Automatic pattern making method and system for personalized clothing customization

Abstract

The present application relates to an automatic pattern making method and system for personalized clothing customization; the method includes: obtaining the net body size data and demand data of the user; inputting the net body size data and demand data into a pre-trained automatic pattern making algorithm model for calculation, and outputting the garment size data; obtaining the human body image data of the user; inputting the human body image data into a pre-trained body type classification algorithm model for processing, and outputting the body type category label; determining the corresponding body type modification rule according to the body type category label, combining the body type modification rule with the initial pattern adjustment rule to output the final pattern adjustment rule; outputting the garment size data, change items and final pattern adjustment rules as a file in a specified format. The scheme of the present application is based on an artificial intelligence machine learning algorithm, which makes the entire clothing customization process intelligent and automated, can eliminate the uncertainty caused by human factors in the actual production process, and improve the accuracy and comfort of the final clothing size.

Description

Technical Field

[0001] The present application relates to the technical field of clothing customization production, and in particular to an automatic pattern making method and system for personalized clothing customization. Background Art

[0002] The current industry situation generally relies on the manual experience of pattern makers to make patterns for customers, and in this process, it relies heavily on the pattern makers' own professional experience. However, different pattern makers have different experiences. Even if the same pattern is used to make a pattern for the same customer, completely different garment data results will be generated. In the process of pattern making, the subjective initiative of many pattern makers is full of uncertainty, which greatly affects the accuracy and stability of the final product, and it is impossible to accurately measure which pattern maker does a better job, because in real life, it is impossible for a customer to place an order for two pattern makers to do it. As for the generation of body type identification, change items and pattern adjustment rules with a higher degree of difficulty, it is even more difficult to achieve empirical and unified.

[0003] Among the related technologies, some technical solutions have complicated steps for inputting clothing pattern parameters and have high data requirements. The method of manually calibrating key points and structural line diagrams also requires the operator to have relatively rich industry experience and knowledge. The operation process is cumbersome and inefficient, and it is impossible to achieve millisecond-level plate making, and the labor cost is high. Due to its own limitations (non-universality) and poor scalability, it is unable to effectively improve the production efficiency of the entire clothing customization industry.

[0004] In addition, the labor cost of pattern makers is very high, and the value of clothing pattern libraries to factories is immeasurable, which also limits the development of the clothing customization industry. The traditional high-end customization process relies heavily on the manual experience of pattern makers, which leads to problems such as low efficiency and high rework.

[0005] In summary, the mapping from the customer's body data to the garment size data requires a tedious pattern-making process by professional pattern makers, which is inefficient and expensive. This pattern-making method can no longer meet the current market demand for large-scale customization. Summary of the invention

[0006] In order to overcome the problems existing in the related art to at least a certain extent, the present application provides an automatic pattern making method and system for personalized clothing customization.

[0007] According to a first aspect of an embodiment of the present application, there is provided an automatic pattern making method for personalized clothing customization, comprising:

[0008] Obtain the user's net body size data and demand data;

[0009] Input the net body size data and demand data into the pre-trained automatic pattern making algorithm model for calculation, and output the garment size data, change items and initial pattern adjustment rules;

[0010] Acquire user's human body image data;

[0011] Input the human body image data into the pre-trained body shape classification algorithm model for processing, and output the body shape category label;

[0012] Determine the corresponding body shape modification rule according to the body shape category label, combine the body shape modification rule with the initial pattern adjustment rule, and output the final pattern adjustment rule;

[0013] Output the garment size data, change items and final pattern adjustment rules into a file of the specified format.

[0014] Furthermore, the net body size data is attribute data of multiple dimensions commonly used in the clothing customization industry, including at least: clothing length, sleeve length, shoulder width, chest circumference, trouser length, waist circumference, and hip circumference;

[0015] The demand data includes: first demand data associated with the size of the ready-made garment, and second demand data associated with the change item; wherein the first demand data at least includes slim fit, loose fit, fitted, slightly longer sleeves, slightly shorter sleeves, cropped trousers, slightly wider trouser legs, slightly thinner trouser legs, slightly narrow shoulders, slightly wider shoulders, slightly longer length, slightly shorter length; the second demand data at least includes rolled-up trouser legs, slightly wider sleeves, and slightly thin sleeves;

[0016] The ready-made garment size data includes ready-made garment attributes of different dimensions required for garment customization.

[0017] The plate adjustment rules are data for changing and adjusting the plate-making CAD according to the demand data.

[0018] Furthermore, the training process of the automatic pattern making algorithm model includes:

[0019] Traverse the garment attributes of different parts respectively, and use the correlation coefficient method to calculate the N net body size data most relevant to the garment attributes of each part;

[0020] The net body size data after feature selection is divided into training data set and test data set by random sampling;

[0021] Through the K-Fold cross-training method, the training data set is evenly divided into several parts, and cross-training and verification are continuously performed;

[0022] Use machine learning regression algorithm to train the model, set the model parameters, select the parameters with the best model effect and save them;

[0023] Use the saved optimal model to test the net body size data in the test data set, output the garment size data, and calculate the mean absolute error (MAE) based on the output garment size data;

[0024] The saved optimal model is iteratively optimized using reinforcement learning method.

[0025] Furthermore, the setting of the model parameters includes: automatically searching for parameters and / or manually setting parameters;

[0026] The method of selecting the parameters with the best model effect includes: using a loss function evaluation method of a regression model to determine the parameters with the best effect; wherein the loss function includes a square root error and / or a mean absolute error.

[0027] Furthermore, the iterative optimization of the saved optimal model by using the reinforcement learning method includes:

[0028] Determine whether the calculated MAE result is credible;

[0029] If it is credible, add the net body size data in the corresponding test data set and the corresponding garment size data output by the model to the training set, and iterate the training model;

[0030] If it is not credible, the template is manually adjusted and modified according to the preset template adjustment rules, and then the MAE is calculated again to determine whether it is credible and iterate again.

[0031] Furthermore, the training process of the body shape classification algorithm model includes:

[0032] Collecting body images of the front, back and side of the human body through multiple channels, and cropping and screening the collected images;

[0033] Classify and label the images according to pre-set body shape features;

[0034] The annotated images are randomly divided into training and test sets according to a predetermined ratio;

[0035] The images of the training set are input into the computer vision AI algorithm model based on deep learning, and the body shape features are trained and fitted using the stochastic gradient descent method to generate a human body shape classification algorithm model;

[0036] The algorithm model is verified through the validation set.

[0037] Furthermore, the inputting of the human body image data into a pre-trained body shape classification algorithm model for processing includes:

[0038] Construct the convolutional neural network of the algorithm, initialize and load the pre-trained network parameters;

[0039] The preprocessed image is input into the convolutional neural network for feature extraction to obtain a feature map;

[0040] The extracted feature region is refined through the RPN network to obtain a refined feature region;

[0041] The RoIAlign operation is performed by bilinear interpolation method to process the feature map so that each feature area generates a feature map of fixed size;

[0042] The classification branch obtains the category information of the target, the regression branch obtains the coordinate information of the target box, and the mask network obtains the mask information;

[0043] The category information, coordinate information, and mask information are applied to the feature map to obtain instance segmentation of human body shape categories and human body shape category labels.

[0044] Furthermore, the pre-set body shape characteristics include normal body and special body;

[0045] The special body types of human frontal images include: sloping shoulders, flat shoulders, low right shoulder and low left shoulder;

[0046] The special body types of human side images include: convex belly, hunchback, hunched back, erect chest, erect chest and supine body, bent waist, round back and flat back;

[0047] The special types of human buttocks images include: perky buttocks, flat buttocks, high hips, low front waist, and wide / narrow between legs;

[0048] Each human body corresponds to a combination of body shape features in three parts: the front, side, and buttocks.

[0049] According to a second aspect of an embodiment of the present application, there is provided an automatic pattern making system for personalized clothing customization, comprising:

[0050] A net body data acquisition unit, used to acquire the net body size data and demand data of the user;

[0051] The pattern making algorithm unit is used to input the net body size data and the demand data into the pre-trained automatic pattern making algorithm model for calculation, and output the garment size data, the change items and the initial pattern adjustment rules;

[0052] An image data acquisition unit, used to acquire the user's human body image data;

[0053] A body shape classification unit is used to input human image data into a pre-trained body shape classification algorithm model for processing and output a body shape category label;

[0054] A data correction unit, used to determine a corresponding body shape modification rule according to the body shape category label, and output a final body shape adjustment rule after combining the body shape modification rule with the initial shape adjustment rule;

[0055] The file output unit is used to output the garment size data, change items and final pattern adjustment rules into a file of a specified format.

[0056] The technical solution provided by the embodiments of the present application has the following beneficial effects:

[0057] The solution of the present application is based on artificial intelligence machine learning algorithms, which makes the entire clothing customization process intelligent and automated, can eliminate the uncertainty caused by human factors in the actual production process, and improve the accuracy and comfort of the final clothing size; the present application is applied to the personalized clothing customization process, and the ready-made clothing size data can be automatically generated within milliseconds, greatly improving the production efficiency of the clothing customization industry; and through an easy-to-promote pattern library, it can give small and medium-sized enterprise factories the ability of advanced customization, and comprehensively promote the development of the clothing customization industry.

[0058] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0060] Figure 1 The present invention is a flowchart of an automatic pattern making method for personalized clothing customization according to an exemplary embodiment.

[0061] Figure 2 The figure is a flow chart of an automatic pattern making algorithm model according to an exemplary embodiment.

[0062] Figure 3 It is a flowchart of a human body shape recognition algorithm according to an exemplary embodiment.

[0063] FIG. 4( a ) shows an original image and a binary image of normal body features of a human shoulder according to an exemplary embodiment.

[0064] FIG4( b ) shows an original image and a binary image of a low feature of a right shoulder of a human body according to an exemplary embodiment.

[0065] FIG4( c ) shows an original image and a binary image of a low feature of a left shoulder of a human body according to an exemplary embodiment.

[0066] FIG. 5( a ) shows a body shape and sample examples of a person with sloping shoulders, stooped back and hunchback according to an exemplary embodiment.

[0067] FIG. 5( b ) shows an example of a body shape and a sample piece of a flat-shouldered convex body according to an exemplary embodiment.

[0068] FIG. 6( a ) is a sample of addition and subtraction of height of a normal body type according to an exemplary embodiment.

[0069] FIG. 6( b ) is a sample of the addition and subtraction of the chest circumference of a normal body type according to an exemplary embodiment.

[0070] FIG. 6( c ) is a sample of adding or subtracting shoulder width of a normal body type according to an exemplary embodiment.

[0071] Figure 7 It is a schematic diagram of the structure of the Mask R-CNN algorithm according to an exemplary embodiment.

[0072] Figure 8 The figure is a schematic diagram of the network topology structure of an automatic pattern making system for personalized clothing customization according to an exemplary embodiment. DETAILED DESCRIPTION

[0073] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of methods and devices consistent with some aspects of the present application as detailed in the appended claims.

[0074] Figure 1 The flowchart of an automatic pattern making method for personalized clothing customization is shown according to an exemplary embodiment. The method may include the following steps:

[0075] Step S1: Obtaining the net body size data and demand data of the user. Specifically, the net body size data and demand data can be extracted from orders from various different channels such as online and offline (for example, online shopping malls, online platforms and offline stores).

[0076] It should be noted that for the various data extracted from clothing orders, data standardization, normalization and data transformation are required to meet the input format requirements of the algorithm model.

[0077] Step S2: Input the net body size data and demand data into the pre-trained automatic pattern making algorithm model for calculation, and output the garment size data, change items and initial pattern adjustment rules.

[0078] Step S3: Acquire the user's body image data.

[0079] Step S4: input the human body image data into the pre-trained body shape classification algorithm model for processing, and output the body shape category label.

[0080] Step S5: Determine the corresponding body shape modification rule according to the body shape category label, combine the body shape modification rule with the initial version adjustment rule, and output the final version adjustment rule.

[0081] Step S6: Output the garment size data, change items and final pattern adjustment rules into a file of a specified format. Specifically, a file corresponding to the order can be generated according to the requirements of the subsequent garment CAD pattern making software, and the output data file format is expandable and customizable.

[0082] Finally, the garment size data file generated in the above steps can be output to a lower-level system or software, and the output interface forms include but are not limited to stand-alone software, networked software, and cloud servers.

[0083] The solution of the present application is based on artificial intelligence machine learning algorithms, which makes the entire clothing customization process intelligent and automated, can eliminate the uncertainty caused by human factors in the actual production process, and improve the accuracy and comfort of the final clothing size; the present application is applied to the personalized clothing customization process, and the ready-made clothing size data can be automatically generated within milliseconds, greatly improving the production efficiency of the clothing customization industry; and through an easy-to-promote pattern library, it can give small and medium-sized enterprise factories the ability of advanced customization, and comprehensively promote the development of the clothing customization industry.

[0084] The solution of this application is further explained below in combination with specific application scenarios.

[0085] In some embodiments, the net body size data is attribute data of multiple dimensions commonly used in the clothing customization industry, including at least: length of clothing, sleeve length, shoulder width, chest circumference, trouser length, waist circumference, and hip circumference.

[0086] The demand data include: first demand data associated with ready-made clothing sizes, and second demand data associated with change items; wherein the first demand data at least includes slim fit, loose fit, fitted, slightly longer sleeves, slightly shorter sleeves, cropped trousers, slightly wider trouser legs, slightly thinner trouser legs, slightly narrow shoulders, slightly wider shoulders, slightly longer length, slightly shorter length; the second demand data at least includes rolled-up trouser legs, slightly wider sleeves, and slightly thinner sleeves.

[0087] The ready-made garment size data includes ready-made garment attributes of different dimensions required for garment customization.

[0088] The master pattern is an original, proprietary clothing pattern summarized and produced by a professional pattern maker, and is used for production on a real high-end custom clothing production line.

[0089] In some embodiments, the version adjustment rule includes: changing the adjustment rule type and adjusting the change position point.

[0090] The above-mentioned pattern adjustment rules can be used for all styles of clothing, including but not limited to suits, shirts, cheongsams, etc. The present invention takes suits and cheongsams as examples for illustration, and other style rule changes are the same, mainly including top changes and bottom changes. Among them, top changes include front curvature, rear shoulder drop, chest width, back width, sleeve front and back, round back, flat back, stooped waist, chest up, front chest, chest up, stooped waist, hunchback, hunchback, convex belly, sloping shoulders, flat shoulders, sleeve depth, cuffs, sleeve fat, shoulder width, hip circumference, important, chest circumference, sleeve length, front length and length of clothes, etc. The changes of bottoms include waist circumference, hip circumference, horizontal crotch, knee circumference, leg opening, trouser length, flat buttocks, perky buttocks, thin and tall hips, convex belly, crotch, front crotch and back crotch, etc.

[0091] The categories of the above-mentioned pattern adjustment rules include not only existing categories in the industry, but also details proposed by the present invention involving categories of special bodies and their pattern adjustment rules, including attribute information and categories such as front curve, front chest, round back, hunched back, upright chest, bent waist, flat back, dropped shoulders, sloping shoulders, raised buttocks, thin and tall hips and flat buttocks.

[0092] like Figure 2 As shown, in some embodiments, the training process of the automatic pattern making algorithm model includes:

[0093] S21: traverse the attributes of the garments in different parts (such as tops or pants) respectively, and use the correlation coefficient method (i.e., the feature selection method in feature engineering) to calculate the N net body size data most relevant to the attributes of the garments in each part;

[0094] Among them, the specific value of N can be set according to the actual application, and the embodiment of this solution adopts eighteen net body size data. The feature engineering of the pre-trained model includes but is not limited to using principal component analysis and correlation analysis methods to determine the various human body dimension attribute data associated with each garment attribute.

[0095] S22: The net body size data after feature selection is divided into a training data set and a test data set by random sampling.

[0096] S23: Through the K-Fold cross-training method, the training data set is evenly divided into several parts, and cross-training and verification are continuously performed.

[0097] S24: training the model using a machine learning regression algorithm, setting model parameters, selecting parameters with the best model effect and saving them;

[0098] In some embodiments, the setting of model parameters includes: automatic parameter search and / or manual parameter setting. The selecting of the parameters with the best model effect includes: using a loss function evaluation method of a regression model to determine the parameters with the best effect; wherein the loss function includes but is not limited to square root error (R_squared) and / or mean absolute error (MAE, Mean Absolute Error).

[0099] The automatic pattern making algorithm model includes but is not limited to all regression algorithms. At the same time, as the number and scale of factories increase, after obtaining more data through the above pattern adjustment rules, the accuracy of the regression model can be improved through enhanced learning methods.

[0100] S25: Use the saved optimal model to test the net body size data in the test data set, output the garment size data, and calculate the mean absolute error MAE based on the output garment size data.

[0101] S26: Use reinforcement learning method to iteratively optimize the saved optimal model.

[0102] The automatic pattern making algorithm model of this application is a machine learning regression algorithm model trained from a large amount of real net body size feature data, garment size data, change items and data, and customer preference needs; based on the existing master pattern and pattern adjustment rules, it can help factories quickly accumulate real customer data, facilitate model optimization and the promotion of the production method of this application; the calculation results of the automatic pattern making algorithm model are automatically generated according to the requirements of the subsequent clothing CAD plate making software. The file of garment size and change items and data of the corresponding order is truly realized, truly realizing "one person, one version". At the same time, with the increase of factory and order data, more credible data can be used to improve the model accuracy through the enhanced learning method model.

[0103] Reference Figure 2 In some embodiments, the iterative optimization of the saved optimal model using the reinforcement learning method includes:

[0104] S261: Determine whether the calculated result MAE is credible;

[0105] S262: If it is credible, the net body size data in the corresponding test data set and the corresponding garment size data output by the model are added to the training set, and the model is trained iteratively;

[0106] S263: If it is not credible, the template is manually adjusted (by a person with professional experience) according to the preset template adjustment rules, and then the MAE is calculated again to determine whether it is credible, and it is iterated again.

[0107] In order to meet the needs of large-scale personalized clothing customization, this solution also uses artificial intelligence algorithms to quickly and accurately identify human body shape information and pattern adjustment rules corresponding to body shape, and integrates body shape-related data into the automatic pattern making algorithm to better improve the accuracy and comfort of ready-made clothing sizes. Figure 3 As shown, the figure shows the process of the human body shape classification algorithm.

[0108] In some embodiments, the training process of the body shape classification algorithm model includes:

[0109] S401: Collecting body images of the front, back and side of a human body through multiple channels, and cropping and filtering the collected images. The multiple channel collection may include website crawling, ImageNet data set and self-owned data sorting.

[0110] S402: Classify and label the images according to pre-set body shape features.

[0111] S403: randomly divide the annotated images into a training set and a test set according to a predetermined ratio. It is necessary to use an image annotation tool, and professionals manually annotate the cropped and filtered images to obtain a binary image of human body shape category features. As shown in Figure 4, Figures 4 (a) to (c) respectively show schematic diagrams of a normal shoulder, a low right shoulder, and a low left shoulder.

[0112] S404: Input the images of the training set into a computer vision AI algorithm model based on deep learning, train and fit the body shape features using the stochastic gradient descent method, and generate a human body shape classification algorithm model. The computer vision AI algorithm model refers to all relevant algorithm models involved in target detection, target segmentation, image classification, and other directions in the field of machine learning and deep learning.

[0113] S405: Verify the algorithm model through the verification set.

[0114] The solution of this application utilizes machine learning and deep learning AI algorithms in computer vision to quantify the previously abstract human judgment of body shape into an explainable algorithm model, solving the problem that traditional body measurement cannot quickly and accurately judge the customer's body shape, resulting in clothing customization being unable to meet customer needs in a timely and efficient manner.

[0115] The present invention proposes classification standards for normal body and special body and corresponding processing rules. In some embodiments, the pre-set body shape features include normal body and special body.

[0116] The special body types of human frontal images include: sloping shoulders, flat shoulders, low right shoulder and low left shoulder. The special body types of human side images include: convex belly, hunchback, hunched back, straight chest, straight chest and supine body, bent waist, round back and flat back. The special body types of human buttocks images include: straight buttocks, flat buttocks, high hips, low front waist and wide / narrow between legs. Each human body corresponds to a combination of body shape features of the three parts of the front, side and buttocks.

[0117] Examples of special body shapes and samples are shown in Figure 5. The example diagram in Figure 5 provides a directional method from the fundamental approach and modification ideas. According to different styles, processes and customer requirements, the parameters can be appropriately changed, the proportions can be adjusted, and additions or deletions can be made conveniently. Moreover, Figure 5 also includes all special body shapes of the upper body and the lower body. Figures 5 (a) to (b) respectively show the body shapes and sample examples of two body types: a sloping shoulder, a hunched back, and a flat shoulder and a convex belly. It should be noted that there are other body types that are not listed.

[0118] The modification range, parameter examples and sample examples of normal body types are shown in the following table, where the X movement and the Y movement refer to the scaling ratio along the X axis and the Y axis. The adjustment range of parameters X and Y includes but is not limited to the range shown in the sample examples, and the pictures are only examples. It should be noted that other body types not listed have corresponding pre-set body modification rules.

[0119]

[0120] In some embodiments, the step of inputting the human body image data into a pre-trained body shape classification algorithm model for processing includes:

[0121] S411: Construct the convolutional neural network of the algorithm, initialize and load the pre-trained network parameters;

[0122] S412: input the preprocessed image into a convolutional neural network to extract features and obtain a feature map;

[0123] S413: Refining the extracted feature region through the RPN network to obtain a refined feature region;

[0124] S414: performing RoIAlign operation by bilinear interpolation method to process the feature map so that each feature region generates a feature map of fixed size;

[0125] S415: Obtain target category information through the classification branch, obtain target frame coordinate information through the regression branch, and obtain mask information through the mask network;

[0126] S416: Apply the category information, coordinate information and mask information to the feature map to obtain instance segmentation of the human body shape category.

[0127] Reference Figure 7 , we will take Mask R-CNN as an example. Mask R-CNN is a deep convolutional neural network with a network architecture like Figure 7 As shown in the figure, it is mainly divided into several stages: convolutional neural network (backbone network), candidate region generation network RPN (region proposal network), candidate region classification network (classification), regression network (boxregression) and mask network (mask branch) of target instance; the last two parts of the network are parallel architectures, that is, target classification and target segmentation of candidate regions are performed simultaneously.

[0128] The above-mentioned shared convolutional layer adopts the ResNet101-FPN architecture (residual network structure) to extract features from the entire image and send the obtained feature maps to the RPN. The candidate region generation network (RPN) uses a sliding window to generate a target box (anchor) with a preset aspect ratio in the shared feature map, that is, the specified position of the algorithm's region of interest ROI (Region of Interest); the candidate region classification network is composed of a fully connected neural network, including two branches, one for target box classification and the other for target box regression; finally, the target instance mask generation network is also composed of a fully convolutional neural network, which is used to generate a binary image mask for each local feature instance of the human body.

[0129] The ResNet101 in the convolutional neural network ResNet101-FPN architecture is a deep residual network, and its specific structure includes five stages of sub-networks. Among them, Stage 1 is the input network composed of CNN, which performs convolution and pooling operations on the input image; Stage 2-Stage 5 are composed of two types of residual blocks respectively. One of the specific structures is called Identity Block; the other structure is called Conv Block.

[0130] The FPN (Feature pyramid networks) in the ResNet101-FPN architecture is a feature pyramid algorithm that independently predicts features at different levels, and then integrates the high-resolution information of low-level features and the high-semantic information of high-level features to improve the model's ability to predict object location information and improve the overall performance of model detection and classification.

[0131] The RPN network obtains region proposals and region scores, and uses the non-maximum suppression (NMS) algorithm to extract overlapping proposals from the region scores, and outputs the top-300 scoring regions to the RoIAlign layer. The RoIAlign layer then uses bilinear interpolation to more accurately find the features corresponding to each region.

[0132] By adopting the above-mentioned human body shape recognition model, the present invention can realize the simple, fast and accurate recognition of human body shape without complicated labeling or manual calculation work. It only needs to use the customer's photo to quickly and conveniently identify the body shape of the customer, and provide it to the clothing customization system to improve the production capacity and efficiency of the clothing customization industry.

[0133] According to a second aspect of an embodiment of the present application, there is provided an automatic pattern making system for personalized clothing customization, comprising:

[0134] The order receiving and verification unit is used to receive customer demand orders or production orders provided through online and offline channels. At the same time, the order content must be filled in according to the specified format, and the obtained order data must be deduplicated, processed and standardized, and prompt information must be returned to prompt the platform whether the order data is complete.

[0135] The net body data acquisition unit is used to acquire the net body size data and demand data of the user.

[0136] The data preprocessing unit is used to preprocess the feature data and customization requirements extracted from the clothing orders.

[0137] The pattern-making algorithm unit is used to input the net body size data and demand data into the pre-trained automatic pattern-making algorithm model for calculation, and output the garment size data, change items and initial pattern adjustment rules.

[0138] The reinforcement learning unit is used to add the credible results generated by the algorithm and their corresponding net body sizes to the model training set, and continuously iterate the training through the reinforcement learning algorithm to improve the model regression fitting accuracy and performance.

[0139] The image data acquisition unit is used to acquire the user's human body image data.

[0140] The body shape classification unit is used to input human image data into a pre-trained body shape classification algorithm model for processing and output a body shape category label.

[0141] The data correction unit is used to determine the corresponding body shape modification rules according to the body shape category labels, and output the final shape adjustment rules after combining the body shape modification rules with the initial shape adjustment rules.

[0142] The file output unit is used to output the garment size data, change items and final pattern adjustment rules into a file of a specified format.

[0143] Reference Figure 8 The workflow of the automatic pattern making system of the present application is: first, the clothing order is received, that is, the automatic pattern making system receives the customer's clothing order from different channels such as online and offline through the clothing order acquisition interface, and verifies the order format, that is, it performs operations such as deduplication of data, deletion of blank lines and return of prompt information to check and prompt whether the order data is valid and available; then, the automatic pattern making system extracts data from the acquired order, that is, the automatic pattern making system extracts clothing type, customer's body net size feature data and customer preference requirements from the acquired order data; then, the automatic pattern making system pre-processes the extracted data, that is, it performs operations such as standardization, normalization and data transformation on the order data to make it meet the input requirements of the algorithm model; then, the automatic pattern making system fits the pre-processed data through the personalized clothing customization automatic pattern making algorithm, and outputs the customer's ready-made clothing size data and change items and pattern adjustment rules; finally, the automatic pattern making system generates a ready-made clothing size data file corresponding to the order according to the requirements of the subsequent plate making software, and ends this automatic pattern making work.

[0144] It can be seen that the solution of the present application has brought about a significant improvement in the efficiency of clothing customization, solving the problem that the previous method could only rely on pattern makers to manually make patterns and provide change experience, but the rework rate and accuracy were still low. It truly proposes to rely on artificial intelligence methods to achieve extremely accurate body classification of individuals with different body shapes and corresponding pattern adjustment rules that suit each person.

[0145] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0146] It should be noted that, in the description of this application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0147] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0148] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0149] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0150] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0151] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0152] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0153] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. An automatic pattern making method for personalized clothing customization, characterized in that: include: Obtain the user's net body size data and demand data; Input the net body size data and demand data into the pre-trained automatic pattern making algorithm model for calculation, and output the garment size data, change items and initial pattern adjustment rules; Acquire user's human body image data; Input the human body image data into the pre-trained body shape classification algorithm model for processing, and output the body shape category label; Determine the corresponding body shape modification rule according to the body shape category label, combine the body shape modification rule with the initial pattern adjustment rule, and output the final pattern adjustment rule; Output the garment size data, change items and final pattern adjustment rules into files of specified format; The training process of the body shape classification algorithm model includes: Collecting body images of the front, back and side of the human body through multiple channels, and cropping and screening the collected images; Classify and label the images according to pre-set body shape features; The annotated images are randomly divided into training and test sets according to a predetermined ratio; The images of the training set are input into the computer vision AI algorithm model based on deep learning, and the body shape features are trained and fitted using the stochastic gradient descent method to generate a human body shape classification algorithm model; Verify the algorithm model through the validation set; The multi-channel collection includes: website crawling, ImageNet data set, and self-owned data collation.

2. The method according to claim 1, characterized in that: The net body size data is attribute data of multiple dimensions commonly used in the clothing customization industry, including at least: length, sleeve length, shoulder width, chest circumference, trouser length, waist circumference, and hip circumference; The demand data includes: first demand data associated with the size of the ready-made garment, and second demand data associated with the change item; wherein the first demand data at least includes slim fit, loose fit, fitted, slightly longer sleeves, slightly shorter sleeves, cropped trousers, slightly wider trouser legs, slightly thinner trouser legs, slightly narrow shoulders, slightly wider shoulders, slightly longer length, slightly shorter length; the second demand data at least includes rolled-up trouser legs, slightly wider sleeves, and slightly thin sleeves; The ready-made garment size data includes ready-made garment attributes of different dimensions required for garment customization.

3. The method according to claim 2, characterized in that The plate adjustment rules are data for changing and adjusting the plate-making CAD according to the demand data.

4. The method according to claim 1, characterized in that: The training process of the automatic pattern making algorithm model includes: Traverse the garment attributes of different parts respectively, and use the correlation coefficient method to calculate the N net body size data most relevant to the garment attributes of each part; The net body size data after feature selection is divided into training data set and test data set by random sampling; Through the K-Fold cross-training method, the training data set is evenly divided into several parts, and cross-training and verification are continuously performed; Use machine learning regression algorithm to train the model, set the model parameters, select the parameters with the best model effect and save them; Use the saved optimal model to test the net body size data in the test data set, output the garment size data, and calculate the mean absolute error based on the output garment size data; The saved optimal model is iteratively optimized using reinforcement learning method.

5. The method according to claim 4, characterized in that The setting of model parameters includes: automatically searching for parameters and / or manually setting parameters; The method of selecting the parameters with the best model effect includes: using a loss function evaluation method of a regression model to determine the parameters with the best effect; wherein the loss function includes a square root error and / or a mean absolute error.

6. The method according to claim 4, characterized in that The method of iteratively optimizing the saved optimal model by using the reinforcement learning method includes: Determine whether the calculated MAE result is credible; If it is credible, add the net body size data in the corresponding test data set and the corresponding garment size data output by the model to the training set, and iterate the training model; If it is not credible, the template is manually adjusted and modified according to the preset template adjustment rules, and then the MAE is calculated again to determine whether it is credible and iterate again.

7. The method according to claim 1, characterized in that The step of inputting the human body image data into a pre-trained body shape classification algorithm model for processing includes: Construct the algorithm's convolutional neural network, initialize and load the pre-trained network parameters; The preprocessed image is input into the convolutional neural network for feature extraction to obtain a feature map; The extracted feature region is refined through the RPN network to obtain a refined feature region; The RoIAlign operation is performed by bilinear interpolation to process the feature map so that each feature area generates a feature map of a fixed size; The classification branch obtains the category information of the target, the regression branch obtains the coordinate information of the target box, and the mask network obtains the mask information; The category information, coordinate information, and mask information are applied to the feature map to obtain instance segmentation of human body shape categories and human body shape category labels.

8. The method according to claim 1, characterized in that The preset body shape characteristics include normal body and special body; The special body types of human frontal images include: sloping shoulders, flat shoulders, low right shoulder and low left shoulder; The special body types of human side images include: convex belly, hunchback, hunched back, erect chest, erect chest, bent back, round back and flat back; The special types of human buttocks images include: perky buttocks, flat buttocks, high hips, low front waist, and wide / narrow between legs; Each human body corresponds to a combination of body shape features in three parts: the front, side, and buttocks.

9. An automatic pattern making system for personalized clothing customization, characterized in that: include: A net body data acquisition unit, used to acquire the net body size data and demand data of the user; The pattern making algorithm unit is used to input the net body size data and the demand data into the pre-trained automatic pattern making algorithm model for calculation, and output the garment size data, the change items and the initial pattern adjustment rules; An image data acquisition unit, used to acquire the user's human body image data; A body shape classification unit is used to input human image data into a pre-trained body shape classification algorithm model for processing and output a body shape category label; A data correction unit, used to determine a corresponding body shape modification rule according to the body shape category label, and output a final body shape adjustment rule after combining the body shape modification rule with the initial shape adjustment rule; A file output unit, used to output the garment size data, change items and final pattern adjustment rules into a file of a specified format; The training process of the body shape classification algorithm model includes: Collecting body images of the front, back and side of the human body through multiple channels, and cropping and screening the collected images; Classify and label the images according to pre-set body shape features; The annotated images are randomly divided into training and test sets according to a predetermined ratio; The images of the training set are input into the computer vision AI algorithm model based on deep learning, and the body shape features are trained and fitted using the stochastic gradient descent method to generate a human body shape classification algorithm model; Verify the algorithm model through the validation set; The multi-channel collection includes: website crawling, ImageNet data set, and self-owned data collation.

Images