A method for predicting garment fit

Abstract

The application discloses a garment fit prediction method, and belongs to the technical field of garment performance evaluation. The method comprises the following steps: obtaining local garment fit scores and overall garment fit scores of key body parts, mapping the scores into fit grades to construct an output data set, constructing an input data set according to the undergarment air layer thickness and garment area factor of the key body parts, combining garment sizes and virtual fabric physical property parameters, constructing a local garment fit grade prediction model of the key body parts by using a multilayer perception algorithm, training the model by using the output data set and the input data set, performing weighted summation on the trained local garment fit grade prediction model of the key body parts according to the feature weights corresponding to the local garment fit grades of the key body parts, and constructing an overall garment fit grade prediction model to predict the overall garment fit. The method can improve the objectivity and accuracy of garment fit prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to a garment fit prediction method, belonging to the technical field of garment performance evaluation. BACKGROUND

[0002] Garment fit is an important indicator to measure the degree of garment fit and wearing comfort, and is the core parameter in garment design, intelligent recommendation system and virtual fitting platform. The input parameters of the current garment fit prediction model have gradually expanded from basic key point garment pressure, garment relaxation amount and other physical parameters to more complex parameters such as three-dimensional human body model and pattern data, and innovatively introduced user satisfaction score and other psychological dimension data. The output target mainly focuses on fit level prediction, and extends to pattern matching and looseness quantification.

[0003] However, for loose-fitting garments, their fit is more affected by spatial inclusiveness and freedom of movement, and traditional fit prediction models have significant limitations in predicting such garments. On the one hand, loose-fitting garments usually have a large amount of relaxation, and cannot generate effective pressure points in a virtual environment, which may cause the fit prediction model to fail. On the other hand, in garment structure design, relaxation amount is mainly designed for chest circumference, waist circumference, hip circumference and other two-dimensional dimensions, while garment fit is the perception of the human body to the fit degree of the garment in three-dimensional space. Simply relying on two-dimensional plane data cannot fully represent the complex spatial relationship between the garment and the human body in three-dimensional dressing state. SUMMARY

[0004] The purpose of the present application is to provide a garment fit prediction method that can improve the objectivity and accuracy of garment fit prediction.

[0005] To achieve the above purpose, the present application provides the following technical solutions:

[0006] In a first aspect, the present application provides a garment fit prediction method, comprising:

[0007] Obtaining local garment fit scores and overall garment fit scores of key body parts through human body wearing test and mapping them into fit levels to construct an output data set;

[0008] Constructing a three-dimensional human body dressing model based on a three-dimensional human body model and a three-dimensional garment model, and dividing it into local human body dressing sub-models corresponding to key body parts, calculating the undergarment air layer thickness and garment area factor of key body parts according to the local human body dressing sub-models corresponding to key body parts, and constructing an input data set combined with garment size and virtual fabric physical property parameters;

[0009] Based on the output data set and the input data set, a local garment fit level prediction model of a human key part is constructed and trained by using a multilayer perception algorithm, and a trained local garment fit level prediction model of the human key part is obtained;

[0010] According to the contribution degree of the local garment fit of the human key part to the overall garment fit, a feature weight corresponding to the local garment fit level of the human key part is set, and a trained local garment fit level prediction model of the human key part is weighted and summed according to the feature weight to construct an overall garment fit level prediction model;

[0011] The overall garment fit level prediction model is used for overall garment fit prediction, and an overall garment fit level is obtained.

[0012] In combination with the first aspect, further, the human wearing test comprises: selecting multiple types of garments covering different materials, sizes and styles as test samples, randomly wearing the test samples by subjects with different body characteristics in a constant temperature and humidity standard test environment, moving according to a plurality of preset working postures, and respectively scoring the local garment fit and the overall garment fit of the human key part and the overall garment fit.

[0013] In combination with the first aspect, further, the construction method of the three-dimensional human model comprises: performing full-range three-dimensional scanning on a naked human body in a standard standing posture, obtaining point cloud data of the surface morphology of the human body, and using reverse engineering technology to encapsulate, hole fill, model simplify and smooth the point cloud data, to generate a three-dimensional human model.

[0014] In combination with the first aspect, further, the construction method of the three-dimensional garment model comprises: using a three-dimensional modeling software to draw a two-dimensional garment template according to the actual size of a real garment, inputting the three-dimensional human model and the two-dimensional garment template into a three-dimensional virtual fitting software, taking the three-dimensional human model as a virtual fitting model, placing the two-dimensional garment template around the virtual model, and completing the three-dimensional shaping of the garment through virtual stitching technology, setting virtual fabric parameters in combination with the fabric properties of the garment, and generating a three-dimensional garment model.

[0015] In combination with the first aspect, further, the three-dimensional human wearing model is constructed based on the three-dimensional human model and the three-dimensional garment model, and is divided into a local human wearing sub-model corresponding to the human key part.

[0016] The three-dimensional human wearing model is generated by using a reverse engineering modeling software to match the spatial positions and align the feature points of the three-dimensional human model and the three-dimensional garment model;

[0017] With the human body joint structure as a reference, the three-dimensional human dressing model is divided into regions according to human anatomy parts by using reverse engineering software, and redundant parts outside the regions are deleted, to generate a local human dressing sub-model corresponding to a key part of the human body.

[0018] In combination with the first aspect, further, the key parts of the human body include a shoulder, a chest, a waist, a crotch, a thigh, a calf, a large arm, a small arm, an arm, an upper body and a lower body.

[0019] According to the local human dressing sub-model corresponding to the key part of the human body, the undergarment air layer thickness and the garment area factor of the key part of the human body are calculated, and the input data set is constructed in combination with the garment size and the virtual fabric physical property parameters, including:

[0020] According to the local human dressing sub-models corresponding to the shoulder, the chest, the waist, the crotch, the thigh, the calf, the large arm and the small arm, the undergarment air layer thicknesses of the shoulder, the chest, the waist, the crotch, the thigh, the calf, the large arm and the small arm are calculated respectively.

[0021] According to the local human dressing sub-models corresponding to the arm, the upper body and the lower body, the garment area factors of the arm, the upper body and the lower body are calculated respectively.

[0022] The undergarment air layer thicknesses of the shoulder, the chest, the waist, the crotch, the thigh, the calf, the large arm and the small arm, the garment size and the virtual fabric physical property parameters, and the garment area factors of the arm, the upper body and the lower body, the garment size and the virtual fabric physical property parameters, jointly constitute the input data set.

[0023] The virtual fabric physical property parameters include warp tensile modulus, weft tensile modulus, bias tensile modulus, warp bending stiffness, weft bending stiffness, bias bending stiffness, dynamic friction coefficient, static friction coefficient, fabric weight and fabric thickness.

[0024] In combination with the first aspect, further, the calculation formula of the undergarment air layer thickness is:

[0025] ;

[0026] Wherein, represents the undergarment air layer thickness, represents the volume of the three-dimensional human model corresponding to the local human dressing sub-model, represents the volume of the three-dimensional garment model corresponding to the local human dressing sub-model, represents the height of the local human dressing sub-model, represents the fabric thickness.

[0027] The calculation formula of the garment area factor is:

[0028] ;

[0029] wherein, represents a garment area factor, represents a lateral surface area of the three-dimensional human body model corresponding to the local human body dressing sub-model, represents a lateral surface area of the three-dimensional garment model corresponding to the local human body dressing sub-model.

[0030] In combination with the first aspect, further, based on the output data set and the input data set, a local garment fit level prediction model of a human key part is constructed and trained by using a multilayer perception algorithm, and a trained local garment fit level prediction model of a human key part is obtained, including:

[0031] The output data set and the input data set are preprocessed and encoded, and the feature variables are standardized and proportionally divided into a training set and a test set;

[0032] A multilayer perception neural network model including an input layer, a hidden layer and an output layer is defined, the number of neurons of the input layer is consistent with the feature dimension of the input data set, the number of neurons of the output layer is consistent with the feature dimension of the output data set, the hidden layer adopts a "pyramid" structure with decreasing number of neurons and the activation function is a ReLU function;

[0033] A cross-entropy loss function, an Adam optimizer and an L2 regularization term are introduced, and the model is trained for several rounds by using a small batch random gradient descent method, the accuracy, precision, recall and F1 score are evaluated on the test set every 10 rounds, if the accuracy of the test set stops improving for more than 10 rounds, the training is terminated in advance, and the multilayer perception neural network model with the best comprehensive performance is selected as the trained local garment fit level prediction model of the human key part.

[0034] In combination with the first aspect, further, the calculation formula of the overall garment fit level is:

[0035] ;

[0036] wherein, represents an overall garment fit level, represents a local garment fit level of a personal human key part, represents a local garment fit level of a personal human key part, represents a corresponding feature weight, represents a corresponding feature weight, represents a total number of human key parts.

[0037] In the second aspect, the present application provides a garment fit prediction device, including:

[0038] an output data set construction module configured to obtain local garment fit scores and overall garment fit scores of key body parts through a human body wearing test and map the scores to fit grades to construct an output data set;

[0039] an input data set construction module configured to construct a three-dimensional human body dressing model based on a three-dimensional human body model and a three-dimensional garment model, divide the three-dimensional human body dressing model into local human body dressing sub-models corresponding to key body parts, calculate undergarment air layer thickness and garment area factors of the key body parts according to the local human body dressing sub-models corresponding to the key body parts, and construct an input data set in combination with garment sizes and virtual fabric physical property parameters;

[0040] a prediction model construction module configured to construct a local garment fit grade prediction model of the key body parts based on the output data set and the input data set, train the local garment fit grade prediction model of the key body parts, obtain a trained local garment fit grade prediction model of the key body parts, set feature weights corresponding to the local garment fit grades of the key body parts according to contribution degrees of the local garment fit grades of the key body parts to overall garment fit, and construct an overall garment fit grade prediction model by weighted summation of the trained local garment fit grade prediction model of the key body parts according to the feature weights;

[0041] a prediction module configured to perform overall garment fit prediction by using the overall garment fit grade prediction model and obtain an overall garment fit grade.

[0042] In a third aspect, the present application provides a computer device, comprising:

[0043] a storage medium configured to store a computer program;

[0044] a processor configured to execute the computer program to implement the garment fit prediction method of the first aspect.

[0045] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the garment fit prediction method of the first aspect.

[0046] In a fifth aspect, the present application provides a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the garment fit prediction method of the first aspect.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] The clothing fit prediction method provided by this invention no longer relies on traditional ease calculation or pressure distribution diagrams, but introduces the spatial morphological characteristics of the air layer under the garment (such as thickness and area factor) as the core evaluation index. It is especially suitable for clothing types with large ease and non-fitted structure, solving the problem that existing technologies cannot accurately evaluate the fit of loose clothing, and improving the objectivity and accuracy of clothing fit prediction.

[0049] The feature system constructed in this invention integrates clothing size information, virtual fabric physical property parameters, and simulation-extracted air layer thickness and clothing area factor, realizing data fusion of three dimensions: human body, clothing, and simulation results, which significantly improves the model's generalization ability and prediction accuracy.

[0050] By using the multilayer perceptron algorithm in deep learning to process high-dimensional feature data, nonlinear relationships can be effectively modeled, and the accuracy of local clothing fit level prediction can reach 92.49%, which is better than traditional machine learning methods such as decision trees and SVM. Attached Figure Description

[0051] Figure 1 This is a flowchart of the clothing fit prediction method provided in the embodiments of the present invention;

[0052] Figure 2 This is a flowchart of the output dataset construction method provided in an embodiment of the present invention;

[0053] Figure 3 This is a flowchart of the input dataset construction method provided in an embodiment of the present invention;

[0054] Figure 4 This is a schematic diagram illustrating the application of the clothing fit prediction method provided in this embodiment of the invention in an online shopping scenario. Detailed Implementation

[0055] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0056] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Unless otherwise specified, embodiments of the present invention and the technical features thereof can be combined with each other.

[0057] This invention provides a method for predicting clothing fit, comprising:

[0058] obtaining local garment fit scores and overall garment fit scores of key body parts through human body wearing experiments and mapping the scores to garment fit grades to construct an output data set;

[0059] constructing a three-dimensional human body dressing model based on a three-dimensional human body model and a three-dimensional garment model and dividing the model into local human body dressing sub-models corresponding to key body parts, calculating undergarment air layer thickness and garment area factors of the key body parts according to the local human body dressing sub-models, and constructing an input data set in combination with garment sizes and virtual fabric physical property parameters;

[0060] based on the output data set and the input data set, constructing a local garment fit grade prediction model of the key body parts by using a multi-layer perception algorithm and training the model to obtain a trained local garment fit grade prediction model of the key body parts;

[0061] setting feature weights corresponding to the local garment fit grades of the key body parts according to the contribution degrees of the local garment fit grades to overall garment fit, and performing weighted summation on the trained local garment fit grade prediction model of the key body parts according to the feature weights to construct an overall garment fit grade prediction model;

[0062] performing overall garment fit prediction by using the overall garment fit grade prediction model to obtain an overall garment fit grade.

[0063] The garment fit prediction method provided by the embodiment of the present application can improve the objectivity and accuracy of garment fit prediction by fusing two spatial form parameters, i.e., undergarment air layer thickness and garment area factor, combining garment sizes and virtual fabric physical property parameters, constructing a local garment fit grade prediction model of key body parts and an overall garment fit grade prediction model, and performing grade prediction on local garment fit and overall garment fit, and is suitable for intelligent recommendation and customized design of loose garments.

[0064] Figure 1 is a garment fit prediction method flowchart provided by the embodiment of the present application, and the flowchart only shows the logical order of the method of the present embodiment, and the steps shown or described can be completed in an order different from that shown in the flowchart on the premise that there is no conflict. Figure 1

[0065] The garment fit prediction method provided by the embodiment of the present application can be applied to a terminal and can be executed by a garment fit prediction device, which can be realized in the form of software and / or hardware and can be integrated in the terminal, such as any tablet computer or computer device with communication function.

[0066] In one possible embodiment, as Figure 2 ​As shown, the human wearing test includes: selecting multiple types of garments covering different materials, sizes and styles as test samples, randomly wearing the test samples by subjects with different body characteristics in a constant temperature and humidity standard test environment, moving according to a plurality of preset working postures, and respectively scoring the local garment fit of the key parts of the human body and the overall garment fit.

[0067] Specifically, in order to ensure the comprehensiveness of the output data set, multiple types of garments covering different materials, sizes and styles are selected as test samples, and a subject group of no less than 21 people with representative body types are recruited to carry out human wearing tests to cover a wide range of body shape differences. The subjects wear different types of garments and complete various work actions (such as raising hands, bending over, squatting, etc.). The subjects need to score the local garment fit of 11 key parts of the human body (shoulders, chest, waist, crotch, thighs, calves, large arms, small arms, arms, upper body and lower body) in each action state during wearing, and score the overall garment fit after completing all work actions. The subjective fit perception is converted into a fit score, and the fit score is mapped to five fit levels.

[0068] Firstly, the subjective evaluation of the local garment fit of the key parts of the human body under various work actions is carried out using a five-level quantitative scale, with scale markings from -2 to 2, corresponding to different local garment fit levels. Among them, -2 represents "very tight or very short", -1 represents "relatively tight or relatively short", 0 represents "moderate", 1 represents "relatively loose or relatively long", and 2 represents "very loose or very long".

[0069] Secondly, the local garment fit scores of the key parts of the human body under various work actions are arithmetically averaged to obtain the local garment fit score of the key parts of the human body, and the local garment fit score and the overall garment fit score are mapped to five fit levels according to the preset level standard. Among them, level 1 represents very tight or very short, level 2 represents relatively tight or relatively short, level 3 represents moderate, level 4 represents relatively loose or relatively long, and level 5 represents very loose or very long.

[0070] Finally, the output data set is composed of the local garment fit level data of 11 key parts of the human body and the overall garment fit level data.

[0071] In one possible embodiment, as shown in Figure 3 The construction method of the three-dimensional human body model includes: performing full-range three-dimensional scanning on a naked human body in a standard standing posture, obtaining point cloud data of the surface morphology of the human body, and using reverse engineering technology to encapsulate, hole fill, model simplify and smooth the point cloud data, to generate a three-dimensional human body model.

[0072] Specifically, a three-dimensional scanner is used to conduct full-range three-dimensional scanning on a naked human body in a standard standing posture, point cloud data of the surface morphology of the human body is collected, and the point cloud data is encapsulated, hole filled, model simplified and smoothed by using reverse engineering technology, so that a high-precision, smooth and complete encapsulated three-dimensional human body model is generated.

[0073] In one possible embodiment, as shown in FIG. 1, the method for constructing a three-dimensional garment model comprises the following steps: Figure 3 Specifically, the actual size of a real garment is measured, a professional garment three-dimensional modeling software is used to accurately draw a digital two-dimensional garment template completely consistent with the size of the real garment, so as to ensure the accuracy and standardization of the version of the two-dimensional garment template. The three-dimensional human body model and the two-dimensional garment template are input into a three-dimensional virtual fitting software, the three-dimensional human body model is converted into a virtual fitting model, the two-dimensional garment template is placed around the virtual model, and the three-dimensional formation of the garment is completed through virtual stitching technology. At the same time, the virtual parameters such as stretching and bending of the virtual fabric are set according to the fabric properties, the color and texture of the garment are adjusted, and finally the three-dimensional garment models of different styles of garments worn by each subject are generated.

[0074] In one possible embodiment, as shown in FIG. 1, the method for constructing a three-dimensional garment model comprises the following steps:

[0075] In one possible embodiment, as shown in FIG. 1, the method for constructing a three-dimensional garment model comprises the following steps: Figure 3 Specifically, the three-dimensional human body model and the three-dimensional garment model are matched in space and position by using reverse engineering modeling software, and are aligned by feature points, so as to realize accurate registration of the three-dimensional human body model and the three-dimensional garment model, and obtain a three-dimensional human body dressing model that fits the real wearing state.

[0076] Step 1: The three-dimensional human body model and the three-dimensional garment model are matched in space and position by using reverse engineering modeling software, and are aligned by feature points, so as to realize accurate registration of the three-dimensional human body model and the three-dimensional garment model, and obtain a three-dimensional human body dressing model that fits the real wearing state.

[0077] Specifically, the three-dimensional human body model and the three-dimensional garment model are matched in space and position by using reverse engineering modeling software, and are aligned by feature points, so as to realize accurate registration of the three-dimensional human body model and the three-dimensional garment model, and obtain a three-dimensional human body dressing model that fits the real wearing state.

[0078] Step 2: Taking the joint structure of the human skeleton as a reference, the three-dimensional human body dressing model is regionally divided according to the anatomical parts of the human body by using reverse engineering software, and the redundant parts outside the region are deleted, so as to generate a local human body dressing sub-model corresponding to the key parts of the human body.

[0079] Specifically, according to human anatomy parts, the three-dimensional human dressing model is divided into multiple independent local areas by using reverse engineering modeling software, and the redundant parts beyond the preset range are deleted, and finally 11 local human dressing sub-models corresponding to the shoulder, chest, waist, hip, thigh, calf, upper arm, forearm, arm, upper body and lower body are obtained.

[0080] In one possible embodiment, the input data set is constructed by calculating the undergarment air layer thickness and the clothing area factor of the human key parts according to the local human dressing sub-models corresponding to the human key parts, and combining the clothing size and the virtual fabric physical property parameters, and specifically includes the following steps:

[0081] Step 1: According to the local human dressing sub-models corresponding to the shoulder, chest, waist, hip, thigh, calf, upper arm and forearm, the undergarment air layer thicknesses of the shoulder, chest, waist, hip, thigh, calf, upper arm and forearm are calculated respectively.

[0082] Specifically, the calculation formula of the undergarment air layer thickness is:

[0083] ;

[0084] Wherein, represents the undergarment air layer thickness, and the unit is mm, represents the volume of the three-dimensional human model corresponding to the local human dressing sub-model, and the unit is mm 3 , represents the volume of the three-dimensional clothing model corresponding to the local human dressing sub-model, and the unit is mm 3 , represents the height of the local human dressing sub-model, and the unit is mm, represents the fabric thickness, and the unit is mm.

[0085] Step 2: According to the local human dressing sub-models corresponding to the arm, upper body and lower body, the clothing area factors of the arm, upper body and lower body are calculated respectively.

[0086] Specifically, the calculation formula of the clothing area factor is:

[0087] ;

[0088] Wherein, represents the clothing area factor, and the unit is dimensionless, represents the lateral surface area of the three-dimensional human model corresponding to the local human dressing sub-model, and the unit is mm 2 , represents the lateral surface area of the three-dimensional clothing model corresponding to the local human dressing sub-model, and the unit is mm2 .

[0089] Step 3: The input data set is composed of the undergarment air layer thickness of the shoulder, chest, waist, crotch, thigh, calf, upper arm and lower arm, the garment size and the virtual fabric physical property parameters, and the garment area factor, garment size and virtual fabric physical property parameters of the arm, upper body and lower body.

[0090] Specifically, the virtual fabric physical property parameters include warp tensile modulus, weft tensile modulus, bias tensile modulus, warp bending stiffness, weft bending stiffness, bias bending stiffness, dynamic friction coefficient, static friction coefficient, fabric weight and fabric thickness.

[0091] In this embodiment, a three-dimensional body model is constructed for each subject using three-dimensional scanning technology. A three-dimensional body-dressed model is constructed for each subject wearing each garment, and is structurally segmented according to anatomical features to obtain 11 local body-dressed sub-models corresponding to key body parts. The undergarment air layer thickness is calculated for the local body-dressed sub-models corresponding to the 8 key body parts of the shoulder, chest, waist, crotch, thigh, calf, upper arm and lower arm. The garment area factor is calculated for the local body-dressed sub-models corresponding to the 3 key body parts of the arm, upper body and lower body. The garment size, virtual fabric physical property parameters and calculated undergarment air layer thickness or garment area factor are integrated to construct the input data set, making the input features more comprehensive.

[0092] In one possible embodiment, based on the output data set and the input data set, a local garment fit level prediction model for the key body parts is constructed and trained using a multi-layer perception algorithm, and the trained local garment fit level prediction model for the key body parts includes the following steps:

[0093] Step 1: Preprocess and encode the output data set and the input data set, and divide them into a training set and a test set in proportion after standardizing the feature variables;

[0094] Specifically, the output data set and the input data set are preprocessed and encoded, and the feature variables are standardized and divided into a training set and a test set in a ratio of 7:3 to ensure the effectiveness of model training and evaluation.

[0095] Step 2: Define a multi-layer perception (MLP) neural network model including an input layer, a hidden layer and an output layer, the number of neurons in the input layer is consistent with the feature dimension of the input data set, the number of neurons in the output layer is consistent with the feature dimension of the output data set, the hidden layer adopts a “pyramid” structure with decreasing number of neurons and the activation function is ReLU function;

[0096] Specifically, the hidden layer is set to 2-3 layers, adopts a "pyramid structure" to decrease the number of neurons, and uses a ReLU function as the activation function of the hidden layer to enhance the nonlinear fitting ability.

[0097] Step 3: Introduce cross-entropy loss function, Adam optimizer and L2 regularization term, train for several rounds by small batch stochastic gradient descent method, evaluate accuracy, precision, recall and F1 score on the test set every 10 rounds, if the accuracy of the test set stops improving for more than 10 rounds, terminate the training in advance, and select the multilayer perceptron neural network model with the best comprehensive performance as the trained local garment fit level prediction model of human key parts.

[0098] Specifically, the cross-entropy loss function (Cross Entropy Loss) is used to measure the difference between the predicted level output by the local garment fit level prediction model of human key parts and the true level. The optimizer uses Adam, and the L2 regularization term (weight_decay parameter) is introduced to prevent overfitting. During training, the small batch (minibatch) stochastic gradient descent method (such as batch size=64) is used, and the total training rounds are set to 100-150. After each round, the training loss and accuracy trend are recorded. Every 10 rounds of training, the accuracy, precision, recall and F1 score are evaluated on the test set, and the loss function and accuracy curve are plotted to monitor overfitting. If the accuracy of the test set stops improving for more than 10 rounds, terminate the training in advance. Finally, select the multilayer perceptron neural network model with the best comprehensive performance on the test set as the trained local garment fit level prediction model of human key parts.

[0099] Based on the measured data of 21 subjects, 11 MLP local garment fit level prediction models of human key parts were constructed for local garment fit level prediction experiments. The experimental results show that the average accuracy is 92.49%, the precision is 89.11%, the recall is 90.96%, and the F1 score is 89.86%, which is significantly better than traditional models such as logistic regression, random forest and support vector machine.

[0100] In this embodiment, the calculation formula of the overall garment fit level is:

[0101] ;

[0102] Among them, represents the overall garment fit level, represents the local garment fit level of the i-th human key part, represents the local garment fit level of the i-th human key part, represents the local garment fit level of the i-th human key part, the corresponding feature weight, represents the total number of human key parts.

[0103] In this embodiment, according to the contribution degree of the local garment fit of the human key part to the overall garment fit, the feature weight corresponding to the local garment fit grade of the human key part is set, and the local garment fit grades of the human key parts are weighted and summed according to the feature weight to obtain the overall garment fit grade, so that the structural combination prediction is realized.

[0104] Specifically, the feature weights corresponding to the local garment fit grades of the shoulder, chest, waist, crotch, thigh, calf, large arm, small arm, arm, upper body and lower body are respectively 0.069, 0.093, 0.108, 0.106, 0.108, 0.105, 0.084, 0.053, 0.091, 0.097 and 0.087.

[0105] The garment fit prediction method provided by the embodiment of the application clearly defines the structural attribution relationship of 11 human key parts in garment fit, and gives different feature weights to each human key part, thereby establishing a combination prediction path of “local fit→overall fit”, which has clearer prediction structure, stronger explanation, and is beneficial to realizing high-precision personalized recommendation in garment e-commerce, virtual fitting platform and industrial garment design.

[0106] The embodiment of the application provides an application of the garment fit prediction method in an online shopping scene, which can realize accurate size recommendation for terminal users, such as Figure 4 As shown in the figure, the specific process is as follows:

[0107] The terminal user selects a garment size on an e-commerce platform, and inputs human size data (such as height, weight, shoulder width, chest circumference, waist circumference, hip circumference, limb length, etc.) through a platform interactive interface; after the system receives the human size data, it automatically calls a three-dimensional modeling software to construct a three-dimensional human model for the terminal user; then, the system imports the three-dimensional human model into virtual fitting software, simultaneously loads a digitalized two-dimensional template of the garment selected by the user and preset virtual fabric physical property parameters, and completes virtual fitting.

[0108] After the virtual fitting is completed, the system performs structured segmentation on the three-dimensional human dressed model through reverse engineering technology, extracts local human dressed sub-models corresponding to 11 human key parts, calculates the air layer thickness of 8 parts of the shoulder, chest, waist, crotch, thigh, calf, upper arm and lower arm, and the clothing area factor of 3 parts of the arm, upper body and lower body; after integrating these air layer morphological characteristic parameters with the clothing size and virtual fabric physical property parameters, the 11 local clothing fit level prediction models of human key parts are input into the trained local clothing fit level prediction models, and the local clothing fit level of 11 human key parts is obtained; then, the local clothing fit level of 11 human key parts is weighted and summed according to the feature weight, and the overall clothing fit level is obtained and fed back to the terminal user.

[0109] If the terminal user is satisfied with the recommended result, the purchase can be directly completed; if not, the system will automatically recommend an adjusted clothing size (such as increasing or decreasing the size) according to the current prediction result, repeat the virtual fitting, feature extraction and fit prediction process until the user obtains a satisfactory fit level, thereby realizing the closed-loop recommendation of “try on-predict-adjust-confirm”, and effectively improving the size selection efficiency and user experience of online shopping.

[0110] In this embodiment, a suitable size of a one-piece garment is selected for a subject with a height of 179 cm and a weight of 93 kg, and 3 to 5 sizes of the same fabric are selected for virtual fitting (all sizes of the cotton fabric garment in this example, including M, L, XL, 2XL and 3XL), the predicted local clothing fit level is obtained, and the overall clothing fit level is finally calculated by weighted summation, and the prediction result is shown in Table 1. It can be concluded that the overall clothing fit level of size 2XL is 3.094, which is the closest to the “moderate” level (level 3), meaning that compared with other sizes, size 2XL is more suitable for this subject. Therefore, the system will recommend him to wear size 2XL one-piece garment.

[0111] Table 1: Prediction results of local clothing fit level and overall clothing fit level

[0112] .

[0113] The embodiment of the present application provides a clothing fit prediction device, which comprises:

[0114] An output data set construction module is configured to obtain local clothing fit scores and overall clothing fit scores of human key parts through human wearing experiments and map them into fit levels to construct an output data set;

[0115] The input data set construction module is configured to construct a three-dimensional human body dressing model based on the three-dimensional human body model and the three-dimensional garment model, divide the three-dimensional human body dressing model into local human body dressing sub-models corresponding to key human body parts, calculate the undergarment air layer thickness and garment area factor of the key human body parts according to the local human body dressing sub-models corresponding to the key human body parts, and construct the input data set in combination with the garment size and virtual fabric physical property parameters.

[0116] The prediction model construction module is configured to construct and train a local garment fit level prediction model of the key human body part based on the output data set and the input data set by using a multilayer perception algorithm, obtain the trained local garment fit level prediction model of the key human body part, set a feature weight corresponding to a local garment fit level of the key human body part according to a contribution degree of the local garment fit level of the key human body part to the overall garment fit, and perform weighted summation on the trained local garment fit level prediction model of the key human body part according to the feature weight to construct an overall garment fit level prediction model.

[0117] The prediction module is configured to perform overall garment fit prediction by using the overall garment fit level prediction model and obtain an overall garment fit level.

[0118] The garment fit prediction device provided in the embodiment of the present application can execute the garment fit prediction method provided in the embodiment of the present application, and has the function modules and beneficial effects corresponding to the execution method.

[0119] The embodiment of the present application provides a computer device, comprising:

[0120] The storage medium is configured to store the computer program.

[0121] The processor is configured to execute the computer program to implement the garment fit prediction method provided in the embodiment of the present application.

[0122] The embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the garment fit prediction method provided in the embodiment of the present application.

[0123] The embodiment of the present application provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the garment fit prediction method provided in the embodiment of the present application.

[0124] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0125] The present application is described in reference to the flowchart and / or block diagrams of the method, apparatus (system) and computer program product according to embodiments of the present application. It will be understood that each block of the flowchart and / or block diagrams, and combinations of blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0126] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0128] The above merely preferred embodiments of the present application and it is to be understood that those still skilled in the art will be able to propose several improvements and modifications to the present application without departing from the technical principles of the present application, and these improvements and modifications are also to be considered as falling within the scope of the present application.

Claims

1. A garment fit prediction method characterized by, The method comprises the following steps: Obtaining the local garment fit scores and the overall garment fit scores of the human body key parts through a human body wearing test and mapping the scores to a fit level to construct an output data set; Based on the three-dimensional human body model and the three-dimensional garment model, a three-dimensional human body dressing model is constructed and divided into local human body dressing sub-models corresponding to the human body key parts. The undergarment air layer thickness and the garment area factor of the human body key parts are calculated according to the local human body dressing sub-models corresponding to the human body key parts. The input data set is constructed by combining the garment size and the virtual fabric physical property parameters; Based on the output data set and the input data set, a local garment fit level prediction model of the human body key parts is constructed and trained by using a multi-layer perception algorithm to obtain the trained local garment fit level prediction model of the human body key parts; According to the contribution degree of the local garment fit of the human body key parts to the overall garment fit, the feature weight corresponding to the local garment fit level of the human body key parts is set, and the trained local garment fit level prediction model of the human body key parts is weighted and summed according to the feature weight to construct an overall garment fit level prediction model; The overall garment fit level prediction model is used to predict the overall garment fit, and the overall garment fit level is obtained. The human body key parts include the shoulder, the chest, the waist, the hip crotch, the thigh, the calf, the large arm, the small arm, the arm, the upper body and the lower body. The undergarment air layer thickness and the garment area factor of the human body key parts are calculated according to the local human body dressing sub-models corresponding to the human body key parts, and the input data set is constructed by combining the garment size and the virtual fabric physical property parameters. The undergarment air layer thickness of the shoulder, the chest, the waist, the hip crotch, the thigh, the calf, the large arm and the small arm is calculated according to the local human body dressing sub-models corresponding to the shoulder, the chest, the waist, the hip crotch, the thigh, the calf, the large arm and the small arm, respectively. The garment area factor of the arm, the upper body and the lower body is calculated according to the local human body dressing sub-models corresponding to the arm, the upper body and the lower body, respectively. The undergarment air layer thickness of the shoulder, the chest, the waist, the hip crotch, the thigh, the calf, the large arm and the small arm, the garment size and the virtual fabric physical property parameters, and the garment area factor of the arm, the upper body and the lower body, the garment size and the virtual fabric physical property parameters, jointly constitute the input data set. The virtual fabric physical property parameters include the warp tensile modulus, the weft tensile modulus, the bias tensile modulus, the warp bending stiffness, the weft bending stiffness, the bias bending stiffness, the dynamic friction coefficient, the static friction coefficient, the fabric weight and the fabric thickness.

2. The garment fit prediction method of claim 1, wherein, The human body wearing test comprises the following steps: selecting multiple types of garments covering different materials, sizes and styles as test samples, randomly wearing the test samples by subjects with different body characteristics in a constant temperature and humidity standard test environment, moving according to a plurality of preset working postures, and scoring the local garment fit and the overall garment fit of the human body key parts and the overall garment fit.

3. The garment fit prediction method of claim 1, wherein, The method for constructing a three-dimensional human body model comprises: performing omnidirectional three-dimensional scanning on a naked human body in a standard standing posture to obtain point cloud data of the surface morphology of the human body, and performing encapsulation, hole filling, model simplification and smoothing processing on the point cloud data by using reverse engineering technology to generate a three-dimensional human body model.

4. The garment fit prediction method of claim 1, wherein, The method for constructing a three-dimensional garment model comprises: drawing a two-dimensional garment template according to the actual size of a real garment by using a three-dimensional modeling software, inputting the three-dimensional human body model and the two-dimensional garment template into a three-dimensional virtual fitting software, taking the three-dimensional human body model as a virtual fitting model, placing the two-dimensional garment template around the virtual model, and completing three-dimensional shaping of the garment by using a virtual stitching technology, setting virtual fabric parameters in combination with the fabric properties of the garment to generate a three-dimensional garment model.

5. The garment fit prediction method of claim 1, wherein, The method for constructing a three-dimensional human body dressed model based on the three-dimensional human body model and the three-dimensional garment model and dividing the three-dimensional human body dressed model into local human body dressed sub-models corresponding to key parts of the human body comprises: Performing spatial position matching and feature point alignment on the three-dimensional human body model and the three-dimensional garment model by using a reverse engineering modeling software to generate a three-dimensional human body dressed model; Taking the joint structure of the human skeleton as a reference, performing regional division on the three-dimensional human body dressed model according to the anatomical parts of the human body by using the reverse engineering software, and deleting redundant parts outside the regions to generate local human body dressed sub-models corresponding to key parts of the human body.

6. The garment fit prediction method of claim 1, wherein, The calculation formula of the thickness of the air layer under the garment is: ； wherein, represents the thickness of the air layer under the garment, represents the volume of the three-dimensional body model to which the local body-dressing sub-model corresponds, represents the volume of the three-dimensional garment model to which the local body-dressing sub-model corresponds, represents the height of the local body-dressing sub-model, represents the fabric thickness; The calculation formula of the garment area factor is: ； wherein, represents a garment area factor, represents a lateral surface area of the three-dimensional human body model to which the local human dressing sub-model corresponds, represents a lateral surface area of the three-dimensional garment model to which the local human dressing sub-model corresponds.

7. The garment fit prediction method of claim 1, wherein, Based on the output data set and the input data set, a local garment fit level prediction model of a key part of the human body is constructed by using a multilayer perception algorithm and is trained to obtain a trained local garment fit level prediction model of a key part of the human body, which comprises: The output data set and the input data set are preprocessed and encoded, and the feature variables are standardized and proportionally divided into a training set and a test set; A multilayer perception neural network model comprising an input layer, a hidden layer and an output layer is defined, the number of neurons of the input layer is consistent with the feature dimension of the input data set, the number of neurons of the output layer is consistent with the feature dimension of the output data set, the hidden layer adopts a "pyramid” structure with decreasing number of neurons and the activation function is a ReLU function; A cross-entropy loss function, an Adam optimizer and an L2 regular term are introduced, and the model is trained for several rounds by using a small batch random gradient descent method, the accuracy, precision, recall and F1 score are evaluated on the test set every 10 rounds, if the accuracy of the test set stops improving for more than 10 rounds, the training is terminated in advance, and the multilayer perception neural network model with the best comprehensive performance is selected as the trained local garment fit level prediction model of a key part of the human body.

8. The garment fit prediction method of claim 1, wherein, The calculation formula of the overall garment fit level is: ； wherein, denotes the overall garment fit grade, denotes the local garment fit grade of the individual body key part, denotes the corresponding feature weight, denotes the total number of body key parts.

9. A garment fit prediction apparatus characterized by, The method comprises: An output data set construction module is configured to obtain local garment fit scores of key parts of the human body and overall garment fit scores by human wearing tests, and map the scores into fit levels to construct an output data set; The input data set construction module is configured to construct a three-dimensional human body dressing model based on the three-dimensional human body model and the three-dimensional garment model, divide the three-dimensional human body dressing model into local human body dressing sub-models corresponding to human body key parts, calculate the undergarment air layer thickness and garment area factor of the human body key parts according to the local human body dressing sub-models corresponding to the human body key parts, and construct the input data set in combination with the garment size and the virtual fabric physical property parameters. The prediction model construction module is configured to construct and train a local garment fit level prediction model of the human body key parts based on the output data set and the input data set by using a multi-layer perception algorithm, obtain the trained local garment fit level prediction model of the human body key parts, set feature weights corresponding to the local garment fit levels of the human body key parts according to the contribution degrees of the local garment fit levels of the human body key parts to the overall garment fit, and construct an overall garment fit level prediction model by weighted summation of the trained local garment fit level prediction model of the human body key parts according to the feature weights. The prediction module is configured to perform overall garment fit prediction by using the overall garment fit level prediction model, and obtain an overall garment fit level. The human body key parts include shoulders, a chest, a waist, a hip crotch, thighs, calves, large arms, small arms, arms, an upper body, and a lower body. The input data set construction module is configured to construct a three-dimensional human body dressing model based on the three-dimensional human body model and the three-dimensional garment model, divide the three-dimensional human body dressing model into local human body dressing sub-models corresponding to human body key parts, calculate the undergarment air layer thickness and garment area factor of the human body key parts according to the local human body dressing sub-models corresponding to the human body key parts, and construct the input data set in combination with the garment size and the virtual fabric physical property parameters. The input data set construction module is configured to construct a three-dimensional human body dressing model based on the three-dimensional human body model and the three-dimensional garment model, divide the three-dimensional human body dressing model into local human body dressing sub-models corresponding to human body key parts, calculate the undergarment air layer thickness and garment area factor of the human body key parts according to the local human body dressing sub-models corresponding to the human body key parts, and construct the input data set in combination with the garment size and the virtual fabric physical property parameters. The input data set construction module is configured to construct a three-dimensional human body dressing model based on the three-dimensional human body model and the three-dimensional garment model, divide the three-dimensional human body dressing model into local human body dressing sub-models corresponding to human body key parts, calculate the undergarment air layer thickness and garment area factor of the human body key parts according to the local human body dressing sub-models corresponding to the human body key parts, and construct the input data set in combination with the garment size and the virtual fabric physical property parameters. The virtual fabric physical property parameters include warp tensile modulus, weft tensile modulus, bias tensile modulus, warp bending stiffness, weft bending stiffness, bias bending stiffness, dynamic friction coefficient, static friction coefficient, fabric weight, and fabric thickness. ​

Images