Intelligent sheet stitching and arranging method and system based on multi-task learning

Abstract

The invention provides an intelligent sheet stitching and arranging method and system based on multi-task learning. The method comprises the steps that a sheet file about clothes is acquired; analyzing the original geometric data of each sheet; inputting the original geometric data into a pre-trained evaluation model, and analyzing semantic information and stitching data of each sheet on the same garment; and in modeling software, intelligent stitching between the sheets is carried out based on stitching data, and the sheets are placed at corresponding positions of the virtual model based on semantic information to realize intelligent arrangement of the sheets. According to the scheme, the neural network model is trained based on a large amount of sheet data including stitching data and structural features, effective information on the sheets is fully utilized, and a reference basis with higher quality is provided for intelligent stitching and automatic arrangement of the clothing sheets; the sheet sewing efficiency, the sewing quality and the automation degree of sheet arrangement are improved, and then rapid modeling of a three-dimensional garment model is achieved.

Description

Technical Field

[0001] The present invention relates to the field of clothing, and in particular to a method and system for intelligently stitching and arranging panels based on multi-task learning. Background Art

[0002] A 3D model is a polygonal representation of an object, usually displayed by a computer or other video device. Anything that exists in the physical world can be represented by a 3D model. Clothing modeling refers to the use of computer technology to create clothing through 3D modeling software to simulate the appearance and structure of real clothing. Clothing modeling can use different methods and techniques, such as geometry-based modeling, physics-based modeling, and style-based modeling.

[0003] Clothing modeling can be used in many fields, such as clothing design, virtual fitting, etc. Through clothing modeling, you can create realistic clothing to better display and evaluate the design and effect of clothing. At the same time, clothing modeling can also be used to simulate and predict the behavior and performance of clothing to help designers better optimize and improve clothing design.

[0004] In the process of garment modeling, it is inevitable to connect different garment panels together to achieve a similar effect of "stitching" physical clothing. This technology is called garment panel stitching. Garment panel stitching is an important part of garment design, especially for digital garment modeling, which links 2D garment design to 3D garment modeling, and connects sample design to garment digitization.

[0005] Among the existing pattern processing technologies, manual stitching and arrangement are still the commonly used methods in clothing modeling software. However, this operation method is undoubtedly cumbersome, complicated and inefficient, and its quality is also extremely dependent on personal experience. Although some software provides automatic stitching and arrangement functions, it does not significantly improve the efficiency and quality of stitching. This means that in the process of clothing modeling, a lot of time and effort are needed to stitch and arrange the patterns to achieve the desired effect of clothing modeling. Summary of the invention

[0006] In view of this, the present invention proposes a method and system for intelligent stitching and arrangement of panels based on multi-task learning. The specific scheme is as follows:

[0007] A method for intelligent stitching and arrangement of panels based on multi-task learning, comprising:

[0008] Get the pattern files of the clothing;

[0009] Parsing out the original geometric data of each plate in the plate file;

[0010] Inputting the original geometric data into a pre-trained evaluation model to analyze the semantic information and stitching data of each panel on the same garment, wherein the evaluation model is a neural network model based on multi-task learning;

[0011] In the modeling software, the intelligent stitching between the patterns is performed based on the stitching data, and the patterns are placed at the corresponding positions of the virtual model based on the semantic information to realize the intelligent arrangement of the patterns;

[0012] The semantic information relates to the position of the plate relative to the virtual model in the software, and the stitching data relates to the stitching position and stitching method on the plate and the stitching relationship between the plates.

[0013] In a specific embodiment, the training of the evaluation model includes:

[0014] Obtain a large amount of pattern data containing semantic information and stitching data, and ensure that the pattern data covers patterns of different styles, sizes and craftsmanship;

[0015] Preprocessing the plate data and dividing it into a training set and a test set;

[0016] In each round of training iteration, the plate data corresponding to the training set is input into the neural network model for training, and the error between the output of the neural network branch on the stitching data and the stitching data in the training set, and the error between the output of the neural network branch on the semantic information and the semantic information in the training set are calculated respectively, and then the errors of the two neural network branches are combined into the total error of the neural network model; based on the total error, the parameters of the neural network model are adjusted by the back propagation algorithm to minimize the loss function;

[0017] The neural network model that meets the preset iteration stopping conditions is used as the evaluation model.

[0018] In a specific embodiment, an evaluation criterion is constructed based on the plate data corresponding to the test set;

[0019] The prediction quality of the evaluation model for stitching data and the recognition accuracy of semantic information are evaluated based on the evaluation criteria.

[0020] In a specific embodiment, the iteration stopping condition includes reaching a maximum number of iterations or convergence of a loss function.

[0021] In a specific embodiment, the preprocessing includes data cleaning, standardization, normalization, and missing value processing.

[0022] In a specific embodiment, additional information of the plate is obtained, wherein the additional information relates to the process and customization requirements of the plate;

[0023] The evaluation model acquires semantic information with the assistance of the additional information.

[0024] In a specific embodiment, points and lines on each plate are determined, geometric relationships including the distance between adjacent points, line length and arc length are analyzed, and annotations are performed to obtain the original geometric data.

[0025] A multi-task learning-based intelligent panel stitching and arrangement system, comprising:

[0026] A pattern acquisition unit, used to acquire pattern files related to clothing;

[0027] A parsing unit, used for parsing the original geometric data of each plate in the plate file;

[0028] An evaluation unit, used for inputting the original geometric data into a pre-trained evaluation model to analyze the semantic information and stitching data of each panel on the same garment, wherein the evaluation model is a neural network model based on multi-task learning;

[0029] The output unit, in the modeling software, performs intelligent stitching between patterns based on stitching data, and places the patterns at the corresponding positions of the virtual model based on semantic information to achieve intelligent arrangement of the patterns;

[0030] The semantic information relates to the position of the plate relative to the virtual model in the software, and the stitching data relates to the stitching position and stitching method on the plate and the stitching relationship between the plates.

[0031] In a specific embodiment, the training of the evaluation model includes:

[0032] Obtain a large amount of pattern data containing semantic information and stitching data, and ensure that the pattern data covers patterns of different styles, sizes and craftsmanship;

[0033] Preprocessing the plate data and dividing it into a training set and a test set;

[0034] In each round of training iteration, the plate data corresponding to the training set is input into the neural network model for training, and the error between the output of the neural network branch on the stitching data and the stitching data in the training set, and the error between the output of the neural network branch on the semantic information and the semantic information in the training set are calculated respectively, and then the errors of the two neural network branches are combined into the total error of the neural network model; based on the error, the parameters of the neural network model are adjusted by the back propagation algorithm to minimize the loss function;

[0035] The neural network model that meets the preset iteration stopping conditions is used as the evaluation model.

[0036] In a specific embodiment, an evaluation criterion is constructed based on the plate data corresponding to the test set; and the prediction quality of the evaluation model for the stitching data and the recognition accuracy of the semantic information are evaluated based on the evaluation criterion.

[0037] Beneficial effects:

[0038] The present invention provides a method and system for intelligent stitching and arrangement of panels based on multi-task learning, which trains a neural network model based on a large amount of panel data including stitching data and structured features, fully utilizes the effective information on the panels, and provides a higher quality reference basis for the intelligent stitching and automatic arrangement of clothing panels, thereby improving the panel stitching efficiency, stitching quality and the degree of automation of panel arrangement, thereby realizing the rapid modeling of three-dimensional clothing models. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flow chart of a method for intelligently stitching and arranging panels according to an embodiment of the present invention;

[0040] Figure 2 is a three-part schematic diagram of an embodiment of the present invention;

[0041] Figure 3 is a schematic diagram of model input and output of an embodiment of the present invention;

[0042] Figure 4 2 is a schematic diagram of a panel intelligent stitching and arrangement system module according to an embodiment of the present invention.

[0043] Reference numerals: 1 - sheet acquisition unit; 2 - analysis unit; 3 - evaluation unit; 4 - output unit. DETAILED DESCRIPTION

[0044] In the following, various embodiments of the present disclosure will be described more fully. The present disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of the present disclosure to the specific embodiments disclosed herein, but the present disclosure should be understood to cover all adjustments, equivalents and / or alternatives that fall within the spirit and scope of the various embodiments of the present disclosure.

[0045] Example 1

[0046] Embodiment 1 of the present invention discloses a method for intelligent stitching and arrangement of panels based on multi-task learning, which uses a neural network algorithm to learn the stitching relationship between clothing panels, improves the efficiency and quality of panel stitching and the degree of automation of panel arrangement, and thus realizes the rapid modeling of three-dimensional clothing models. The specific process is as shown in the attached manual. Figure 1 The specific plan is as follows:

[0047] A method for intelligent stitching and arrangement of panels based on multi-task learning comprises the following steps:

[0048] 101. Get the pattern file about clothing;

[0049] 102. Analyze the original geometric data of each plate in the plate file;

[0050] 103. The original geometric data is input into a pre-trained evaluation model to analyze the semantic information and stitching data of each panel on the same garment. The evaluation model is a neural network model based on multi-task learning. The semantic information involves the position of the panel relative to the virtual model in the software, and the stitching data involves the stitching position and stitching method on the panel as well as the stitching relationship between the panels.

[0051] 104. In the modeling software, intelligent stitching between panels is performed based on stitching data, and panels are placed at corresponding positions on the virtual model based on semantic information to achieve intelligent arrangement of panels.

[0052] In order to make full use of the effective information of the panels, this embodiment proposes a multi-task based neural network learning technology solution. By learning a large amount of stitching data and panel structure features, the semantic information of the panels and the optimal stitching relationship between the panels can be obtained.

[0053] The pattern file of this embodiment includes DXF and other formats. By analyzing the pattern file, the geometric parameters on the pattern can be known, such as clothing length, shape and other data. Specifically, the points and lines on each pattern are determined, the geometric relationship including the spacing between adjacent points, line length and arc length is analyzed, and the original geometric data is obtained by marking.

[0054] The solution of this embodiment mainly includes three parts: the first is the data level, the second is the model level, and the third is the deployment and application of the model. Figure 2 shown.

[0055] The data level mainly refers to the collection and processing of preliminary training data. The evaluation model of this embodiment requires a large amount of plate data as training data for training. The plate data contains rich structured information, such as plate information, plate structural characteristics, and stitching relationship data between plates. These data should cover as many plates of different styles, sizes, and processes as possible to meet the universal and customized requirements of plate stitching.

[0056] The network model used in this embodiment is a multi-task based neural network model, which optimizes and learns the semantic and other structural information of each plate and the stitching relationship between plates to improve the efficiency and quality of plate stitching. The model level is divided into a model training phase and a model testing phase.

[0057] The training of the evaluation model includes: obtaining a large amount of plate data containing semantic information and stitching data, and ensuring that the plate data covers plates of different styles, sizes and processes; preprocessing the plate data and dividing it into training sets and test sets; in each round of training iteration, inputting the plate data corresponding to the training set into the neural network model for training, respectively calculating the error between the output of the neural network branch about the stitching data and the stitching data in the training set, and the error between the output of the neural network branch about the semantic information and the semantic information in the training set, and then merging the errors of the two neural network branches into the total error of the neural network model; adjusting the parameters of the neural network model through the back propagation algorithm based on the total error to minimize the loss function; and using the neural network model that meets the preset iteration stop condition as the evaluation model. Preferably, the iteration stop condition includes reaching the maximum number of iterations or the convergence of the loss function.

[0058] In the testing phase of the model, an evaluation standard is constructed based on the plate data corresponding to the test set; the prediction quality of the evaluation model for stitched data and the recognition accuracy of semantic information are evaluated based on the evaluation standard. The output quality of the model is fed back through the evaluation standard to obtain a network model that meets expectations.

[0059] Before training, the structured plate data needs to be preprocessed to improve data quality and accuracy, so as to better perform data analysis and machine learning tasks. Preprocessing includes data cleaning, standardization, normalization, and missing value processing. Data cleaning refers to removing or repairing data sets that contain noise, abnormalities, duplications, or useless data to improve data quality. Cleaning steps may include data screening, outlier removal, noise removal, and data transformation. Standardization is to scale or normalize the data in a data set so that the data has the same scale or range. This can help solve the problem of size differences between different features, thereby improving the performance and accuracy of machine learning algorithms. Normalization is to map the data in a data set to a specified range, usually [0,1] or [-1,1]. Normalization can help solve the problem of size differences between features so that each feature is on the same scale, thereby improving the performance and accuracy of machine learning algorithms.

[0060] After the model training is completed, the model can be deployed and applied, and the board data can be input into the trained neural network model for intelligent analysis. The input and output of the model are shown in the attached figure. Figure 3As shown. Among them, the obtained plate semantic information can be effectively used for the automatic arrangement of plates, and the stitching relationship between plates is used for intelligent plate stitching connection. Among them, the semantic information of plates mainly refers to the category information of plates, such as front plate, back plate, sleeves, collar, etc. Based on this, the position arrangement relative to the clothing model can be constructed in the software. The arrangement in the modeling software refers to placing the plates in the corresponding positions of the virtual model so that simulation can be performed, and the semantic information provides this position information. The stitching data involves relevant information about stitching, including where to stitch the plates, the length of the stitching, the stitching method, and which plates to stitch with. Among them, the most important thing in the stitching data is the stitching relationship between plates, which involves the linkage with the semantic information.

[0061] The training data will also involve information such as the process and material of the plate, and these additional information will also affect the output of the model to a certain extent. For example, the process of the plate can assist the model in predicting semantic information to a certain extent. For another example, there are special customization requirements, which will be different from the normal plate stitching habits. In order to retain the individuality of the model, this embodiment deliberately incorporates the customization requirements into the additional information. Preferably, additional information of the plate is obtained, and the additional information involves the process and customization requirements of the plate; the evaluation model acquires semantic information with the assistance of the additional information. This embodiment incorporates the process and customization requirements of the plate into the additional information to assist the model in making predictions.

[0062] This embodiment proposes a method for intelligent stitching and arrangement of panels based on multi-task learning. A neural network model is trained based on a large amount of panel data including stitching data and structured features, and effective information on the panels is fully utilized to provide a higher quality reference basis for the intelligent stitching and automatic arrangement of clothing panels, thereby improving the efficiency and quality of panel stitching and the degree of automation of panel arrangement, thereby realizing rapid modeling of three-dimensional clothing models.

[0063] Example 2

[0064] Embodiment 2 of the present invention discloses a panel intelligent stitching and arrangement system based on multi-task learning, which systematizes the panel intelligent stitching and arrangement method based on multi-task learning in embodiment 1. The overall structure diagram of the panel intelligent stitching and arrangement system is shown in the attached manual. Figure 4 The specific plan is as follows:

[0065] A multi-task learning-based intelligent panel stitching and arrangement system, comprising:

[0066] The pattern acquisition unit 1 is used to acquire the pattern file of the clothing;

[0067] Parsing unit 2, used for parsing the original geometric data of each plate in the plate file;

[0068] Evaluation unit 3, used to input the original geometric data into the pre-trained evaluation model to analyze the semantic information and stitching data of each panel on the same garment. The evaluation model is a neural network model based on multi-task learning;

[0069] Output unit 4, in the modeling software, performs intelligent stitching between the panels based on the stitching data, and places the panels at corresponding positions of the virtual model based on the semantic information to realize intelligent arrangement of the panels;

[0070] Among them, the semantic information involves the position of the pattern relative to the virtual model in the software, and the stitching data involves the stitching position and stitching method on the pattern and the stitching relationship between the patterns.

[0071] Among them, the training of the evaluation model includes:

[0072] A large amount of plate data containing semantic information and stitching data is obtained, and it is ensured that the plate data covers plates of different styles, sizes and craftsmanship; the plate data is preprocessed and divided into training sets and test sets; in each round of training iteration, the plate data corresponding to the training set is input into the neural network model for training, and the error between the output of the neural network branch about stitching data and the stitching data in the training set, and the error between the output of the neural network branch about semantic information and the semantic information in the training set are calculated respectively, and then the errors of the two neural network branches are combined into the total error of the neural network model; based on the error, the parameters of the neural network model are adjusted through the back propagation algorithm to minimize the loss function; the neural network model that meets the preset iteration stop conditions is used as the evaluation model.

[0073] An evaluation criterion is constructed based on the plate data corresponding to the test set. The prediction quality of the evaluation model for stitching data and the recognition accuracy of semantic information are evaluated based on the evaluation criterion.

[0074] This embodiment proposes a panel intelligent stitching and arrangement system based on multi-task learning, which systematizes the intelligent stitching method of Embodiment 1 to make it more practical.

[0075] The present invention provides a method and system for intelligent stitching and arrangement of panels based on multi-task learning, which trains a neural network model based on a large amount of panel data including stitching data and structured features, fully utilizes the effective information on the panels, and provides a higher quality reference basis for the intelligent stitching and automatic arrangement of clothing panels, thereby improving the panel stitching efficiency, stitching quality and the degree of automation of panel arrangement, thereby realizing the rapid modeling of three-dimensional clothing models.

[0076] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required for implementing the present invention. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario may be distributed in the devices in the implementation scenario according to the description of the implementation scenario, or may be changed accordingly and located in one or more devices different from the present implementation scenario.

Claims

1. A method for intelligent stitching and arrangement of panels based on multi-task learning, characterized in that: include: Get the pattern files of the clothing; Parsing out the original geometric data of each plate in the plate file; Inputting the original geometric data into a pre-trained evaluation model to analyze the semantic information and stitching data of each panel on the same garment, wherein the evaluation model is a neural network model based on multi-task learning; In the modeling software, the intelligent stitching between the patterns is performed based on the stitching data, and the patterns are placed at the corresponding positions of the virtual model based on the semantic information to realize the intelligent arrangement of the patterns; The semantic information relates to the position of the plate relative to the virtual model in the software, and the stitching data relates to the stitching position and stitching method on the plate and the stitching relationship between the plates.

2. The panel intelligent stitching and arrangement method according to claim 1, characterized in that: The training of the evaluation model includes: Obtain a large amount of pattern data containing semantic information and stitching data, and ensure that the pattern data covers patterns of different styles, sizes and craftsmanship; Preprocessing the plate data and dividing it into a training set and a test set; In each round of training iteration, the plate data corresponding to the training set is input into the neural network model for training, and the error between the output of the neural network branch on the stitching data and the stitching data in the training set, and the error between the output of the neural network branch on the semantic information and the semantic information in the training set are calculated respectively, and then the errors of the two neural network branches are combined into the total error of the neural network model; based on the total error, the parameters of the neural network model are adjusted by the back propagation algorithm to minimize the loss function; The neural network model that meets the preset iteration stopping conditions is used as the evaluation model.

3. The panel intelligent stitching and arrangement method according to claim 2, characterized in that: Establishing an evaluation standard based on the plate data corresponding to the test set; The prediction quality of the evaluation model for stitching data and the recognition accuracy of semantic information are evaluated based on the evaluation criteria.

4. The panel intelligent stitching and arrangement method according to claim 2, characterized in that: The iteration stopping condition includes reaching the maximum number of iterations or the loss function converging.

5. The panel intelligent stitching and arrangement method according to claim 2, characterized in that: The preprocessing includes data cleaning, standardization, normalization, and missing value processing.

6. The panel intelligent stitching and arrangement method according to claim 2, characterized in that: Obtaining additional information of the plate, wherein the additional information relates to the process and customization requirements of the plate; The evaluation model acquires semantic information with the assistance of the additional information.

7. The panel intelligent stitching and arrangement method according to claim 1, characterized in that: The points and lines on each plate are determined, the geometric relationships including the spacing between adjacent points, line length and arc length are analyzed, and they are marked to obtain the original geometric data.

8. A multi-task learning-based intelligent panel stitching and arrangement system, characterized in that: include: A pattern acquisition unit, used to acquire pattern files related to clothing; A parsing unit, used for parsing the original geometric data of each plate in the plate file; An evaluation unit, used for inputting the original geometric data into a pre-trained evaluation model to analyze the semantic information and stitching data of each panel on the same garment, wherein the evaluation model is a neural network model based on multi-task learning; The output unit, in the modeling software, performs intelligent stitching between patterns based on stitching data, and places the patterns at the corresponding positions of the virtual model based on semantic information to achieve intelligent arrangement of the patterns; The semantic information relates to the position of the plate relative to the virtual model in the software, and the stitching data relates to the stitching position and stitching method on the plate and the stitching relationship between the plates.

9. The panel intelligent stitching and arrangement system according to claim 8, characterized in that: The training of the evaluation model includes: Obtain a large amount of pattern data containing semantic information and stitching data, and ensure that the pattern data covers patterns of different styles, sizes and craftsmanship; Preprocessing the plate data and dividing it into a training set and a test set; In each round of training iteration, the plate data corresponding to the training set is input into the neural network model for training, and the error between the output of the neural network branch on the stitching data and the stitching data in the training set, and the error between the output of the neural network branch on the semantic information and the semantic information in the training set are calculated respectively, and then the errors of the two neural network branches are combined into the total error of the neural network model; based on the error, the parameters of the neural network model are adjusted by the back propagation algorithm to minimize the loss function; The neural network model that meets the preset iteration stopping conditions is used as the evaluation model.

10. The panel intelligent stitching and arrangement system according to claim 9, characterized in that: Establishing an evaluation standard based on the plate data corresponding to the test set; The prediction quality of the evaluation model for stitching data and the recognition accuracy of semantic information are evaluated based on the evaluation criteria.

Images