A method and system for automatic generation of garment patterns based on user interaction

Abstract

The application relates to the technical field of garment design pattern generation, and discloses a garment pattern automatic generation method and system based on user interaction, wherein the method comprises the following steps: acquiring cross-platform global trend pattern data in real time; receiving a user text keyword and generating a final condition embedding vector; outputting a two-dimensional pattern piece set by a multi-modal pattern generation model; performing affine mapping optimization and color difference minimization processing on the two-dimensional pattern piece set; and generating a final complete garment pattern. Compared with the prior art, the technical problem that the aesthetic preference of a target market and the garment contour adaptation requirement cannot be simultaneously met under cross-cultural trend conditions, resulting in pattern mispositioning and joint texture interruption, is solved. Due to the synergistic process of trend weight calibration, multi-modal condition embedding and joint continuity constraint optimization, high-consistency garment pattern automatic generation under different cultural aesthetics and various garment patterns is realized, and the automatic garment pattern design level is improved.

Description

Technical Field

[0001] This invention relates to the field of clothing design graphic generation technology, and in particular to a method and system for automatically generating clothing patterns based on user interaction. Background Technology

[0002] Currently, the design and production of clothing patterns mostly rely on a combination of human creativity and static material libraries. Although deep learning and generative models (such as GAN and Diffusion Model) have been widely used in the field of image generation in recent years, there are still significant shortcomings in areas such as dynamic adaptation to cross-cultural trends, understanding of multimodal interactions, intelligent layout of clothing silhouettes, and control of seam continuity.

[0003] For example, in the trend analysis stage, existing solutions often rely solely on a single data source (such as sales rankings on a particular e-commerce platform or popular hashtags on a single social media platform) for popularity statistics, lacking cross-platform trend integration and time decay mechanisms, and failing to capture the dynamic characteristics of design element weight changes over time under different cultural backgrounds. Especially in the global market, different regions have significant differences in preferences for colors, patterns, symbols, and typography, making it difficult for traditional methods to accurately reflect the aesthetic weights of the target cultural region during the generation stage. In the user intent parsing stage, most existing systems only support single-round text description input, lacking multimodal fusion capabilities, and cannot simultaneously process multi-source information such as keywords, reference images, and historical interaction records. Furthermore, they cannot perform semantic disambiguation and weight correction based on a target cultural feature lexicon, leading to style deviations, cultural misuse, and even symbolic taboo conflicts in the generated results. In the pattern generation stage, while existing deep generation models can output high-quality patterns, their conditional control granularity is insufficient, lacking layer-by-layer feature generation strategies based on style modulation and seamless tiling constraints. The generated patterns often fail to maintain consistent texture direction and continuous seams after cropping and splicing. For ready-to-wear garments that need to be mapped to different silhouettes (such as A-line skirts, hoodies, oversized shirts, etc.), most existing methods use uniform scaling or simple cutting, ignoring local non-rigid deformation and seam direction constraints, resulting in problems such as pattern misalignment, uneven color difference, and local stretching after garment assembly.

[0004] Therefore, there is an urgent need for a user-interactive automatic garment pattern generation method to achieve automatic generation of multimodal garment patterns oriented towards target cultural aesthetics, while maintaining continuous seams and consistent colors, so as to improve the matching degree between garment patterns and market demands, the consistency of garment pattern adaptation, and the automation level of garment pattern design. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose an automatic clothing pattern generation method based on user interaction. This method aims to solve the technical problem in existing technologies where the aesthetic preferences of the target market and the requirements for clothing silhouette adaptation cannot be simultaneously met under cross-cultural trend conditions, resulting in pattern misalignment and interrupted seam textures.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a method for automatically generating clothing patterns based on user interaction.

[0007] The method for automatically generating clothing patterns based on user interaction includes:

[0008] Step S10: Obtain the clothing pattern dataset, perform the design element recognition task based on the clothing pattern dataset, and extract the design category element i; obtain the target cultural region c, and process the design category element i and the target cultural region c using a time-sensitive cross-platform trend weight calibration mechanism to output the design category element weight set E;

[0009] Step S20: Receive user interaction input, and based on the user interaction input and the design category element weight set E, perform text processing, image processing, design category element mapping processing and fusion processing in sequence, and output the final conditional embedding vector F;

[0010] Step S30: Input the final condition embedding vector F into the pre-trained multimodal clothing pattern generation model. The multimodal clothing pattern generation model outputs a set of two-dimensional pattern pieces of the target clothing silhouette. ;

[0011] Step S40: Based on the two-dimensional cut piece set A multi-scale affine mapping optimization method with seam continuity constraints is used to process cross-pattern splicing and color difference minimization, outputting an optimized pattern mapping set. ;

[0012] Step S50: Map the output pattern pieces to a set. The data is input into a preset virtual garment assembly network, which then outputs the final complete garment pattern.

[0013] Preferably, in step S10, the clothing pattern dataset includes high-frequency tagged patterns from the Pinterest platform, best-selling patterns from e-commerce sales data, and fashion image data from social media; the design category elements include floral symbols, geometric symbols, animal symbols, and traditional symbols.

[0014] Preferably, step S10, which involves using a time-sensitive cross-platform trend weight calibration mechanism to process design category element i and target cultural region c, and outputting the design category element weight set E, specifically includes:

[0015] For each design category element i, the frequency of its occurrence is counted on multiple preset trend data platforms in the target cultural region c, and the platform frequency set is output.

[0016] Obtain the active user ratio and the proportion of clothing-related topics, determine the platform influence coefficient based on the active user ratio and the proportion of clothing-related topics, and use the platform influence coefficient to weight the platform frequency set to obtain the merged cross-platform weighted frequency result;

[0017] A time decay factor is introduced, which is determined based on the time interval between the occurrence time of design category element i and the current time. The time decay factor is used to perform time correction processing on the fused cross-platform weighted frequency results, and the time-corrected weighted frequency results are output.

[0018] The design category element i is calculated within a preset time window based on the time-corrected weighted frequency results. The frequency change rate within the range, when the frequency change rate exceeds the set trend growth threshold, multiplies the weighted frequency result of design category element i by a preset trend amplification factor. Then, the frequency results of all elements are normalized, and the set of element weights for the design category is output as E.

[0019] Preferably, step S20, which involves receiving user interaction input and sequentially performing text processing, image processing, design category element mapping processing, and fusion processing based on the user interaction input and the design category element weight set E, to output the final conditional embedding vector F, specifically includes:

[0020] Receive users within the target cultural region c User interaction input, including text keywords Reference image Refining instructions through multi-turn dialogue interaction ;

[0021] Text processing: Semantic parsing method based on bidirectional encoder representation for text keywords. Refining instructions through multi-turn dialogue interaction Encode the text to obtain text feature vectors;

[0022] Image Processing: Utilizing a global color quantization method based on perceptual hashing for reference images Color texture analysis was performed, and a local structure extraction method based on morphological gradient and edge spectrum was used to analyze the reference image. Shape analysis is performed, and the results of color texture analysis and shape analysis are fused to obtain the image feature vector;

[0023] Design category element mapping processing: The principal component feature-based reconstruction method is used to map the design category elements to the corresponding style basis vectors. The target cultural style vector is constructed based on the design category element weight set E and the corresponding style basis vectors.

[0024] Fusion processing: Based on the text feature vector, image feature vector and target cultural style vector, an adaptive attention weighted fusion method is used to construct the final conditional embedding vector F.

[0025] Preferably, in step S30, the multimodal clothing pattern generation model includes an input layer for receiving the final conditional embedding vector F; a feature expansion layer for expanding and multi-scale decomposing the final conditional embedding vector F within a preset spatial style mixing feature space; a cross-modal attention fusion layer for performing cross-channel attention calculations among text, image, and style basis vector features to enhance key information; a residual convolution generation layer for generating a high-resolution pattern feature map while preserving global style consistency; a style modulation decoding layer for performing spatial transformation and pattern adaptation on the high-resolution pattern feature map based on the pattern outline information of the target clothing silhouette; and an output layer for outputting a two-dimensional pattern set of the target clothing silhouette. .

[0026] Preferably, in step S40, based on the two-dimensional cut piece set A multi-scale affine mapping optimization method with seam continuity constraints is used to process cross-pattern splicing and color difference minimization, outputting an optimized pattern mapping set. The steps specifically include:

[0027] Step S401: Based on the two-dimensional cut piece set Establish an initial mapping relationship set between clipping and pattern using the corresponding high-resolution pattern feature map. and the initial mapping set As input for subsequent processing;

[0028] Step S402: Based on the initial mapping relationship set Extract the color gradient and texture direction features of adjacent fabric pieces at the shared seam, construct a seam continuity constraint set C, and then map the seam continuity constraint set C to the initial mapping relationship set. Together they serve as inputs for affine parameter optimization;

[0029] Step S403: Apply the initial mapping relation set at a global scale. Affine parameter optimization is performed to minimize the objective value of the seam continuity constraint C, and local non-rigid fine-tuning is performed at shared seams to obtain the updated set of mappings. ;

[0030] Step S404: Update the mapping set Color difference minimization and color consistency correction are performed at shared seams, outputting an optimized set of pattern mappings for the cut pieces. .

[0031] Preferably, in step S50, the preset virtual garment assembly network includes a pattern mesh mapping unit. The pattern mesh mapping unit adopts a UV mapping method based on per-vertex texture coordinate reconstruction to map the output optimized pattern pattern set. Each piece of fabric is mapped to the mesh surface of the target garment silhouette, ensuring that the texture maintains consistent proportions and avoids stretching deformation during the mapping process. The seam continuity detection unit uses a dual-channel detection method based on boundary neighborhood feature matching to simultaneously detect the geometric position continuity and color gradient consistency of the seam neighborhood on the mesh surface, generating a set of seam continuity deviation indices. The local deformation adjustment unit uses a constraint optimization-based mesh node and texture synchronous adjustment method to collaboratively fine-tune the mesh node positions and corresponding texture coordinates at shared seams when the seam continuity deviation indices in the set exceed a preset deviation threshold, thereby reducing misalignment and color difference. The curved surface rendering unit uses a PBR-based texture lighting synthesis method to perform high-fidelity rendering of the garment surface after the deformation adjustment in the local deformation adjustment unit, generating a visual model of the complete garment pattern.

[0032] This invention also provides a user-interactive-based automatic clothing pattern generation system, comprising:

[0033] The trend pattern data processing module is used to acquire clothing pattern datasets, perform design element recognition tasks based on the clothing pattern datasets, extract design category elements i, acquire target cultural regions c, and process design category elements i and target cultural regions c using a time-sensitive cross-platform trend weight calibration mechanism to output design category element weight set E.

[0034] The multimodal feature construction module is used to receive user data from the target cultural region c. Based on the user interaction input and the design category element weight set E, text processing, image processing, design category element mapping processing and fusion processing are performed in sequence to output the final conditional embedding vector F.

[0035] The garment pattern generation module is used to input the final condition embedding vector F into the pre-trained multimodal garment pattern generation model. The multimodal garment pattern generation model outputs a set of two-dimensional pattern pieces of the target garment silhouette. ;

[0036] The pattern optimization module is used for patterns based on two-dimensional pattern sets. A multi-scale affine mapping optimization method with seam continuity constraints is used to process cross-pattern splicing and color difference minimization, outputting an optimized pattern mapping set. ;

[0037] The garment assembly module is used to map the output pattern pieces to a set. Input is fed into a preset virtual garment assembly network, and the final complete garment pattern is output.

[0038] The present invention also provides a user-interactive clothing pattern automatic generation device, comprising: a memory, a processor, and a user-interactive clothing pattern automatic generation program stored in the memory and executable on the processor. When the user-interactive clothing pattern automatic generation program is executed by the processor, it implements a user-interactive clothing pattern automatic generation method.

[0039] The present invention also provides a computer program product, including a user-interactive automatic clothing pattern generation program, which, when executed by a processor, implements the user-interactive automatic clothing pattern generation method.

[0040] The beneficial effects of this invention are as follows: By introducing a time-sensitive cross-platform trend weight calibration mechanism and a multimodal conditional embedding generation strategy, this invention achieves accurate adaptation of clothing patterns to the aesthetic preferences of different cultural regions and target markets. Compared with traditional methods that rely on static trends or single-platform data, it significantly improves the market matching degree and personalization of the generated patterns.

[0041] This invention combines layer-by-layer feature generation based on conditional diffusion and style modulation, multi-scale affine mapping optimization with seam continuity constraints, and closed-loop correction for virtual garment assembly. This ensures high continuity and color consistency of the generated pattern after splicing different garment silhouettes and pieces, effectively avoiding seam misalignment and texture interruption problems, and improving the appearance quality of the garment and the level of design automation. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart illustrating the first embodiment of a method for automatically generating clothing patterns based on user interaction according to the present invention.

[0044] Figure 2This is a schematic diagram of the linear gradient blending effect of a first embodiment of a method for automatically generating clothing patterns based on user interaction according to the present invention.

[0045] Figure 3 This is a schematic diagram of the color distribution matching effect of a first embodiment of a method for automatically generating clothing patterns based on user interaction according to the present invention.

[0046] Figure 4 This is a schematic diagram comparing the misalignment distance and color difference height effects of a first embodiment of the automatic clothing pattern generation method based on user interaction of the present invention.

[0047] Figure 5 This is a schematic diagram of a device for an automatic clothing pattern generation method based on user interaction according to the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the automatic clothing pattern generation method based on user interaction of the present invention, which presents the first embodiment of the automatic clothing pattern generation method based on user interaction of the present invention.

[0050] In the first embodiment, the method for automatically generating clothing patterns based on user interaction includes:

[0051] Step S10: Obtain the clothing pattern dataset, perform the design element recognition task based on the clothing pattern dataset, and extract the design category element i; obtain the target cultural region c, and process the design category element i and the target cultural region c using a time-sensitive cross-platform trend weight calibration mechanism to output the design category element weight set E;

[0052] It should be noted that in step S10, the step of using a time-sensitive cross-platform trend weight calibration mechanism to process the design category element i and the target cultural region c, and outputting the design category element weight set E, specifically includes:

[0053] For each design category element i, the frequency of its occurrence is counted on multiple preset trend data platforms in the target cultural region c, and the platform frequency set is output.

[0054] Obtain the active user ratio and the proportion of clothing-related topics, determine the platform influence coefficient based on the active user ratio and the proportion of clothing-related topics, and use the platform influence coefficient to weight the platform frequency set to obtain the merged cross-platform weighted frequency result;

[0055] A time decay factor is introduced, which is determined based on the time interval between the occurrence time of design category element i and the current time. The time decay factor is used to perform time correction processing on the fused cross-platform weighted frequency results, and the time-corrected weighted frequency results are output.

[0056] The design category element i is calculated within a preset time window based on the time-corrected weighted frequency results. The frequency change rate within the range, when the frequency change rate exceeds the set trend growth threshold, multiplies the weighted frequency result of design category element i by a preset trend amplification factor. Then, the frequency results of all elements are normalized, and the set of element weights for the design category is output as E.

[0057] Understandably, through this mechanism, the present invention can obtain a high-precision priority distribution of design elements for the target cultural region before pattern generation, thereby making the subsequently generated clothing patterns more in line with the visual habits and aesthetic preferences of the target market. This preprocessing method can reduce the risk of style deviation and cultural mismatch in the early stages of pattern generation, and improve the market adaptability of the generated results.

[0058] It should be understood that, compared to traditional methods based on single platforms or static label statistics, this mechanism introduces joint modeling with a time dimension and cross-platform weighting, which can effectively offset the problem of distorted results caused by abnormal data from a single platform or short-term viral trends.

[0059] Step S20: Receive user interaction input, and based on the user interaction input and the design category element weight set E, perform text processing, image processing, design category element mapping processing and fusion processing in sequence, and output the final conditional embedding vector F;

[0060] It should be noted that "text processing" refers to structuring information such as style, color, and texture in a user-input natural language description into text feature vectors that correspond to design category elements by combining semantic parsing and keyword extraction; "image processing" refers to performing feature extraction and style analysis on reference patterns uploaded or selected by the user to generate image feature vectors of the same dimension as the text features; "design category element mapping processing" refers to matching the design category element weight set E obtained in step S10 with a preset style base vector library to obtain a target cultural style vector; and "fusion processing" refers to weighted fusion of text features, image features, and target cultural style vectors to form a final conditional embedding vector F that can simultaneously reflect the user's personalized needs and target cultural trends.

[0061] Understandably, this step unifies users' explicit needs (expressed through text and images) with implicit trend information (cultural preferences represented by E) into a multimodal vector space, providing complete contextual conditions for subsequent clothing pattern generation. This approach enables the generative model to not only respond to immediate user input but also automatically incorporate current popular elements in the market, improving the commercial value and aesthetic appeal of the generated results.

[0062] It should be understood that, compared to the traditional method of directly feeding user input into the generative model, this method introduces E-based style mapping processing before fusion, thereby avoiding the appearance of design elements in the generated results that are inconsistent with the target culture or outdated in terms of trends. At the same time, the multimodal fusion mechanism ensures the complementarity of information from different input forms (text, images, trend weights) and reduces the risk of pattern distortion caused by the lack of information from a single modality.

[0063] Step S30: Input the final condition embedding vector F into the pre-trained multimodal clothing pattern generation model. The multimodal clothing pattern generation model outputs a set of two-dimensional pattern pieces of the target clothing silhouette. ;

[0064] It should be noted that in step S30, the multimodal clothing pattern generation model includes an input layer for receiving the final conditional embedding vector F; a feature expansion layer for expanding and multi-scale decomposing the final conditional embedding vector F within a preset spatial style mixing feature space; a cross-modal attention fusion layer for performing cross-channel attention calculations among text, image, and style basis vector features to enhance key information; a residual convolution generation layer for generating a high-resolution pattern feature map while preserving global style consistency; a style modulation decoding layer for performing spatial transformation and pattern adaptation on the high-resolution pattern feature map based on the pattern outline information of the target clothing silhouette; and an output layer for outputting a two-dimensional pattern set of the target clothing silhouette. .

[0065] It should be noted that "layer-by-layer feature generation based on conditional diffusion" refers to using F as a conditional input and gradually restoring a high-resolution pattern feature map through the reverse reasoning process of the diffusion model. In the process, style modulation parameters are introduced at each layer, so that the pattern details dynamically adapt to the target cultural style and the user's personalized requirements during the generation process. The "piece adaptation mechanism" refers to performing spatial transformation and boundary constraints on the pattern according to the piece outline information of the target garment silhouette while generating the feature map, so as to ensure that the generated two-dimensional pieces can be seamlessly aligned and maintain design continuity in subsequent splicing.

[0066] Understandably, through this step, the generated model can not only generate high-resolution patterns that meet the user's style and cultural trend requirements, but also directly output a set of two-dimensional pattern pieces that match the target garment silhouette, thus eliminating the traditional process of "generating patterns → then manually matching pattern pieces", thereby improving the level of automation and production efficiency.

[0067] It should be understood that, compared to the traditional method of first generating a complete pattern and then cutting it according to rules to fit the silhouette, this method introduces the outline constraint of the pattern piece at the generation stage, so that the pattern and the edge of the pattern piece blend naturally, avoiding common problems such as misaligned seams, broken patterns, or unbalanced proportions. At the same time, the style modulation mechanism ensures that each pattern piece maintains the same style as the whole in local details, which greatly improves the overall aesthetics and consistency of the garment pattern.

[0068] For example, in experiments targeting three different silhouettes—A-line skirts, hoodies, and oversized shirts—the average seam texture misalignment rate was 17% when using the traditional cutting and fitting method. However, after adopting the layer-by-layer feature generation and pattern fitting mechanism in this step, the misalignment rate dropped to below 3%, and the visual consistency score of the samples in the finished garment state improved by an average of 1.5 points (out of 5).

[0069] Step S40: Based on the two-dimensional cut piece set A multi-scale affine mapping optimization method with seam continuity constraints is used to process cross-pattern splicing and color difference minimization, outputting an optimized pattern mapping set. ;

[0070] It should be noted that the "multi-scale affine mapping optimization method with seam continuity constraints" refers to: using the two-dimensional pattern set output in step S30 as input, establishing affine mapping relationships at three scales: pixel level, texture block level, and overall pattern level. The seam edge detection algorithm extracts the splicing boundaries of adjacent patterns, and the continuity of texture gradients at the boundaries is used as a constraint to adjust the parameters of the patterns in spatial position, rotation angle, scaling, and local deformation. Simultaneously, a distribution matching and gradient fusion strategy is employed in the color channels to reduce color differences and brightness unevenness between different patterns, achieving seamless integration of the overall visual appearance. "Mapping relationship" refers to the correspondence between the two-dimensional pattern pattern and the target outline pattern area. Traditional methods directly scale and cut a complete pattern, simply establishing coordinate alignment relationships. This invention establishes mapping relationships at three scales: pixel level, texture block level, and overall pattern level: pixel level: ensuring continuity of local details (such as petal edges and stripe direction) at the seam; texture block level: ensuring consistent orientation of medium-sized texture blocks between adjacent patterns; overall pattern level: ensuring that the overall composition does not shift due to local adjustments. This multi-scale mapping is more flexible than single scaling, preserving both overall and local naturalness. Seam edge detection is used to identify the boundary lines at the joints of adjacent fabric pieces and analyze the texture direction and color gradient on both sides of the boundary. Canny edge detection or the Sobel operator can be used to extract the boundaries; based on this, a gradient direction histogram is used to measure whether the textures on both sides are consistent; and then color distribution statistics (RGB or LAB color space) are combined to determine if the color difference is too large.

[0071] "Distribution matching" refers to comparing and correcting the color statistical characteristics of the fabric pieces on both sides of the seam. In practical applications, a common method is to statistically analyze the color histogram of the seam's neighborhood (e.g., the mean, variance, and distribution curve of the RGB or LAB color space). By matching these statistical characteristics, the color distribution on one side is adjusted to be closer to that on the other side, thereby reducing the abrupt feeling of "sudden brightening" or "uneven warm and cool" at the seam. This is similar to "white balance correction" or "color normalization," but it is done locally on the seam area. "Gradient blending" refers to not making a direct hard switch in the seam area, but rather using a transition area to allow the color and texture to gradually and smoothly transition from one side to the other. A certain width of area can be taken on both sides of the seam as a "transition zone"; within the transition zone, a weighted average (e.g., a linear gradient weight from 0% to 100%) is used to gradually transition the color information from one side to the other. The effect of this is to make the seam no longer look like a "hard line," but like a natural color transition, much like feathering or gradient masks in Photoshop.

[0072] Understandably, this step allows for visual continuity and naturalness at the seams of adjacent pieces while maintaining the original design style and shape of the cut pieces. It also ensures that the color representation of different pieces is consistent, thereby enhancing the overall integrity and aesthetics of the final garment pattern and providing a high-quality pattern foundation for subsequent virtual assembly of the garment.

[0073] It should be understood that, compared with the traditional single-scale pattern splicing method, this method simultaneously optimizes the geometric position and color distribution at multiple scales, and introduces seam continuity as a strong constraint. This makes the splicing result not only accurately fit the silhouette structure in macro layout, but also maintain a natural transition in micro texture. It effectively avoids common problems such as seam misalignment, texture breakage, and color jump, and improves the visual consistency of the generated pattern in the finished garment state.

[0074] For example, such as Figure 2 and Figure 3 As shown, distribution matching alone can only ensure that the overall color characteristics are similar, but slight abrupt changes may still occur at the boundaries. Gradient blending alone can result in a blurry or "dirty" effect if the color difference between the two sides is too large. The method of this invention combines the two: first, distribution matching is used to align the overall color tone, and then gradient blending is used to eliminate abrupt transitions in certain areas, thus maximizing naturalness and consistency. For example, at the junction of the front and sleeves of a patchwork sweatshirt: if the fabric is directly patched, the front fabric is light blue, and the sleeve fabric is grayish-blue, making the seam appear abrupt. Using "distribution matching" will brighten the overall grayish-blue of the sleeves, making its statistical distribution closer to light blue. Then, "gradient blending" is used to create a color transition within a 10-pixel width at the seam, allowing the light blue to gradually connect to the grayish-blue. The final visual effect is a unified, continuous color.

[0075] Step S50: Map the output pattern pieces to a set. The data is input into a preset virtual garment assembly network, which then outputs the final complete garment pattern.

[0076] It should be noted that after processing the 2D pattern pieces, they can be transformed into 3D for display from different angles. Specifically, by introducing a UV mapping method based on per-vertex texture coordinate reconstruction, the pattern piece pattern can automatically maintain its original proportional relationship when mapped to a 3D mesh, effectively avoiding pattern deformation caused by surface stretching. Furthermore, by using a lighting compositing method based on PBR (Physically Based Rendering), the realism of the clothing fabric can be further enhanced, including details such as luster, wrinkles, and reflections, improving the sensory fidelity of the final visualized image.

[0077] It should be understood that, compared to traditional texture mapping methods for garment display, this invention not only achieves geometric fitting of cut pieces but also constructs a dynamic adjustment and closed-loop rendering process with "seam continuity as the core indicator," thereby realizing an integrated virtual garment generation chain from "planar design -> mesh splicing -> seam detection -> local optimization -> realistic rendering." This architecture effectively avoids the pattern distortion problems caused by seam misalignment and texture stretching in traditional methods. It is especially suitable for design scenarios with multiple patterns, multiple materials, and complex patterns (such as multi-segment splicing and gradient fabrics), significantly improving the quality and efficiency of garment visualization modeling and reducing material waste and labor costs caused by physical trial and error.

[0078] For example, such as Figure 4 As shown, using the industry software's default UV splicing and seam treatment methods as a reference, and taking asymmetrical garment pieces, areas with abrupt changes in curvature, and horizontally spliced ​​fabrics as experimental benchmarks, the results compare the traditional texture mapping garment display method with the method of this invention in terms of seam misalignment distance and color difference. The figures show that the traditional method exhibits significant seam misalignment and large color difference, while the method of this invention, through innovative seam optimization and gradient blending strategies, significantly reduces color difference in the seam area and lowers the misalignment to a more ideal range.

[0079] Example 2: Furthermore, the present invention provides a user-interactive automatic clothing pattern generation system, which employs a user-interactive automatic clothing pattern generation method from the above embodiments, and can solve a technical problem related to user-interactive automatic clothing pattern generation. Compared with the prior art, the beneficial effects of the user-interactive automatic clothing pattern generation system provided by the present invention are the same as those of the user-interactive automatic clothing pattern generation method provided in the above embodiments, and other technical features of the user-interactive automatic clothing pattern generation system are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0080] Example 3: This invention provides a user-interactive automatic clothing pattern generation device. Please refer to... Figure 5An automatic clothing pattern generation device based on user interaction includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform the automatic clothing pattern generation method based on user interaction described in Embodiment 1 above. An automatic clothing pattern generation device based on user interaction in this embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. An automatic clothing pattern generation device based on user interaction is merely an example and should not impose any limitations on the functionality and scope of use of this embodiment. An automatic clothing pattern generation device based on user interaction may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. Random access memory 1004 also stores various programs and data required for the operation of a user-interactive automatic garment pattern generation device. Processing device 1001, read-only memory 1002, and random access memory 1004 are interconnected via bus 1005. I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows a user-interactive automatic garment pattern generation device to communicate wirelessly or wiredly with other devices to exchange data. Although a user-interactive automatic garment pattern generation device with various systems is shown in the figures, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0081] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for automatically generating clothing patterns based on user interaction. The computer program product provided by this invention can solve the technical problem of automatically generating clothing patterns based on user interaction. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the method for automatically generating clothing patterns based on user interaction provided in the above embodiments, and will not be repeated here.

[0082] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0083] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0084] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for automatically generating clothing patterns based on user interaction, characterized in that, The methods include: Step S10: Obtain the clothing pattern dataset, perform a design element recognition task based on the clothing pattern dataset, and extract the design category element i; obtain the target cultural region c, and process the design category element i and the target cultural region c using a time-sensitive cross-platform trend weight calibration mechanism to output the design category element weight set E; the step of processing the design category element i and the target cultural region c using a time-sensitive cross-platform trend weight calibration mechanism to output the design category element weight set E specifically includes: Count the frequency of occurrence of element i in the design category and output the set of platform frequencies; Obtain the active user ratio and the proportion of clothing-related topics, determine the platform influence coefficient based on the active user ratio and the proportion of clothing-related topics, and use the platform influence coefficient to weight the platform frequency set to obtain the merged cross-platform weighted frequency result; A time decay factor is introduced, and the time correction processing is performed on the fused cross-platform weighted frequency results based on the time decay factor, and the time-corrected weighted frequency results are output. The design category element i is calculated within a preset time window based on the time-corrected weighted frequency results. The frequency change rate within the range, when the frequency change rate exceeds the set trend growth threshold, multiplies the weighted frequency result of design category element i by a preset trend amplification factor. Then, the frequency results of all elements are normalized, and the set of element weights for the design category is output as E. Step S20: Receive user interaction input, and based on the user interaction input and the design category element weight set E, perform text processing, image processing, design category element mapping processing and fusion processing in sequence, and output the final conditional embedding vector F; Step S30: Input the final condition embedding vector F into the pre-trained multimodal clothing pattern generation model. The multimodal clothing pattern generation model outputs a set of two-dimensional pattern pieces of the target clothing silhouette. The multimodal clothing pattern generation model includes an input layer for receiving the final conditional embedding vector F; a feature expansion layer for expanding and multi-scale decomposing the final conditional embedding vector F within a predefined spatial style mixing feature space; a cross-modal attention fusion layer for performing cross-channel attention calculations among text, image, and style basis vector features to enhance key information; a residual convolution generation layer for generating high-resolution pattern feature maps while preserving global style consistency; a style modulation decoding layer for spatial transformation and pattern adaptation of the high-resolution pattern feature maps based on the pattern outline information of the target clothing silhouette; and an output layer for outputting a two-dimensional pattern set of the target clothing silhouette. ; Step S40: Based on the two-dimensional cut piece set A multi-scale affine mapping optimization method with seam continuity constraints is used to process cross-pattern splicing and color difference minimization, outputting an optimized pattern mapping set. Among them, based on two-dimensional pattern set A multi-scale affine mapping optimization method with seam continuity constraints is used to process cross-pattern splicing and color difference minimization, outputting an optimized pattern mapping set. The steps specifically include: Based on the two-dimensional cut piece set Establish an initial mapping relationship set between clipping and pattern using the corresponding high-resolution pattern feature map. and the initial mapping set As input for subsequent processing; Based on the initial mapping relationship set Extract the color gradient and texture direction features of adjacent fabric pieces at the shared seam, construct a seam continuity constraint set C, and then map the seam continuity constraint set C to the initial mapping relationship set. Together they serve as inputs for affine parameter optimization; The initial mapping set at the global scale Affine parameter optimization is performed to minimize the objective value of the seam continuity constraint C, and local non-rigid fine-tuning is performed at shared seams to obtain the updated set of mappings. ; For the updated mapping set Color difference minimization and color consistency correction are performed at shared seams, outputting an optimized set of pattern mappings for the cut pieces. ; Step S50: Map the output pattern pieces to a set. The data is input into a preset virtual garment assembly network, which then outputs the final complete garment pattern.

2. The method for automatically generating clothing patterns based on user interaction as described in claim 1, characterized in that, In step S10, the clothing pattern dataset includes high-frequency tagged patterns from the Pinterest platform, best-selling patterns from e-commerce sales data, and fashion image data from social media; design category elements include floral symbols, geometric symbols, animal symbols, and traditional symbols.

3. The method for automatically generating clothing patterns based on user interaction as described in claim 1, characterized in that, In step S20, the steps of receiving user interaction input, performing text processing, image processing, design category element mapping processing, and fusion processing sequentially based on the user interaction input and the design category element weight set E, and outputting the final conditional embedding vector F, specifically include: Receive users within the target cultural region c User interaction input, including text keywords Reference image Refining instructions through multi-turn dialogue interaction ; Text processing: Semantic parsing method based on bidirectional encoder representation for text keywords. Refining instructions through multi-turn dialogue interaction Encode the text to obtain text feature vectors; Image Processing: Utilizing a global color quantization method based on perceptual hashing for reference images Color texture analysis was performed, and a local structure extraction method based on morphological gradient and edge spectrum was used to analyze the reference image. Shape analysis is performed, and the results of color texture analysis and shape analysis are fused to obtain the image feature vector; Design category element mapping processing: The principal component feature-based reconstruction method is used to map the design category elements to the corresponding style basis vectors. The target cultural style vector is constructed based on the design category element weight set E and the corresponding style basis vectors. Fusion processing: Based on the text feature vector, image feature vector and target cultural style vector, an adaptive attention weighted fusion method is used to construct the final conditional embedding vector F.

4. The method for automatically generating clothing patterns based on user interaction as described in claim 1, characterized in that, In step S50, the preset virtual garment assembly network includes a pattern mesh mapping unit. The pattern mesh mapping unit adopts a UV mapping method based on per-vertex texture coordinate reconstruction to map the output optimized pattern pattern set. Each piece of fabric is mapped to the grid surface of the target garment silhouette, ensuring that the texture remains proportionally consistent during the mapping process and avoiding stretching and deformation; The seam continuity detection unit employs a dual-channel detection method based on boundary neighborhood feature matching to simultaneously detect the geometric position continuity and color gradient consistency of the seam neighborhood on the mesh surface, generating a set of seam continuity deviation indices. The local deformation adjustment unit employs a constraint optimization-based mesh node and texture synchronous adjustment method to collaboratively fine-tune the mesh node positions and corresponding texture coordinates at the shared seam when the seam continuity deviation indices in the set exceed a preset deviation threshold, thereby reducing misalignment and color difference. The surface rendering unit uses a PBR-based texture lighting synthesis method to perform high-fidelity rendering of the garment surface after the deformation adjustment is completed in the local deformation adjustment unit, generating a visual model of the complete garment pattern.

5. A user-interactive automatic clothing pattern generation system, applied to the user-interactive automatic clothing pattern generation method according to any one of claims 1 to 4, characterized in that, The user-interaction-based automatic clothing pattern generation system includes: The trend pattern data processing module is used to acquire a clothing pattern dataset, perform a design element recognition task based on the clothing pattern dataset, extract design category elements i, and acquire the target cultural region c. A time-sensitive cross-platform trend weight calibration mechanism is applied to design category elements i and target cultural region c to output a set of design category element weights E. Specifically, the steps of applying the time-sensitive cross-platform trend weight calibration mechanism to design category elements i and target cultural region c to output the set of design category element weights E include: Count the frequency of occurrence of element i in the design category and output the set of platform frequencies; Obtain the active user ratio and the proportion of clothing-related topics, determine the platform influence coefficient based on the active user ratio and the proportion of clothing-related topics, and use the platform influence coefficient to weight the platform frequency set to obtain the merged cross-platform weighted frequency result; A time decay factor is introduced, and the time correction processing is performed on the fused cross-platform weighted frequency results based on the time decay factor, and the time-corrected weighted frequency results are output. The design category element i is calculated within a preset time window based on the time-corrected weighted frequency results. The frequency change rate within the range, when the frequency change rate exceeds the set trend growth threshold, multiplies the weighted frequency result of design category element i by a preset trend amplification factor. Then, the frequency results of all elements are normalized, and the set of element weights for the design category is output as E. The multimodal feature construction module is used to receive user data from the target cultural region c. Based on the user interaction input and the design category element weight set E, text processing, image processing, design category element mapping processing and fusion processing are performed in sequence to output the final conditional embedding vector F. The garment pattern generation module is used to input the final condition embedding vector F into the pre-trained multimodal garment pattern generation model. The multimodal garment pattern generation model outputs a set of two-dimensional pattern pieces of the target garment silhouette. The multimodal clothing pattern generation model includes an input layer for receiving the final conditional embedding vector F; a feature expansion layer for expanding and multi-scale decomposing the final conditional embedding vector F within a predefined spatial style mixing feature space; a cross-modal attention fusion layer for performing cross-channel attention calculations among text, image, and style basis vector features to enhance key information; a residual convolution generation layer for generating high-resolution pattern feature maps while preserving global style consistency; a style modulation decoding layer for spatial transformation and pattern adaptation of the high-resolution pattern feature maps based on the pattern outline information of the target clothing silhouette; and an output layer for outputting a two-dimensional pattern set of the target clothing silhouette. ; The pattern optimization module is used for patterns based on two-dimensional pattern sets. A multi-scale affine mapping optimization method with seam continuity constraints is used to process cross-pattern splicing and color difference minimization, outputting an optimized pattern mapping set. Among them, based on two-dimensional pattern set A multi-scale affine mapping optimization method with seam continuity constraints is used to process cross-pattern splicing and color difference minimization, outputting an optimized pattern mapping set. The steps specifically include: Based on the two-dimensional cut piece set Establish an initial mapping relationship set between clipping and pattern using the corresponding high-resolution pattern feature map. and the initial mapping set As input for subsequent processing; Based on the initial mapping relationship set Extract the color gradient and texture direction features of adjacent fabric pieces at the shared seam, construct a seam continuity constraint set C, and then map the seam continuity constraint set C to the initial mapping relationship set. Together they serve as inputs for affine parameter optimization; The initial mapping set at the global scale Affine parameter optimization is performed to minimize the objective value of the seam continuity constraint C, and local non-rigid fine-tuning is performed at shared seams to obtain the updated set of mappings. ; For the updated mapping set Color difference minimization and color consistency correction are performed at shared seams, outputting an optimized set of pattern mappings for the cut pieces. ; The garment assembly module is used to map the output pattern pieces to a set. Input is fed into a preset virtual garment assembly network, and the final complete garment pattern is output.

6. A device for automatically generating clothing patterns based on user interaction, characterized in that, The user-interactive clothing pattern automatic generation device includes: a memory, a processor, and a user-interactive clothing pattern automatic generation program stored in the memory and executable on the processor. When the user-interactive clothing pattern automatic generation program is executed by the processor, it implements a user-interactive clothing pattern automatic generation method according to any one of claims 1 to 4.

7. A computer program product, characterized in that, The computer program product includes a user-interactive automatic clothing pattern generation program, which, when executed by a processor, implements a user-interactive automatic clothing pattern generation method according to any one of claims 1 to 4.

Images