Virtual clothes fitting effect image generation method and system

Abstract

This invention discloses a method and system for generating virtual clothing try-on effect images, comprising: acquiring a full-body photo of a user standing upright and a flat image of the clothing to be tried on; preprocessing the photo and image, and performing background stripping on the flat image of the clothing to be tried on; outputting a human body template image based on the preprocessed full-body photo of the user standing upright; and extracting key point coordinates and contour curvature based on the human body template image. This invention relates to the field of virtual image data processing technology. This method and system for generating virtual clothing try-on effect images optimizes the processing of user photos and try-on clothing images, calculates and corrects errors, achieves accurate segmentation of the torso, limb contours, and the original clothing area, outputs a high-definition human body template image without redundancy, accurately extracts key body points and core body shape parameters such as shoulder width and chest circumference, and generates a three-dimensional human body model that highly adapts to the user's real body shape.

Description

Technical Field

[0001] This invention relates to the field of virtual image data processing technology, specifically to a method and system for generating virtual clothing try-on effect images. Background Technology

[0002] To enhance the online clothing shopping experience, a corresponding method for generating virtual clothing try-on images is needed to provide users with a virtual clothing try-on experience. Chinese patent number "CN119359972B" discloses a virtual try-on method, virtual try-on device, electronic device, and storage medium, including: acquiring the current user image and the current clothing image to be tried on; generating a current 3D human body mask based on the current user image. As shown in the aforementioned patent, existing methods and systems for generating virtual clothing try-on images have the following shortcomings: First, they often use a single semantic segmentation model, resulting in insufficient accuracy in extracting the contours of the torso and limbs, and significant residual segmentation of the original clothing area, leading to problems such as blurred edges and redundant pixels in the generated human body template image; second, the pose estimation and body modeling stages are disconnected, only extracting basic body key point coordinates, and the calculation of core body shape parameters such as shoulder width and chest circumference is relatively coarse, resulting in a low degree of fit between the generated personalized 3D human body model and the user's real body shape, leading to poor fit during try-on. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method and system for generating virtual clothing try-on effect images, which solves problems such as low extraction accuracy of torso and limb contours, obvious residual parts of the original clothing area segmentation, coarse calculation of body shape parameters, and poor fit between the three-dimensional human body model and the real body shape.

[0004] To achieve the above objectives, the present invention provides a method for generating virtual clothing try-on effect images, specifically comprising the following steps:

[0005] S1. Obtain a full-body photo of the user standing facing forward and a flat image of the clothing to be tried on. Preprocess the photo and image, and perform background stripping on the flat image of the clothing to be tried on.

[0006] S2. Based on the preprocessed full-body photo of the user standing upright, output a human template image;

[0007] S3. Extract key point coordinates and contour curvature based on human body template images;

[0008] S4. Calculate the body shape parameters based on the coordinates of the key points;

[0009] S5. Correct the perspective error and measurement error based on the body shape parameters respectively;

[0010] S6. Construct a 3D model of the user's human body based on the contour curvature, corrected perspective parameters, and measurement parameters;

[0011] S7. Generate an image of the user wearing the clothing based on the user's 3D human body model and the tiled image of the clothing to be tried on;

[0012] S8. Extract light and shadow features and texture features from the human template image and the human clothing image respectively, generate a seamlessly synthesized virtual try-on image and output it;

[0013] S9. While supporting multi-angle rotation display and quick switching between multiple garments, display virtual try-on images and score the displayed virtual try-on effect images through a comprehensive scoring mechanism.

[0014] Preferably, in step S2, the human body template image is obtained through semantic segmentation, edge refinement, and removal of residual clothing areas. The semantic segmentation is used to achieve four types of semantic segmentation: torso, limbs, original clothing, and background, and outputs a semantic segmentation map.

[0015] The edge refinement is used for output. 1. Limb outline diagram;

[0016] The clothing area cleaning process removes clothing residue and outputs a human body template image with no redundant pixels.

[0017] Preferably, in step S4, the shoulder width, chest circumference, waist circumference, and hip circumference in the human body template are calculated based on the coordinates of 24 key points in the human body template image, and in step S5, a linear regression calibration model is constructed and the perspective error and measurement error of the initial parameters are corrected.

[0018] Preferably, in step S1, a model for preprocessing the flat lay image of clothing is constructed; in step S7, the preprocessed flat lay image of clothing is mapped onto a 3D human body model using a bonding formula to obtain a human body clothing wearing image; and in step S8, a clothing wrinkle strength model and a clothing wrinkle morphology model are constructed based on the human body clothing wearing image and the elasticity and stiffness of the clothing fabric, respectively, and a clothing deformation image is output.

[0019] Preferably, in step S6, the process of constructing the human 3D model is as follows: constructing a training dataset, designing a loss function, and iteratively solving for parameters, wherein the training dataset is constructed by collecting human samples covering different body types, postures, and shooting angles and completing data preprocessing.

[0020] Preferably, in step S8, the light and shadow alignment loss and texture preservation loss are derived from the extracted light and shadow features and texture features, and a mathematical model of the fusion loss function is constructed based on the light and shadow alignment loss and texture preservation loss.

[0021] Preferably, in step S8, the light and shadow features and texture features of the human template image and the human clothing image are extracted and seamlessly synthesized into a virtual try-on image using a fusion loss function.

[0022] Preferably, the mathematical model of the comprehensive scoring mechanism in S9 is:

[0023] ,

[0024] In the formula: For comprehensive scoring, Rate user subjective satisfaction. To score the objective results, , , The weighting coefficient representing user subjective satisfaction. The weighting coefficients representing the objective effect score. and The constraint relationships are as follows:

[0025] This invention also discloses a virtual clothing try-on effect image generation system for implementing the virtual clothing try-on effect image generation method described above. The system includes an image acquisition module, a preprocessing module, a segmentation module, a human body modeling module, a clothing deformation module, an image fusion module, an effect output module, and an interaction module, wherein:

[0026] The image acquisition module is used to acquire full-body photos of the user standing upright and flat images of their clothing.

[0027] The preprocessing module is used to preprocess a full-body standing photo of a user and to perform background stripping processing based on a flat lay image of clothing.

[0028] The segmentation module is used to process a preprocessed full-body photo of a user standing upright and outputs a human template image;

[0029] The human body modeling module is used to extract key point coordinates and contour curvature based on human body template images, calculate body shape parameters based on key point coordinates, correct perspective and measurement errors based on the calculated body shape parameters, and construct the user's human body 3D model based on contour curvature, corrected perspective parameters, and measurement parameters.

[0030] The clothing deformation module is used to transform pre-processed flat images of clothing into images that fit the three-dimensional human body model, based on human body template images.

[0031] The image fusion module performs fusion processing based on human template images and clothing wearing images to output seamlessly synthesized virtual try-on images;

[0032] The effect output module is used to display virtual try-on images based on seamless synthesis;

[0033] The interactive module is used to generate a comprehensive score based on the displayed virtual try-on images.

[0034] Preferably, the system is compatible with the deployment of smart terminals, VR devices, and cloud servers, and supports rapid switching between multiple garments.

[0035] Beneficial effects

[0036] This invention provides a method and system for generating virtual clothing try-on effect images. Compared with existing technologies, it has the following advantages:

[0037] (1) The virtual clothing try-on effect image generation method and system optimizes the preprocessing process of user photos and try-on clothing images, and combines calculation and error correction to achieve accurate segmentation of the torso, limb contours and the original clothing area, outputting a human body template image without redundancy and with high definition. At the same time, it can accurately extract key body points and core body shape parameters such as shoulder width and chest circumference to generate a personalized human body three-dimensional model that is highly adapted to the user's real body shape. It also integrates perspective transformation and physical-driven wrinkle simulation technology to finally generate a clothing wearing form that fits the user's body shape and is natural and realistic.

[0038] (2) The virtual clothing try-on effect image generation method and system maps the pre-processed clothing tiled image onto the surface of the three-dimensional human body model. After deformation and seamless fusion processing, the clothing wearing image and the light and shadow and texture in the three-dimensional human body model image are aligned, realizing the seamless fusion of clothing and human body, ensuring the integrity of texture details and the consistency of light and shadow environment, and improving the realism of try-on.

[0039] (3) The virtual clothing try-on effect image generation method and system, by setting a comprehensive evaluation mechanism of subjective scoring and objective indicators, supports multi-angle rotation display of try-on images and quick switching of multiple clothing, which not only makes it easier for users to intuitively observe the try-on effect, but also improves the accuracy of the evaluation and optimizes the user interaction experience. Attached Figure Description

[0040] Figure 1 This is a flowchart of the steps of the present invention;

[0041] Figure 2 This is the main principle block diagram of the system of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] refer to Figures 1-2 The present invention provides the following three technical solutions:

[0044] The first implementation method: a method for generating virtual clothing try-on effect images, specifically including the following steps:

[0045] S1. Obtain a full-body photo of the user standing in front and a flat image of the clothing to be tried on, and preprocess the photo and image respectively. The specific preprocessing operations are as follows: standardize the size, equalize the brightness and denoise the full-body photo of the user standing in front; perform background stripping on the flat image of the clothing to be tried on in order to purify the core information of the clothing and eliminate image interference, so as to lay the foundation for the subsequent virtual try-on process.

[0046] S2. Based on the preprocessed full-body photo of the user standing upright, semantic segmentation, edge refinement, and removal of residual clothing areas are performed sequentially to output a human body template image. The model formula used for semantic segmentation is as follows:

[0047] ,

[0048] In the formula: For semantic segmentation graphs, For channel attention module, This is a two-dimensional convolution operation. Softmax is the classification activation function used to achieve pixel-level class probability output. I represents a full-body photo of the user standing upright. It is a U-shaped structure consisting of an encoder, a bottleneck layer, and a decoder. The decoder is used to extract human body regions and clothing textures. The bottleneck layer is the connecting unit between the encoder and the decoder. The decoder is used for human body segmentation and seamless integration of clothing and human body.

[0049] The edge strength calculation model is as follows:

[0050] ,

[0051] In the formula: This is a contour edge map. For k-layer edge convolution operations, Here, K represents the edge convolution weights of the k-th layer, and K is the number of edge convolution layers. for Activation function, used to normalize edge intensity to interval;

[0052] The model formula used for the human body template image is:

[0053] ,

[0054] Human body template image, For the final segmentation result, This is a contour edge map. For non-clothing areas, "in " indicates element-wise multiplication;

[0055] S3. Extract the coordinates of 24 key points from the human body template image. The output of the coordinates of the 24 key points in the human body template image is as follows:

[0056] ,

[0057] ,

[0058] Where: P is the set of coordinates of 24 key points. For the i-th key point, Let x be the x-coordinate of the i-th key point. Let y be the ordinate of the i-th key point;

[0059] S4. Calculate body shape parameters based on key point coordinates. The calculation of body shape parameters in the human body template is as follows:

[0060] ,

[0061] ,

[0062] ,

[0063] ,

[0064] ,

[0065] In the formula: Shoulder width, For bust measurement, Waist circumference, Hip circumference, The Euclidean distance between the two points. , , , Perspective correction factor for a single photograph;

[0066] S5. Correction of perspective and measurement errors based on body shape parameters. The linear regression calibration model formula for perspective and measurement errors of body shape parameters is as follows:

[0067] ,

[0068] ,

[0069] ,

[0070] In the formula: The calibrated parameter vector, This is the initial body shape parameter vector. To calibrate the weight matrix, b is the bias term. Human body template image, Shoulder width, For bust measurement, Waist circumference, Hip circumference, These are the calibrated shoulder width, calibrated chest circumference, calibrated waist circumference, and calibrated hip circumference, respectively.

[0071] S6. Extract contour curvature from human template images, and construct a 3D human body model based on contour curvature, corrected perspective parameters, and measurement parameters. The construction process of the 3D human body model is as follows: construct a training dataset, design a loss function, and iteratively solve the parameters. The training dataset is constructed by collecting human body samples covering different body types, poses, and shooting angles and completing data preprocessing.

[0072] The formula for the loss function is:

[0073] ,

[0074] In the formula: Let be the initial parameter vector for the j-th sample. Let j be the true parameter vector of the j-th sample. It is an L2 norm;

[0075] The update formula for iteratively solving the parameters is:

[0076] ,

[0077] ,

[0078] In the formula: =0.001, Let W be the partial derivative. Let W be the partial derivative of b, W be the calibration weight matrix, and b be the bias term vector. W and b are updated using the gradient descent algorithm, iterating until the loss function converges.

[0079] The formula for constructing a 3D human body model is:

[0080] ,

[0081] For a personalized 3D human body model, The calibrated parameter vector, The body contour features are extracted from the human template image T. This is a lightweight 3D modeling function used to fuse 2D contour features with 3D body shape parameters to generate a lightweight 3D human body model.

[0082] S7. Based on the user's 3D human body model, the preprocessed clothing lay-up image is mapped onto the surface of the human body model. The clothing image is then subjected to shape transformation and seamless fusion processing to generate a clothing wearing image that fits the human body model.

[0083] The model formula for preprocessing garment flat lay images is:

[0084] ,

[0085] In the formula: This is a preprocessing function used to remove the background and extract clothing information. To input a flat lay image of the garment, For the outline of the clothing, Features of clothing texture;

[0086] The formula for the fit model of a garment flat lay image mapped onto the human body surface through perspective transformation is:

[0087] ,

[0088] In the formula: For perspective transformation operation, for Perspective transformation matrix, Images of clothing designed to fit the human body;

[0089] The models for garment fold strength and shape are as follows:

[0090] ,

[0091] ,

[0092] In the formula: , , The weighting coefficients are satisfied. , For fold strength, The pressure of contact between the human body and clothing. To enhance the elasticity of clothing fabrics. For the bending stiffness of clothing fabrics. Images of deformed clothing. The wrinkled texture generated by the algorithm, The clothing image is designed to fit the surface of the human body. This is a texture overlay operation used to integrate natural wrinkles into clothing images that fit the human body surface, achieving precise segmentation of the torso, limb contours and the original clothing area, outputting a high-definition human body template image without redundancy, accurately extracting key body points and core body shape parameters such as shoulder width and chest circumference, generating a personalized 3D human body model that is highly adapted to the user's real body shape, and integrating perspective transformation and physically driven wrinkle simulation technology to generate a natural and realistic clothing wearing form that fits the user's body shape.

[0093] The second implementation differs from the first in that it adds step S8, which extracts lighting and texture features from the human template image and the clothing image, respectively. The optimization process of the network parameters is constrained by a fusion loss function, outputting a seamlessly synthesized virtual try-on image. The fusion loss function consists of a lighting alignment loss function and a texture preservation loss function, used to ensure the lighting consistency and texture integrity of the fused image. The mathematical model of the fusion loss function is as follows:

[0094] ,

[0095] In the formula: To preserve the weight coefficients for the texture, take This is used to balance the weight of light and shadow on its preservation of texture. This represents the light and shadow alignment loss value. Preserve the loss value for the texture. , The calculation formula is:

[0096] ,

[0097] ,

[0098] In the formula: It is the L2 norm. To integrate the texture features of clothing in the output image, The light and shadow feature vector of the human body template image. The light and shadow feature vector of the deformed clothing image. The texture feature vector of the deformed clothing image;

[0099] right , , Extract the data using the following formula:

[0100] ,

[0101] ,

[0102] In the formula: For feature extraction function, The texture feature vector of the human template image. Human body template image, Images of deformed clothing;

[0103] The calculation process is as follows: The clothing region is segmented from the fused output image to obtain a pure clothing local image, and then... Logically consistent LBP operators (Local Binary Pattern, LBP), 2D convolution, and normalization operations are used to extract initial texture features, which are then normalized to... The interval is adjusted to 256 dimensions to obtain... The mathematical model for seamlessly synthesized virtual try-on images is:

[0104] ,

[0105] ,

[0106] In the formula: For fusion function, For network training parameters, To seamlessly synthesize virtual try-on images, Images of deformed clothing. Human body template image, This is the core weight matrix of the network's convolutional layers. The convolutional layer bias term vector is initialized to... , ,in Gradient calculation is based on total loss. Differentiate, , b is the bias term vector. For the total loss, weighted The partial derivatives, The total loss is relative to the bias term. The partial derivatives; the convergence criterion is ,in, By aligning the light and shadow and texture in the clothing wearing image with the human body 3D model image, a seamless integration of clothing and human body is achieved, ensuring the integrity of texture details and the consistency of light and shadow environment, and improving the realism of try-on.

[0107] The third implementation differs from the second in that it adds step S9. While supporting multi-angle rotation and rapid switching between multiple garments, it displays the virtual try-on image and scores it using a comprehensive scoring mechanism. When the score is ≥85, the virtual try-on image is output as the final try-on image. The mathematical model for the comprehensive scoring mechanism is as follows:

[0108] ,

[0109] In the formula: For comprehensive scoring, Rate user subjective satisfaction. To score the objective results, , The objective effect score is calculated based on edge transition error, texture clarity, and light and shadow consistency indicators, with edge transition error ≤ 1 pixel;

[0110] The formula for multi-angle rotation display is: In the formula: To preset key angles ( , For lightweight projection functions, For a personalized 3D human body model, For the projected image at the corresponding angle, the rotation delay is ≤200ms;

[0111] The formula for quickly switching between multiple outfits is: In the formula: Flat lay image of the new clothing. The system generates composite images of new clothing try-on with a switching latency of ≤300ms. While achieving the output of try-on effects, it supports multi-angle rotation display and quick switching between multiple clothing items, making it easier for users to observe and improving the accuracy of evaluations. It is also compatible with the needs of multiple terminals of Softcom.

[0112] This invention also discloses a virtual clothing try-on effect image generation system, including an image acquisition module, a preprocessing module, a segmentation module, a human body modeling module, a clothing deformation module, an image fusion module, an effect output module, and an interaction module. The image acquisition module is used to acquire a full-body photo of the user standing upright and a flat image of the clothing. The preprocessing module is used to perform size standardization, brightness equalization, and noise reduction based on the full-body photo of the user standing upright, and to perform background stripping processing based on the flat image of the clothing.

[0113] The segmentation module performs semantic segmentation, edge refinement, and removal of residual clothing areas based on a preprocessed full-body photo of the user standing upright, and outputs a human body template image. The human body modeling module extracts key point coordinates and contour curvature from the human body template image, calculates body shape parameters based on the key point coordinates, corrects perspective and measurement errors based on the calculated body shape parameters, and constructs a 3D human body model of the user based on the contour curvature, corrected perspective parameters, and measurement parameters. The clothing deformation module transforms a preprocessed flat-lay image of clothing into a wearing shape image that fits the 3D human body model based on the human body template image.

[0114] The image fusion module fuses human template images and clothing wearing images to output seamlessly synthesized virtual try-on images; the effect output module displays the seamlessly synthesized virtual try-on images; the interaction module performs a comprehensive score based on the displayed virtual try-on images, enabling interaction and working with the effect output module to output qualified virtual try-on images. The system is adaptable to lightweight deployment on multiple terminals such as smart terminals, VR devices, and cloud servers. It combines user subjective scores with objective effect indicators to comprehensively evaluate the try-on effect and support effect optimization and iteration.

[0115] Furthermore, all content not described in detail in this specification is existing technology known to those skilled in the art, and the model parameters of each electrical appliance are not specifically limited; conventional equipment can be used.

[0116] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0117] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for generating virtual clothing try-on effect images, characterized in that, Specifically, the following steps are included: S1. Obtain a full-body photo of the user standing facing forward and a flat image of the clothing to be tried on. Preprocess the photo and image, and perform background stripping on the flat image of the clothing to be tried on. S2. Based on the preprocessed full-body photo of the user standing upright, output a human template image; S3. Extract key point coordinates and contour curvature based on human body template images; S4. Calculate the body shape parameters based on the coordinates of the key points; S5. Correct the perspective error and measurement error based on the body shape parameters respectively; S6. Construct a 3D model of the user's human body based on the contour curvature, corrected perspective parameters, and measurement parameters; S7. Generate an image of the user wearing the clothing based on the user's 3D human body model and the tiled image of the clothing to be tried on; S8. Extract light and shadow features and texture features from the human template image and the human clothing image respectively, generate a seamlessly synthesized virtual try-on image and output it; S9. While supporting multi-angle rotation display and quick switching between multiple garments, display virtual try-on images and score the displayed virtual try-on effect images through a comprehensive scoring mechanism.

2. The method for generating virtual clothing try-on effect images according to claim 1, characterized in that, In S2, the human body template image is obtained through semantic segmentation, edge refinement, and removal of residual clothing areas. The semantic segmentation is used to achieve four types of semantic segmentation: torso, limbs, original clothing, and background, and outputs a semantic segmentation map. The edge refinement is used to output the outline edge map of the torso and limbs; The clothing area cleaning process removes clothing residue and outputs a human body template image with no redundant pixels.

3. The method for generating virtual clothing try-on effect images according to claim 1, characterized in that, In step S4, the shoulder width, chest circumference, waist circumference, and hip circumference in the human body template are calculated based on the coordinates of 24 key points in the human body template image. In step S5, a linear regression calibration model is constructed and the perspective error and measurement error of the initial parameters are corrected.

4. The method for generating virtual clothing try-on effect images according to claim 1, characterized in that, In step S1, a model for preprocessing the flat lay image of clothing is constructed. In step S7, the preprocessed flat lay image of clothing is mapped onto the three-dimensional human body model using a bonding formula to obtain a human body clothing wearing image. In step S8, a clothing wrinkle strength model and a clothing wrinkle shape model are constructed based on the human body clothing wearing image and the elasticity and stiffness of the clothing fabric, respectively, and a clothing deformation image is output.

5. The method for generating virtual clothing try-on effect images according to claim 1, characterized in that, In S6, the construction process of the human body 3D model is as follows: constructing a training dataset, designing a loss function, and iteratively solving for parameters. The training dataset is constructed by collecting human body samples covering different body types, postures, and shooting angles and completing data preprocessing.

6. The method for generating virtual clothing try-on effect images according to claim 1, characterized in that, In step S8, the light and shadow alignment loss and texture preservation loss are derived by extracting the light and shadow features and texture features, and a mathematical model of the fusion loss function is constructed based on the light and shadow alignment loss and texture preservation loss.

7. The method for generating virtual clothing try-on effect images according to claim 1, characterized in that, In step S8, the light and shadow features and texture features of the human body template image and the human body clothing image are extracted and seamlessly synthesized into a virtual try-on image by using a fusion loss function.

8. The method for generating virtual clothing try-on effect images according to claim 7, characterized in that: The mathematical model for the comprehensive scoring mechanism described in S9 is as follows: ， In the formula: For comprehensive scoring, Rate user subjective satisfaction. To score the objective results, , , The weighting coefficient representing user subjective satisfaction. The weighting coefficients representing the objective effect score. and The constraint relationships are as follows: .

9. A virtual clothing try-on effect image generation system, characterized in that, To implement the virtual clothing try-on effect image generation method as described in any one of claims 1-8, the system includes an image acquisition module, a preprocessing module, a segmentation module, a human body modeling module, a clothing deformation module, an image fusion module, an effect output module, and an interaction module, wherein: The image acquisition module is used to acquire full-body photos of the user standing upright and flat images of their clothing. The preprocessing module is used to preprocess a full-body standing photo of a user and to perform background stripping processing based on a flat lay image of clothing. The segmentation module is used to process a preprocessed full-body photo of a user standing upright and outputs a human template image; The human body modeling module is used to extract key point coordinates and contour curvature based on human body template images, calculate body shape parameters based on key point coordinates, correct perspective and measurement errors based on the calculated body shape parameters, and construct the user's human body 3D model based on contour curvature, corrected perspective parameters, and measurement parameters. The clothing deformation module is used to transform pre-processed flat images of clothing into images that fit the three-dimensional human body model, based on human body template images. The image fusion module performs fusion processing based on human template images and clothing wearing images to output seamlessly synthesized virtual try-on images; The effect output module is used to display virtual try-on images based on seamless synthesis; The interactive module is used to generate a comprehensive score based on the displayed virtual try-on images.

10. The virtual clothing try-on effect image generation system according to claim 9, characterized in that, The system is compatible with smart terminals, VR devices, and cloud servers, and supports rapid switching between multiple outfits.

Images