A virtual fitting method for clothing using limited AIGC to achieve real-life pull-on effect transfer

Abstract

The present invention discloses a method for virtual fitting of clothing that utilizes limited AIGC to achieve transfer of real pull-on effects. The method comprises the following steps: storing pull-on image data in a structured manner in a real pull-on effect library; forming a data pair of a target garment to be virtually fitted; determining an effect transfer reference map; performing feature region segmentation and base color extraction on the effect transfer reference map consistent with the data pair of the target garment to be virtually fitted, thereby obtaining an intermediate grayscale image that retains the original wrinkles and spatial structure; performing color coordinate inversion on the base color RGB data in the data pair of the target garment to be virtually fitted, and performing regional fusion with the intermediate grayscale image; and performing color coordinate restoration processing on the synthesized inverted image to generate a virtual fitting effect map of the target garment on a target body shape. The present invention realizes an automatic closed-loop conversion from two-dimensional garment images plus structural parameters to virtual fitting effect maps, enhancing the versatility and deployment efficiency of the method.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual fitting technology, and in particular to a clothing virtual fitting method for realizing the migration of a real pull-on effect by utilizing limited AIGC. Background Art

[0002] With the development of artificial intelligence and image generation technology, virtual fitting has gradually become an important technical direction for clothing e-commerce platforms to enhance user experience and optimize product recommendations. Traditional virtual fitting methods are mostly based on image segmentation, key point recognition and texture fusion visual processing methods. By geometrically matching and visually fusing user images with clothing images, the effect of a certain clothing being worn on the user can be achieved online. However, in existing technologies, the core goals of virtual fitting are mainly focused on "aesthetic display" and "style recommendation", rather than "fit assessment" and "size screening", which poses obvious technical deficiencies in clothing e-commerce scenarios.

[0003] Currently, typical virtual fitting systems rely on the SegFormer B2 model within the mmsegmentation framework. This model segments the human body and clothing regions in user-uploaded images and then overlays the processed clothing images onto the user's body contours. This approach emphasizes visual integration and symmetrical aesthetics, but fails to consider the standard sizing structure of clothing and accurately model the user's actual body shape. The resulting "fit" is often an algorithm-forced conformation to the user's contours, rather than reflecting the actual wearer's size. Therefore, it struggles to provide users with effective sizing references and cannot support the real-world sizing recommendations required by e-commerce platforms.

[0004] On the other hand, while 3D modeling-based virtual fitting can theoretically achieve three-dimensional interaction between clothing and the human body, due to the extremely high requirements for fabric material, mechanical properties, and the simulation accuracy of wrinkle deformation, current simulations are mostly based on spring-mass models and rigid body physics engines. This results in significant distortion in fabric wrapping, stretching, and folding. Modeling is time-consuming, parameter adjustment is difficult, and real-time response and batch deployment are difficult. In addition, while physical fitting can demonstrate realistic effects, it relies on repeated fitting and filming of physical clothing with real people or robots. This leads to high setup costs, cumbersome operational processes, and a strong dependence on spatial resources. It cannot meet the needs of fast online display of multiple styles and sizes in the e-commerce environment.

[0005] In summary, existing virtual fitting technology has obvious shortcomings in terms of realistic fit expression, size recommendation accuracy, and batch adaptation efficiency. It is difficult to support a high-precision virtual fitting experience that matches user body shape and clothing size one-to-one. There is an urgent need for a new virtual fitting method based on the combination of real fitting data and image migration mechanism to solve the bottleneck problems of current technology in fit assessment and wearing space effect generation. Summary of the Invention

[0006] One purpose of the present invention is to propose a virtual clothing fitting method that uses limited AIGC to achieve the migration of real fitting effects. The present invention realizes an automatic conversion closed loop from two-dimensional clothing images plus structural parameters to virtual fitting effect images, thereby enhancing the versatility and deployment efficiency of the method.

[0007] According to an embodiment of the present invention, a virtual clothing fitting method for implementing a real pull-on effect transfer using limited AIGC includes the following steps:

[0008] S1. Collect and store a variety of clothing samples and body data of the fitting robot, through the fitting robot real wear clothing samples of various sizes and shoot the formation of wear image data, the wear image data is stored in a structured manner in the real wear effect gallery;

[0009] S2. For the target garment to be tried on virtually, a target garment template image is shot flat in a wrinkle-free state to form a data pair of the target garment to be tried on virtually;

[0010] S3. In the real pull-on effect image library, based on the size parameters of the target garment and the user's body parameters, search and select matching pull-on image data, extract a set of pull-on images that are the same size as the target garment and similar to the target user's body type, and determine the effect migration reference image;

[0011] S4. Perform feature region segmentation and base color extraction on the effect migration reference image consistent with the data of the target garment to be virtually tried on, and obtain an intermediate grayscale image that retains the original folds and spatial structure;

[0012] S5. The color coordinates of the primary color RGB data of the target clothing to be virtually tried on are inverted, and the region is fused with the intermediate grayscale image to obtain a synthetic inverted image;

[0013] S6. Perform color coordinate restoration processing on the synthesized inverted image to generate a virtual try-on effect image of the target garment on the target body shape.

[0014] Optionally, the S1 includes the following steps:

[0015] S11. Get the basic size parameter information of the clothing sample, including the chest circumference of the sample , sample waist , sample hip circumference , sample length Sleeve length , forming the clothing size vector And each size sample is individually numbered and bound with parameters to form a multi-size sample set;

[0016] S12. Set up a standard fitting robot and collect the standard fitting robot's body parameter information, including the standard fitting robot's chest circumference , Standard waist circumference of fitting robot , Standard fitting robot hip circumference , Standard fitting robot shoulder width and the height of a standard fitting robot , forming the robot body parameter vector The standard fitting robot body parameter information corresponds to a plurality of standard robot models constructed by simulating real human body shapes;

[0017] S13. Each size sample is put on a standard fitting robot of different body types, and the fitting effect is photographed from multiple angles using uniform lighting conditions and background settings to generate an image set. ,in Indicates the sample number. Indicates the standard fitting robot number, Represents clothing size vector Put on the robot body parameter vector The image on the image is formed into a clothing size vector and the robot body parameter vector Pull-through image data generated by the combination of ;

[0018] S14. Put all the image data into Store in the real pull-on effect gallery to form a real pull-on effect gallery :

[0019] ;

[0020] in, is the total number of garment samples, This is the number of standard fitting robot models.

[0021] Optionally, the S2 includes the following steps:

[0022] S21. Spread the target garment to be virtually tried on flatly on the shooting platform, so that the target garment to be virtually tried on is in an ideal shooting state without wrinkles or obvious shadows. Use a standard light source and a fixed shooting angle to obtain a template image of the target garment. ;

[0023] S22. Measure the actual size of the target garment to obtain the size parameters of the target garment, including the chest circumference of the target garment. Target clothing waistline Target clothing hip circumference Target clothing length And target clothing sleeve length , forming the target clothing size vector ;

[0024] S23. The target clothing template image with the target clothing size vector Binding to form the data pair of the target clothing to be virtually tried on .

[0025] Optionally, S3 includes the following steps:

[0026] S31. Based on the data of the target clothing to be virtually tried on Target clothing size vector in and the user's body parameter vector , in the real pull-on effect gallery The pull-on image data are retrieved and matched in the dataset to determine the subset of pull-on image data that has the same size as the target garment and is similar to the target user's body shape. :

[0027] ;

[0028] in, represents the Euclidean distance between vectors, Indicates the allowed matching error threshold of the user's body shape parameters;

[0029] S32. The pull-through image data subset Comprehensively evaluate the structural feature area, base color RGB data, and actual wear wrinkle morphology to determine the optimal pull-on image data As a benchmark for effect migration:

[0030] ;

[0031] in, 、 are weight coefficients, representing the importance of body shape similarity and image quality, Indicates pull-through image data The image quality evaluation index of the real wearing wrinkle morphology. The smaller the index value, the higher the quality of the pull-on image.

[0032] Optionally, the S4 includes the following steps:

[0033] S41. Benchmark chart for effect migration Execute data comparison with the target garment to be virtually tried on Consistent clothing feature area mask segmentation to obtain clothing feature area mask image of effect migration benchmark ;

[0034] S42. Clothing feature region mask image based on effect transfer benchmark image , respectively extract the base color RGB data of each clothing feature area in the effect migration benchmark image to form a base color RGB data set ;

[0035] S43. Base color RGB data set for each clothing feature area of the effect migration benchmark image Decolorization is performed by inverting the color coordinates to obtain an intermediate grayscale image that retains the original folds and spatial structure. :

[0036] ;

[0037] in, The coordinates in the effect migration benchmark diagram are The original pixel value, Indicates the corresponding coordinates Feature area The base color RGB data, Indicates taking the absolute value of the color difference between each RGB channel.

[0038] Optionally, the S5 includes the following steps:

[0039] S51. Data pair for virtual try-on of target clothing The base color RGB data set of each clothing feature area Perform color coordinate inversion processing to form an inverted base color RGB data set ;

[0040] S52. The inverted primary color RGB data set With the middle grayscale image Perform pixel-by-pixel regional fusion processing to form a synthetic inverted image , the synthetic inverted image By integrating the color style and effect of the target garment, the real wearing spatial structure information in the benchmark image is transferred to fully express the real wearing folds and fitting spatial effects of the target garment:

[0041] ;

[0042] in, Indicates that the coordinates in the synthetic inverted image are The pixel value at Represents an intermediate grayscale image The median coordinate is The pixel value of Indicates the corresponding coordinates of the target garment to be virtually tried on Feature area The inverted base color RGB data.

[0043] Optionally, the S6 includes the following steps:

[0044] S61. Synthesize inverted image Perform color coordinate restoration processing on all pixel points in the image, convert the inverted color information back to the real display color value, and generate a virtual try-on effect diagram of the target clothing on the target body shape. ;

[0045] S62. The virtual try-on effect diagram The visual effects of each characteristic area of the target clothing meet the following conditions:

[0046] The color feature is consistent with the base color of the corresponding area in the target clothing template image;

[0047] The spatial wrapping feature completely corresponds to the grayscale of the folds of the same-sized garment on a similar body shape in the effect transfer benchmark image;

[0048] Fit expression based on size vector User body parameters If the following conditions are met, the virtual try-on effect diagram Rendered in a fitted visual state:

[0049] ;

[0050] ;

[0051] ;

[0052] in, They are the target clothing bust, waist and hip circumference respectively. They are the user's chest circumference, waist circumference, and hip circumference respectively. Error thresholds set for experience;

[0053] If any of the above conditions are not met, then The corresponding area in the middle is shown as loose or tight.

[0054] The beneficial effects of the present invention are:

[0055] (1) The present invention systematically collects real wearing images of various clothing sizes on fitting robots of various body shapes, and stores the images together with the corresponding size parameters and body shape parameters in a structured manner to construct a complete real fitting library. By retrieving sample images that are highly matched with the target clothing size and user body shape from the real fitting library as migration reference images, the final virtual fitting image is no longer based on geometric mapping, but is derived from a real physical fitting scene, thereby significantly improving the accuracy and credibility of the fitting performance.

[0056] (2) This invention introduces a color coordinate inversion strategy to separate the color features of the target clothing image from the structural texture information of the wearing image. By decolorizing the real wearing image, only the wrinkles and wrapping structure are retained. The color information of the target clothing is then superimposed on the grayscale image to complete the image synthesis of the new clothing in the real wearing form. This avoids the loss of lighting, shadows and fabric details caused by traditional direct mapping, and can accurately restore the wrapping state, tightness and spatial hierarchy of the clothing on different body shapes.

[0057] (3) The present invention constructs a complete image generation chain by uniformly modeling clothing template images, size parameters, structural masks and body parameters to form a standardized input format, and combines the fusion and inverse transformation processing of real wearing images. The chain does not rely on three-dimensional modeling and complex physical simulation. It can automatically output virtual fitting effect images with spatial realism and style consistency without changing the fitting environment. It is suitable for e-commerce, school uniform customization, remote size selection and many other scenarios, greatly improving the response speed and large-scale deployment capabilities of virtual fitting. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0059] Figure 1 This is a flowchart of a virtual clothing fitting method proposed by the present invention that uses limited AIGC to achieve real pull-on effect migration. DETAILED DESCRIPTION

[0060] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0061] refer to Figure 1 A virtual fitting method for clothing using limited AIGC to achieve real pull-on effect transfer includes the following steps:

[0062] S1. Collect and store a variety of clothing samples and body data of the fitting robot, through the fitting robot real wear clothing samples of various sizes and shoot the formation of wear image data, the wear image data is stored in a structured manner in the real wear effect gallery;

[0063] S2. For the target garment to be tried on virtually, a target garment template image is shot flat in a wrinkle-free state to form a data pair of the target garment to be tried on virtually;

[0064] S3. In the real pull-on effect image library, based on the size parameters of the target garment and the user's body parameters, search and select matching pull-on image data, extract a set of pull-on images that are the same size as the target garment and similar to the target user's body type, and determine the effect migration reference image;

[0065] S4. Perform feature region segmentation and base color extraction on the effect migration reference image consistent with the data of the target garment to be virtually tried on, and obtain an intermediate grayscale image that retains the original folds and spatial structure;

[0066] S5. The color coordinates of the primary color RGB data of the target clothing to be virtually tried on are inverted, and the region is fused with the intermediate grayscale image to obtain a synthetic inverted image;

[0067] S6. Perform color coordinate restoration processing on the synthesized inverted image to generate a virtual try-on effect image of the target garment on the target body shape.

[0068] In this embodiment, S1 includes the following steps:

[0069] S11. Get the basic size parameter information of the clothing sample, including the chest circumference of the sample , sample waist , sample hip circumference , sample length Sleeve length , forming the clothing size vector And each size sample is individually numbered and bound with parameters to form a multi-size sample set;

[0070] S12. Set up a standard fitting robot and collect the standard fitting robot's body parameter information, including the standard fitting robot's chest circumference , Standard waist circumference of fitting robot , Standard fitting robot hip circumference , Standard fitting robot shoulder width and the height of a standard fitting robot , forming the robot body parameter vector The standard fitting robot body parameter information corresponds to a plurality of standard robot models constructed by simulating real human body shapes;

[0071] S13. Each size sample is put on a standard fitting robot of different body types, and the fitting effect is photographed from multiple angles using uniform lighting conditions and background settings to generate an image set. ,in Indicates the sample number. Indicates the standard fitting robot number, Represents clothing size vector Put on the robot body parameter vector The image on the image is formed into a clothing size vector and the robot body parameter vector Pull-through image data generated by the combination of ;

[0072] S14. Put all the image data into Store in the real pull-on effect gallery to form a real pull-on effect gallery :

[0073] ;

[0074] in, is the total number of garment samples, This is the number of standard fitting robot models.

[0075] This embodiment systematically collects real-life images of various clothing sizes worn on fitting robots of various body shapes, and structures the images together with the corresponding size parameters and body shape parameters to construct a complete real-life fitting image library. By retrieving sample images that highly match the target clothing size and user body shape from the real-life fitting image library as migration reference images, the final virtual fitting image is no longer based on geometric mapping, but is derived from a real physical fitting scene, thereby significantly improving the accuracy and credibility of the fitting performance.

[0076] In this embodiment, S2 includes the following steps:

[0077] S21. Spread the target garment to be virtually tried on flatly on the shooting platform, so that the target garment to be virtually tried on is in an ideal shooting state without wrinkles or obvious shadows. Use a standard light source and a fixed shooting angle to obtain a template image of the target garment. ;

[0078] S22. Measure the actual size of the target garment to obtain the size parameters of the target garment, including the chest circumference of the target garment. Target clothing waistline Target clothing hip circumference Target clothing length And target clothing sleeve length , forming the target clothing size vector ;

[0079] S23. The target clothing template image with the target clothing size vector Binding to form the data pair of the target clothing to be virtually tried on .

[0080] This implementation method ensures a unified input basis for subsequent segmentation and fusion processing by standardizing the capture of wrinkle-free target clothing template images and accurately recording size parameters, thereby improving the accuracy of image structure extraction and the fit performance of virtual try-on effects, and achieving data consistency and accuracy in the virtual fitting process.

[0081] In this embodiment, S3 includes the following steps:

[0082] S31. Based on the data of the target clothing to be virtually tried on Target clothing size vector in and the user's body parameter vector , in the real pull-on effect gallery The pull-on image data are retrieved and matched in the dataset to determine the subset of pull-on image data that has the same size as the target garment and is similar to the target user's body shape. :

[0083] ;

[0084] in, represents the Euclidean distance between vectors, Indicates the allowed matching error threshold of the user's body shape parameters;

[0085] The formula screens and matches data from a library of real-world pull-on effects, ensuring that the selected image data not only matches the garment size, but also that the robot's body shape resembles the user's. The resulting set of realistic pull-on images "matches the user's body shape and garment size," laying a reliable data foundation for subsequent effect transfer. Body shape differences are measured using Euclidean distance, and an error threshold is set to ensure flexibility while maintaining precision in the matching process. This parameterized selection method is a key step that distinguishes this invention from traditional "visual fit" methods. This approach achieves a "visible fit," thus supporting size selection.

[0086] S32. The pull-through image data subset Comprehensively evaluate the structural feature area, base color RGB data, and actual wear wrinkle morphology to determine the optimal pull-on image data As a benchmark for effect migration:

[0087] ;

[0088] in, 、 are weight coefficients, representing the importance of body shape similarity and image quality, Indicates pull-through image data The image quality evaluation index of the real wearing wrinkle morphology. The smaller the index value, the higher the quality of the pull-on image.

[0089] A formula is used to select the optimal transfer benchmark from candidate images, taking into account both the user's body shape and the robot model's proximity, as well as image quality, to ensure the transfer result is as perceptually close to a real-life try-on. The final image will serve as the "grayscale reference template" for this transfer, accounting for spatial fit, wrinkles, and fabric wrapping. This process essentially involves solving a multi-objective optimization problem, selecting the optimal template while maintaining body shape similarity while also taking into account image quality (background clarity, garment edge integrity). This decision-making logic ensures a balance between aesthetics and authenticity in the transfer result, which is key to enhancing the user's try-on experience.

[0090] This implementation introduces a two-way matching mechanism between size parameters and body shape parameters, enabling accurate screening of reference images with a high degree of fit matching from a real pull-on image library, significantly improving the authenticity and personalization of virtual try-on renderings in terms of size fit and body shape restoration.

[0091] In this embodiment, S4 includes the following steps:

[0092] S41. Benchmark chart for effect migration Execute data comparison with the target garment to be virtually tried on Consistent clothing feature area mask segmentation to obtain clothing feature area mask image of effect migration benchmark ;

[0093] In the mask segmentation part of the clothing feature area, this implementation method first calls templates A and B, identifies the base colors of their respective feature color blocks and records the RGB data. The recognition technology calls open source AI: mmsegmentation is an open source semantic segmentation library based on PyTorch, which belongs to one of the OpenMMLab series of projects. It is not a dedicated open source clothing model, but it can be applied to clothing-related image segmentation tasks.

[0094] S42. Clothing feature region mask image based on effect transfer benchmark image , respectively extract the base color RGB data of each clothing feature area in the effect migration benchmark image to form a base color RGB data set ;

[0095] S43. Base color RGB data set for each clothing feature area of the effect migration benchmark image Decolorization is performed by inverting the color coordinates to obtain an intermediate grayscale image that retains the original folds and spatial structure. :

[0096] ;

[0097] in, The coordinates in the effect migration benchmark diagram are The original pixel value, Indicates the corresponding coordinates Feature area The base color RGB data, Indicates taking the absolute value of the color difference between each RGB channel.

[0098] This embodiment performs structural region segmentation and base color inversion processing on the effect migration reference image to generate an intermediate grayscale image that retains the real wrinkles and spatial fitting form, thereby achieving the decoupling of the real wearing and wrapping effect of the clothing and the target clothing style.

[0099] In this embodiment, S5 includes the following steps:

[0100] S51. Data pair for virtual try-on of target clothing The base color RGB data set of each clothing feature area Perform color coordinate inversion processing to form an inverted base color RGB data set ;

[0101] S52. The inverted primary color RGB data set With the middle grayscale image Perform pixel-by-pixel regional fusion processing to form a synthetic inverted image , the synthetic inverted image By integrating the color style and effect of the target garment, the real wearing spatial structure information in the benchmark image is transferred to fully express the real wearing folds and fitting spatial effects of the target garment:

[0102] ;

[0103] in, Indicates that the coordinates in the synthetic inverted image are The pixel value at Represents an intermediate grayscale image The median coordinate is The pixel value of Indicates the corresponding coordinates of the target garment to be virtually tried on Feature area The inverted base color RGB data.

[0104] This implementation incorporates a color coordinate inversion strategy to separate the color features of the target garment image from the structural and texture information of the worn image. By desaturating the real-world garment image, only the wrinkles and wrapping structure are retained. The target garment's color information is then superimposed onto the grayscale image to create a synthetic image of the new garment in its true-to-life form. This avoids the loss of lighting, shadows, and fabric detail associated with traditional direct mapping, accurately reproducing the garment's wrapping, tightness, and spatial layering on different body types.

[0105] In this embodiment, S6 includes the following steps:

[0106] S61. Synthesize inverted image Perform color coordinate restoration processing on all pixel points in the image, convert the inverted color information back to the real display color value, and generate a virtual try-on effect diagram of the target clothing on the target body shape. ;

[0107] S62. The virtual try-on effect diagram The visual effects of each characteristic area of the target clothing meet the following conditions:

[0108] The color feature is consistent with the base color of the corresponding area in the target clothing template image;

[0109] The spatial wrapping feature completely corresponds to the grayscale of the folds of the same-sized garment on a similar body shape in the effect transfer benchmark image;

[0110] Fit expression based on size vector User body parameters If the following conditions are met, the virtual try-on effect diagram Rendered in a fitted visual state:

[0111] ;

[0112] ;

[0113] ;

[0114] in, They are the target clothing bust, waist and hip circumference respectively. They are the user's chest circumference, waist circumference, and hip circumference respectively. Error thresholds set for experience;

[0115] If any of the above conditions are not met, then The corresponding area in the middle is shown as loose or tight.

[0116] This implementation method unifies the modeling of clothing template images, size parameters, structural masks, and body shape parameters to form a standardized input format, and combines the fusion and inverse transformation processing of real wearing images to construct a complete image generation chain. The chain does not rely on three-dimensional modeling and complex physical simulation. Without changing the fitting environment, it can automatically output virtual fitting renderings with spatial realism and style consistency. It is suitable for e-commerce, school uniform customization, remote size selection, and many other scenarios, greatly improving the response speed and large-scale deployment capabilities of virtual fitting.

[0117] Example 1:

[0118] The following is a specific embodiment of the present invention, used to illustrate the feasibility and effectiveness of the virtual fitting method for clothing using limited AIGC to achieve real pull-on effect transfer in actual scenarios:

[0119] In October 2024, Primary School A and Primary School B jointly launched a pilot program for an AI-based remote school uniform size selection and fitting system. The goal is to help students and parents complete the fitting experience and size selection of school uniforms of different styles and sizes online without offline fittings.

[0120] The user scenario is as follows: The parent of a fifth-grade student at Elementary School A submitted a fitting request through the school's "showme-AI virtual fitting platform." The user uploaded an image of a target uniform template provided by the school (i.e., Elementary School A 2024 Spring Uniform Style 002) and entered the student's current standard body measurements: chest circumference 56cm, waist circumference 53cm, hip circumference 60cm, shoulder width 30cm, and height 135cm. The user requested to view virtual fitting images of the student's body in different sizes (130, 140, and 150) so they could select the appropriate size.

[0121] After receiving the request, the system automatically identified the target garment size in the template image as size 140, with a size vector of (chest 58cm, waist 54cm, hip 62cm, length 52cm, sleeve length 48cm). The system then performed a reverse match search using a library of real-world pull-on effects. The system detected that the structure of the B Elementary School 2024 Spring School Uniform, Style 001, matched the target uniform and that real-world pull-on image data for all sizes (120-160) existed. The library also contained images of the size 140 garment pulled on by robots of various body types.

[0122] The system automatically selects a real-world pull-on image Bi of a "robot size 17" (body parameters: chest 56cm, waist 53cm, hip 60cm, shoulder width 30cm, height 135cm) as the reference image for effect transfer. This image fully captures the wrinkles, fabric wrapping, and spatial fit of the size 140 sample garment on the target body type. The system then uses the mmsegmentation model to perform feature region segmentation and base color extraction on the target uniform template and the reference image. The segmented structural regions include the collar, sleeves, chest, waist, and hem, and their corresponding base color RGB data is identified. The system then performs color inversion and region fusion operations to merge the target uniform style with the reference wearing form, restoring color coordinates and ultimately generating a virtual try-on image Ai of the target uniform size 140 on the target body type.

[0123] Taking the A Elementary School uniform, style 002, as an example, the method presented in this paper, using only the template image before the uniform is completed, leverages existing real-life data from the B Elementary School uniform, style 001, with the same structure and size. This enables the transfer of realistic cross-brand sample fit. The resulting virtual fit image Ai shows a high degree of consistency in wrinkles, stretch, and spatial fit with the actual fit image Bi. Comparative scoring by image experts shows over 90% consistency between the two images in terms of realistic wrapping and light and shadow reproduction.

[0124] In addition, data collected through feedback from parent volunteers showed that 94 out of the 100 parents participating in this pilot program made clear size selection decisions after viewing the multi-size virtual try-on images provided by the system; 87% of users reported that "the degree of looseness or tightness can be clearly seen in the renderings." Previously, in the try-on system that used traditional virtual mapping, users mainly focused on the color and style of clothing, and it was difficult to judge the degree of fit.

[0125] The system then provided virtual try-on images for sizes 130 and 150, highlighting key areas (sleeves that were too short, hems that were too long, and tight shoulder pleats) to help users more intuitively compare how different sizes performed. Ultimately, the student's parents selected size 140 as their order, and subsequent offline try-on feedback showed that the fit was highly consistent with the virtual try-on images.

[0126] The examples demonstrate the practical application capabilities of the method of the present invention in school uniform customization and e-commerce size selection scenarios, solving the problem that traditional methods cannot display the true fit. It has the significant advantages of high precision, strong adaptability, fast response, and low resource dependence, and can effectively bring experience improvement and business value optimization to the virtual fitting scenario of clothing.

[0127] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A virtual fitting method for clothing using limited AIGC to achieve real pull-on effect transfer, characterized in that: The steps include: S1. Collect and store a variety of clothing samples and body data of the fitting robot, through the fitting robot real wear clothing samples of various sizes and shoot the formation of wear image data, the wear image data is stored in a structured manner in the real wear effect gallery; S2. For the target garment to be tried on virtually, shoot a target garment template image in a flat, wrinkle-free state, bind the target garment template image to the target garment size to form a data pair for the target garment to be tried on virtually; S3. In the real pull-on effect image library, based on the size parameters of the target garment and the user's body parameters, search and select matching pull-on image data, extract a set of pull-on images that are the same size as the target garment and similar to the target user's body type, and determine the effect migration reference image; S4. Using the semantic segmentation model, the effect transfer benchmark image is segmented and base color extracted, consistent with the data of the target garment to be virtually tried on, to obtain an intermediate grayscale image that retains the original wrinkles and spatial structure. S5. The color coordinates of the primary color RGB data of the target clothing to be virtually tried on are inverted, and the region is fused with the intermediate grayscale image to obtain a synthetic inverted image; S6. Perform color coordinate restoration processing on the synthesized inverted image to generate a virtual try-on effect image of the target clothing on the target user's body shape.

2. The method for virtual fitting of clothing using limited AIGC to achieve real pull-on effect transfer according to claim 1, characterized in that: Said S1 comprises the following steps: S11. Get the basic size parameter information of the clothing sample, including the chest circumference of the sample , sample waist , sample hip circumference , sample length Sleeve length , forming the clothing size vector And each size sample is individually numbered and bound with parameters to form a multi-size sample set; S12. Set up a standard fitting robot and collect the standard fitting robot's body parameter information, including the standard fitting robot's chest circumference , Standard waist circumference of fitting robot , Standard fitting robot hip circumference , Standard fitting robot shoulder width and the height of a standard fitting robot , forming the robot body parameter vector The body parameter information of the standard fitting robot corresponds to a plurality of standard fitting robot models constructed by simulating the body shapes of real people; S13. Each size sample is put on a standard fitting robot of different body types, and the fitting effect is photographed from multiple angles using uniform lighting conditions and background settings to generate an image set. ,in Indicates the sample number. Indicates the standard fitting robot number, Represents clothing size vector Put on the robot body parameter vector The image on the image is formed into a clothing size vector and the robot body parameter vector Pull-through image data generated by the combination of ; S14. Put all the image data into Store in the real pull-on effect gallery to form a real pull-on effect gallery : ； in, is the total number of garment samples, This is the number of standard fitting robot models.

3. The method for virtual fitting of clothing using limited AIGC to achieve real pull-on effect transfer according to claim 2, characterized in that: The S2 comprises the following steps: S21. Spread the target garment to be virtually tried on flat on the shooting platform so that the target garment to be virtually tried on is in a wrinkle-free shooting state. Use a standard light source and a fixed shooting angle to obtain a template image of the target garment. ; S22. Measure the actual size of the target garment to obtain the size parameters of the target garment, including the chest circumference of the target garment. Target clothing waistline Target clothing hip circumference Target clothing length And target clothing sleeve length , forming the target clothing size vector ; S23. The target clothing template image with the target clothing size vector Binding to form the data pair of the target clothing to be virtually tried on .

4. The method for virtual fitting of clothing using limited AIGC to achieve real pull-on effect transfer according to claim 3, characterized in that: The S3 includes the following steps: S31. Based on the data of the target clothing to be virtually tried on Target clothing size vector in and the user's body parameter vector , in the real pull-on effect gallery The pull-on image data are retrieved and matched in the dataset to determine the subset of pull-on image data that has the same size as the target garment and is similar to the target user's body shape. : ； in, represents the Euclidean distance between vectors, Indicates the allowed matching error threshold of the user's body shape parameters; S32. The pull-through image data subset Comprehensively evaluate the structural feature area, base color RGB data, and actual wear wrinkle morphology to determine the optimal pull-on image data As a benchmark for effect migration: ； in, 、 are weight coefficients, representing the importance of body shape similarity and image quality, Indicates pull-through image data The image quality evaluation index of the real wearing wrinkle morphology. The smaller the index value, the higher the quality of the pull-on image.

5. The method for virtual fitting of clothing using limited AIGC to achieve real pull-on effect transfer according to claim 4, characterized in that: The S4 comprises the following steps: S41. Using semantic segmentation model to transfer effect benchmark Execute data comparison with the target garment to be virtually tried on Consistent clothing feature area mask segmentation to obtain clothing feature area mask image of effect migration benchmark ; S42. Clothing feature region mask image based on effect transfer benchmark image , respectively extract the base color RGB data of each clothing feature area in the effect migration benchmark image to form a base color RGB data set ; S43. Base color RGB data set for each clothing feature area of the effect migration benchmark image Decolorization is performed by inverting the color coordinates to obtain an intermediate grayscale image that retains the original folds and spatial structure. : ； in, Represents an intermediate grayscale image The median coordinate is The pixel value of The coordinates in the effect migration benchmark diagram are The original pixel value, Indicates the corresponding coordinates Feature area The base color RGB data, Indicates taking the absolute value of the color difference between each RGB channel.

6. The method for virtual fitting of clothing using limited AIGC to achieve real pull-on effect transfer according to claim 5, characterized in that: The S5 comprises the following steps: S51. Data pair for virtual try-on of target clothing The base color RGB data set of each clothing feature area Perform color coordinate inversion processing to form an inverted base color RGB data set ; S52. The inverted primary color RGB data set With the middle grayscale image Perform pixel-by-pixel regional fusion processing to form a synthetic inverted image , the synthetic inverted image By integrating the color style and effect of the target garment, the real wearing spatial structure information in the benchmark image is transferred to fully express the real wearing folds and fitting spatial effects of the target garment: ； in, Indicates that the coordinates in the synthetic inverted image are The pixel value at Indicates the corresponding coordinates of the target garment to be virtually tried on Feature area The inverted base color RGB data.

7. The method for virtual fitting of clothing using limited AIGC to achieve real pull-on effect transfer according to claim 6, characterized in that: The S6 comprises the following steps: S61. Synthesize inverted image Perform color coordinate restoration processing on all pixel points in the image, convert the inverted color information back to the real display color value, and generate a virtual try-on effect diagram of the target clothing on the target user's body shape. ; S62. The virtual try-on effect diagram The visual effects of each characteristic area of the target clothing meet the following conditions: The color feature is consistent with the base color of the corresponding area in the target clothing template image; The spatial wrapping feature completely corresponds to the grayscale of the folds of the same-sized garment on a similar body shape in the effect transfer benchmark image; Fit expression based on size vector User body parameters The relative difference between the two, if the following conditions are met, in the virtual try-on effect diagram It is presented in a fitted visual state: ； ； ； in, They are the target clothing bust, waist and hip circumference respectively. They are the user's chest, waist and hip circumference respectively. Error thresholds set for experience; If any of the above conditions are not met, then The corresponding area in the middle is shown as loose or tight.

Images