VAE-based virtual fitting matching system and method

Abstract

The invention relates to the technical field of virtual fitting, particularly provides a VAE-based virtual fitting matching system and method, and aims to solve the problems that a virtual fitting technical scheme in the prior art is relatively time-consuming and the fitting effect lacks a sense of reality. According to the invention, a three-dimensional garment generation module generates three-dimensional garments through a VAE model, a three-dimensional body generation module generates a three-dimensional body model through a body image input by a user, and a garment matching model screens out the three-dimensional garments meeting the dressing preference of the user through dressing preference information input by the user. And the virtual fitting module fits the screened three-dimensional clothes on the three-dimensional body model to complete virtual fitting. The textures on the three-dimensional clothes generated through the VAE model are finer and more vivid, the generation cost is lower, the efficiency is higher, the three-dimensional clothes can be better attached to the three-dimensional body model, the sense of reality of the visual effect is enhanced, and the more real and vivid fitting effect is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of virtual fitting, and in particular relates to a virtual fitting matching system and method based on VAE. Background Art

[0002] Existing virtual fitting technologies mainly include two solutions: the first solution is to create a three-dimensional model of clothing through manual or semi-automatic methods, and match it with the user's body data to achieve fitting in a virtual environment. The second solution is to use image processing and computer vision technology to identify and mark key points of the human body, fit the clothing image to the user's image, and simulate the fitting effect.

[0003] However, both of the above virtual fitting technology solutions have some disadvantages. The traditional 3D modeling method of the first solution requires a lot of manual work and professional skills, which is time-consuming and costly. The image processing-based method of the second solution is usually difficult to handle complex clothing details and textures, and the fitting effect lacks realism. Summary of the invention

[0004] The present invention provides a virtual fitting matching system and method based on VAE, aiming to solve the problem that the technical solution of virtual fitting in the prior art is relatively time-consuming and the fitting effect lacks realism.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] In a first aspect, the present invention provides a virtual fitting and matching system based on VAE, comprising:

[0007] 3D clothing generation module, which generates 3D clothing through VAE model;

[0008] A three-dimensional clothing database storing three-dimensional clothing generated by the three-dimensional clothing generation module;

[0009] A receiving module, receiving body images and clothing preference information input by a user;

[0010] A three-dimensional body generation module analyzes the body image received by the receiving module through a computer vision algorithm to obtain body data; and constructs a three-dimensional body model according to the body data;

[0011] a clothing matching module, which selects three-dimensional clothing that meets the clothing preference information in the three-dimensional clothing database according to the clothing preference information received by the receiving module;

[0012] The virtual fitting module fits the three-dimensional clothing selected by the clothing matching module onto the three-dimensional body model to complete the virtual fitting.

[0013] Further solution: In the three-dimensional clothing generation module, the steps of generating the three-dimensional clothing by using the VAE model are:

[0014] The VAE model receives a three-dimensional texture pattern;

[0015] The VAE model generates a three-dimensional garment with a corresponding texture according to the received three-dimensional texture style.

[0016] Based on the above scheme, the VAE model can optimize the texture and map the optimized texture to the corresponding position on the three-dimensional clothing, thereby making the texture on the three-dimensional clothing generated by the VAE model more real and realistic.

[0017] Further solution: In the three-dimensional clothing generation module, the steps of generating the three-dimensional clothing by using the VAE model are:

[0018] The VAE model receives a two-dimensional texture pattern;

[0019] The VAE model generates a two-dimensional clothing image with a corresponding texture according to the received two-dimensional texture pattern;

[0020] The two-dimensional clothing image generated by the VAE model is reconstructed into three-dimensional clothing through three-dimensional reconstruction technology.

[0021] Based on the above scheme, the VAE model optimizes the texture style on the two-dimensional clothing image, so that when the two-dimensional clothing image is subsequently reconstructed into three-dimensional clothing through three-dimensional reconstruction technology, the texture on the three-dimensional clothing is more fitting and more realistic.

[0022] A further solution: the VAE model includes an encoder and a decoder;

[0023] The encoder is used to extract features of the three-dimensional texture pattern or the two-dimensional texture pattern, and convert the three-dimensional texture pattern and the two-dimensional texture pattern into representations in the spatial domain;

[0024] The decoder is used to reconstruct the three-dimensional texture pattern represented in the spatial domain of the encoder into a three-dimensional garment with a corresponding texture; or, the decoder is used to reconstruct the two-dimensional texture pattern represented in the spatial domain of the encoder into a two-dimensional garment image with a corresponding texture.

[0025] A further solution is: before generating the three-dimensional clothing through the VAE model, the VAE model is first trained;

[0026] Collecting three-dimensional clothing images with different textures to form a three-dimensional clothing image dataset; or collecting two-dimensional clothing images with different textures to form a two-dimensional clothing image dataset;

[0027] Inputting the three-dimensional clothing image dataset or the two-dimensional clothing image dataset into the encoder of the VAE model;

[0028] The encoder extracts features of three-dimensional clothing in the three-dimensional clothing image dataset and converts the three-dimensional clothing into a spatial vector representation in the spatial domain; or, the encoder extracts features of two-dimensional clothing in the two-dimensional clothing image dataset and converts the two-dimensional clothing into a spatial vector representation in the spatial domain;

[0029] The encoder inputs the spatial vector representation of the three-dimensional clothing or the spatial vector representation of the two-dimensional clothing into the decoder;

[0030] The decoder reconstructs the spatial vector representation of the three-dimensional garment into a three-dimensional garment; or, the decoder reconstructs the spatial vector representation of the two-dimensional garment into a two-dimensional garment image in an image domain;

[0031] Complete the training of the VAE model.

[0032] Based on the above scheme, by training the VAE model, the VAE model learns the potential feature representation of the texture, shape and style of the three-dimensional clothing image or the two-dimensional clothing image, continuously optimizes the parameters inside the VAE, and continuously learns to capture the diversity and detail features of clothing, so that the VAE model can reconstruct high-quality three-dimensional clothing images or two-dimensional clothing images.

[0033] A further solution: During the training of the VAE model, the loss functions adopted by the encoder and the decoder include a reconstruction error function and a KL divergence function.

[0034] A further solution: in the three-dimensional body generation module, the body data includes skin color, joint positions, height, and body shape;

[0035] The steps of analyzing the body image received by the receiving module by a computer vision algorithm to obtain body data are as follows:

[0036] extracting skin color in the body image using a computer vision algorithm;

[0037] Detect the size of the head, shoulder width and leg length in the body image by using a computer vision algorithm, and calculate the height and joint position of the user by combining knowledge of human body proportions;

[0038] The body contour in the body image is matched with the body shape template defined in the learning model through a deep learning model to obtain the body shape of the user.

[0039] Based on the above scheme, the user's body shape is obtained through computer vision algorithms and deep learning models, and then the user's three-dimensional body model is constructed based on the user's body shape as a "carrier" for trying on clothes, thereby completing virtual fitting.

[0040] A further solution is that in the three-dimensional body generation module, a posture estimation model in three-dimensional human body modeling or deep learning is adopted, and a three-dimensional body model is constructed according to the body data.

[0041] Based on the above scheme, the posture estimation model in 3D human body modeling or deep learning is relatively simple and easy to use.

[0042] A further solution: in the clothing matching module, a classification model or a regression model in deep learning is used to construct a user profile based on the user's historical purchase records, browsing behaviors, and data on likes and shares on social media;

[0043] A clothing preference feature that matches the clothing preference information input by the user is extracted from the user portrait, and three-dimensional clothing that matches the clothing preference feature is screened out from the three-dimensional clothing database based on the extracted clothing preference feature of the user.

[0044] In a second aspect, the present invention provides a virtual fitting matching method based on VAE, which is run on any virtual fitting matching system based on VAE described in the first aspect, and comprises the following steps:

[0045] S100, generating a three-dimensional garment through a VAE model, and storing the three-dimensional garment in the three-dimensional garment database;

[0046] S200, receiving body image and clothing preference information input by a user;

[0047] S300, analyzing the body image received by the receiving module through a computer vision algorithm to obtain body data; and constructing a three-dimensional body model according to the body data;

[0048] S400, screening out three-dimensional clothing that meets the clothing preference information in the three-dimensional clothing database according to the clothing preference information received by the receiving module;

[0049] S500: Fit the three-dimensional clothing selected by the clothing matching module onto the three-dimensional body model to complete virtual fitting.

[0050] The beneficial effects of the present invention are:

[0051] 1. In the present invention, the texture style on three-dimensional clothing is generated and optimized through the VAE model, which can process complex clothing details and textures, thereby generating three-dimensional clothing with finer, more fitting, realistic and diverse textures.

[0052] 2. The VAE model in the present invention automatically generates three-dimensional clothing. Compared with the prior art of creating three-dimensional clothing models by manual or semi-automatic methods, the VAE model in the present invention generates three-dimensional clothing at a lower cost, with higher efficiency and better effect.

[0053] 3. The present invention constructs a three-dimensional body model that is closer to the user's reality through the received user's body image, and selects highly matching three-dimensional clothing in the three-dimensional clothing database through the received user's clothing preference information; the three-dimensional clothing is fitted on the three-dimensional body model to simulate clothing try-on. In the present invention, the three-dimensional clothing can be well fitted on the three-dimensional body model, enhancing the realism of the visual effect, thereby achieving a more realistic and lifelike try-on effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 It is a structural schematic diagram of a virtual fitting and matching system based on VAE of the present invention;

[0056] Figure 2 It is a flow chart of a virtual fitting and matching method based on VAE of the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work belong to the protection scope of the present invention.

[0058] Embodiment 1:

[0059] like Figure 1 As shown, this embodiment provides a virtual fitting and matching system based on VAE, including:

[0060] 3D clothing generation module, which generates 3D clothing through VAE model;

[0061] A three-dimensional clothing database storing three-dimensional clothing generated by the three-dimensional clothing generation module;

[0062] A receiving module, receiving body images and clothing preference information input by a user;

[0063] A three-dimensional body generation module analyzes the body image received by the receiving module through a computer vision algorithm to obtain body data; and constructs a three-dimensional body model according to the body data;

[0064] a clothing matching module, which selects three-dimensional clothing that meets the clothing preference information in the three-dimensional clothing database according to the clothing preference information received by the receiving module;

[0065] The virtual fitting module fits the three-dimensional clothing selected by the clothing matching module onto the three-dimensional body model to complete the virtual fitting.

[0066] For the three-dimensional clothing generation module, the VAE generates the three-dimensional clothing in different ways according to different texture styles received by the VAE;

[0067] In the first case, when the VAE receives a 3D texture pattern,

[0068] In the three-dimensional clothing generation module, the steps of generating the three-dimensional clothing through the VAE model are:

[0069] The VAE model receives a three-dimensional texture pattern;

[0070] The VAE model generates a three-dimensional garment with a corresponding texture according to the received three-dimensional texture style.

[0071] In the second case, when the VAE receives a 2D texture pattern,

[0072] In the three-dimensional clothing generation module, the steps of generating the three-dimensional clothing through the VAE model are:

[0073] The VAE model receives a two-dimensional texture pattern;

[0074] The VAE model generates a two-dimensional clothing image with a corresponding texture according to the received two-dimensional texture pattern;

[0075] The two-dimensional clothing image generated by the VAE model is reconstructed into three-dimensional clothing through three-dimensional reconstruction technology.

[0076] Among them, the three-dimensional reconstruction technology includes image-based modeling, or deep learning methods, etc.

[0077] In addition, in addition to inputting the two-dimensional data texture pattern number into the VAE model, and the VAE model generating the two-dimensional clothing image according to the two-dimensional texture pattern, this case also provides a method of generating a two-dimensional clothing image with a new texture by randomly sampling points in the latent space in the VAE model.

[0078] The VAE model includes an encoder and a decoder;

[0079] The encoder is used to extract features of the three-dimensional texture pattern or the two-dimensional texture pattern, and convert the three-dimensional texture pattern and the two-dimensional texture pattern into representations in the spatial domain;

[0080] The decoder is used to reconstruct the three-dimensional texture pattern represented in the spatial domain of the encoder into a three-dimensional garment with a corresponding texture; or, the decoder is used to reconstruct the two-dimensional texture pattern represented in the spatial domain of the encoder into a two-dimensional garment image with a corresponding texture.

[0081] A more specific example for the encoder and the decoder is that both the encoder and the decoder use convolutional neural network (CNN) layers.

[0082] The encoder compresses the received data (in this embodiment, the data received by the encoder is a three-dimensional texture pattern or a two-dimensional texture pattern) into a potential variable space, and the encoder outputs the mean (μ) and standard deviation (δ) in the potential space, which are used to define the Gaussian distribution; in the potential space, the input data (three-dimensional texture pattern or two-dimensional texture pattern) is represented as the parameters (mean and standard deviation) of the Gaussian distribution. The potential vector z is sampled from the Gaussian distribution in the potential space and input into the decoder.

[0083] The decoder receives the latent vector z and decodes the latent vector z back to the space of the original data; the decoder generates new data (a three-dimensional garment with corresponding texture or a two-dimensional garment image with corresponding texture) having similar features to the input data (a three-dimensional texture pattern or a two-dimensional texture pattern).

[0084] In order to enable the VAE model to generate more realistic three-dimensional clothing with more fitting textures on the clothing, the VAE model is first trained before generating the three-dimensional clothing through the VAE model;

[0085] Collecting three-dimensional clothing images with different textures to form a three-dimensional clothing image dataset; or collecting two-dimensional clothing images with different textures to form a two-dimensional clothing image dataset;

[0086] Inputting the three-dimensional clothing image dataset or the two-dimensional clothing image dataset into the encoder of the VAE model;

[0087] The encoder extracts features of three-dimensional clothing in the three-dimensional clothing image dataset and converts the three-dimensional clothing into a spatial vector representation in the spatial domain; or, the encoder extracts features of two-dimensional clothing in the two-dimensional clothing image dataset and converts the two-dimensional clothing into a spatial vector representation in the spatial domain;

[0088] The encoder inputs the spatial vector representation of the three-dimensional clothing or the spatial vector representation of the two-dimensional clothing into the decoder;

[0089] The decoder reconstructs the spatial vector representation of the three-dimensional garment into a three-dimensional garment; or, the decoder reconstructs the spatial vector representation of the two-dimensional garment into a two-dimensional garment image in an image domain;

[0090] Complete the training of the VAE model.

[0091] Specifically, the sources of collecting three-dimensional clothing images and two-dimensional images with different textures include professional clothing image databases, fashion websites, social media, etc. Then, the three-dimensional clothing image dataset and the two-dimensional clothing image dataset are preprocessed; wherein, the preprocessing content of the three-dimensional clothing image dataset includes operations such as cleaning, denoising, graying, and standardized posture to eliminate the noise and residual information of the texture data; the preprocessing content of the two-dimensional clothing image dataset includes operations such as size adjustment, cropping, denoising, and normalization; finally, the preprocessed three-dimensional clothing image dataset and the two-dimensional clothing image dataset are input into the VAE model to train the VAE model.

[0092] Based on the above scheme, in the process of training the VAE model, the loss functions adopted by the encoder and the decoder include a reconstruction error function and a KL divergence function.

[0093] It should be noted that the reconstruction error function can be a mean square error at the pixel level, which is used to measure the difference between the data generated by the decoder and the received data; the KL divergence function is the KL divergence between the prior distribution and the posterior distribution of the latent space of the VAE model, which is used to measure the difference between the latent distribution and the standard normal distribution to ensure the regularization of the latent space.

[0094] In the receiving module, the body image received by the receiving module may be a frontal photo and a side photo of the user's body, or a video of the user's body. The clothing preference received by the receiving module includes color preference, style preference, material preference, etc.

[0095] A more specific example of the three-dimensional body generation module is: in the three-dimensional body generation module, the body data includes skin color, joint position, height, and body shape;

[0096] First, in the three-dimensional body generation module, the body image received by the receiving module is analyzed by a computer vision algorithm to obtain body data in the following steps:

[0097] extracting skin color in the body image using a computer vision algorithm;

[0098] Detect the size of the head, shoulder width and leg length in the body image by using a computer vision algorithm, and calculate the height and joint position of the user by combining knowledge of human body proportions;

[0099] The body contour in the body image is matched with the body shape template defined in the learning model through the deep learning model to obtain the user's body shape (for example: thin, fat or normal, etc.).

[0100] When the computer vision algorithm analyzes the received body image, it is necessary to pre-process the body image first, and the pre-processing includes denoising, adjusting resolution, etc. Then, the skin color, head, shoulder width and leg length in the body image are identified, so as to extract the user's skin color, and calculate the user's height and joint position.

[0101] Additionally, if the body image (video with the user's body) provides enough perspective variation, 3D reconstruction can be used to more accurately measure the user's height.

[0102] When obtaining the user's body shape through the deep learning model, the sizes of various parts of the user's body, such as waist circumference, hip circumference, chest circumference, etc., can be further analyzed to obtain more detailed body shape information.

[0103] Secondly, in the three-dimensional body generation module, a posture estimation model in three-dimensional human body modeling or deep learning is adopted, and a three-dimensional body model is constructed according to the body data.

[0104] A more specific example of the clothing matching module is as follows: in the clothing matching module, a classification model or a regression model in deep learning is used to build a user profile based on the user's historical purchase records, browsing behaviors, likes and sharing data on social media;

[0105] The clothing preference features (e.g., color, style, texture or material, etc.) that match the clothing preference information input by the user are extracted from the user portrait, and three-dimensional clothing that matches the clothing preference features is screened out from the three-dimensional clothing database based on the extracted user clothing preference features.

[0106] For example, the user's clothing preference information received by the receiving module is: "white", "long sleeves", and "cotton". The clothing matching module constructs a user portrait (i.e., white cotton long-sleeved clothing that the user has purchased, browsed, liked, or shared) based on the three keyword information "white", "long sleeves", and "cotton" through a classification model or a regression model in deep learning, and then selects clothing preference features from the user portrait (i.e., extracting the texture, decorative design, etc. on the white cotton long-sleeved clothing), and finally selects three-dimensional clothing that meets the clothing preference features in the three-dimensional clothing database based on the extracted user clothing preference features.

[0107] In the virtual fitting module, image processing technology and three-dimensional rendering technology (such as texture mapping, lighting simulation and shadow processing, etc.) can be used to fit the three-dimensional clothing selected by the clothing matching module onto the three-dimensional body model to complete the virtual fitting.

[0108] On the basis of the above scheme, a VAE-based virtual fitting matching system also includes a user feedback module, which can receive user feedback on the virtual fitting effect; the virtual fitting matching system is continuously optimized according to the user feedback, thereby improving the accuracy of the virtual fitting, and the VAE model generates clothing that better meets the user's preferences.

[0109] Embodiment 2:

[0110] like Figure 2 As shown, this embodiment provides a virtual fitting matching method based on VAE, which is run on a virtual fitting matching system based on VAE described in Embodiment 1, and includes the following steps:

[0111] S100, generating a three-dimensional garment through a VAE model, and storing the three-dimensional garment in the three-dimensional garment database;

[0112] S200, receiving body image and clothing preference information input by a user;

[0113] S300, analyzing the body image received by the receiving module through a computer vision algorithm to obtain body data; and constructing a three-dimensional body model according to the body data;

[0114] S400, screening out three-dimensional clothing that meets the clothing preference information in the three-dimensional clothing database according to the clothing preference information received by the receiving module;

[0115] S500: Fit the three-dimensional clothing selected by the clothing matching module onto the three-dimensional body model to complete virtual fitting.

[0116] The present invention is not limited to the above-mentioned optional implementation modes. Anyone can derive other various forms of products under the inspiration of the present invention. However, no matter what changes are made in the shape or structure, all technical solutions that fall within the scope defined by the claims of the present invention fall within the protection scope of the present invention.

Claims

1. A virtual fitting matching system based on VAE, characterized in that: include: 3D clothing generation module, which generates 3D clothing through VAE model; A three-dimensional clothing database storing three-dimensional clothing generated by the three-dimensional clothing generation module; A receiving module, receiving body images and clothing preference information input by a user; A three-dimensional body generation module analyzes the body image received by the receiving module through a computer vision algorithm to obtain body data; and constructs a three-dimensional body model according to the body data; a clothing matching module, which selects three-dimensional clothing that meets the clothing preference information in the three-dimensional clothing database according to the clothing preference information received by the receiving module; The virtual fitting module fits the three-dimensional clothing selected by the clothing matching module onto the three-dimensional body model to complete the virtual fitting.

2. A VAE-based virtual fitting matching system according to claim 1, characterized in that: In the three-dimensional clothing generation module, the steps of generating the three-dimensional clothing through the VAE model are: The VAE model receives a three-dimensional texture pattern; The VAE model generates a three-dimensional garment with a corresponding texture according to the received three-dimensional texture style.

3. The VAE-based virtual fitting matching system according to claim 1, characterized in that: In the three-dimensional clothing generation module, the steps of generating the three-dimensional clothing through the VAE model are: The VAE model receives a two-dimensional texture pattern; The VAE model generates a two-dimensional clothing image with a corresponding texture according to the received two-dimensional texture pattern; The two-dimensional clothing image generated by the VAE model is reconstructed into three-dimensional clothing through three-dimensional reconstruction technology.

4. A VAE-based virtual fitting matching system according to claim 2 or 3, characterized in that: The VAE model includes an encoder and a decoder; The encoder is used to extract features of the three-dimensional texture pattern or the two-dimensional texture pattern, and convert the three-dimensional texture pattern and the two-dimensional texture pattern into representations in the spatial domain; The decoder is used to reconstruct the three-dimensional texture pattern represented in the spatial domain of the encoder into a three-dimensional garment with a corresponding texture; or, the decoder is used to reconstruct the two-dimensional texture pattern represented in the spatial domain of the encoder into a two-dimensional garment image with a corresponding texture.

5. The VAE-based virtual fitting system according to claim 4, characterized in that: Before generating the three-dimensional clothing through the VAE model, training the VAE model; Collecting three-dimensional clothing images with different textures to form a three-dimensional clothing image dataset; or collecting two-dimensional clothing images with different textures to form a two-dimensional clothing image dataset; Inputting the three-dimensional clothing image dataset or the two-dimensional clothing image dataset into the encoder of the VAE model; The encoder extracts features of three-dimensional clothing in the three-dimensional clothing image dataset and converts the three-dimensional clothing into a spatial vector representation in the spatial domain; or, the encoder extracts features of two-dimensional clothing in the two-dimensional clothing image dataset and converts the two-dimensional clothing into a spatial vector representation in the spatial domain; The encoder inputs the spatial vector representation of the three-dimensional clothing or the spatial vector representation of the two-dimensional clothing into the decoder; The decoder reconstructs the spatial vector representation of the three-dimensional garment into a three-dimensional garment; or the decoder reconstructs the spatial vector representation of the two-dimensional garment into a two-dimensional garment image in an image domain; Complete the training of the VAE model.

6. The VAE-based virtual fitting system according to claim 5, characterized in that: In the process of training the VAE model, the loss functions adopted by the encoder and the decoder include a reconstruction error function and a KL divergence function.

7. The VAE-based virtual fitting system according to claim 1, characterized in that: In the three-dimensional body generation module, the body data includes skin color, joint positions, height, and body shape; The steps of analyzing the body image received by the receiving module by a computer vision algorithm to obtain body data are as follows: extracting skin color in the body image using a computer vision algorithm; Detect the size of the head, shoulder width and leg length in the body image by using a computer vision algorithm, and calculate the height and joint position of the user by combining knowledge of human body proportions; The body contour in the body image is matched with the body shape template defined in the learning model through a deep learning model to obtain the body shape of the user.

8. The VAE-based virtual fitting matching system according to claim 7, characterized in that: In the three-dimensional body generation module, a posture estimation model in three-dimensional human body modeling or deep learning is adopted, and a three-dimensional body model is constructed according to the body data.

9. The VAE-based virtual fitting system according to claim 1, characterized in that: In the clothing matching module, a user portrait is constructed based on the user's historical purchase records, browsing behaviors, likes and sharing data on social media through a classification model or regression model in deep learning; A clothing preference feature that matches the clothing preference information input by the user is extracted from the user portrait, and three-dimensional clothing that matches the clothing preference feature is screened out from the three-dimensional clothing database based on the extracted clothing preference feature of the user.

10. A virtual fitting matching method based on VAE, characterized in that: Running on a VAE-based virtual fitting matching system as described in any one of claims 1 to 9, comprising the following steps: S100, generating a three-dimensional garment through a VAE model, and storing the three-dimensional garment in the three-dimensional garment database; S200, receiving body image and clothing preference information input by a user; S300, analyzing the body image received by the receiving module through a computer vision algorithm to obtain body data; and constructing a three-dimensional body model according to the body data; S400, screening out three-dimensional clothing that meets the clothing preference information in the three-dimensional clothing database according to the clothing preference information received by the receiving module; S500: Fit the three-dimensional clothing selected by the clothing matching module onto the three-dimensional body model to complete virtual fitting.

Images