A simulation optimization design method and system for underwear

Abstract

The present invention relates to the field of simulation design, and in particular to a simulation optimization design method and system for underwear. A simulation optimization design system for underwear comprises: a multi-view character image acquisition module, a multi-view chest area image construction module, a chest area model construction module, a chest area feature vector extraction module, an underwear feature vector simulation calculation module, and an underwear model construction module. The present invention constructs a user's chest area model through the user's multi-view character image, and extracts a chest area feature vector corresponding to the chest area model through a chest feature extraction model; on the basis of the chest area feature vector, a group optimization algorithm is used to simulate and calculate the underwear feature vector corresponding to the underwear model that best matches the chest area feature vector, and an underwear model adapted to the user is constructed based on the underwear feature vector, thereby realizing personalized settings for the user's underwear.

Description

Technical Field

[0001] The present invention relates to the field of simulation design, and in particular to a simulation optimization design method and system for underwear. Background Art

[0002] In the field of modern underwear design, personalized customization is increasingly favored by consumers because it can provide a more comfortable and fitting wearing experience based on the unique needs of individuals. However, the existing personalized underwear design process relies heavily on manual operation. Designers need to combine rich experience and professional knowledge to conduct detailed analysis of the collected data in order to develop a design plan that meets individual needs, which affects the efficiency of personalized underwear design. Summary of the invention

[0003] The present invention constructs a chest region model of a user through a multi-view character picture of the user, and extracts a chest region feature vector corresponding to the chest region model through a chest feature extraction model; on the basis of the chest region feature vector, a group optimization algorithm is used to simulate and calculate an underwear feature vector corresponding to an underwear model that is most suitable for the chest region feature vector, and an underwear model suitable for the user is constructed based on the underwear feature vector, thereby realizing personalized settings for the user's underwear, and only needs to obtain the multi-view character picture of the user, without the need for an operator to perform additional operations, thereby improving the efficiency of underwear simulation optimization design.

[0004] A simulation optimization design method for underwear, comprising:

[0005] Get the user's multi-view character images;

[0006] Perform image segmentation operations on the multi-view character images to construct multi-view chest area images;

[0007] Performing a 3D modeling operation based on the multi-view chest region images to construct a chest region model;

[0008] Obtaining a chest region point cloud data set corresponding to the chest region model, and then sending the chest region point cloud data set to a chest feature extraction model for feature extraction operation to construct a chest region feature vector;

[0009] Based on the chest region feature vector, the underwear feature vector is simulated and calculated by a swarm optimization algorithm to construct the optimal underwear feature vector. During the simulation calculation process, the fitness evaluation is performed based on the chest region feature vector and the underwear feature vector, and the evaluated fitness is used as the fitness;

[0010] The optimal underwear feature vector is sent to the underwear modeling model for processing, an underwear point cloud data set is output, and an underwear model is constructed based on the underwear point cloud data set.

[0011] Preferably, the chest feature extraction model is established based on the PointNet model;

[0012] The training of the chest feature extraction model specifically includes the following steps: adding a chest feature fully connected layer and a chest feature output layer at the end of the chest feature extraction model to form a chest feature extraction extension model, obtaining a number of chest feature training samples, the chest feature training samples including a chest area point cloud data set, annotating the chest feature training samples by a first target feature vector, forming a chest feature training set with all annotated chest feature training samples, and then sending the chest feature training set to a parameter-initialized chest feature extraction extension model for training, during which the first target feature vector is used as a target, a first loss value is calculated, and it is determined whether the first loss value is within a first preset range. If the first loss value is within the first preset range, the trained chest feature extraction extension model is output, and the chest feature extraction model is extracted from the trained chest feature extraction extension model; if the first loss value is not within the first preset range, the chest feature training set is continued to be sent to the chest feature extraction extension model for iterative training.

[0013] Preferably, based on the chest region feature vector, the underwear feature vector is simulated and calculated by a group optimization algorithm to construct an optimal underwear feature vector, which specifically includes the following steps:

[0014] Construct several simulated underwear feature vectors, group all simulated underwear feature vectors into a simulated population set, set the maximum number of iterations, and construct the simulated underwear feature vectors as follows: determine the value range corresponding to each feature item in the simulated underwear feature vector through the underwear feature extraction model, traverse from 1 to N, N is the length of the simulated underwear feature vector, and for each traversed value n, n=1, 2, 3, ..., N, generate a random number within the value range corresponding to the nth feature item; until traversing to N, all random numbers are sequentially grouped into a simulated underwear feature vector;

[0015] The corresponding fitness is calculated for each simulated underwear feature vector. The specific operation is as follows: the simulated underwear feature vector and the chest region feature vector are sent to the fitness evaluation model for processing, and the output fitness is used as the fitness calculated by the simulated underwear feature vector;

[0016] Based on the fitness corresponding to the characteristic vector of each simulated underwear, the simulated population set is iteratively simulated and calculated through the group optimization algorithm;

[0017] Until the number of iterations reaches the maximum number of iterations, the simulated underwear feature vector with the largest fitness is output as the optimal underwear feature vector.

[0018] Preferably, determining the value range corresponding to each feature item in the simulated underwear feature vector by using an underwear feature extraction model specifically includes the following steps:

[0019] The underwear feature extraction model is also built based on the PointNet model, and the underwear feature vector is extracted through the underwear feature extraction model;

[0020] The training of the underwear feature extraction model specifically includes the following steps: adding an underwear feature fully connected layer and an underwear feature output layer at the end of the underwear feature extraction model to form an underwear feature extraction extension model, obtaining a number of underwear feature training samples, the underwear feature training samples including an underwear model point cloud data set, annotating the underwear feature training samples by a second target feature vector, forming an underwear feature training set with all the annotated underwear feature training samples, and then sending the underwear feature training set to the underwear feature extraction extension model with parameter initialization for training, during which the second target feature vector is used as a target, calculating a second loss value, judging whether the second loss value is within a second preset range, and if the second loss value is within the second preset range, outputting the trained underwear feature extraction extension model, and extracting the underwear feature extraction model from the trained underwear feature extraction extension model; if the second loss value is not within the second preset range, continuing to send the underwear feature training set to the underwear feature extraction extension model for iterative training;

[0021] Obtain a trained underwear feature extraction model, input each underwear feature training sample in the underwear feature training set into the trained underwear feature extraction model one by one, obtain the corresponding underwear feature vector, traverse from 1 to N, and for each traversed value n, perform the following operation, the minimum value and the maximum value of the nth feature item in all underwear feature vectors form the value range corresponding to the nth feature item in the underwear feature vector; until traversing to N, the value range corresponding to each feature item in the underwear feature vector is used as the value range corresponding to each feature item in the simulated underwear feature vector.

[0022] Preferably, the fitness evaluation model is established based on BP neural network;

[0023] The training of the fitness evaluation model includes the following contents: obtaining several positive samples and negative samples for fitness evaluation. The positive samples for fitness evaluation include the chest region feature vector and underwear feature vector corresponding to the fully adapted chest region model and underwear model. The positive samples for fitness evaluation are labeled by the positive target fitness. The negative samples for fitness evaluation include the chest region feature vector and underwear feature vector corresponding to the fully unfit chest region model and underwear model. The negative samples for fitness evaluation are labeled by the negative target fitness. The noted fitness assessment positive samples and fitness assessment negative samples constitute a fitness assessment training set, and the fitness assessment training set is sent to the fitness assessment model with parameter initialization for training. During the training, the positive target fitness or the negative target fitness is used as the target, and the third loss value is calculated to determine whether the third loss value is within the third preset range. If the third loss value is within the third preset range, the trained fitness assessment model is output; if the third loss value is not within the third preset range, the fitness assessment training set is continued to be sent to the fitness assessment model for iterative training.

[0024] Preferably, the underwear modeling model is established based on a variational autoencoder;

[0025] The training of the underwear modeling model specifically includes the following steps: after completing the training of the underwear feature extraction model, the underwear feature vector is used as the underwear modeling training sample, and the underwear model point cloud data set corresponding to the underwear feature vector is used as the target point cloud data set to annotate the underwear modeling training sample, and the annotated underwear modeling training samples are formed into an underwear modeling training set, and then the underwear modeling training set is sent to the underwear modeling model with parameter initialization for training. During the training, the target point cloud data set is used as the target, the fourth loss value is calculated, and it is determined whether the fourth loss value is within the fourth preset range. If the fourth loss value is within the fourth preset range, the trained underwear modeling model is output; if the fourth loss value is not within the fourth preset range, the underwear modeling training set is continued to be sent to the underwear modeling model for iterative training.

[0026] Preferably, the population optimization algorithm adopts a genetic algorithm.

[0027] A simulation optimization design system for underwear, comprising:

[0028] A multi-view character image acquisition module is used to acquire a user's multi-view character image;

[0029] A multi-view chest region image construction module is used to perform image segmentation operations on the multi-view character images to construct a multi-view chest region image;

[0030] A chest region model building module, used to perform 3D modeling operations based on multi-view chest region images to build a chest region model;

[0031] A chest region feature vector extraction module is used to obtain a chest region point cloud data set corresponding to a chest region model, and then send the chest region point cloud data set to a chest feature extraction model for feature extraction to construct a chest region feature vector;

[0032] The underwear feature vector simulation calculation module is used to simulate and calculate the underwear feature vector based on the chest region feature vector through a group optimization algorithm to construct an optimal underwear feature vector. During the simulation calculation process, the fitness evaluation is performed based on the chest region feature vector and the underwear feature vector, and the evaluated fitness is used as the fitness;

[0033] The underwear model building module is used to send the optimal underwear feature vector to the underwear modeling model for processing, output the underwear point cloud data set, and build the underwear model based on the underwear point cloud data set.

[0034] The present invention has the following advantages:

[0035] The present invention constructs a chest region model of a user through a multi-view character picture of the user, and extracts a chest region feature vector corresponding to the chest region model through a chest feature extraction model; on the basis of the chest region feature vector, a group optimization algorithm is used to simulate and calculate an underwear feature vector corresponding to an underwear model that is most suitable for the chest region feature vector, and an underwear model suitable for the user is constructed based on the underwear feature vector, thereby realizing personalized settings for the user's underwear, and only needs to obtain the multi-view character picture of the user, without the need for an operator to perform additional operations, thereby improving the efficiency of underwear simulation optimization design. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the structure of the underwear simulation optimization design system adopted in the embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0038] Example 1

[0039] A simulation optimization design method for underwear, comprising:

[0040] Obtaining multi-view person images of the user. It should be noted that, if the user needs underwear customization service, the user's person images from different perspectives, i.e., multi-view person images, will be taken through a camera with the user's consent;

[0041] Perform image segmentation operations on the multi-view character images to construct multi-view chest area images. By obtaining the chest area images, it is convenient to subsequently model the user's chest area;

[0042] Perform 3D modeling operations based on multi-view chest area images to build a chest area model. The chest area model is actually a series of point cloud data obtained through surface fitting. In the process of performing 3D modeling, firstly, feature point detection is performed on the chest area image of each view through SIFT and other algorithms, and feature point alignment is performed. Then, a dense depth map of the chest area is constructed using stereo matching technology. Based on the dense depth map, corresponding point cloud data is generated. Finally, the chest area model is constructed based on the point cloud data through Poisson reconstruction or other surface reconstruction technologies.

[0043] A chest region point cloud data set corresponding to the chest region model is obtained. In the process of obtaining the chest region point cloud data set, in order to reduce a certain amount of calculation, voxel division can be performed on the chest region model, and a certain number of point cloud data can be selected in each voxel according to a set strategy. The chest region point cloud data set is then sent to a chest feature extraction model for feature extraction operations to construct a chest region feature vector.

[0044] Based on the chest region feature vector, the underwear feature vector is simulated and calculated by a group optimization algorithm to construct the optimal underwear feature vector. It should be noted that the underwear feature vector reflects the specific structure of the underwear model, and the specific form of the underwear feature vector is also obtained through training of a large number of underwear models. In the simulation calculation process, the fitness evaluation will be performed based on the chest region feature vector and the underwear feature vector, and the evaluated fitness will be used as the fitness to automatically obtain the underwear model that best matches the chest region feature vector, without the need for operators to calculate based on experience, thereby increasing the efficiency of underwear simulation optimization design;

[0045] The optimal underwear feature vector is sent to the underwear modeling model for processing, an underwear point cloud data set is output, and an underwear model is constructed based on the underwear point cloud data set.

[0046] The present application constructs a chest region model of the user through multi-view character pictures of the user, and extracts a chest region feature vector corresponding to the chest region model through a chest feature extraction model; based on the chest region feature vector, a swarm optimization algorithm is used to simulate and calculate an underwear feature vector corresponding to an underwear model that best matches the chest region feature vector, and an underwear model adapted to the user is constructed based on the underwear feature vector, thereby achieving personalized settings for the user's underwear, and only requires obtaining multi-view character pictures of the user, without the need for the operator to perform additional operations, thereby improving the efficiency of underwear simulation optimization design.

[0047] The chest feature extraction model is based on the PointNet model, which is a deep learning model for processing point cloud data. It can learn the feature information of three-dimensional point cloud data and map it to a string of feature vectors.

[0048] The training of the chest feature extraction model specifically includes the following steps: adding a chest feature fully connected layer and a chest feature output layer at the end of the chest feature extraction model to form a chest feature extraction extension model, obtaining several chest feature training samples, the chest feature training samples include a chest area point cloud data set, it should be noted that the chest area point cloud data set is constructed based on the chest area model in the biometric library that is publicly disclosed in accordance with legal regulations, annotating the chest feature training samples through the first target feature vector, and the first target feature vector is the feature information obtained by the developer through feature extraction based on the chest area model, for example, by combining Python with Open3D, Trimesh and other means to automatically calculate the key size of the model (chest circumference, etc.), forming all the annotated chest feature training samples into a chest feature training set, and then sending the chest feature training set to the parameter The chest feature extraction extension model is initialized for training, during which the first target feature vector is used as the target, the first loss value is calculated, and it is determined whether the first loss value is within a first preset range. The first preset range is determined by the developer based on experience. If the first loss value is within the first preset range, the trained chest feature extraction extension model is output, and the chest feature extraction model is extracted from the trained chest feature extraction extension model; if the first loss value is not within the first preset range, the chest feature training set is continued to be sent to the chest feature extraction extension model for iterative training; it should be noted that the chest region feature vector is much shorter than the first target feature vector, and the chest region feature vector can reflect the first target feature vector that characterizes the actual parameters of the chest region model. The much shorter chest region feature vector can require less computing power in the subsequent simulation calculation process.

[0049] Based on the chest region feature vector, the underwear feature vector is simulated and calculated by a group optimization algorithm to construct the optimal underwear feature vector, which specifically includes the following steps:

[0050] Construct several simulated underwear feature vectors, group all simulated underwear feature vectors into a simulated population set, set the maximum number of iterations, and construct the simulated underwear feature vectors as follows: determine the value range corresponding to each feature item in the simulated underwear feature vector through the underwear feature extraction model, traverse from 1 to N, N is the length of the simulated underwear feature vector, and for each traversed value n, n=1, 2, 3, ..., N, generate a random number within the value range corresponding to the nth feature item; until traversing to N, all random numbers are sequentially grouped into a simulated underwear feature vector;

[0051] The corresponding fitness is calculated for each simulated underwear feature vector. The specific operation is as follows: the simulated underwear feature vector and the chest region feature vector are sent to the fitness evaluation model for processing, and the output fitness is used as the fitness calculated by the simulated underwear feature vector;

[0052] Based on the fitness corresponding to each simulated underwear feature vector, the simulated population set is iteratively simulated and calculated by genetic algorithm, during which selection, recombination and mutation operations are performed;

[0053] Until the number of iterations reaches the maximum number of iterations, the simulated underwear feature vector with the largest fitness is output as the optimal underwear feature vector.

[0054] The present application calculates the degree of fit as fitness by simulating underwear feature vectors and chest area feature vectors, and executes a genetic algorithm to search for the underwear feature vector that is most suitable for the chest area feature vector. There is no need for operators to judge the fit of the chest model and the underwear model based on experience and observation, and the automation of underwear model design can be achieved and the efficiency of underwear model design can be improved.

[0055] Determining the value range corresponding to each feature item in the simulated underwear feature vector by using the underwear feature extraction model specifically includes the following steps:

[0056] The underwear feature extraction model is also built based on the PointNet model, and the underwear feature vector is extracted through the underwear feature extraction model;

[0057] The training of the underwear feature extraction model specifically includes the following steps: adding an underwear feature fully connected layer and an underwear feature output layer at the end of the underwear feature extraction model to form an underwear feature extraction extension model, obtaining several underwear feature training samples, and the underwear feature training samples include an underwear model point cloud data set. It should be noted that the underwear model point cloud data set is obtained by the development user based on a large number of underwear models. The underwear model here can be modeled according to the actual underwear, or it can be obtained by the development user through simulation design using simulation software. The underwear feature training samples are labeled by the second target feature vector. The second target feature vector is the feature information obtained by the developer according to the underwear model through feature extraction. For example, the curvature distribution of the underwear model surface is calculated by the Gaussian curvature algorithm. All labeled underwear feature training samples are composed of underwear feature vectors. The underwear feature training set is then sent to the underwear feature extraction extension model with parameter initialization for training, during which the second target feature vector is used as the target, the second loss value is calculated, and it is determined whether the second loss value is within a second preset range. The second preset range is determined by the developer based on experience. If the second loss value is within the second preset range, the trained underwear feature extraction extension model is output, and the underwear feature extraction model is extracted from the trained underwear feature extraction extension model; if the second loss value is not within the second preset range, the underwear feature training set is continuously sent to the underwear feature extraction extension model for iterative training; similarly, the underwear feature vector is much shorter than the second target feature vector, and the underwear feature vector can reflect the second target feature vector that represents the actual parameters of the underwear model; and it is convenient to perform the recombination and mutation operations of the genetic algorithm;

[0058] Obtain a trained underwear feature extraction model, input each underwear feature training sample in the underwear feature training set into the trained underwear feature extraction model one by one, obtain the corresponding underwear feature vector, traverse from 1 to N, and for each traversed value n, perform the following operation, the minimum value and the maximum value of the nth feature item in all underwear feature vectors form the value range corresponding to the nth feature item in the underwear feature vector; until traversing to N, the value range corresponding to each feature item in the underwear feature vector is used as the value range corresponding to each feature item in the simulated underwear feature vector.

[0059] The fitness evaluation model is established based on BP neural network;

[0060] The training of the fitness evaluation model includes the following contents: obtaining several positive samples and negative samples for fitness evaluation. The positive samples for fitness evaluation include the chest region feature vector and underwear feature vector corresponding to the fully adapted chest region model and underwear model. It should be noted that the chest region feature vector and underwear feature vector are extracted by the chest feature extraction model and underwear feature extraction model respectively, and the full adaptation is determined according to the user's evaluation. The positive samples for fitness evaluation are labeled by the positive target fitness. Generally, the positive target fitness is set to 1. The negative samples for fitness evaluation include the chest region feature vector and underwear feature vector corresponding to the fully unfit chest region model and underwear model. The negative samples of fitness evaluation are labeled with standard fitness. Generally, the negative target fitness is set to 0. All labeled fitness evaluation positive samples and fitness evaluation negative samples form a fitness evaluation training set. The fitness evaluation training set is sent to the fitness evaluation model with parameter initialization for training. During this period, the positive target fitness or the negative target fitness is used as the target, and the third loss value is calculated to determine whether the third loss value is within the third preset range. The third preset range is determined by the developer based on experience. If the third loss value is within the third preset range, the trained fitness evaluation model is output; if the third loss value is not within the third preset range, the fitness evaluation training set is continued to be sent to the fitness evaluation model for iterative training.

[0061] The underwear modeling model is built based on variational autoencoder;

[0062] The training of the underwear modeling model specifically includes the following steps. After completing the training of the underwear feature extraction model, the underwear feature vector is used as the underwear modeling training sample, and the underwear model point cloud data set corresponding to the underwear feature vector is used as the target point cloud data set to annotate the underwear modeling training sample. The annotated underwear modeling training samples form an underwear modeling training set, and then the underwear modeling training set is sent to the underwear modeling model with parameter initialization for training. During the training, the target point cloud data set is used as the target, and the fourth loss value is calculated to determine whether the fourth loss value is within a fourth preset range. The fourth preset range is determined by the developer based on experience. If the fourth loss value is within the fourth preset range, the trained underwear modeling model is output; if the fourth loss value is not within the fourth preset range, the underwear modeling training set is continued to be sent to the underwear modeling model for iterative training.

[0063] Example 2

[0064] A simulation optimization design system for underwear, such as Figure 1 As shown, including:

[0065] The multi-view character image acquisition module is used to acquire the multi-view character images of the user. It should be noted that, when the user needs to carry out underwear customization service, the user's character images from different perspectives, i.e., multi-view character images, will be captured by the camera with the user's consent.

[0066] A multi-view chest area image construction module is used to perform image segmentation operations on the multi-view character images to construct a multi-view chest area image. By obtaining the chest area image, it is convenient to subsequently model the user's chest area;

[0067] The chest region model building module is used to perform 3D modeling operations based on multi-view chest region images to build a chest region model. The chest region model is actually a series of point cloud data obtained through surface fitting. In the process of executing 3D modeling, the feature points of the chest region image of each view are first detected and aligned through algorithms such as SIFT, and then a dense depth map of the chest region is constructed using stereo matching technology. Based on the dense depth map, the corresponding point cloud data is generated. Finally, the chest region model is constructed based on the point cloud data through Poisson reconstruction or other surface reconstruction technologies;

[0068] The chest region feature vector extraction module is used to obtain the chest region point cloud data set corresponding to the chest region model. In the process of obtaining the chest region point cloud data set, in order to reduce a certain amount of calculation, the chest region model can be divided into voxels, and a certain number of point cloud data can be selected in each voxel according to a set strategy. The chest region point cloud data set is then sent to the chest feature extraction model for feature extraction operations to construct a chest region feature vector;

[0069] The underwear feature vector simulation calculation module is used to simulate and calculate the underwear feature vector based on the chest area feature vector through a group optimization algorithm to construct the optimal underwear feature vector. It should be noted that the underwear feature vector reflects the specific structure of the underwear model, and the specific form of the underwear feature vector is also obtained through training of a large number of underwear models. In the simulation calculation process, the fitness evaluation will be performed based on the chest area feature vector and the underwear feature vector, and the evaluated fitness will be used as the fitness;

[0070] The underwear model building module is used to send the optimal underwear feature vector to the underwear modeling model for processing, output the underwear point cloud data set, and build the underwear model based on the underwear point cloud data set.

[0071] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A method for simulation optimization design of underwear, characterized in that: include: Get the user's multi-view character images; Perform image segmentation operations on the multi-view character images to construct multi-view chest area images; Performing a 3D modeling operation based on the multi-view chest region images to construct a chest region model; Obtaining a chest region point cloud data set corresponding to the chest region model, and then sending the chest region point cloud data set to a chest feature extraction model for feature extraction operation to construct a chest region feature vector; Based on the chest region feature vector, the underwear feature vector is simulated and calculated by a swarm optimization algorithm to construct the optimal underwear feature vector. During the simulation calculation process, the fitness evaluation is performed based on the chest region feature vector and the underwear feature vector, and the evaluated fitness is used as the fitness; Send the optimal underwear feature vector to the underwear modeling model for processing, output the underwear point cloud data set, and build the underwear model based on the underwear point cloud data set; Based on the chest region feature vector, the underwear feature vector is simulated and calculated by a group optimization algorithm to construct the optimal underwear feature vector, which specifically includes the following steps: Construct several simulated underwear feature vectors, group all simulated underwear feature vectors into a simulated population set, set the maximum number of iterations, and construct the simulated underwear feature vectors as follows: determine the value range corresponding to each feature item in the simulated underwear feature vector through the underwear feature extraction model, traverse from 1 to N, N is the length of the simulated underwear feature vector, and for each traversed value n, n = 1, 2, 3, ..., N, generate a random number within the value range corresponding to the nth feature item; until traversing to N, all random numbers are sequentially grouped into a simulated underwear feature vector; The corresponding fitness is calculated for each simulated underwear feature vector. The specific operation is as follows: the simulated underwear feature vector and the chest region feature vector are sent to the fitness evaluation model for processing, and the output fitness is used as the fitness calculated by the simulated underwear feature vector; Based on the fitness corresponding to the characteristic vector of each simulated underwear, the simulated population set is iteratively simulated and calculated through the group optimization algorithm; Until the number of iterations reaches the maximum number of iterations, the simulated underwear feature vector with the largest fitness is output as the optimal underwear feature vector.

2. The underwear simulation optimization design method according to claim 1, characterized in that: The chest feature extraction model is built based on the PointNet model; The training of the chest feature extraction model specifically includes the following steps: adding a chest feature fully connected layer and a chest feature output layer at the end of the chest feature extraction model to form a chest feature extraction extension model, obtaining a number of chest feature training samples, the chest feature training samples including a chest area point cloud data set, annotating the chest feature training samples by a first target feature vector, forming a chest feature training set with all annotated chest feature training samples, and then sending the chest feature training set to a parameter-initialized chest feature extraction extension model for training, during which the first target feature vector is used as a target, a first loss value is calculated, and it is determined whether the first loss value is within a first preset range. If the first loss value is within the first preset range, the trained chest feature extraction extension model is output, and the chest feature extraction model is extracted from the trained chest feature extraction extension model; if the first loss value is not within the first preset range, the chest feature training set is continued to be sent to the chest feature extraction extension model for iterative training.

3. The underwear simulation optimization design method according to claim 2, characterized in that: Determining the value range corresponding to each feature item in the simulated underwear feature vector by using the underwear feature extraction model specifically includes the following steps: The underwear feature extraction model is also built based on the PointNet model, and the underwear feature vector is extracted through the underwear feature extraction model; The training of the underwear feature extraction model specifically includes the following steps: adding an underwear feature fully connected layer and an underwear feature output layer at the end of the underwear feature extraction model to form an underwear feature extraction extension model, obtaining a number of underwear feature training samples, the underwear feature training samples including an underwear model point cloud data set, annotating the underwear feature training samples by a second target feature vector, forming an underwear feature training set with all the annotated underwear feature training samples, and then sending the underwear feature training set to the underwear feature extraction extension model with parameter initialization for training, during which the second target feature vector is used as a target, calculating a second loss value, judging whether the second loss value is within a second preset range, and if the second loss value is within the second preset range, outputting the trained underwear feature extraction extension model, and extracting the underwear feature extraction model from the trained underwear feature extraction extension model; if the second loss value is not within the second preset range, continuing to send the underwear feature training set to the underwear feature extraction extension model for iterative training; Obtain a trained underwear feature extraction model, input each underwear feature training sample in the underwear feature training set into the trained underwear feature extraction model one by one, obtain the corresponding underwear feature vector, traverse from 1 to N, and for each traversed value n, perform the following operation, the minimum value and the maximum value of the nth feature item in all underwear feature vectors form the value range corresponding to the nth feature item in the underwear feature vector; until traversing to N, the value range corresponding to each feature item in the underwear feature vector is used as the value range corresponding to each feature item in the simulated underwear feature vector.

4. The underwear simulation optimization design method according to claim 3, characterized in that: The fitness evaluation model is established based on BP neural network; The training of the fitness evaluation model includes the following contents: obtaining several positive samples and negative samples for fitness evaluation. The positive samples for fitness evaluation include the chest region feature vector and underwear feature vector corresponding to the fully adapted chest region model and underwear model. The positive samples for fitness evaluation are labeled by the positive target fitness. The negative samples for fitness evaluation include the chest region feature vector and underwear feature vector corresponding to the fully unfit chest region model and underwear model. The negative samples for fitness evaluation are labeled by the negative target fitness. The noted fitness assessment positive samples and fitness assessment negative samples constitute a fitness assessment training set, and the fitness assessment training set is sent to the fitness assessment model with parameter initialization for training. During the training, the positive target fitness or the negative target fitness is used as the target, and the third loss value is calculated to determine whether the third loss value is within the third preset range. If the third loss value is within the third preset range, the trained fitness assessment model is output; if the third loss value is not within the third preset range, the fitness assessment training set is continued to be sent to the fitness assessment model for iterative training.

5. The underwear simulation optimization design method according to claim 4, characterized in that: The underwear modeling model is built based on variational autoencoder; The training of the underwear modeling model specifically includes the following steps: after completing the training of the underwear feature extraction model, the underwear feature vector is used as the underwear modeling training sample, and the underwear model point cloud data set corresponding to the underwear feature vector is used as the target point cloud data set to annotate the underwear modeling training sample, and the annotated underwear modeling training samples are formed into an underwear modeling training set, and then the underwear modeling training set is sent to the underwear modeling model with parameter initialization for training. During the training, the target point cloud data set is used as the target, the fourth loss value is calculated, and it is determined whether the fourth loss value is within the fourth preset range. If the fourth loss value is within the fourth preset range, the trained underwear modeling model is output; if the fourth loss value is not within the fourth preset range, the underwear modeling training set is continued to be sent to the underwear modeling model for iterative training.

6. The underwear simulation optimization design method according to claim 5, characterized in that: The population optimization algorithm adopts genetic algorithm.

7. A simulation optimization design system for underwear, characterized in that: The system applies a simulation optimization design method for underwear according to any one of claims 1 to 6, comprising: A multi-view character image acquisition module is used to acquire a user's multi-view character image; A multi-view chest region image construction module is used to perform image segmentation operations on the multi-view character images to construct a multi-view chest region image; A chest region model building module, used to perform 3D modeling operations based on multi-view chest region images to build a chest region model; A chest region feature vector extraction module is used to obtain a chest region point cloud data set corresponding to a chest region model, and then send the chest region point cloud data set to a chest feature extraction model for feature extraction to construct a chest region feature vector; The underwear feature vector simulation calculation module is used to simulate and calculate the underwear feature vector based on the chest region feature vector through a group optimization algorithm to construct an optimal underwear feature vector. During the simulation calculation process, the fitness evaluation is performed based on the chest region feature vector and the underwear feature vector, and the evaluated fitness is used as the fitness; The underwear model building module is used to send the optimal underwear feature vector to the underwear modeling model for processing, output the underwear point cloud data set, and build the underwear model based on the underwear point cloud data set.

Images