Garment pattern prediction method and system based on garment image and body dimensions

Abstract

The application provides a garment pattern prediction method and system based on a garment image and body size, which comprises the following steps: extracting a silhouette semantic description vector from an unstructured garment reference image by using a deep learning model; constructing a user body shape feature vector based on body size data of the user; fusing the silhouette semantic description vector and the user body shape feature vector and taking them as inputs of a pre-trained garment pattern prediction model to output a pattern parameter modification amount of a corresponding target garment of the garment reference image; wherein the garment pattern prediction model is pre-trained through a training set constructed based on historical garment reference images, historical body size data and corresponding historical pattern parameter modification amounts; a corresponding basic pattern template is retrieved and called from a pattern template library according to the category of the target garment, the basic pattern template is adjusted by using the pattern parameter modification amount, and structured garment pattern data which is adapted to the body shape of the user and reflects the silhouette style of the garment reference image is obtained.

Description

Technical Field

[0001] This invention relates to the field of garment pattern prediction technology, and in particular to a method and system for garment pattern prediction based on garment images and body dimensions. Background Technology

[0002] As the global apparel industry transforms towards personalization, flexibility, and mass customization, consumers' demands for diverse styles, better fits, and personalized aesthetics are growing. However, the traditional pattern design process, the core link between design concepts and finished garments, still heavily relies on the experience, intuition, and craftsmanship of senior pattern makers. This model is already struggling with small-batch, high-value-added customization; when facing the demands of large-scale personalized customization, its inherent inefficiencies, high costs, insufficient standardization, and over-reliance on expert experience have become key bottlenecks hindering the industry's development.

[0003] Currently, the industry has tried various technological approaches to overcome this bottleneck, but all of them have significant limitations: 1) Parametric Template Adjustment Method: This method uses a series of preset basic pattern templates and adjusts parameters such as bust, waist, and hip measurements to suit different user body types. The advantage of this method is its relatively standardized operation and ability to ensure a basic fit, making it particularly suitable for standardized products with a single style. The limitation is its inability to respond to unstructured visual aesthetic input. For example, when a user wants to customize clothing with a specific silhouette (e.g., oversized, fitted, and three-dimensional) based on an unstructured image such as a fashion photo or street style picture, this method cannot extract and quantify these abstract silhouette semantics from the image, nor can it translate them into fine-grained adjustment instructions for the template, thus failing to meet the user's personalized aesthetic needs.

[0004] 2) Virtual try-on and image deformation method: This method simulates the wearing effect by overlaying clothing images or models onto a user's portrait or virtual human body. The limitation of this method is that it only simulates at the visual level and does not generate manufacturable pattern data. The image deformation algorithm only geometrically distorts pixels, without addressing the internal structure and engineering geometry of the clothing. Therefore, the generated visual effect cannot be directly converted into a pattern file to guide cutting and sewing, and when there is a significant difference between the user's body shape and the clothing's preset body shape, distortion and deformation easily occur, losing their guiding significance for actual production.

[0005] 3) Image-based garment pattern recommendation method: This method retrieves visually similar ready-to-wear patterns or historical patterns from a database based on reference images uploaded by the user. The limitation of this method is that it is essentially still a retrieval method rather than a generation method. The retrieved patterns are not adapted to the user's specific body shape, and extensive manual modifications by garment pattern makers are still required during customization. This method fails to fundamentally reduce reliance on human experience and improve pattern generation efficiency.

[0006] 4) Deep Learning-Based Automatic Pattern Generation Method: This method applies the popular generative artificial intelligence (such as Generative Adversarial Networks, GANs) to garment pattern generation. The current limitation of this method is that it generally focuses on learning the visual or statistical features of the pattern, failing to effectively model the strict geometric constraints necessary for garment manufacturing (such as the matching of arc lengths at seams between fabric pieces, continuity of curvature at joints, and symmetry). This results in geometrically flawed generated patterns that cannot be directly used in production, requiring extensive and time-consuming manual corrections, severely restricting its practical application value.

[0007] In conclusion, existing methods for automatically recommending or generating garment patterns have failed to build an automated bridge between unstructured visual inspiration and structured pattern data, and still rely heavily on the personal experience of senior garment pattern makers.

[0008] Therefore, how to provide an automated method for generating clothing patterns that can systematically solve the problems existing in the above-mentioned technologies is a technical problem that urgently needs to be solved. Summary of the Invention

[0009] In view of this, embodiments of the present invention provide a method and system for predicting clothing patterns based on clothing images and body dimensions, in order to eliminate or improve one or more defects existing in the prior art.

[0010] One aspect of the present invention provides a method for predicting clothing patterns based on clothing images and body dimensions. The method includes the following steps: extracting a silhouette semantic description vector from an unstructured clothing reference image using a deep learning model; wherein the silhouette semantic description vector includes some or all of the following: overall silhouette classification probability distribution, key size proportion features, local contour quantification indicators, and style intensity scores; constructing a user body shape feature vector based on the user's body size data; fusing the silhouette semantic description vector and the user body shape feature vector and using them as input to a pre-trained clothing pattern prediction model, outputting the pattern parameter modification amount for the corresponding target clothing in the clothing reference image; wherein the clothing pattern prediction model is pre-trained using a training set constructed based on historical clothing reference images, historical body size data, and corresponding historical pattern parameter modification amounts; retrieving and calling the corresponding basic pattern template from a pattern template library according to the category of the target clothing, adjusting the basic pattern template using the pattern parameter modification amount, and obtaining structured clothing pattern data that adapts to the user's body shape and reflects the silhouette style of the clothing reference image.

[0011] In some embodiments of the present invention, the step of extracting silhouette semantic description vectors from unstructured clothing reference images using a deep learning model includes: identifying human body contour bounding boxes and corresponding clothing region masks in the clothing reference image; extracting clothing foreground regions from the clothing reference image based on human body contour bounding boxes and corresponding clothing region masks; inputting the clothing foreground regions into a pre-trained deep learning model and outputting structured silhouette semantic description vectors; wherein the pre-trained deep learning model is obtained through supervised pre-training using a dataset containing clothing reference images and their overall silhouette classification data, key size ratio data, local contour quantification indicators, and style intensity score annotations.

[0012] In some embodiments of the present invention, the method further includes the step of acquiring the user's body size data, specifically including: acquiring body size data manually input by the user; or, acquiring images or videos of the user's front and side views, and calculating the user's body size data using AI body measurement technology; or, acquiring three-dimensional point cloud data or mesh model of the user's surface through 3D human body scanning, and calculating the user's body size data based on the three-dimensional point cloud data or the mesh model.

[0013] In some embodiments of the present invention, the clothing pattern prediction model is a gradient boosting decision tree model, and the method further includes a step of pre-training the clothing pattern prediction model, specifically including: constructing a training set based on historical clothing reference images, historical body size data, and corresponding historical pattern parameter modification amounts; wherein, each training sample in the training set is a triple consisting of user body size data, clothing reference images, and their corresponding pattern parameter modification amount annotations; constructing a loss function based on the difference between the prediction output of the gradient boosting decision tree model and the pattern parameter modification amount annotations; and using the constructed training set to iteratively train the initially constructed gradient boosting decision tree model to obtain the trained clothing pattern prediction model.

[0014] In some embodiments of the present invention, the step of adjusting the basic pattern template using the pattern parameter modification amount to obtain structured garment pattern data that adapts to the user's body shape and reflects the silhouette style of the garment reference image specifically includes: calculating the new coordinates of each key control point included in the basic pattern template based on the pattern parameter modification amount according to predefined parameter-geometric mapping rules; regenerating the contour curves of each piece in the basic pattern template using a spline interpolation algorithm based on the new coordinates of each key control point; and organizing the contour curves of all the regenerated pieces into a standard two-dimensional geometric data form to obtain structured garment pattern data in the form of a 2D pattern diagram.

[0015] In some embodiments of the present invention, after obtaining structured garment pattern data in the form of a 2D pattern diagram, the method further includes: initially mapping each piece of fabric into a three-dimensional space based on the natural position and posture of each piece in the garment structure according to the natural position and posture of each piece in the structured garment pattern data in the form of a 2D pattern diagram; and stitching the corresponding edges of the pieces together in a three-dimensional space according to the stitching relationship between the pieces using a fast mesh construction algorithm to obtain structured garment pattern data in the form of a 3D garment model; wherein the 3D garment model is in the form of a mesh.

[0016] In some embodiments of the present invention, the structured garment pattern data includes a 2D pattern drawing and / or a 3D garment model; after obtaining the structured garment pattern data, the method further includes: presenting the 2D pattern drawing and / or the 3D garment model to the user using an interactive interface, and adjusting the structured garment pattern data based on fine-tuning instructions received from the user.

[0017] Corresponding to the above methods, the present invention also provides a clothing pattern prediction system based on clothing images and body dimensions, including a processor, a memory, and a computer program / instructions stored in the memory. The processor is used to execute the computer program / instructions, and when the computer program / instructions are executed, the system implements the steps of any of the methods described in the above embodiments.

[0018] In accordance with the above methods, the present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implements the steps of the method as described in any of the above embodiments.

[0019] Corresponding to the above methods, the present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the method as described in any of the above embodiments.

[0020] The clothing pattern prediction method and system based on clothing images and body dimensions proposed in this invention can reduce the over-reliance on the personal experience of clothing pattern makers, significantly improve the conversion efficiency and consistency from visual inspiration to structured (manufacturable) pattern data, and help improve the automated pattern making efficiency of personalized clothing customization.

[0021] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.

[0022] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. In the drawings: Figure 1 This is a flowchart of a clothing pattern prediction method based on clothing images and body dimensions in one embodiment of the present invention.

[0024] Figure 2 This is a flowchart of extracting the semantic description vector of the profile in one embodiment of the present invention.

[0025] Figure 3 This is a flowchart of obtaining structured clothing pattern data in one embodiment of the present invention.

[0026] Figure 4 This is a schematic diagram of the computer equipment included in the system. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0028] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0029] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0030] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0031] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0032] To overcome the problems existing in the prior art, this invention proposes a method and system for predicting clothing patterns based on clothing images and body dimensions. This scheme introduces a deep learning model to extract the silhouette semantic description from the clothing reference image, and introduces a machine learning model to predict the modification amount of pattern parameters based on the fusion of silhouette semantic description and user reminder features, thereby obtaining structured clothing pattern data that adapts to the user's body shape and reflects the silhouette style of the clothing reference image.

[0033] Figure 1 This is a flowchart of a clothing pattern prediction method based on clothing images and body dimensions according to an embodiment of the present invention. The method includes the following steps: Step S100: Extract silhouette semantic description vectors from unstructured clothing reference images using a deep learning model; wherein the silhouette semantic description vectors include some or all of the overall silhouette classification probability distribution, key size proportion features, local contour quantification indicators, and style intensity scores.

[0034] The clothing reference images can be uploaded by users and are unstructured. The technical objective of this solution is to build an automated conversion bridge between unstructured visual inspiration and structured pattern data. Optionally, the aforementioned deep learning model can combine human body topology reconstruction to extract fine-grained silhouette semantic description vectors from the clothing reference images.

[0035] Step S200: Construct a user body shape feature vector based on the user's body size data.

[0036] Optionally, in step S200, key body shape feature values ​​can be obtained based on the user's body size data. Standard body size data can be obtained by completing the body size data. The standard body size data and key body shape feature values ​​are then concatenated and combined in a predefined and fixed order to form a user body shape feature vector. This user body shape feature vector can be a one-dimensional numerical array, where each dimension corresponds to a specific body size or body shape feature value. The user body shape features are key indicators reflecting the user's body proportions, shape, and characteristics, including some or all of the circumference proportion features, length proportion features, and width proportion features.

[0037] Step S300: The silhouette semantic description vector and the user body shape feature vector are fused and used as input to the pre-trained clothing pattern prediction model, and the output is the modification amount of the pattern parameters of the corresponding target clothing in the clothing reference image; wherein, the clothing pattern prediction model is pre-trained by a training set constructed based on historical clothing reference images, historical body size data and corresponding historical pattern parameter modification amounts.

[0038] Optionally, the garment pattern prediction model can be a machine learning model or a deep learning model.

[0039] Step S400: Based on the category of the target garment, retrieve and call the corresponding basic pattern template from the pattern template library, adjust the basic pattern template using the pattern parameter modification amount, and obtain structured garment pattern data that adapts to the user's body shape and reflects the silhouette style of the garment reference image.

[0040] The garment pattern prediction method based on garment images and body dimensions proposed in this invention can reduce the over-reliance on the personal experience of garment pattern makers, significantly improve the conversion efficiency and consistency from visual inspiration to structured (manufacturable) pattern data, and help improve the automated pattern making efficiency of personalized garment customization.

[0041] Figure 2 This is a flowchart of extracting the semantic description vector of the profile in one embodiment of the present invention, as follows: Figure 2 As shown, in some embodiments of the present invention, the step of extracting silhouette semantic description vectors from unstructured clothing reference images using a deep learning model includes: Step S110: Identify the human body outline bounding box and the corresponding clothing area mask in the clothing reference image.

[0042] Step S120: Extract the foreground region of the clothing from the clothing reference image based on the human body contour bounding box and the corresponding clothing region mask.

[0043] Step S130: Input the foreground region of the clothing into a pre-trained deep learning model and output a structured silhouette semantic description vector; wherein, the pre-trained deep learning model is obtained by supervised pre-training using a dataset containing clothing reference images and their overall silhouette classification data, key size ratio data, local contour quantification indicators and style intensity rating labels.

[0044] Optionally, after receiving the clothing reference image, its format and resolution can be checked first to ensure that it meets the input requirements of the subsequent image processing module, and then its silhouette semantic description vector can be extracted.

[0045] By employing this embodiment of the invention, fine-grained and quantifiable silhouette semantics can be extracted from clothing images using a deep learning model. This helps to significantly improve the efficiency and consistency of the conversion from visual data to manufacturable pattern data in subsequent steps based on the extracted fine-grained and quantifiable silhouette semantics.

[0046] In some embodiments of the present invention, the method further includes the step of acquiring the user's body size data, specifically including: acquiring body size data manually input by the user; or, acquiring images or videos of the user's front and side views, and calculating the user's body size data using AI body measurement technology; or, acquiring three-dimensional point cloud data or mesh model of the user's surface through 3D human body scanning, and calculating the user's body size data based on the three-dimensional point cloud data or the mesh model.

[0047] By employing this embodiment of the invention, users' body size data can be automatically acquired in various ways, which helps subsequent steps to fuse the user's specific body size data with fine-grained semantic information.

[0048] In some embodiments of the present invention, the clothing pattern prediction model is a gradient boosting decision tree model, and the method further includes a step of pre-training the clothing pattern prediction model, specifically including: constructing a training set based on historical clothing reference images, historical body size data, and corresponding historical pattern parameter modification amounts; wherein, each training sample in the training set is a triple consisting of user body size data, clothing reference images, and their corresponding pattern parameter modification amount annotations; constructing a loss function based on the difference between the prediction output of the gradient boosting decision tree model and the pattern parameter modification amount annotations; and using the constructed training set to iteratively train the initially constructed gradient boosting decision tree model to obtain the trained clothing pattern prediction model.

[0049] Optionally, the training data for the pre-trained garment pattern prediction model is constructed using a hybrid virtual-real approach based on 3D physical simulation. Since real-aligned triplet data containing unstructured reference images, accurate body shapes, and final pattern parameters is extremely scarce in the industry, it can be generated in batches using 3D garment CAD software through reverse parameterization: automatically sampling parameter modifications and standardized body size data for various basic patterns, assigning fabric attributes, and then performing 3D physical simulation and image rendering to generate a massive virtual pairing dataset, thus solving the cold start problem in model training. The learning objective of the pre-trained garment pattern prediction model is to establish a non-linear mapping relationship between user body shape and garment style semantics to the optimal pattern parameter configuration, thereby encoding the garment pattern maker's experiential knowledge into the parameters of the garment pattern prediction model.

[0050] Each training sample is a complete triplet of data, which specifically includes: simulated or real user body size data; design requirement data (synthetic reference image rendered by 3D simulation model and its extracted contour semantic description vector); and a set of structured pattern parameter modifications that precisely correspond to the synthetic reference image.

[0051] Using this embodiment of the invention, a training set can be constructed using historical data, and the original gradient boosting decision tree model can be trained using the constructed training set to obtain a trained clothing pattern prediction model.

[0052] Figure 3 This is a flowchart of obtaining structured clothing pattern data in one embodiment of the present invention, such as... Figure 3 As shown, in some embodiments of the present invention, the step of adjusting the basic pattern template using the pattern parameter modification amount to obtain structured garment pattern data that adapts to the user's body shape and reflects the silhouette style of the garment reference image specifically includes: Step S410: Calculate the new coordinates of each key control point included in the basic pattern template based on the predefined parameter-geometric mapping rules and the pattern parameter modification amount.

[0053] Step S420: Based on the new coordinates of each key control point, the spline interpolation algorithm is used to regenerate the outline curves of each piece in the basic pattern template.

[0054] Step S430: Organize the outline curves of all the regenerated cut pieces into a standard two-dimensional geometric data form to obtain structured garment pattern data in the form of a 2D pattern diagram.

[0055] Using this embodiment of the invention, the basic pattern template can be adjusted according to the pattern parameter modification amount generated by the garment pattern prediction model, thereby constructing structured garment pattern data in the form of a 2D pattern diagram. Optionally, the structured garment pattern data includes a 2D pattern diagram and / or a 3D garment model.

[0056] In some embodiments of the present invention, after obtaining structured garment pattern data in the form of a 2D pattern diagram, the method further includes: mapping each piece of fabric into a three-dimensional space based on the natural position and posture of each piece in the garment structure according to the natural position and posture of each piece in the structured garment pattern data in the form of a 2D pattern diagram; and stitching the corresponding edges of the pieces together in the three-dimensional space according to the stitching relationship between the pieces using a fast mesh construction algorithm to obtain structured garment pattern data in the form of a 3D garment model; wherein the 3D garment model is in the form of a mesh.

[0057] Using this embodiment of the invention, a 3D clothing model in three-dimensional space can be further constructed based on the 2D pattern drawing, which helps to provide different visualization methods in 2D and 3D in subsequent steps.

[0058] In some embodiments of the present invention, the structured garment pattern data includes a 2D pattern drawing and / or a 3D garment model. After obtaining the structured garment pattern data, the method further includes: presenting the 2D pattern drawing and / or the 3D garment model to a user using an interactive interface, and adjusting the structured garment pattern data based on fine-tuning instructions received from the user.

[0059] This invention enables the visualization of structured garment pattern data to users through an interactive interface, while simultaneously allowing for real-time modifications based on user feedback and fine-tuning instructions. This facilitates garment pattern makers in viewing automated results and making real-time adjustments. The output structured pattern data file serves as the starting point for garment pattern makers to perform efficient fine-tuning, significantly reducing workload.

[0060] In a specific embodiment of the present invention, the clothing reference image can be preprocessed first to obtain the clothing foreground region. Then, a deep learning model is used to perform multi-layer convolution and pooling operations on the clothing foreground region (image) to extract deep convolution features. Based on the extracted deep convolution features, a structured silhouette semantic description vector can be extracted using a pre-trained deep learning model.

[0061] The unstructured clothing reference images can be user-uploaded images, such as street photos taken with a mobile phone or camera, fashion magazine images, product catalog images, or social media images. The pre-trained deep learning model used to extract structured silhouette semantic description vectors is a model built with a high-performance convolutional neural network backbone, and its specific implementation can adopt ResNet, EfficientNet, or Vision Transformer architectures.

[0062] Optionally, the steps described above for extracting silhouette semantic description vectors from unstructured clothing reference images using deep learning models may specifically include: 1) Receive unstructured clothing reference images uploaded by users.

[0063] The received image can be formatted and its resolution checked first to ensure that it meets the input requirements of the subsequent image processing module.

[0064] 2) Detect the formation of a mask for the human body and clothing area.

[0065] For the validated clothing reference images, a deep learning-based object detection and segmentation model can be used to identify and initially locate the human body and clothing regions. The object detection and segmentation model outputs the bounding box of the human body contour and the corresponding clothing region mask in the image, thereby initially separating the foreground (human body and clothing) from the complex background. The object detection and segmentation model can be an instance segmentation model pre-trained on a large-scale dataset containing human body and clothing annotations, and its specific implementation can employ a Mask R-CNN model or a YOLO series model.

[0066] 3) Divide the foreground area of ​​the clothing.

[0067] Based on the human body contour bounding box and the corresponding clothing region mask obtained in the preceding steps, the clothing region is precisely segmented at the pixel level using an image segmentation model to extract the clothing foreground region. The clothing foreground region typically appears as one or more consecutive pixel regions in the image, its contour corresponding to the actual outer contour of the clothing in the image, including all visible parts of the clothing, such as the bodice, sleeves, and collar of a top, and the trouser legs and hem of a bottom. The image segmentation model can be a semantic segmentation model, more specifically, a U-Net model or a DeepLab series model. Guided by the human body and clothing region mask, the semantic segmentation model performs refined segmentation of the clothing region, removing pixel interference from the image background, human skin, accessories, and other non-clothing elements, outputting a clean binary mask image containing only the visual information of the clothing itself. Applying this binary mask image to the original clothing reference image extracts the clothing foreground region.

[0068] 4) Extract depthwise convolution features.

[0069] The extracted foreground region (image) of the clothing is input into a deep learning model pre-trained for a clothing vision task. The deep learning model extracts high-dimensional depth features related to the clothing's silhouette, structure, texture, and style by performing multi-layer convolution and pooling operations on the input foreground region. The pre-trained deep learning model is obtained through supervised pre-training using a dataset containing clothing reference images and their overall silhouette classification data, key size proportion data, local contour quantification indicators, and style intensity rating annotations. The pre-training process optimizes the loss function between the model's predicted output and the ground truth annotations (e.g., using cross-entropy loss for classification tasks and mean squared error loss for regression tasks) to adapt the model to the task of extracting fine-grained silhouette semantic description vectors from clothing reference images. The deep learning model is built on a high-performance convolutional neural network backbone, and its specific implementation can employ ResNet, EfficientNet, or Vision Transformer architectures.

[0070] 5) Construct fine-grained profile semantic description vectors.

[0071] Based on the extracted high-dimensional deep features related to clothing silhouette, structure, texture, and style, the structured, fine-grained silhouette semantic description vector is calculated and output through the fully connected layers or specific task heads included in the aforementioned pre-trained deep learning model.

[0072] Optionally, the silhouette semantic description vector, as a high-dimensional numerical vector, encodes some or all of the following four types of quantitative semantic information strongly related to pattern design: ① Overall silhouette classification probability distribution -- used to quantitatively describe the shape tendency of the overall external contour of the garment; the overall silhouette classification probability distribution is implemented through a classification output layer, which outputs the probability value of the garment belonging to a predefined basic silhouette category (such as one or more of X-type, A-type, H-type, T-type, and O-type), and the probability distribution form expresses the silhouette type to which the garment in the image is most likely to belong and its degree of approximation. ② Key Dimension Proportional Features – Used to quantitatively describe the relative dimensional relationships between different key parts of the garment; these key dimensional proportional features are obtained by constructing an inverse solver network based on a 3D human body parametric model (such as the SMPL model), reconstructing the 3D topological relationship between the human skeleton and the garment from a 2D image, and extracting measurements in the reconstructed 3D space to eliminate measurement errors caused by camera perspective and posture distortion; the specifically calculated proportions include at least one or more of the following: the ratio of shoulder width to hip width, the ratio of chest width to waist width, the ratio of garment length to shoulder width, the ratio of sleeve length to garment length, and the ratio of neckline width to shoulder width; these ratios reflect the basic structure and shape proportions of the garment pattern. ③ Local contour quantification index -- used to provide a refined and numerical description of specific local design features of clothing; the local contour quantification index is also obtained based on the analysis of the reconstructed 3D topological relationship and the foreground area of ​​the clothing. Its specific index includes at least: the relative height of the waist position (expressed as a percentage from the shoulder point to the garment length), the indentation strength of the waist (quantified by the curvature of the waistline or the maximum deviation distance from the reference straight line), the ratio of the width of the narrowest part of the waist to the bust or hip circumference, the expansion of the hem (calculated by the ratio of the hem circumference to the bust or hip circumference), the shape characteristics of the hem (such as the classification or parameter description of straight, A-line, and fishtail), the inclination angle of the shoulder line, and one or more of the roundness or squareness of the shoulder contour. ④ Style Intensity Score – This is used to objectively quantify the subjective perceived style dimensions of clothing. The style intensity score is achieved through a regression output layer. Based on the style perception mapping relationship learned by the model from expert-annotated data, it outputs the intensity score for one or more style dimensions such as “fit”, “looseness”, “three-dimensionality”, “drape”, and “crispness”. The score is usually represented by a continuous numerical value (e.g., 0-10 points). The higher the value, the more significant the style feature.

[0073] 6) Conduct confidence assessment and feedback on the semantic description vector of the profile.

[0074] Calculate the overall confidence score of the fine-grained silhouette semantic description vector (e.g., the probability entropy based on the output of the Softmax function of a deep learning model, or the reliability score detected by key features); compare the confidence score with a preset quality threshold; if the confidence score is lower than the quality threshold, prompt the user through the interactive interface, suggesting that the user resubmit a clearer, more frontal, or more complete clothing reference image.

[0075] In some embodiments of the present invention, the steps for obtaining the user's body size data specifically include any of the following: ① Manual input: The user manually inputs the size values ​​of various parts of their body according to the measurement instructions; ② AI body measurement acquisition: Using a smartphone camera or a dedicated camera device, by capturing images or videos of the user's front and side views, the user's body size is automatically estimated non-contactly based on computer vision and deep learning algorithms; the AI ​​body measurement technology calculates multiple dimensions, including but not limited to height, chest circumference, waist circumference, hip circumference, shoulder width, arm length, and leg length, by recognizing key points and contours of the human body in the image and combining them with a depth estimation model; ③ 3D scanning acquisition: The three-dimensional point cloud data or mesh model of the user's body surface is obtained through a 3D human body scanning device, and then various body dimensions are accurately extracted from the three-dimensional data.

[0076] In some embodiments of the present invention, after obtaining the user's body size data, the body size data can be cleaned and standardized to form standard body size data. Specifically, this includes: ① Unit unification: converting size data from different sources or with different unit systems into a preset standard unit of measurement (e.g., centimeters); ② Outlier detection and processing: based on common sense in anthropometrics and statistical methods, identifying and marking abnormal size data that clearly exceeds a reasonable range; the outlier detection includes, but is not limited to: checking whether the absolute value of a single size is within a physiologically reasonable range (e.g., adult male height is usually between 150cm and 220cm), and identifying extreme values ​​that may contain input errors; ③ Logical consistency verification: verifying whether the logical relationship between different size data conforms to basic human proportion rules; the consistency verification includes, but is not limited to: verifying that chest circumference should be greater than waist circumference, shoulder width should be less than chest circumference, arm length should be less than height, etc.; when a logical conflict is detected, the system triggers a prompt to request user confirmation or remeasurement.

[0077] Using this embodiment helps ensure the reliability, consistency, and comparability of data.

[0078] In some embodiments of the present invention, the circumference proportion features include some or all of the following: waist-to-hip ratio (the ratio of waist circumference to hip circumference), chest-to-waist difference (the absolute difference between chest circumference and waist circumference), and shoulder-to-chest ratio (the ratio of shoulder width to chest circumference); the length proportion features include some or all of the following: the ratio of leg length to height, the ratio of arm length to height, and the ratio of back length to height; and the width proportion features include the ratio of shoulder width to hip width, etc.

[0079] Optionally, considering the lack of some size dimensions in practical applications (such as missing arm length data), an association mapping model based on an anthropometric database can be pre-established for sparse data feature completion. The association mapping model uses high-confidence data such as height, weight, and chest circumference as conditions to perform regression prediction and numerical completion of missing features, thereby ensuring that the input vector has no empty values. During use, the completed standardized body size data and key body shape feature values ​​are concatenated and combined in a predefined and fixed order to form a user body shape feature vector. The user body shape feature vector is a one-dimensional numerical array, where each dimension corresponds to a specific body size or body shape feature value, and its value represents the standardized or calculated value of that dimension. The user body shape feature vector provides structured and numerical input of user physiological conditions for subsequent pattern parameter prediction.

[0080] In some embodiments of the present invention, the silhouette semantic description vector and the user body shape feature vector are fused. Generally, this involves concatenating the two vectors to form a joint feature vector. This is achieved by directly connecting all dimensions of the two vectors in a predetermined order to form a higher-dimensional feature vector. This ensures that style information and body shape information are fed equally and completely into the subsequent clothing pattern prediction model. The joint feature vector serves as the input to the pre-trained clothing pattern prediction model. Based on the input joint feature vector, the clothing pattern prediction model outputs a numerical vector, where each element corresponds to an adjustment amount for a predefined key pattern control parameter.

[0081] The garment pattern prediction model can be a machine learning model, preferably a Gradient Boosting Decision Tree (GBDT) model, specifically XGBoost, LightGBM, etc. It has excellent modeling capabilities and prediction accuracy for tabular data, high-dimensional sparse features, and complex feature interactions, and also boasts high training and inference efficiency and relatively good interpretability, making it a suitable core algorithm.

[0082] Optionally, the numerical vector output by the garment pattern prediction model can be parsed into a set of structured, executable pattern parameter modifications to form a pattern parameter modification set. The pattern parameter modifications include at least one or more of width-type parameters, length and position-type parameters, and structural parameters; wherein the width-type parameters include at least one of bust circumference ease, waist circumference ease, hip circumference ease, shoulder width adjustment, and hem circumference ease; the length and position-type parameters include at least one of garment length adjustment, sleeve length adjustment, waistline position offset, neckline depth and width adjustment, and armhole depth adjustment; and the structural parameters include at least one of dart amount, seam line position offset, and pleat depth.

[0083] The definition of the set of pattern parameter modification amounts specifically includes: Parameter System: The key control parameters for the pattern are a predefined parameter system based on garment engineering and pattern design principles, covering at least the following three categories of parameters: Width parameters—controlling the ease allowance of various circumference dimensions of the garment, directly determining the fit and looseness of the garment, such as bust ease allowance, waist ease allowance, hip ease allowance, shoulder width adjustment allowance, and hem ease allowance. Length and position parameters—controlling the overall proportions and local structural positions of the garment, such as garment length adjustment allowance, sleeve length adjustment allowance, waistline position (vertical distance from the shoulder point) offset, neckline depth and width adjustment allowance, and armhole depth adjustment allowance. Structural parameters—controlling the three-dimensional shape and internal shaping structure of the garment, such as the position, direction, and amount of darts, the position offset of princess seams and other dividing lines, and the depth and number of pleats.

[0084] Numerical Format: The modification amount of the pattern parameters is usually given in incremental form, that is, the increase or decrease of the corresponding parameter value relative to a certain standard base pattern template. The modification amount of the pattern parameters can be a floating-point number, accurate to one or two decimal places, thereby ensuring the accuracy and operability of the adjustment instructions.

[0085] Interpretability: Pre-trained clothing pattern prediction models (e.g., gradient boosting decision trees) have feature importance analysis capabilities to indicate which silhouette semantic features (e.g., "high fit score") or which user body shape features (e.g., "large waist-to-hip ratio") have a key impact on the final parameter adjustment amount (e.g., "reduced waist circumference relaxation") in this prediction, thereby enhancing the transparency and credibility of the entire prediction process.

[0086] In other embodiments of the present invention, a 3D clothing model can be generated based on the 2D pattern drawing, and the physical property parameters of the target clothing fabric can be obtained. The human body model can be deformed by combining the physical property parameters of the fabric with the user body shape feature vector to match the user body shape. The 3D clothing model can then be worn on the deformed human body model to perform dynamic calculations and form a visual preview.

[0087] In a specific embodiment of the present invention, the basic pattern template is invoked as follows: based on the target garment category (e.g., suit, shirt, dress, trench coat, etc.) specified by the user or automatically identified by the system, the corresponding basic pattern template is retrieved and invoked from the pre-built pattern template library. The pattern template library stores parameterized basic patterns for various garment categories, and each basic pattern template is defined in a structured data format. The structured data format definition includes: the identifiers of all independent cut pieces constituting the garment (e.g., front piece, back piece, left / right sleeve piece, collar piece, etc.); the two-dimensional coordinates of a series of key control points on the outline of each cut piece; curve parameters used to describe the outline curve of the cut piece (e.g., Bézier curve control points, arc radius, etc.); and the preset connection and sewing relationship information between each cut piece.

[0088] In a specific embodiment of the present invention, the predicted modification amount of the pattern parameters is applied to the called basic pattern template. The pattern calculation module calculates and updates the geometric contours of each piece, thereby generating an updated 2D pattern drawing. This process specifically includes: ① Control point coordinate update: Based on the modification amount of the pattern parameters (mostly in incremental form), the new coordinates of each key control point in the basic pattern template are calculated according to predefined parameter-geometric mapping rules. ② Contour curve reconstruction: Based on the updated key control point coordinates, a smooth and continuous contour curve of each piece is regenerated using spline interpolation algorithms (such as Bézier curves and B-spline curve fitting). ③ Maintenance of process constraint consistency: During the contour reconstruction process, the pattern calculation module integrates process constraint checking logic to verify and maintain the manufacturability of the generated pattern. For example, it ensures that the side pairs to be sewn (such as sleeve cap arc and armhole arc) match in length, the internal angles of the pieces are within a reasonable process range, and symmetrical pieces maintain geometric consistency. ④ 2D Pattern Output: Update the outlines of all the calculated cut pieces, organize them into a standard two-dimensional geometric data format, and generate a 2D pattern. This pattern contains the outlines, key markers, cuts and seams of all the cut pieces, forming geometric data that can be directly used by downstream computer-aided design (CAD) software for pattern making, layout or further editing.

[0089] Furthermore, in some embodiments of the present invention, a simplified 3D garment model can be constructed based on a 2D pattern drawing and according to preset stitching and mapping rules for 3D visualization preview. The specific steps for constructing the 3D garment model include: ① Mapping from 2D pieces to 3D space: Each piece in the 2D pattern drawing is initially mapped to 3D space according to its natural position and posture within the garment structure. ② Virtual stitching and mesh generation: Based on the stitching relationship between the pieces, a fast mesh construction algorithm is used to stitch and connect the corresponding edges in 3D space, forming triangular mesh surfaces, which together constitute the 3D garment model. ③ Physical simulation and dynamic calculation: A mass spring model or finite element analysis physics engine is introduced to read the physical property parameters of the target fabric (at least including density, tensile stiffness, shear stiffness, and bending stiffness) to drive the dynamic calculation of the 3D mesh in 3D space, realistically simulating the drape and stiffness of the garment. ④ Model optimization: The number and distribution of triangular mesh surfaces can be optimized to significantly reduce the computational load of real-time rendering while ensuring the accuracy of the dynamic calculation.

[0090] Furthermore, in some embodiments of the present invention, a personalized human body model can be generated based on the 3D model, allowing users to more intuitively see the effect of clothing on their own body shape. Using the user body shape feature vector constructed in the aforementioned steps, a standard human body model is personalized by deformation to generate a personalized human body model that matches the user's actual body shape. The steps for generating a personalized human body model by personalized deformation of the standard human body model specifically include: ① Difference calculation: Comparing key dimensions (such as chest circumference, waist circumference, hip circumference, shoulder width, height, etc.) in the user body shape feature vector with the corresponding preset standard dimensions of the standard human body model to calculate the dimensional differences of each part. ② Model deformation: Using free deformation or linear hybrid skinning technology based on skeletal line weights, the mesh vertices of the standard human body model are driven according to the calculated dimensional differences, causing them to expand, contract, or stretch in the corresponding parts, thereby adapting to the user's body shape. In addition, the deformation process considers the proportional coordination of various parts of the human body to ensure that the deformed model appears visually natural and reasonable.

[0091] In some embodiments of the present invention, virtual try-on and dynamic previews can be generated from 2D pattern drawings and / or the 3D clothing model. Using a preset clothing simulation algorithm, the garment is worn onto the generated personalized human body model, thus forming a dynamic visual preview. Furthermore, this visual preview can display the garment from multiple angles in three-dimensional space, including: the actual presentation of the overall silhouette style of the clothing (such as X-shape, A-shape, etc.) on the user's body shape; the fit of key areas (such as the closeness or looseness of the shoulders, chest, waist, and hips); and the overall proportions and visual effect of the clothing.

[0092] In an easy-to-understand way, virtual try-on and dynamic preview can be performed through a real-time rendering engine, allowing users or garment pattern makers to rotate the view and zoom to view details, thereby providing an intuitive and comprehensive evaluation of the pattern design effect.

[0093] In some embodiments of the present invention, the interactive interface includes a preview area for displaying dynamic previews, for visually displaying 2D pattern diagrams and / or 3D garment models, and supports users in adjusting the pattern parameters. After adjustment, the system updates the 2D pattern diagrams and / or 3D garment models in real time. The interactive interface can present the interactive environment to the user through a graphical user interface (GUI), which is logically divided into multiple functional areas for synchronously displaying all information and visualization results related to the current customization task.

[0094] In other embodiments of the present invention, the functional areas of the interactive interface include: ① Reference image and semantic analysis result area: displaying the original clothing reference image uploaded by the user, and overlaying the key silhouette semantic tags parsed by the system (such as "X-shaped silhouette probability: 92%", "fitness score: 9.5 / 10", etc.); ② 2D pattern display area: clearly displaying the 2D pattern in the form of engineering drawings, including the outlines of all pieces, key control points and size annotations, supporting operations such as panning, scaling, and layer control; ③ Pattern parameter modification amount and adjustment panel area: clearly listing all predicted pattern parameter modification amounts in the form of structured tables or lists (such as "bust ease: +2.5 cm", "waistline position offset: -1.0 cm"), and associating interactive controls (such as numerical input boxes, slider controls) with each adjustable parameter, allowing users to directly modify parameter values; some interfaces can also provide direct geometric drag-and-drop editing functions for specific control points or contour curves in the 2D pattern. ④ 3D Visual Preview Area (Optional): Through an embedded 3D rendering engine, a virtual try-on preview is dynamically displayed, which is the effect of the 3D clothing model worn on a personalized human body model. Users can freely rotate the view and zoom to view details.

[0095] Using this embodiment of the invention, users can fine-tune parameters. Specifically, when a user modifies any pattern parameter value or performs a geometric drag operation through controls on the interactive interface, the system obtains the adjustment instruction; when the adjustment instruction carries the identifier of the modified parameter and its new target value or geometric offset, a real-time processing flow is triggered.

[0096] Optionally, the real-time processing flow includes: ① Parameter update: Updating the user-adjusted parameter values ​​to the internal pattern parameter dataset. ② Pattern recalculation: Calling the pattern calculation module (same as S42), applying the updated complete parameter set to the current basic pattern template, quickly recalculating the outlines of all fabric pieces, and generating updated 2D pattern geometry data. ③ 3D model and preview regeneration (optional): Based on the updated 2D pattern geometry data, updating the mesh shape of the 3D garment model, and re-wearing it onto the personalized human body model, re-rendering it in real time through the 3D rendering engine to generate a new dynamic visual preview.

[0097] Furthermore, the 2D pattern and 3D visualization preview results, which are regenerated after real-time processing, are pushed back to the interactive interface, triggering a refresh of the corresponding display areas: the 2D pattern display area is updated with the new cut outline; the 3D visualization preview area is updated with the new virtual try-on effect.

[0098] By employing this embodiment of the invention, a closed loop of parameter adjustment, pattern recalculation, and effect preview can be achieved through a real-time update and feedback mechanism. This allows users to intuitively and instantly observe the direct impact of each fine-tuning on the garment's pattern outline and wearing effect, thereby supporting efficient iterative optimization.

[0099] In addition, all parameter adjustments made by the user during the current session can be automatically recorded in the background, forming an adjustment history. Each record includes at least: the adjustment time, the identifier of the modified parameter, the value before adjustment, and the value after adjustment. Users can easily revert to any previous adjustment state or compare the pattern and preview effect under different parameter schemes through the "Undo / Redo" function provided in the interface, to assist in decision-making.

[0100] Understandably, the interactive fine-tuning process terminates when the user or pattern maker confirms the adjusted pattern design (e.g., by clicking the "Confirm Final Design" button). The process then executes and generates a structured pattern data file, specifically including: encapsulating all finalized pattern parameters (including initial predicted values ​​and all manually adjusted values), corresponding 2D pattern geometry data, cut piece information, seam relationships, user information, and reference image identifiers, according to a predefined structured format, to generate a structured pattern data file. The preferred file format is a lightweight data exchange format, such as JSON or XML.

[0101] In a specific embodiment of the present invention, taking a slim-fit suit as an example, the process of predicting the garment pattern is as follows.

[0102] First, the system receives a reference image of a suit uploaded by the user: a street photograph. Background interference is removed through object detection and segmentation to extract a clean foreground image of the suit. A deep learning model is then used to infer meaning from this image, outputting a fine-grained silhouette semantic description vector, specifically including: Overall silhouette classification: X-type probability 0.92, H-type probability 0.05.

[0103] Size proportions: shoulder width / hip width = 0.98, chest width / waist width = 1.15, garment length / shoulder width = 1.60, sleeve length / garment length = 0.75.

[0104] Local contour indicators: the waistline is at 60% of the garment length, the indentation strength is 0.25, the hem expansion is 0.05, and the shoulder line slopes at 15 degrees.

[0105] Style intensity rating: Slim fit 9.5, Loose fit 1.5, Three-dimensionality 9.0, Drape 3.0, Crispness 8.8.

[0106] Then, the user's body dimensions were obtained using AI-powered body measurement: height 178.5cm, chest circumference 98.2cm, waist circumference 79.5cm, hip circumference 95.8cm, and shoulder width 46.1cm. Key body shape features were calculated: waist-to-hip ratio 0.83, chest-to-waist difference 18.7cm, shoulder-to-hip ratio 1.04, and leg length / height = 0.59. Based on these, a user body shape feature vector was constructed.

[0107] Then, the aforementioned silhouette semantic vector (25 dimensions) and body shape feature vector (12 dimensions) are concatenated into a joint feature vector (37 dimensions), which is then input into an XGBoost model trained based on historical suit orders. The model predicts the amount of modification to the published shape parameters, including: chest circumference ease +2.5cm, waist circumference ease +1.0cm, hem circumference ease +0.5cm, garment length adjustment -1.0cm, sleeve length adjustment +0.8cm, shoulder width adjustment +0.3cm, and dart adjustment 1.5cm.

[0108] Next, based on the clothing category, a basic suit pattern template is called. The predicted modification amount is applied to the template through the pattern calculation module, automatically updating the outline of each piece and generating a 2D pattern drawing. Further, a 3D clothing model is generated based on the 2D pattern drawing. After deforming the human body model according to the user's body shape data, a virtual try-on is performed, creating a visual preview.

[0109] Finally, the system synchronously displays reference images, analytical labels, 2D pattern diagrams, 3D previews, and parameter modification tables through an interactive interface. Pattern makers can fine-tune parameters using sliders and other controls, and the system updates the pattern and preview in real time. Upon confirmation, a structured JSON file containing all pattern piece geometry data, parameters, and user information is output, which can be directly imported into CAD software for subsequent production.

[0110] If a traditional process is used to generate a slim-fit suit pattern, firstly, a pattern maker needs to manually interpret the style of the reference images (such as silhouette, proportions, and details) and communicate with the client to confirm understanding. Then, the pattern maker must rely on personal experience to translate the style intention into specific pattern parameters, manually adjust the basic pattern in CAD software, and confirm with the client again. This traditional process typically requires 4-6 rounds of communication and takes 5-7 working days. It is highly dependent on the pattern maker's core experience, and the decision-making process for pattern parameters is highly subjective and lacks consistency. In contrast, the process based on this solution only requires 1-2 rounds of communication and takes 1-2 working days. It has a low to medium reliance on the pattern maker's core experience, and the decision-making process for pattern parameters is less subjective and more consistent. Clearly, this solution can improve the overall efficiency of garment pattern prediction / generation by more than 70%, reduce over-reliance on specific scarce human resources, enable ordinary technicians to produce professional solutions with the help of the system, facilitate knowledge transfer and scaling, and ensure stable and consistent parameter suggestions for the same input (image + body shape) based on a unified model and data, which helps to standardize customized services and improve the controllability of brand quality and the stability of customer experience.

[0111] This invention combines intelligent image analysis, data-driven prediction, and parametric pattern generation, which not only compresses the traditional weekly customization cycle to the daily level, but also brings fundamental improvements that are difficult to achieve with traditional methods in terms of reducing reliance on experience, improving solution consistency, enhancing customer experience, and achieving scalability.

[0112] In another specific embodiment of the present invention, taking a men's shirt as an example, the process of predicting the garment pattern is as follows.

[0113] First, the system receives a reference image of a shirt uploaded by the user: a business casual style image. Background interference is removed through object detection and segmentation to extract a clean foreground image of the shirt. A deep learning model is then used to infer meaning from this image, outputting a fine-grained silhouette semantic description vector, specifically including: Overall silhouette classification: H-shape (straight-cut) probability 0.78, slim-fit probability 0.20.

[0114] Size proportions: shoulder width / chest width = 0.92, chest width / waist width = 1.08, garment length / shoulder width = 1.75, sleeve length / garment length = 0.82.

[0115] Local contour indicators: the waist is positioned at 55% of the garment length (slightly tapered), the waist indentation strength is 0.15, the hem width / chest width = 0.98, and the neckline width / shoulder width = 0.25.

[0116] Style intensity rating: Fit 7.5, Looseness 3.0, Crispness 8.5, Drape 2.0, Neckline formality 8.0.

[0117] Then, the user inputs the following data: height 180.0cm, chest circumference 100.0cm, waist circumference 86.0cm, hip circumference 96.0cm, shoulder width 48.0cm, arm length 63.0cm, and neck circumference 40.0cm. Key body shape features are calculated: chest-to-waist difference 14.0cm, shoulder-to-waist ratio 1.12, arm length / height = 0.35, and neck-to-shoulder angle 15 degrees. Based on this, a user body shape feature vector is constructed.

[0118] Then, the aforementioned silhouette semantic vector (28 dimensions) and body shape feature vector (15 dimensions) are concatenated into a joint feature vector (43 dimensions), which is then input into the LightGBM model trained based on historical shirt orders. The model predicts the amount of modification to the published parameters, including: chest circumference ease +4.0cm, waist circumference ease +2.0cm, hem circumference ease +1.5cm, garment length adjustment -2.0cm (to adapt to leg length), sleeve length adjustment +1.2cm, shoulder width adjustment +0.5cm, neckline adjustment +1.0cm, and cuff width +0.3cm.

[0119] Next, based on the clothing category, a basic men's shirt pattern template is called. The predicted amount of modification is applied to the template through the pattern calculation module, automatically updating the outlines of the front and back pieces, sleeves, and collar to generate a 2D pattern drawing. A 3D clothing model is then generated based on the 2D pattern drawing, and after deforming the human body model according to the user's body shape data, a virtual try-on is performed, creating a visual preview.

[0120] Finally, the system synchronously displays reference images, analytical labels, 2D pattern diagrams, 3D previews, and parameter modification tables through an interactive interface. Pattern makers can fine-tune parameters (such as neckline and sleeve length) using sliders and other controls, with the system updating the pattern and preview in real time. Upon confirmation, a structured JSON pattern file containing all pattern geometric data, parameters, and user information is output, which can be directly imported into CAD software for subsequent production.

[0121] Corresponding to the above method, the present invention also provides a clothing pattern prediction system based on clothing images and body dimensions. The system includes a computer device, which includes a processor and a memory. The memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps of the method described above.

[0122] Figure 4 See the schematic diagram of the computer equipment included in the system. Figure 4The computer device 00 includes: a processor 01, a memory 02, and a computer program stored on the memory 02 and executable on the processor 01. When the processor 01 executes the computer program, it implements the method steps proposed in any of the above embodiments.

[0123] The processor 01 is connected to the memory 02, such as via a bus 03. The processor 01 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor 01 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The bus 03 may include a pathway for transmitting information between the aforementioned components. The bus 03 can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 03 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 The text uses only a single thick line to represent a bus, but this does not imply that there is only one bus or one type of bus. Memory 02 stores a computer program corresponding to the human factors data server access control method described in the above embodiments of this application. This computer program is executed under the control of processor 01. Processor 01 executes the computer program stored in memory 02 to implement the content shown in the aforementioned method embodiments.

[0124] Corresponding to the methods described above, the present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implements the steps of the method as described in any of the above embodiments. The computer-readable storage medium may be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art.

[0125] Corresponding to the above methods, the present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the method as described in any of the above embodiments.

[0126] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0127] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0128] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0129] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting clothing patterns based on clothing images and body dimensions, characterized in that, The method includes: A deep learning model is used to extract silhouette semantic description vectors from unstructured clothing reference images; wherein, the silhouette semantic description vectors include some or all of the overall silhouette classification probability distribution, key size proportion features, local contour quantification indicators, and style intensity scores; Construct a user body shape feature vector based on the user's body size data; The silhouette semantic description vector and the user body shape feature vector are fused and used as input to a pre-trained clothing pattern prediction model, which outputs the pattern parameter modification amount of the corresponding target clothing in the clothing reference image; wherein, the clothing pattern prediction model is pre-trained through a training set constructed based on historical clothing reference images, historical body size data and corresponding historical pattern parameter modification amounts. Based on the category of the target garment, the corresponding basic pattern template is retrieved from the pattern template library and called. The basic pattern template is adjusted using the pattern parameter modification amount to obtain structured garment pattern data that adapts to the user's body shape and reflects the silhouette style of the garment reference image.

2. The method according to claim 1, characterized in that, The step of extracting silhouette semantic description vectors from unstructured clothing reference images using a deep learning model includes: Identify the human body outline bounding box and the corresponding clothing area mask in the clothing reference image; The foreground region of clothing is extracted from the clothing reference image based on the human body outline bounding box and the corresponding clothing region mask. The foreground region of the garment is input into a pre-trained deep learning model, which outputs a structured silhouette semantic description vector. The pre-trained deep learning model is obtained through supervised pre-training using a dataset containing garment reference images and their overall silhouette classification data, key size ratio data, local contour quantification indicators, and style intensity rating labels.

3. The method according to claim 1, characterized in that, The method also includes a step of acquiring the user's body size data, specifically including: The system can acquire body size data manually input by the user; or acquire images or videos of the user's front and side views and calculate the user's body size data using AI body measurement technology; or acquire three-dimensional point cloud data or a mesh model of the user's surface through 3D human body scanning and calculate the user's body size data based on the three-dimensional point cloud data or the mesh model.

4. The method according to claim 1, characterized in that, The garment pattern prediction model is a gradient boosting decision tree model. The method further includes a step of pre-training the garment pattern prediction model, specifically including: A training set is constructed based on historical clothing reference images, historical body size data, and corresponding historical pattern parameter modification amounts. Each training sample in the training set is a triple consisting of user body size data, clothing reference images, and their corresponding pattern parameter modification amount annotations. A loss function is constructed based on the difference between the predicted output of the gradient boosting decision tree model and the labeled modification amount of the pattern parameters. The constructed training set is used to iteratively train the initially constructed gradient boosting decision tree model to obtain the trained clothing pattern prediction model.

5. The method according to claim 1, characterized in that, The step of adjusting the basic pattern template using the pattern parameter modification amount to obtain structured garment pattern data that adapts to the user's body shape and reflects the silhouette style of the garment reference image specifically includes: According to the predefined parameter-geometric mapping rules, the new coordinates of each key control point included in the basic pattern template are calculated based on the modification amount of the pattern parameters; Based on the new coordinates of each key control point, the spline interpolation algorithm is used to regenerate the outline curves of each piece in the basic pattern template; The outline curves of all the regenerated cut pieces are organized into a standard two-dimensional geometric data form to obtain structured garment pattern data in the form of a 2D pattern diagram.

6. The method according to claim 5, characterized in that, After obtaining the structured garment pattern data in the form of a 2D pattern diagram, the method further includes: mapping each piece of fabric into a three-dimensional space based on the natural position and posture of each piece in the garment structure according to the natural position and posture of each piece in the structured garment pattern data in the form of a 2D pattern diagram; and stitching the corresponding edges of the pieces together in the three-dimensional space according to the stitching relationship between the pieces using a fast mesh construction algorithm to obtain the structured garment pattern data in the form of a 3D garment model; wherein the 3D garment model is in the form of a mesh.

7. The method according to claim 1, characterized in that, The structured garment pattern data includes 2D pattern diagrams and / or 3D garment models; After obtaining the structured garment pattern data, the method further includes: presenting the 2D pattern diagram and / or the 3D garment model to the user using an interactive interface, and adjusting the structured garment pattern data based on the fine-tuning instructions received from the user.

8. A garment pattern prediction system based on garment images and body measurements, comprising a processor, a memory, and a computer program / instructions stored in the memory, characterized in that, The processor is configured to execute the computer program / instructions, and when the computer program / instructions are executed, the system implements the steps of the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

Images