, to ensure the integrity of the mesh and computational stability; in other embodiments, if the original model is a parametric surface, it needs to be discretized into a triangular mesh before data extraction, which is not limited in this application.

[0024] It should be noted that the three-dimensional surface model of the target garment described in this application refers to a virtual garment surface characterized in the form of a digital triangular mesh, which has a specific style and pattern structure. It is a three-dimensional spatial carrier for pattern mapping and effect display, thereby providing a precise geometric basis for the visual transformation from two-dimensional patterns to three-dimensional garments. In addition, the geometric and topological data refers to the set of basic digital information extracted from the three-dimensional surface model to describe its spatial shape and connection relationship.

[0025] In specific implementation, the conformal parameterization optimization problem is solved using the geometric and topological data. The coordinate transformation mapping the three-dimensional surface to the two-dimensional plane can be achieved in the following way: based on the extracted triangular mesh data, the least squares conformal mapping algorithm is used to construct and solve the conformal parameterization optimization problem. The core of this algorithm is to establish a quadratic optimization problem with the objective of minimizing the conformal distortion energy (i.e., angular deformation) of all triangular faces in the mesh. First, select one or a set of vertices as the boundary for the three-dimensional mesh and map it to the boundary of a simple convex polygon (such as a rectangle) on the two-dimensional plane to fix the boundary conditions. Then, the remaining internal... The two-dimensional coordinates of the vertices are taken as unknown variables. Based on the constraint that each triangular facet should keep its three interior angles unchanged as much as possible (conformalism), a set of linear equations about these unknown variables is established. Finally, by solving this large sparse linear equation system, the optimal coordinates of all interior vertices on the two-dimensional plane are obtained. This entire mapping from three-dimensional vertices to their corresponding two-dimensional coordinates constitutes the coordinate transformation. As a preferred embodiment, the selection of the boundary vertices can be determined based on the geodesic distance of the grid or principal component analysis to optimize the unfolded shape. In other embodiments, conformal parameterization methods based on centroid coordinates or harmonic mapping can also be used for solving, and this application does not limit this.

[0026] It should be noted that the coordinate transformation mentioned in this application refers to a complete set of mathematical mapping rules from the coordinates of vertices of a three-dimensional surface to the coordinates of points in a two-dimensional plane, obtained by solving the conformal parameterization problem.

[0027] In specific implementation, the generation of a two-dimensional plate planar diagram, the bidirectional mapping relationship between the three-dimensional surface and the two-dimensional plate, and the local area deformation factor during the mapping process based on the coordinate transformation results can be achieved in the following way: First, all three-dimensional vertices are drawn on the plane according to the solved two-dimensional coordinates, and the points are connected in the two-dimensional plane strictly according to the original triangular facet topological connection relationship (defined by the vertex index list) to form a planar region filled with triangular facets. The overall outer contour of this region constitutes the two-dimensional plate planar diagram. At the same time, a mapping relationship table is established and stored. This table records the unique identifier, three-dimensional coordinates, and calculation of each three-dimensional vertex and its corresponding triangular facet. The calculated two-dimensional coordinates establish a bidirectional mapping relationship from the three-dimensional surface to the two-dimensional plate (through coordinate transformation) and from the two-dimensional plate back to the three-dimensional surface (through index lookup). Further, each triangular facet is traversed, and its area on the original three-dimensional surface and the area of ​​its corresponding projected region on the two-dimensional plate are calculated. The ratio of the two-dimensional area to the three-dimensional area is used as the local area deformation factor to measure the degree of deformation of the local region. In a preferred embodiment, the area deformation factor can be attached as attribute data to each two-dimensional triangular facet for subsequent visualization or analysis. In other embodiments, the average area deformation of each vertex can also be calculated as its local deformation index; this application does not limit this approach.

[0028] It should be noted that the two-dimensional pattern planar view mentioned in this application refers to the two-dimensional polygonal region obtained after unfolding the three-dimensional garment surface through conformal parameterization. Its shape reflects the true outline and size of the garment pattern in a planar state. The bidirectional mapping relationship between the three-dimensional surface and the two-dimensional pattern is a data structure that accurately records the one-to-one correspondence between three-dimensional mesh elements (vertices, faces) and their projection positions on the two-dimensional pattern, establishing a reversible, point-to-point geometric link between the surface and the plane in digital space. The local area deformation factor refers to the dimensionless value of the area scaling ratio of each tiny local area on the three-dimensional surface when mapped to the two-dimensional plane during the conformal parameterization process, used to characterize the degree of area distortion introduced by the mapping process.

[0029] In step 103, the pattern layout on the two-dimensional pattern sheet is automatically planned according to the garment process rules to obtain the pattern layout scheme on the garment piece.

[0030] In some embodiments, the automatic planning on the two-dimensional pattern drawing according to garment manufacturing rules to obtain the pattern layout scheme on the garment piece can be achieved through the following steps: Based on the garment manufacturing process rules and the properties of the two-dimensional planar pattern, geometric constraints and optimization objectives for the pattern arrangement are generated. Using the closed outline of the two-dimensional plate planar diagram as the container area and the two-dimensional planar pattern as the object to be placed, a two-dimensional irregular shape nested optimization model is constructed by combining the geometric constraints and optimization objectives. Solve the nested optimization model of the two-dimensional irregular shape to obtain the layout scheme of the pattern on the garment piece.

[0031] In specific implementation, based on the garment manufacturing process rules and the attributes of the two-dimensional planar pattern, generating geometric constraints and optimization objectives for the pattern layout can be achieved in the following ways: First, read the layout rule set related to the current target garment type (such as a shirt or dress) from a pre-set garment manufacturing process knowledge base. The rule set defines, in the form of structured data, the allowed position range of the pattern on the garment piece, the prohibited rotation direction, the pattern orientation that must be maintained, and the maximum allowable deformation ratio of different pattern semantic regions (such as "main graphic" and "background texture"). At the same time, parse the attributes of the two-dimensional planar pattern from the generated semantic segmentation map, and extract its outer contour polygon, semantic region mask, and region area. Subsequently, transform the manufacturing process rules into specific mathematical constraints, such as transforming "the pattern cannot be inverted" into a condition related to the overall rotation of the pattern. The constraint of the angle range transforms the requirement that "the main graphic should be located in the center area of ​​the front piece" into a requirement that the center of gravity of the main graphic mask must fall inside a preset target polygon in the pattern planar diagram. Finally, the design goal is transformed into a calculable optimization goal, such as setting the optimization goal to maximize the fabric utilization rate of the pattern (i.e., the ratio of the total area of ​​the placed pattern to the area of ​​the pattern), or to minimize the expected visual deformation caused by the local area deformation factor when the pattern is mapped back to the three-dimensional curved surface. As a preferred embodiment, different deformation tolerances can be set for different semantic regions in the pattern, setting stricter deformation constraints for the main graphic and a looser deformation allowance range for the background texture. In other embodiments, alignment rules (such as aligning multiple patterns with the sewing lines of the garment pieces) can also be introduced as additional optimization goals or hard constraints, which are not limited in this application.

[0032] It should be noted that the garment process rules mentioned in this application refer to a set of experiential knowledge or design requirements encoded in a structured form regarding how patterns should be placed and presented on specific garment styles; the attributes of the two-dimensional planar pattern refer to a set of parameters parsed from the generated planar pattern and its semantic segmentation map, used to describe the geometric and semantic features of the pattern itself, which is a digital description of the pattern as an object to be laid out; the geometric constraints and optimization objectives refer to specific expressions that transform the garment process rules and design objectives into mathematical models, which are used to limit the feasibility boundaries of pattern layout and measure the quality of layout schemes, thereby standardizing the intuitive design problem into a mathematical optimization problem that can be solved by a computer.

[0033] In specific implementation, the closed contour of the two-dimensional plate planar image is used as the container region, and the two-dimensional planar pattern is used as the object to be placed. Combining the geometric constraints and optimization objectives, the two-dimensional irregular shape nested optimization model can be constructed in the following way: for example, the outer closed polygon of the two-dimensional plate planar image is defined as the container boundary of the nested problem; the contour polygon (or its semantic sub-region contour) of the two-dimensional planar pattern is defined as the object to be placed; the decision variables of the optimization model are the two-dimensional translation coordinates, rotation angle, and an optional uniform scaling factor for each pattern instance to be placed; the constraints of the model include: the contours of all pattern instances after translation and rotation must be completely within the container boundary polygon; overlap between the contours of any two pattern instances is not allowed; and all geometric constraints generated in step 1 regarding position, orientation, and semantic region deformation must be obtained. The optimization objective of the model is the defined objective function. To efficiently handle collision detection of irregular shapes, a geometric reasoning method based on No-Fit Polygons (NFPs) can be used. The NFPs between the publishing container and the pattern, as well as between the patterns, are pre-calculated. The overlap detection between polygons is transformed into the positional relationship judgment between points and NFPs, thereby transforming the complex geometric constraints into a series of linear or nonlinear inequalities, and completing the mathematical construction of the two-dimensional irregular shape nested optimization model. As a preferred embodiment, when the pattern contains multiple separable semantic sub-regions, each sub-region can be treated as an independent object, and a constraint that its relative position must remain fixed can be added to maintain the integrity of the pattern. In other embodiments, a grid-based or distance field-based representation method can also be used to approximate the shape and construct the model. This application does not limit this.

[0034] It should be noted that the two-dimensional irregular shape nested optimization model described in this application refers to a mathematical programming model that uses the plate outline as a container and the pattern outline as an object, and incorporates geometric constraints and optimization objectives. It abstracts the automatic pattern layout problem into a constrained irregular shape packing problem.

[0035] In specific implementation, solving the nested optimization model of the two-dimensional irregular shape to obtain the pattern layout scheme on the garment piece can be achieved in the following ways, for example: using a heuristic optimization algorithm to solve the constructed nested optimization model of the two-dimensional irregular shape; taking a genetic algorithm as an example, each possible layout (i.e., the set of position, rotation, and scaling parameters of all pattern instances) is encoded as a chromosome, and an initial population containing multiple such chromosomes is randomly generated; then, a fitness function is designed to evaluate each chromosome, the value of which is mainly determined by the objective function value of the model optimization, and a penalty term proportional to the degree of violation is applied to individuals that violate constraints (such as exceeding the boundary or overlapping) to reduce their fitness; subsequently, selection, crossover, and mutation operations are iteratively performed to evolve the population, wherein the mutation operation... The process includes making small random perturbations to the position and angle of the pattern; in each generation of evolution, pre-calculated NFP or real-time geometric calculations are used to verify the degree to which new individuals meet the constraints; when the evolution reaches a preset number of iterations or the fitness converges to a stable threshold, the individual with the highest fitness is selected from the final population, and its chromosome is decoded to obtain the precise position coordinates, rotation angle and scaling factor of each pattern instance on the two-dimensional plate planar diagram. This set of parameters constitutes the layout scheme of the pattern on the garment piece; as a preferred embodiment, simulated annealing algorithm can be combined to enhance the local search capability to escape possible local optima; in other embodiments, for problems of smaller scale or with highly simplified constraints, an accurate mathematical programming solver can also be used to solve them, and this application does not limit this.

[0036] It should be noted that the layout scheme mentioned in this application refers to the solution of the nested optimization model, which is specifically represented by the set of precise placement parameters of each pattern instance on the two-dimensional plate plan, and is the decision result connecting the pattern design intention with the final spatial position of the garment carrier.

[0037] In step 104, the two-dimensional planar pattern is attached to the corresponding area of ​​the two-dimensional plate planar image according to the layout scheme, and then differentiated texture mapping is performed on different semantic regions identified by the semantic segmentation map in the two-dimensional planar pattern according to the local area deformation factor to obtain a pre-adapted two-dimensional plate planar pattern.

[0038] In some embodiments, the following steps can be used to attach the two-dimensional planar pattern to the corresponding area of ​​the two-dimensional plate planar image according to the layout scheme: The layout scheme is analyzed to obtain the target position, rotation angle, and scaling factor of the two-dimensional planar pattern on the two-dimensional plate planar diagram; Based on the target position, rotation angle, and scaling factor, determine the corresponding texture coordinates of each sampling point on the two-dimensional plate planar image in the two-dimensional planar pattern texture space; The two-dimensional planar pattern is sampled and transformed according to the texture coordinates, thereby fitting the two-dimensional planar pattern to the corresponding area of ​​the two-dimensional plate planar image.

[0039] In specific implementation, parsing the layout scheme to obtain the target position, rotation angle, and scaling factor of the two-dimensional planar pattern on the two-dimensional plate planar diagram can be achieved in the following way: for example, the layout scheme is a set of parameters consisting of the position coordinates, rotation angle, and scaling factor of each pattern instance; for each pattern instance in the scheme, its position coordinates are extracted as the target position, its rotation angle is extracted as the rotation angle, and its scaling factor is extracted as the scaling factor; this process completes the parsing of the layout scheme and obtains the explicit parameters required for geometric transformation; as a preferred embodiment, the logical validity of the parameters can be basically verified after parsing; in other embodiments, if the layout scheme is stored in a non-parametric form, equivalent transformation parameters need to be derived from it through calculation methods, which is not limited in this application.

[0040] It should be noted that the target position, rotation angle and scaling factor mentioned in this application are the three elements constituting the placement state of a pattern instance in the layout scheme, which together define the rigid body transformation of the pattern from its original state to its final state on the plate.

[0041] In specific implementation, the corresponding texture coordinates of each sampling point on the two-dimensional plate planar image in the two-dimensional planar pattern texture space can be determined according to the target position, rotation angle, and scaling factor. This can be achieved in the following way: First, define a regular pixel grid or arbitrary set of sampling points covering the entire area of ​​the two-dimensional plate planar image; then, define a normalized texture coordinate system for the original image of the two-dimensional planar pattern, with its origin usually located at a corner of the image and its coordinate range typically [0, 1]; next, construct an affine transformation from the plate planar coordinates to the pattern texture coordinates; this transformation is determined by the target position (translation). The rotation angle (rotation) and scaling factor (scaling) are defined together. For each sampling point on the plate plane, its coordinates are first translated to a relative coordinate system with the target position as the origin, then the reverse rotation is applied, and finally the scaling factor is divided and the origin offset of the pattern texture coordinate system is added to calculate the texture coordinates corresponding to the sampling point in the two-dimensional planar pattern texture space. As a preferred embodiment, the plate plane coordinates can be normalized before calculation to simplify the transformation operation. In other embodiments, the entire transformation matrix can be pre-calculated, and the coordinate transformation of all sampling points can be completed by a single matrix multiplication. This application does not limit this.

[0042] It should be noted that the corresponding texture coordinates mentioned in this application refer to the normalized sampling position of each point on the two-dimensional plate planar image, which is calculated according to the layout scheme and is located on the original two-dimensional planar pattern texture image.

[0043] In specific implementation, sampling and image transformation of the two-dimensional planar pattern based on the texture coordinates to fit the two-dimensional planar pattern onto the corresponding area of ​​the two-dimensional plate planar image can be achieved in the following way: For example, create a blank output image that matches the size and spatial range of the two-dimensional plate planar image area; traverse each pixel of the output image or each point in the sampling point set to obtain its texture coordinates calculated by the aforementioned steps; then, based on the texture coordinates, sample the original image data of the two-dimensional planar pattern to obtain the corresponding color value; the sampling process can use bilinear interpolation. The method is to smooth out the jagged effect that may be caused by the non-integerization of coordinates; finally, the sampled color value is assigned to the corresponding pixel position of the output image; this process is repeated until all pixels are processed, and the generated output image is the result image after the two-dimensional planar pattern is pasted onto the corresponding area of ​​the two-dimensional plate planar image according to the layout scheme; as a preferred embodiment, when the texture coordinates exceed the effective range [0, 1] of the pattern texture space, a preset boundary processing strategy can be adopted, such as clamping, repeating or transparent filling; in other embodiments, this process can be accelerated in parallel using a graphics processor, which is not limited in this application.

[0044] It should be noted that the sampling and image transformation described in this application refers to the calculation process of obtaining color values ​​from the original pattern image and filling them into the corresponding positions of the plate plan based on the calculated texture coordinates. This is a key image processing step for performing texture bonding and visualizing the layout scheme, thereby generating an intermediate texture image without deformation compensation.

[0045] In some embodiments, reference Figure 3 As shown in the figure, this is a schematic diagram of the process of implementing texture mapping according to some embodiments of this application. Differentiated texture mapping is performed on different semantic regions identified by the semantic segmentation map in the two-dimensional planar pattern according to the local area deformation factor to obtain a pre-adapted two-dimensional plate pattern. This can be achieved by the following steps: Based on the semantic segmentation map, a mapping relationship is established between different semantic regions and preset deformation resistance parameters; The differentiated texture coordinate correction amount is determined based on the deformation resistance parameter and the corresponding local area deformation factor of the semantic region to which each local region of the two-dimensional plate planar image belongs. By locally adjusting the texture coordinates of the intermediate texture image using the differentiated texture coordinate correction amount, a pre-adapted two-dimensional plate pattern is obtained.

[0046] In specific implementation, establishing the mapping relationship between different semantic regions and preset deformation resistance parameters based on the semantic segmentation map can be achieved in the following way: First, the semantic segmentation map is obtained, which identifies the semantic category to which each pixel in the pattern belongs in the form of pixel labels; then, the preset deformation resistance parameter values ​​for different semantic categories are read from the preset design rule library. The deformation resistance parameter is a dimensionless scalar coefficient used to quantify the tendency of the semantic region to resist or compensate for geometric deformation during texture mapping. The larger the value, the more the region should maintain the original texture and resist deformation as much as possible, and the smaller the value, the more texture adaptation deformation is allowed; then, a direct mapping dictionary from semantic category labels to deformation resistance parameter values ​​is established as the mapping relationship. For example, "main graphic" is mapped to a higher resistance parameter, and "background texture" is mapped to a lower resistance parameter; as a preferred embodiment, the deformation resistance parameter can be preset according to the artistic importance or visual salience of the pattern, and a higher resistance value is set for the visual focus area; in other embodiments, users can also fine-tune the resistance parameters of different categories through interface interaction, which is not limited in this application.

[0047] It should be noted that the mapping relationship described in this application specifically refers to the association rule established based on the semantic segmentation map, which connects the semantic category label of the pattern pixel to the preset deformation resistance parameter value, and is used to assign differentiated deformation processing preferences to pattern regions with different semantic functions.

[0048] In specific implementation, determining the differentiated texture coordinate correction amount based on the deformation resistance parameter and the corresponding local area deformation factor of the semantic region to which each local region of the two-dimensional plate planar image belongs can be achieved in the following way: For example, traverse each local region of the two-dimensional plate planar image, where a local region is defined as each triangular facet or a regular grid unit constituting the plate planar image; for each local region, firstly, through the bidirectional mapping relationship, find the original facet on its corresponding three-dimensional surface and obtain the local area deformation factor of the facet; simultaneously, based on the position of the local region on the two-dimensional plate planar image, map the coordinates to the corresponding position of the semantic segmentation image, query the main semantic category of the pixels it covers, and then obtain the deformation resistance parameter corresponding to the semantic category through the aforementioned mapping relationship; next, based on the local area deformation factor and the deformation resistance parameter, calculate the texture coordinate correction amount of the local region through a preset correction function; the correction function defines how to generate an offset vector for adjusting the texture sampling position based on the area deformation factor (representing geometric compression or stretching) and the resistance parameter (representing deformation compensation intention); for example, the correction amount can be calculated as: the offset vector and (1 - The product of (area deformation factor) and (1 - deformation resistance parameter) is proportional, so that in areas with large area compression and low resistance parameter, the texture coordinates are adjusted more significantly to compensate for visual deformation. As a preferred embodiment, the correction function can be designed as a piecewise linear or nonlinear function to finely control the deformation compensation behavior. In other embodiments, smoothing filtering can also be introduced to ensure the continuity of the correction amount in adjacent areas and avoid texture tearing. This application does not limit this.

[0049] It should be noted that the differentiated texture coordinate correction amount mentioned in this application refers to a texture coordinate adjustment vector dynamically calculated based on the area deformation factor of each local region and the deformation resistance parameter of its semantic region. It is the core control variable for realizing different degrees of deformation pre-compensation for different semantic regions.

[0050] In specific implementation, the local adjustment of the texture coordinates of the intermediate texture image using the differentiated texture coordinate correction amount to obtain the pre-adapted 2D plate pattern can be achieved in the following way: First, obtain the intermediate texture image, that is, the image in which the 2D planar pattern has been fitted to the corresponding area of ​​the 2D plate planar image but without deformation compensation, and the original texture coordinates corresponding to each pixel or each local area on the intermediate texture image; then, for each target pixel position of the pre-adapted image to be output, find the local area to which it belongs, and apply the texture coordinate correction amount calculated for that area to correct the original texture coordinates to obtain the adjusted new texture coordinates; specifically, if the original texture coordinates are (u, v) and the correction amount is (Δu, Δv), then the new coordinates are (u+Δu, v). v+Δv); Subsequently, using the adjusted new texture coordinates, texture sampling is performed again from the original two-dimensional planar pattern generated in step 101 to obtain color values ​​and fill them into the corresponding pixel positions of the pre-adapted image; this process is repeated until all pixels are processed, and finally a two-dimensional plate pattern whose texture has been differentially adjusted to pre-compensate for the expected geometric deformation during three-dimensional mapping is generated, namely the pre-adapted two-dimensional plate pattern; as a preferred embodiment, texture sampling can use bilinear or bicubic interpolation to ensure visual quality; in other embodiments, this process can be efficiently implemented on a graphics processor through fragment shader programming, which is not limited in this application.

[0051] It should be noted that the pre-adapted two-dimensional pattern described in this application refers to the final two-dimensional pattern obtained after applying differential texture coordinate correction to the intermediate texture image. Its texture content has been intelligently pre-distorted according to the expected three-dimensional deformation. It is a planar texture version customized to achieve a "natural fit" visual effect on a three-dimensional curved surface.

[0052] In step 105, the pre-adapted two-dimensional pattern is wrapped back onto the three-dimensional curved surface model of the target garment through the bidirectional mapping relationship, generating a three-dimensional garment design rendering in which the pattern naturally fits the curved surface of the garment.

[0053] In some embodiments, the process of wrapping the pre-adapted two-dimensional pattern back onto the three-dimensional surface model of the target garment through the bidirectional mapping relationship to generate a three-dimensional garment design rendering in which the pattern naturally fits the garment surface can be achieved through the following steps: Based on the bidirectional mapping relationship, the texture information on the pre-adapted two-dimensional pattern is reverse-mapped to the corresponding surface of the target garment's three-dimensional curved surface model; The mapping results are processed with material and lighting settings and graphic rendering to generate a 3D clothing design rendering in which the pattern and the curved surface of the garment fit together naturally.

[0054] In specific implementation, mapping each pixel on the pre-adapted 2D pattern to its corresponding 3D spatial position on the target garment's 3D curved surface model based on the bidirectional mapping relationship can be achieved in the following way: First, obtain the bidirectional mapping relationship, which records the coordinates of each vertex (and / or triangular facet) of the 3D curved surface mesh and its corresponding point on the 2D pattern plane; for each vertex on the 3D curved surface model or each surface sampling point that needs to be textured, find its corresponding 2D texture coordinates on the pre-adapted 2D pattern by querying the mapping relationship; this query process utilizes the "from 3D to 2D" mapping direction in the mapping relationship. If the query target is a 3D vertex, its corresponding 2D coordinates can be obtained directly; if it is a curved surface... For any interpolation point on the surface, its texture coordinates need to be calculated by interpolating the centroid coordinates using the corresponding two-dimensional coordinates of the three vertices of the triangular facet. Subsequently, using the calculated texture coordinates, texture sampling is performed on the image data of the pre-adapted two-dimensional plate pattern to obtain the corresponding color value, and this color value is assigned as texture information to the three-dimensional vertex or sampling point, thereby completing the reverse mapping and attachment of texture information from the two-dimensional plane to the three-dimensional surface. As a preferred embodiment, texture sampling can use bilinear interpolation to smooth the texture and ensure that the pattern of the three-dimensional surface after mapping is continuous and natural. In other embodiments, if the resolution of the pre-adapted pattern does not match the density of the surface mesh, texture filtering techniques such as Mipmap can be used to optimize the visual quality at different viewing distances. This application does not limit this.

[0055] In practical implementation, the mapping results are processed with material and lighting settings and graphics rendering to generate a 3D clothing design rendering where the pattern naturally fits the garment's curved surface. This can be achieved in the following way: In a 3D graphics engine or rendering environment, material properties are configured for the target garment's 3D curved surface model, which already has textures attached. These material properties include at least the specular coefficient and roughness coefficient used to define the surface's response to light. Simultaneously, one or more virtual light sources are set up in the 3D scene, defining their type (e.g., parallel light, point light), color, intensity, and direction of illumination. A virtual camera is also set up, and its viewing position, perspective, and projection method (e.g., perspective projection) are determined. Then, the graphics engine's lighting and rendering pipeline is invoked, and the settings are applied according to the parameters. The material properties, light source parameters, and camera parameters are used to perform real-time lighting calculations and rasterization rendering on the 3D clothing model. This rendering process calculates the interaction between light and each pixel on the model surface, and combines it with the basic texture color to generate a 2D image with realistic details such as brightness, shadows, and highlights. The final output 2D image is a 3D clothing design rendering that shows the natural fit between the pattern and the curved surface of the clothing. As a preferred embodiment, simple background and ambient light occlusion can be added to the scene to enhance the sense of three-dimensionality and realism. In other embodiments, physically based rendering processes or offline rendering technologies such as path tracing can also be used to obtain more realistic static renderings or dynamic display videos, which is not limited in this application.

[0056] It should be noted that the three-dimensional clothing design renderings mentioned in this application refer to two-dimensional display images generated by wrapping a pre-adapted two-dimensional pattern back into the three-dimensional clothing surface through a two-way mapping relationship and then rendering it using graphics rendering techniques such as material and lighting.

[0057] Furthermore, in another aspect of this application, in some embodiments, this application provides an artificial intelligence-based clothing design and customization system, referencing... Figure 4 The figure is a schematic diagram of the structure of an AI-based clothing design and customization system according to some embodiments of this application. The AI-based clothing design and customization system 200 includes a generation module 201, a processing module 202, and an execution module 203, which are described below: The generation module 201 in this application is mainly used to receive the pattern theme description input by the user, and then generate a two-dimensional planar pattern that conforms to the description and its corresponding semantic segmentation map. Processing module 202, in this application, is mainly used to perform conformal parametric unfolding of the three-dimensional curved surface model of the target garment to obtain a two-dimensional pattern planar view, a two-way mapping relationship between the three-dimensional curved surface and the two-dimensional pattern, and a local area deformation factor in the mapping process. In addition, the processing module 202 in this application is also used to automatically plan on the two-dimensional pattern drawing according to the garment process rules to obtain the pattern layout scheme on the garment piece; In addition, the processing module 202 in this application is also used to attach the two-dimensional planar pattern to the corresponding area of ​​the two-dimensional plate planar image according to the layout scheme, and then perform differentiated texture mapping on different semantic regions identified by the semantic segmentation map in the two-dimensional planar pattern according to the local area deformation factor to obtain a pre-adapted two-dimensional plate pattern. The execution module 203 in this application is mainly used to wrap the pre-adapted two-dimensional pattern back onto the three-dimensional curved surface model of the target garment through the bidirectional mapping relationship, thereby generating a three-dimensional garment design rendering in which the pattern and the curved surface of the garment are naturally aligned.

[0058] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described artificial intelligence-based clothing design and customization method.

[0059] In some embodiments, reference Figure 5 The figure is an internal structural diagram of a computer device implementing an AI-based clothing design and customization method according to some embodiments of this application. The AI-based clothing design and customization method in the above embodiments can be implemented through... Figure 5 The computer device shown is used to implement this, and the computer device 300 includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0060] The processor 301 may be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of the AI-based clothing design and customization method described in this application.

[0061] The communication bus 302 is used to transmit information between the aforementioned components.

[0062] Memory 303 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 303 may exist independently and be connected to processor 301 via communication bus 302. Memory 303 may also be integrated with processor 301.

[0063] The memory 303 stores program code for executing the solution of this application, and its execution is controlled by the processor 301. The processor 301 executes the program code stored in the memory 303. The program code may include one or more software modules. In the above embodiments, the artificial intelligence-based clothing design customization method can be implemented by the processor 301 and one or more software modules in the program code in the memory 303.

[0064] Communication interface 304 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0065] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core processor or a multi-core processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0066] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device may be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0067] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described artificial intelligence-based clothing design and customization method.

[0068] In summary, the AI-based clothing design and customization method and system disclosed in this application receives a pattern theme description input by the user and generates a two-dimensional planar pattern that conforms to the description and its corresponding semantic segmentation map; the three-dimensional curved surface model of the target garment is unfolded using conformal parametric methods to obtain a two-dimensional pattern planar diagram, a bidirectional mapping relationship between the three-dimensional curved surface and the two-dimensional pattern planar diagram, and a local area deformation factor during the mapping process; automatic planning is performed on the two-dimensional pattern planar diagram according to clothing process rules to obtain a pattern layout scheme on the garment piece; the two-dimensional planar pattern is attached to the corresponding area of ​​the two-dimensional pattern planar diagram according to the layout scheme, and then differentiated texture mapping is performed on different semantic regions in the two-dimensional planar pattern identified by the semantic segmentation map according to the local area deformation factor to obtain a pre-adapted two-dimensional pattern planar diagram; the pre-adapted two-dimensional pattern planar diagram is wrapped back onto the three-dimensional curved surface model of the target garment through the bidirectional mapping relationship to generate a three-dimensional clothing design effect diagram in which the pattern and the garment surface are naturally attached; adaptive differentiated texture mapping can be achieved by combining the pattern semantic structure under the deformation differences in the high curvature area of ​​the garment surface model.

[0069] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0070] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for clothing design and customization based on artificial intelligence, characterized in that, Includes the following steps: It receives a pattern theme description input by the user and then generates a two-dimensional planar pattern that matches the description and its corresponding semantic segmentation map; The three-dimensional surface model of the target garment is unfolded using conformal parametric method to obtain a two-dimensional pattern planar diagram, a two-way mapping relationship between the three-dimensional surface and the two-dimensional pattern, and a local area deformation factor during the mapping process. Based on the garment manufacturing rules, the pattern layout on the two-dimensional pattern is automatically planned to obtain the pattern layout scheme on the garment piece. According to the layout scheme, the two-dimensional planar pattern is attached to the corresponding area of ​​the two-dimensional plate planar image, and then the local area deformation factor is used to perform differentiated texture mapping on different semantic regions in the two-dimensional planar pattern identified by the semantic segmentation image to obtain a pre-adapted two-dimensional plate planar pattern. The pre-adapted two-dimensional pattern is wrapped back onto the three-dimensional curved surface model of the target garment through the two-way mapping relationship, generating a three-dimensional garment design rendering in which the pattern naturally fits the curved surface of the garment.

2. The method as described in claim 1, characterized in that, The three-dimensional surface model of the target garment is conformally parametrically unfolded to obtain a two-dimensional pattern planar diagram, a two-way mapping relationship between the three-dimensional surface and the two-dimensional pattern, and local area deformation factors during the mapping process, specifically including: Extract the geometric and topological data of the 3D surface model of the target garment; By solving the conformal parameterization optimization problem using the geometric and topological data, a coordinate transformation is obtained that maps a three-dimensional surface to a two-dimensional plane. Based on the coordinate transformation results, a two-dimensional plate planar diagram, a two-way mapping relationship between the three-dimensional curved surface and the two-dimensional plate, and a local area deformation factor during the mapping process are generated.

3. The method as described in claim 1, characterized in that, Based on the garment manufacturing process rules, automatic planning is performed on the two-dimensional pattern design to obtain the pattern layout scheme on the garment pieces. Specifically, this includes: Based on the garment manufacturing process rules and the properties of the two-dimensional planar pattern, geometric constraints and optimization objectives for the pattern arrangement are generated. Using the closed outline of the two-dimensional plate planar diagram as the container area and the two-dimensional planar pattern as the object to be placed, a two-dimensional irregular shape nested optimization model is constructed by combining the geometric constraints and optimization objectives. Solve the nested optimization model of the two-dimensional irregular shape to obtain the layout scheme of the pattern on the garment piece.

4. The method as described in claim 1, characterized in that, The specific steps of fitting the two-dimensional planar pattern to the corresponding area of ​​the two-dimensional plate planar image according to the layout scheme include: The layout scheme is analyzed to obtain the target position, rotation angle, and scaling factor of the two-dimensional planar pattern on the two-dimensional plate planar diagram; Based on the target position, rotation angle, and scaling factor, determine the corresponding texture coordinates of each sampling point on the two-dimensional plate planar image in the two-dimensional planar pattern texture space; The two-dimensional planar pattern is sampled and transformed according to the texture coordinates, thereby fitting the two-dimensional planar pattern to the corresponding area of ​​the two-dimensional plate planar image.

5. The method as described in claim 1, characterized in that, Based on the local area deformation factor, differentiated texture mapping is performed on different semantic regions identified by the semantic segmentation map in the two-dimensional planar pattern to obtain a pre-adapted two-dimensional plate pattern, specifically including: Based on the semantic segmentation map, a mapping relationship is established between different semantic regions and preset deformation resistance parameters; The differentiated texture coordinate correction amount is determined based on the deformation resistance parameter and the corresponding local area deformation factor of the semantic region to which each local region of the two-dimensional plate planar image belongs. By locally adjusting the texture coordinates of the intermediate texture image using the differentiated texture coordinate correction amount, a pre-adapted two-dimensional plate pattern is obtained.

6. The method as described in claim 1, characterized in that, The process of wrapping the pre-adapted two-dimensional pattern back onto the three-dimensional curved surface model of the target garment through the bidirectional mapping relationship to generate a three-dimensional garment design rendering in which the pattern naturally fits the curved surface of the garment specifically includes: Based on the bidirectional mapping relationship, the texture information on the pre-adapted two-dimensional pattern is reverse-mapped to the corresponding surface of the target garment's three-dimensional curved surface model; The mapping results are processed with material and lighting settings and graphic rendering to generate a 3D clothing design rendering in which the pattern and the curved surface of the garment fit together naturally.

7. The method as described in claim 1, characterized in that, The user-input pattern description refers to any short natural language sentence that can define the visual content of the pattern.

8. An artificial intelligence-based clothing design and customization system, characterized in that, include: The generation module is used to receive the pattern theme description input by the user, and then generate a two-dimensional planar pattern that conforms to the description and its corresponding semantic segmentation map; The processing module is used to perform conformal parametric unfolding of the three-dimensional curved surface model of the target garment to obtain a two-dimensional pattern planar diagram, a two-way mapping relationship between the three-dimensional curved surface and the two-dimensional pattern, and a local area deformation factor during the mapping process. The processing module is also used to automatically plan on the two-dimensional pattern drawing according to the garment process rules to obtain the pattern layout scheme on the garment piece; The processing module is further configured to attach the two-dimensional planar pattern to the corresponding area of ​​the two-dimensional plate planar image according to the layout scheme, and then perform differentiated texture mapping on different semantic regions identified by the semantic segmentation map in the two-dimensional planar pattern according to the local area deformation factor to obtain a pre-adapted two-dimensional plate planar pattern. The execution module is used to wrap the pre-adapted two-dimensional pattern back onto the three-dimensional curved surface model of the target garment through the bidirectional mapping relationship, generating a three-dimensional garment design rendering in which the pattern and the garment surface naturally fit together.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the artificial intelligence-based clothing design and customization method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based clothing design and customization method as described in any one of claims 1 to 7.

Images