A method, device and medium for garment pattern measurement based on artificial intelligence
By generating structured profiles of target users based on artificial intelligence and simulating fitting feedback behavior, combined with historical data for hybrid simulation evaluation, the high cost and low efficiency of traditional clothing style testing methods are solved, achieving efficient and accurate clothing market forecasting.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- QINGDAO KUTESMART CO LTD
- Filing Date
- 2026-03-06
- Publication Date
- 2026-06-02
AI Technical Summary
Existing apparel testing methods rely on traditional or semi-digital approaches, which are costly and time-consuming. They are difficult to assess the market acceptance of new apparel on a large scale and at low cost, and cannot effectively integrate new design elements and meet the needs of new target customer groups.
By generating structured profiles of target users based on artificial intelligence, loading them into virtual character agents to simulate try-on feedback behavior, combining historical style data for hybrid simulation evaluation, generating evaluation results, and adjusting design data based on the results.
It enables large-scale, low-cost virtual try-on evaluation, shortens the product testing cycle, reduces costs, improves the reliability and market accuracy of evaluation results, and avoids the errors and insufficient representativeness of traditional methods.
Smart Images
Figure CN122134434A_ABST
Abstract
Description
Technical Field
[0001] This specification relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based method, device and medium for measuring clothing patterns. Background Technology
[0002] The apparel industry is currently characterized by high competition, high inventory risk, and high timeliness requirements. Therefore, accurately predicting market acceptance of new clothing items before their launch is a crucial factor in determining subsequent marketing strategies.
[0003] Current apparel testing primarily relies on traditional and semi-digital methods. The traditional method involves designers creating physical samples, which are then tried on and reviewed by an internal team, buyers, or a small group of consumers. However, this physical fitting and offline review process is costly and time-consuming, and due to the small size of the review group, its representativeness is limited and it is easily influenced by subjective preferences, making it difficult to scale. The semi-digital method analyzes sales data from past best-selling or slow-moving items to summarize patterns in elements such as color and style. While this method has a certain data foundation, it cannot effectively evaluate innovative apparel styles that incorporate new design elements or target new customer groups, and it cannot make predictions before sample production. Furthermore, methods such as creating digital samples using 3D modeling software or collecting user feedback through online questionnaires and virtual fitting rooms are also limited by sample size, user participation willingness and expressive ability, and cannot be conducted without physical samples.
[0004] Therefore, there is a need for a method of apparel testing based on artificial intelligence to simulate large-scale user feedback. Summary of the Invention
[0005] To address the aforementioned issues, this specification provides one or more embodiments of an artificial intelligence-based method, device, and medium for garment pattern measurement.
[0006] One or more embodiments of this specification employ the following technical solutions: This specification provides one or more embodiments of an artificial intelligence-based method for clothing measurement, the method including: Obtain the design data of the current test garment to determine the target group that matches the design data; wherein, the design data includes partial design data and overall design data; Based on the structured distribution information corresponding to the target group, structured profile information of multiple target users is generated; wherein the structured distribution information represents the proportion of similar feature data in the target group, and the structured profile information, based on the feature data, determines that it includes at least: body shape feature data, aesthetic feature data, and behavioral feature data; The structured profile information of each target user is sequentially loaded into the virtual role agent deployed in the pre-set simulation environment to simulate the fitting feedback behavior of each target user and obtain virtual feedback results. Based on the aforementioned proportion, the virtual feedback results are integrated with similar historical style business data to conduct a hybrid simulation evaluation of the current test garment and obtain the evaluation results. In response to the evaluation results or the change in external demand corresponding to the current test garment, the design data of the current test garment is adjusted.
[0007] Optionally, in one or more embodiments of this specification, obtaining current apparel design data to determine a target group matching the design data specifically includes: Obtain partial and overall design data of the current test garment; The key parts of the local design data are identified to obtain one or more design elements corresponding to the key parts, and the color data and garment silhouette data of the overall design data are extracted; wherein, the key parts correspond to local structures that have functional or visual distinctiveness. Based on the similarity matching of the design elements, color data, and garment silhouette data with the data corresponding to each garment style tag, the garment style tag of the current test garment is determined. Based on the nodes in the pre-built knowledge graph associated with the clothing style tags, a target group matching the design data is determined; wherein, the pre-built knowledge graph is obtained based on historical sales data statistics.
[0008] Optionally, in one or more embodiments of this specification, based on the structured distribution information corresponding to the target group, structured profile information of multiple target users is generated, specifically including: Obtain the distribution characteristics of the target group in the feature space and the proportion of different feature combinations; Based on the relationship between the preset number of virtual character agents and the proportion, the number of characters to be generated for each feature is determined. Based on the distribution characteristics, sampling is performed in the feature space to generate a combination of feature values corresponding to the number of roles to be generated, thereby obtaining structured profile information of multiple target users.
[0009] Optionally, in one or more embodiments of this specification, the structured profile information of each target user is sequentially loaded into a virtual role agent deployed in a pre-set simulation environment to simulate the try-on feedback behavior of each target user and obtain virtual feedback results, specifically including: The structured profile information of each target user is sequentially loaded into the virtual character agent deployed in the pre-set simulation environment to obtain the body shape feature data of each target user; Based on the body shape feature data of each target user, a three-dimensional human body model corresponding to the virtual character agent is determined; the three-dimensional human body model is obtained by matching the body shape feature data with existing models. Based on the design data of the current test garment and the three-dimensional human body model, the fitting state is obtained, and the virtual feedback result is generated based on the fitting state and the aesthetic and behavioral characteristic data of the target user.
[0010] Optionally, in one or more embodiments of this specification, the three-dimensional human body model corresponding to the virtual character agent is determined based on the body shape feature data of each target user: Based on the size information of the current test garment to be put on the shelves, determine the measurement area corresponding to the current test garment; The body shape feature data is analyzed, and the circumference parameters and key morphological parameters associated with the body measurement area are extracted to construct a standardized body shape query vector. The standardized body shape query vector is matched with an existing model, and the existing model corresponding to the standardized body shape query vector is obtained based on the matching result as the base model. Based on the individual preference parameter values corresponding to the aesthetic feature data, the basic model is adaptively adjusted for non-body type visualization features to generate a three-dimensional human body model.
[0011] Optionally, in one or more embodiments of this specification, the fit error of each key part is determined based on the fitting state, and the virtual feedback result is generated by combining the aesthetic and behavioral characteristic data of the target user, specifically including: The static fitting deviation value is determined by the average Euclidean distance between the three-dimensional mesh data and the three-dimensional human body model at the key parts; Based on the clothing type corresponding to the current evaluation clothing, a preset simulated action sequence corresponding to the current evaluation clothing is determined. Based on the preset simulated action sequence, the maximum displacement of the key parts is determined, and the maximum displacement is used as the dynamic activity tolerance. The static fitting deviation value and the dynamic activity tolerance are weighted and fused to generate the fitting error of the key parts; The degree of aesthetic overlap is determined based on the degree of overlap between the target user's aesthetic feature data and the design elements, and between the color data and the clothing silhouette data. The behavioral compatibility is obtained by matching the behavioral characteristic data of the target user with the pre-set data of the current test clothing. The virtual feedback result is generated by combining the fitting error, the aesthetic overlap, and the behavioral compatibility.
[0012] Optionally, in one or more embodiments of this specification, the virtual feedback result is fused with similar historical style business data based on the stated proportion to perform a hybrid simulation evaluation of the current tested garment and obtain evaluation results, specifically including: Based on the proportion of the feature cluster to which each target user belongs, the corresponding virtual feedback results are weighted and aggregated to generate virtual feedback indicators for the target user in each dimension. The virtual feedback indicators and the historical style business data of the same type are normalized under the same dimensions, and the processed data are weighted and fused to obtain a hybrid simulation evaluation score. The hybrid simulation evaluation score is compared with the preset evaluation score and the preset threshold range to output the multi-dimensional evaluation result of the current garment.
[0013] Optionally, in one or more embodiments of this specification, in response to the evaluation results or an external demand change event corresponding to the current test garment, the design data of the current test garment is adjusted, specifically including: The multi-dimensional evaluation results are analyzed to identify target dimensions that are below a preset feasible threshold, and the bottleneck parameters corresponding to the target dimensions are determined. Based on the bottleneck parameters, or the design attributes corresponding to the external demand change event, the design data of the current test garment is matched to determine the design data to be adjusted. Based on the type of the bottleneck parameter or the constraints of the external demand change event, the design data to be adjusted is parametrically modified to generate one or more candidate design data. The candidate design data is evaluated based on the virtual character agent to obtain updated current test clothing design data.
[0014] This specification provides one or more embodiments of an artificial intelligence-based garment measurement device, the device comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described above.
[0015] This specification provides one or more embodiments of a non-volatile computer storage medium storing computer-executable instructions, the computer-executable instructions being configured to execute any of the methods described above.
[0016] The above-described at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: Based on the structured distribution information of the target group, structured profiles of multiple target users are generated. This quantifies group characteristics into body shape, aesthetic, and behavioral data, enabling a multi-dimensional description of target users and avoiding the problem of traditional product testing failing to comprehensively cover diverse user needs. By loading the structured profiles of target users into virtual role agents within a pre-built simulation environment, and simulating the try-on feedback behavior of real users, large-scale virtual try-on evaluations can be completed efficiently and cost-effectively before clothing goes on sale. This replaces the traditional offline live try-on process, significantly shortening the product testing cycle and reducing costs, while avoiding evaluation errors caused by insufficient live sample sizes. Based on hybrid simulation evaluation, the system balances simulation rationality with market practicality, making the evaluation results closer to real market performance and improving the reliability of clothing product testing decisions. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A schematic diagram of an artificial intelligence-based clothing measurement method provided in the embodiments of this specification; Figure 2 A schematic diagram of the structure of an artificial intelligence-based garment measurement device provided in the embodiments of this specification; Figure 3 This is a schematic diagram of the structure of a non-volatile storage medium provided in the embodiments of this specification. Detailed Implementation
[0018] This specification provides an artificial intelligence-based method, device, and medium for garment pattern measurement.
[0019] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.
[0020] like Figure 1 As shown in one or more embodiments of this specification, an artificial intelligence-based method for measuring clothing styles specifically includes the following steps: S101: Obtain the design data of the current test garment to determine the target group that matches the design data; wherein, the design data includes partial design data and overall design data.
[0021] In traditional apparel testing processes, the definition of the target group typically relies on the subjective experience and judgment of product planners or designers. However, individuals with different experience backgrounds may arrive at different customer group positioning judgments for the same garment, leading to a lack of a unified and verifiable decision-making basis for subsequent sales strategies and inventory preparation. Furthermore, apparel design data itself possesses rich semantic information, and traditional methods cannot structurally correlate these fine-grained design features with the implicit group preferences in historical consumption data, easily resulting in a disconnect between apparel design and the target consumer group. Moreover, when an apparel incorporates innovative elements across styles and categories, experience-based judgments often fail, making it impossible to accurately predict which potential customer groups might accept this new combination. To address these issues, this embodiment automatically identifies the target group matching the currently tested apparel. The embodiment acquires the design data of the current tested apparel, which includes both partial design data and overall design data. Partial design data refers to the structured design description of key parts of the garment. Key parts are functionally or visually distinctive local structures such as collar type, sleeve type, placket, waist structure, and hem shape. Overall design data refers to the overall visual style characteristics of the garment, including at least: color data, pattern data, garment silhouette data, and fabric texture category. After obtaining the design data of the current test garment, the target audience that matches the design data is identified.
[0022] In one feasible approach, the target group matching the design data is determined by the following method: For each key part, semantic mapping is performed using a pre-designed design element tag library to obtain one or more design elements constituting the current test garment, forming a set of design elements. It should also be noted that this key part identification process can be automatically completed using a pre-trained computer vision model, such as a deep learning-based keypoint detection network or semantic segmentation model, or it can be obtained by the designer through structured import after annotation in a CAD system. Simultaneously, color data and garment silhouette data of the overall design data are extracted. In a pre-set knowledge graph of clothing styles, one or more clothing style tags corresponding to the design elements, color data, and garment silhouette data are retrieved based on similarity. The consumer group nodes associated with the clothing style tags in the pre-set style knowledge graph are traversed to determine the target group matching the design data. The pre-set style knowledge graph is constructed from historical sales data, user profile tags, style tags, and style semantic relationships, where nodes include design elements, color combinations, silhouette types, style tags, and corresponding consumer group profile clusters.
[0023] S102: Based on the structured distribution information corresponding to the target group, generate structured profile information for multiple target users; wherein the structured distribution information represents the proportion of similar feature data in the target group, and the structured profile information, based on the feature data, determines that it includes at least: body shape feature data, aesthetic feature data, and behavioral feature data.
[0024] After obtaining the target group corresponding to the current test garment based on the above steps, traditional design evaluation often relies on user assumptions or subjective experience, which cannot reflect the diversity and distribution patterns of the target group in terms of body shape, aesthetics, and consumption behavior. This is especially true for new designs that integrate cross-styles and multiple customer groups. Therefore, to achieve accurate conversion from group statistical data to simulateable individual virtual users and improve the authenticity of the test results, in this embodiment, structured profiles of multiple target users are generated based on the structured distribution information corresponding to the target group. This facilitates subsequent virtual simulation of users' fitting feedback using these structured profiles. The structured distribution information characterizes the proportion of similar data within the target group, i.e., what types of people make up this group and what proportion of each type. It should be noted that the structured distribution information does not describe any specific individual, but rather the compositional proportion of the group. Structured user profile information refers to data information that provides an individual description of a virtual user. It is determined based on feature data and includes at least: body shape features, aesthetic features, and behavioral features. Therefore, structured user profile information can be understood as describing a complete set of attributes of a virtual role agent that can be loaded into a simulation environment, perform try-on actions, and output personalized feedback. The virtual role agent is a digital intelligent entity deployed in a pre-set simulation environment to represent a target user individual and perform try-on actions and generate feedback. Each virtual role agent is initialized with a set of structured user profile information and is a runtime instance of the structured user profile in the simulation environment.
[0025] Specifically, in one or more embodiments of this specification, structured profile information of multiple target users is generated based on the structured distribution information corresponding to the target group, specifically including: Based on the aforementioned process, the target group corresponding to the current test garment and its composition ratio, i.e., structured distribution information, were determined. However, this structured distribution information, which includes the types of people and the proportion of each type, is a description of the group. Therefore, in order to determine the profile of individual target users, this embodiment of the specification obtains the distribution characteristics of the target group in the feature space and the proportion of different feature combinations. For example, cluster analysis is performed on the historical sales data corresponding to the target group. The clustering results divide the target group into K feature clusters, and users within each feature cluster have high similarity in three dimensions: body shape, aesthetics, and behavior. Then, for each feature cluster, the percentage of historical users it contains is calculated to represent the proportion of the target group's total users, which is used as the proportion of that feature cluster within the target group. At the same time, feature dimension statistics are performed on each feature cluster to jointly constitute the distribution characteristics of the target group based on the clustering results, the proportion of each feature cluster, and the statistical distribution of feature data within each feature cluster. The number of characters to be generated for each feature is determined by rounding down the product of the number of pre-generated virtual character agents and this proportion. For each feature cluster, based on the aforementioned distribution characteristics, sampling is performed in the feature space to generate feature value combinations corresponding to the number of characters to be generated, thus obtaining structured profile information for multiple target users. Specifically, spatial sampling is performed based on the statistical distribution model corresponding to each distribution feature, randomly generating feature value combinations corresponding to the number of characters to be generated. Each feature value combination contains complete body shape feature data, aesthetic feature data, and behavioral feature data, forming a complete structured profile information for a virtual user within that feature cluster.
[0026] S103: The structured profile information of each target user is sequentially loaded into the virtual role agent deployed in the pre-set simulation environment to simulate the try-on feedback behavior of each target user and obtain virtual feedback results.
[0027] After obtaining the structured profile information of the target users based on the above steps, in order to obtain the feedback results of each target user on the current test garment, the structured profile information of each target user will be loaded sequentially into the virtual role agent deployed in the pre-set simulation environment, so as to simulate the fitting feedback behavior of the target user corresponding to each structured profile information on the current test garment through the virtual role agent, and obtain virtual feedback results.
[0028] Specifically, in one or more embodiments of this specification, the structured profile information of each target user is sequentially loaded into a virtual role agent deployed in a pre-set simulation environment to simulate the try-on feedback behavior of each target user and obtain virtual feedback results. The specific process includes the following: Based on the body shape feature data of each target user, the 3D human body model is matched with existing 3D human body models to determine the 3D human body model corresponding to the virtual character agent. For example, a standardized query vector is constructed based on the body shape feature data of the target user. The Euclidean distance between this query vector and the body shape feature vector of each existing model in the model library is calculated. Then, the existing model with the smallest distance is selected as the 3D human body model corresponding to the target user. In this process, by matching with existing 3D human body models, the 3D human body model corresponding to the virtual character agent is determined, thereby transforming the body shape feature data in the structured portrait information into a loadable 3D human body model in the simulation environment, so as to facilitate accurate simulation of subsequent try-on states.
[0029] Based on the design data of the current test garment, 3D mesh data of the current test garment is obtained, and this 3D mesh data is mapped onto the 3D human body model to obtain the fitting state. The fitting state is determined based on the distance between the 3D mesh data of key areas and the 3D human body model. The 3D mesh data describes the geometry of the garment in the form of a triangular mesh, including vertex coordinates, triangle face indices, and normal directions. The process of mapping this 3D mesh data to the 3D human body model can invoke a pre-built physics simulation engine to perform the garment-to-human mapping process. It should be noted that this physics simulation engine is implemented based on positional dynamics or the finite element method, and can simulate the realistic physical behavior of fabric under gravity, collision, constraints, etc. By mapping the 3D mesh data to the 3D human body model, the distance between the 3D mesh data of key areas and the 3D human body model can be obtained. Specifically, for each key area, the average distance from all mesh vertices of that area to the surface of the human body model can be calculated. The set of distances formed by the average distances from all mesh vertices of that area to the surface of the human body model reflects the fitting state corresponding to the virtual target user.
[0030] Then, the fit error of each key part is determined based on the fitting status, and the virtual feedback result is generated by combining the aesthetic and behavioral characteristic data of the target user. That is, for each key part, the deviation from the ideal fit distance is calculated based on the distance obtained above. It should be noted that in one feasible approach, for each key part, a basic ideal fit distance is determined based on the current garment category (e.g., fitted or loose), and then dynamically adjusted to obtain the user's personalized ideal fit distance by combining the fit preference parameters included in the user's body shape characteristic data. In another feasible approach, the ideal fit distance is not a fixed value, but rather a basic ideal fit distance dynamically set by the designer in the garment design data based on the garment style and user preferences. For example, the ideal fit distance is smaller for fitted styles and larger for loose styles. Simultaneously, this basic ideal fit distance is adjusted based on the fit preferences in the target user's body shape characteristic data to obtain the ideal fit distance. Since fit error only reflects the physical fit of the garment, consumer satisfaction is also influenced by aesthetic factors. Therefore, based on the fitting error, and combining the aesthetic and behavioral characteristic data of the target user, the virtual feedback result is generated. Here, by combining the aesthetic and behavioral characteristic data of the target user, the cosine similarity between the feature vector corresponding to the design features of the current test garment and the user's aesthetic preference vector is calculated to determine the aesthetic matching degree. The pre-set listing data of the current test garment is matched with the user's behavioral characteristic data, such as by performing rule matching or logistic regression scoring between the pre-set listing data and the user's behavioral characteristic data, to calculate the behavioral compatibility. The higher the compatibility, the more the garment matches the user's purchasing habits and capabilities. The final virtual feedback result is generated by combining the fitting error, aesthetic characteristic data, and behavioral characteristic data mentioned above.
[0031] Furthermore, in one or more embodiments of this specification, a three-dimensional human body model corresponding to the virtual character agent is determined based on the body shape feature data of each target user: Based on the size information of the garment to be listed for sale, the key measurement areas to focus on during the fitting and pattern matching process are determined. Different sizes and types of garments rely on different body parts for pattern adaptation. The size information to be listed for sale clearly identifies key measurement areas such as bust, waist, hips, shoulder width, and garment length, thus providing a defined range for subsequent 3D human body model matching and avoiding excessive computational resource consumption for overall matching.
[0032] The system analyzes the target user's body shape feature data, extracting circumference parameters (including but not limited to chest, waist, hip, and neck circumference) and key morphological parameters (including shoulder shape, waist shape, proportions, and posture features) associated with the measurement area. These circumference and key morphological parameters are then processed in a unified format to construct a standardized body shape query vector. This standardized body shape query vector is matched with existing models, using the matching results to obtain the corresponding existing model as the base model. This ensures that the base model is highly consistent with the target user in terms of body shape. Based on the individual preference parameter values corresponding to the aesthetic feature data, the base model undergoes adaptive adjustment of non-body shape visualization features to generate a 3D human body model. These non-body shape visualization features include skin color, hairstyle, posture visual expression, and detailed appearance—features that do not affect the clothing fit and try-on effect. Through adaptive adjustment, the generated 3D human body model is consistent with the virtual target user in body shape and conforms to the virtual target user's aesthetic preferences in appearance, ultimately resulting in a 3D human body model suitable for virtual try-on simulation.
[0033] Furthermore, in one feasible embodiment of this specification, the process of determining the fit error of each key part based on the fitting status, and combining the aesthetic and behavioral characteristic data of the target user to generate the virtual feedback result, specifically includes the following steps: The static fit deviation value is determined by the average Euclidean distance between the 3D mesh data and the 3D human body model at key locations. This can be understood as follows: by acquiring the 3D mesh data of the current garment and the spatial coordinates of the 3D human body model at various key locations, the average Euclidean distance between the 3D mesh and the corresponding locations on the human body model is calculated. This distance value quantifies the tightness of the garment's fit to the human body in a static state, yielding the static fit deviation value, which is used to characterize the fit during static try-on.
[0034] Since clothing exhibits different fitting states due to varying actions in real-world scenarios, this embodiment retrieves a pre-defined simulated action sequence appropriate to the clothing type being evaluated. For example, it simulates raising an arm and bending over for tops, and raising a leg and squatting for trousers. The 3D human model is driven to execute this simulated action sequence in a simulation environment, calculating the maximum displacement of key clothing parts during the action. This maximum displacement is used as the dynamic activity tolerance to characterize the required movement allowance of the clothing during human activity. Then, to generate the fitting error of key parts, the previously obtained static fitting deviation value and the dynamic activity tolerance are weighted and fused to generate the fitting error of key parts. To further integrate the aesthetic characteristic data of the target user, the degree of aesthetic overlap is determined based on the overlap between the target user's aesthetic characteristic data and design elements, color data, and clothing silhouette data. This allows for the characterization of the user's preference for the appearance of the clothing based on the degree of aesthetic overlap. The behavioral characteristics data of the target users, such as clothing scene preferences, wearing frequency, consumption habits, and historical dressing behavior, are matched with the pre-set listing data of the current test garment, such as applicable scenarios, target audience, sales channels, and pricing range, to obtain behavioral compatibility. The virtual feedback result is generated by combining the aforementioned fit error, aesthetic overlap, and behavioral compatibility. This comprehensive method can be a weighted fusion of the aforementioned fit error, aesthetic overlap, and behavioral compatibility.
[0035] S104: Based on the stated proportion, the virtual feedback result is integrated with similar historical style business data to perform a hybrid simulation evaluation of the current test garment and obtain the evaluation result.
[0036] Because the virtual feedback results are not compared with historical performance in the real market, and without real historical data as a calibration benchmark, this single virtual feedback suffers from a lack of market validation. Furthermore, historical business data is lagging. Therefore, to address this issue and combine the forward-looking nature of the virtual feedback with the authenticity of historical data, this embodiment of the specification will integrate the virtual feedback results with business data of similar historical styles based on the aforementioned obtained proportions to conduct a hybrid simulation evaluation of the current tested garment and obtain the evaluation results.
[0037] Specifically, in one or more embodiments of this specification, the virtual feedback results are fused with historical business data of similar styles based on the proportion, in order to perform a hybrid simulation evaluation of the current garment and obtain the evaluation results, specifically including: Based on the proportion of each target user's characteristic cluster, the corresponding virtual feedback results are weighted and aggregated to generate virtual feedback indicators for the target user across various dimensions. Different characteristic clusters correspond to different user group types. The virtual feedback results of users within the same cluster are weighted according to the proportion of users in each characteristic cluster and their importance, resulting in virtual feedback indicators for the tested garment across multiple dimensions such as fit, aesthetics, comfort, and acceptance. This achieves quantitative aggregation from single-user feedback to group-level feedback.
[0038] Virtual feedback metrics are normalized along with historical business data of similar styles across the same dimensions. The processed data is then weighted and fused to obtain a hybrid simulation evaluation score. By normalizing and weighting the processed data, the influence of differences in units and numerical ranges can be eliminated, resulting in a hybrid simulation evaluation score that integrates virtual simulation effects with real business experience. This ensures the evaluation results are both accurate and practical. The hybrid simulation evaluation score is then compared with preset threshold ranges to output a multi-dimensional evaluation result for the current tested garment. For example, in a certain scenario, the hybrid simulation evaluation score can be compared with preset threshold ranges such as qualified, good, and excellent. Judgments can be made on multiple dimensions, including fit, appearance acceptance, market adaptability potential, and overall recommendation level, thus outputting a multi-dimensional evaluation result indicating whether the current tested garment is suitable for sale, whether fit adjustments are needed, and its expected market performance.
[0039] S105: In response to the evaluation results or the external demand change event corresponding to the current test garment, adjust the design data of the current test garment.
[0040] In order to improve the design based on the evaluation results and enhance the market performance of the product after its launch, this embodiment of the specification will adjust the design data of the current test garment in response to the evaluation results or the external demand change event corresponding to the current test garment, such as changes in cost targets, fabric supply, or fashion trends.
[0041] Specifically, in one or more embodiments of this specification, in response to the evaluation results or an external demand change event corresponding to the current test garment, the design data of the current test garment is adjusted, specifically including: Analyzing the aforementioned multi-dimensional evaluation results, target dimensions with scores below a preset feasible threshold are identified. Based on the evaluation content corresponding to each target dimension, bottleneck parameters causing insufficient scores in that dimension are located and determined. Bottleneck parameters include, but are not limited to, quantitative parameters directly related to clothing design, such as insufficient fit, dimensional deviations in key areas, mismatched appearance styles, and inadequate comfort during activities. These bottleneck parameters, or the design attributes corresponding to external demand change events, are matched item by item with the existing design data of the current test garment to determine the specific items in the design data that need modification—that is, the design data to be adjusted. Then, based on the type of bottleneck parameter or the constraints of the external demand change event, such as size requirements, style requirements, scene requirements, and cost requirements, the design data to be adjusted is parametrically modified to generate one or more candidate design data. In one approach, the company's pre-set design rule library can be called, and parameters can be adjusted according to empirical formulas. In another approach, if multiple bottleneck parameters exist, the Pareto optimization algorithm can be used to balance conflicting objectives and obtain candidate design data. The virtual character agent in the aforementioned process evaluates the revised candidate design data again to obtain updated design data for the current test garment, thereby optimizing the design data for the current test garment.
[0042] like Figure 2 As shown in the diagram, this specification provides a structural schematic of an artificial intelligence-based garment measurement device. Figure 2 As can be seen, in one or more embodiments of this specification, the device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described above.
[0043] like Figure 3 As shown in the diagram, this specification provides a schematic diagram of the structure of a non-volatile storage medium. Figure 3 As can be seen, in one or more embodiments of this specification, a non-volatile storage medium stores computer-executable instructions 301, which are capable of executing any of the methods described above.
[0044] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.
[0045] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.
[0046] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.
Claims
1. A method for apparel pattern measurement based on artificial intelligence, characterized in that, The method includes: Obtain the design data of the current test garment to determine the target group that matches the design data; wherein, the design data includes partial design data and overall design data; Based on the structured distribution information corresponding to the target group, structured profile information of multiple target users is generated; wherein the structured distribution information represents the proportion of similar feature data in the target group, and the structured profile information, based on the feature data, determines that it includes at least: body shape feature data, aesthetic feature data, and behavioral feature data; The structured profile information of each target user is sequentially loaded into the virtual role agent deployed in the pre-set simulation environment to simulate the fitting feedback behavior of each target user and obtain virtual feedback results. Based on the aforementioned proportion, the virtual feedback results are integrated with similar historical style business data to conduct a hybrid simulation evaluation of the current test garment and obtain the evaluation results. In response to the evaluation results or the change in external demand corresponding to the current test garment, the design data of the current test garment is adjusted.
2. The method for measuring clothing patterns based on artificial intelligence according to claim 1, characterized in that, Obtain the current apparel design data to identify the target group that matches the design data, specifically including: Obtain partial and overall design data of the current test garment; The key parts of the local design data are identified to obtain one or more design elements corresponding to the key parts, and the color data and garment silhouette data of the overall design data are extracted; wherein, the key parts correspond to local structures that have functional or visual distinctiveness. Based on the similarity matching of the design elements, color data, and garment silhouette data with the data corresponding to each garment style tag, the garment style tag of the current test garment is determined. Based on the nodes in the pre-built knowledge graph associated with the clothing style tags, a target group matching the design data is determined; wherein, the pre-built knowledge graph is obtained based on historical sales data statistics.
3. The method for measuring clothing patterns based on artificial intelligence according to claim 1, characterized in that, Based on the structured distribution information corresponding to the target group, structured profile information of multiple target users is generated, specifically including: Obtain the distribution characteristics of the target group in the feature space and the proportion of different feature combinations; Based on the relationship between the preset number of virtual character agents and the proportion, the number of characters to be generated for each feature is determined. Based on the distribution characteristics, sampling is performed in the feature space to generate a combination of feature values corresponding to the number of roles to be generated, thereby obtaining structured profile information of multiple target users.
4. The method for measuring clothing patterns based on artificial intelligence according to claim 2, characterized in that, The structured profile information of each target user is sequentially loaded into a virtual role agent deployed in a pre-set simulation environment to simulate the try-on feedback behavior of each target user and obtain virtual feedback results, specifically including: The structured profile information of each target user is sequentially loaded into the virtual character agent deployed in the pre-set simulation environment to obtain the body shape feature data of each target user; Based on the body shape feature data of each target user, a three-dimensional human body model corresponding to the virtual character agent is determined; the three-dimensional human body model is obtained by matching the body shape feature data with existing models. Based on the design data of the current test garment and the three-dimensional human body model, the fitting state is obtained, and the virtual feedback result is generated based on the fitting state and the aesthetic and behavioral characteristic data of the target user.
5. The method for measuring clothing patterns based on artificial intelligence according to claim 4, characterized in that, Based on the body shape feature data of each target user, the three-dimensional human body model corresponding to the virtual character agent is determined: Based on the size information of the current test garment to be put on the shelves, determine the measurement area corresponding to the current test garment; The body shape feature data is analyzed, and the circumference parameters and key morphological parameters associated with the body measurement area are extracted to construct a standardized body shape query vector. The standardized body shape query vector is matched with an existing model, and the existing model corresponding to the standardized body shape query vector is obtained based on the matching result as the base model. Based on the individual preference parameter values corresponding to the aesthetic feature data, the basic model is adaptively adjusted for non-body type visualization features to generate a three-dimensional human body model.
6. The method for measuring clothing patterns based on artificial intelligence according to claim 4, characterized in that, Based on the fitting status, the fit error of each key part is determined. Combined with the aesthetic and behavioral characteristic data of the target user, the virtual feedback result is generated, specifically including: The static fitting deviation value is determined by the average Euclidean distance between the three-dimensional mesh data and the three-dimensional human body model at the key parts; Based on the clothing type corresponding to the current evaluation clothing, a preset simulated action sequence corresponding to the current evaluation clothing is determined. Based on the preset simulated action sequence, the maximum displacement of the key parts is determined, and the maximum displacement is used as the dynamic activity tolerance. The static fitting deviation value and the dynamic activity tolerance are weighted and fused to generate the fitting error of the key parts; The degree of aesthetic overlap is determined based on the degree of overlap between the target user's aesthetic feature data and the design elements, and between the color data and the clothing silhouette data. The behavioral compatibility is obtained by matching the behavioral characteristic data of the target user with the pre-set data of the current test clothing. The virtual feedback result is generated by combining the fitting error, the aesthetic overlap, and the behavioral compatibility.
7. The method for measuring clothing patterns based on artificial intelligence according to claim 1, characterized in that, Based on the stated proportion, the virtual feedback results are integrated with historical business data of similar styles to conduct a hybrid simulation evaluation of the current test garment, obtaining evaluation results, specifically including: Based on the proportion of the feature cluster to which each target user belongs, the corresponding virtual feedback results are weighted and aggregated to generate virtual feedback indicators for the target user in each dimension. The virtual feedback indicators and the historical style business data of the same type are normalized under the same dimensions, and the processed data are weighted and fused to obtain a hybrid simulation evaluation score. The hybrid simulation evaluation score is compared with the preset evaluation score and the preset threshold range to output the multi-dimensional evaluation result of the current garment.
8. The method for measuring clothing patterns based on artificial intelligence according to claim 7, characterized in that, In response to evaluation results or changes in external demand corresponding to the currently tested garment, the design data of the currently tested garment is adjusted, specifically including: The multi-dimensional evaluation results are analyzed to identify target dimensions that are below a preset feasible threshold, and the bottleneck parameters corresponding to the target dimensions are determined. Based on the bottleneck parameters, or the design attributes corresponding to the external demand change event, the design data of the current test garment is matched to determine the design data to be adjusted. Based on the type of the bottleneck parameter or the constraints of the external demand change event, the design data to be adjusted is parametrically modified to generate one or more candidate design data. The candidate design data is evaluated based on the virtual character agent to obtain updated current test clothing design data.
9. An artificial intelligence-based garment measurement device, the device comprising: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in any one of claims 1-8.
10. A non-volatile storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of performing the method described in any one of claims 1-8.