Clothing image generation method, apparatus, device, and storage medium

Abstract

The application discloses a clothing image generation method and device, equipment and a storage medium, and relates to the technical field of image processing. The method comprises the following steps: obtaining a target model image of a clothing to be put online and a target three-dimensional clothing image corresponding to the target model image, wherein the target three-dimensional clothing image is a three-dimensional clothing image conforming to a model posture in the target model image; extracting a clothing region image from the target model image; generating a key point image based on the target three-dimensional clothing image; and generating a high-quality clothing image corresponding to the clothing to be put online based on the clothing region image and the key point image through a preset clothing image generation model. The application can generate a high-quality clothing image based on a clothing region image and a key point image of a clothing to be put online through a preset clothing image generation model, thereby solving the technical problem that the production of clothing commodity pictures in the prior art depends on post-image processing, and the efficiency is low and the quality is not high.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to methods, apparatus, devices and storage media for generating clothing images. Background Technology

[0002] With the rapid development of the e-commerce industry, the visual appeal of product displays has significantly influenced consumers' purchasing decisions. In the apparel e-commerce sector, the quality and presentation of product images are particularly crucial.

[0003] Currently, the production of clothing product images usually relies on post-processing. This process is not only time-consuming and labor-intensive, but also limited by various factors such as model poses and shooting environment, resulting in low efficiency and low quality in clothing image generation. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, device, and storage medium for generating clothing images, aiming to solve the technical problem that the production of clothing product images in the prior art relies on post-processing, which results in low efficiency and low quality.

[0005] To achieve the above objectives, this application proposes a method for generating clothing images, the method comprising:

[0006] Obtain the target model image of the clothing to be launched and the target 3D clothing image corresponding to the target model image, wherein the target 3D clothing image is a 3D clothing image that matches the model's pose in the target model image;

[0007] Extract the clothing area image from the target model image;

[0008] Generate key point images based on the target 3D clothing image;

[0009] A high-quality clothing image corresponding to the clothing to be launched is generated based on the clothing area image and the key point image using a preset clothing image generation model.

[0010] In one embodiment, the step of generating a key point image based on the target 3D clothing image includes:

[0011] The key points of the human body in the target 3D clothing image are determined by using a key point detection model to detect key points in the target 3D clothing image.

[0012] A key point image is generated based on the aforementioned human body key points.

[0013] In one embodiment, the step of generating a key point image based on the human body key points includes:

[0014] An initial human skeleton diagram is generated based on the key points of the human body using a preset image drawing method.

[0015] The human skeleton model is constructed by taking the key points of the human body in the initial human skeleton diagram as nodes and the connection relationships between the key points as edges.

[0016] The skeleton graph model is associated with key points using a pre-defined graph convolutional network to generate key point images.

[0017] In one embodiment, the step of extracting the clothing area image from the target model image includes:

[0018] The target model image is input into a preset semantic segmentation model, which includes a feature extraction layer.

[0019] The target model image is classified by pixels through the feature extraction layer;

[0020] The clothing area in the target model image is determined based on the classification results;

[0021] The clothing area in the target model image is cropped to obtain a clothing area image.

[0022] In one embodiment, the step of generating a high-quality clothing image corresponding to the clothing to be uploaded based on the clothing region image and the key point image using a preset clothing image generation model includes:

[0023] The clothing area image and the key point image are input into a preset clothing image generation model, which includes a feature extraction module and a texture diffusion module.

[0024] The feature extraction module extracts multi-scale image texture features corresponding to the clothing area image;

[0025] The texture diffusion module generates a high-quality clothing image corresponding to the clothing to be launched based on the multi-scale image texture features and the key point image.

[0026] In one embodiment, the feature extraction module includes a texture encoder; the step of extracting multi-scale image texture features corresponding to the clothing area image through the feature extraction module includes:

[0027] The texture encoder is used to extract multi-scale features from the clothing area image to obtain image texture features at different scales in the clothing area image.

[0028] The image texture features at each scale are stacked to form multi-scale image texture features corresponding to the clothing area image.

[0029] In one embodiment, the texture diffusion module includes a cross-attention submodule; the step of generating a high-quality garment image corresponding to the garment to be launched based on the multi-scale image texture features and the key point image through the texture diffusion module includes:

[0030] Feature extraction is performed on the key point image to generate a pose feature map;

[0031] The multi-scale image texture features are used as keys and key values, and the pose feature map is used as a query, which are then input into the cross-attention submodule.

[0032] The cross-attention submodule generates an attention weight graph based on the key, the key value, and the query.

[0033] An enhanced image texture feature map is generated based on the attention weight map and the pose feature map;

[0034] A high-quality clothing image corresponding to the clothing to be launched is generated based on the enhanced image texture feature map.

[0035] Furthermore, to achieve the above objectives, this application also proposes a clothing image generation apparatus, the apparatus comprising:

[0036] The image acquisition module is used to acquire a target model image of the clothing to be uploaded and a target 3D clothing image corresponding to the target model image, wherein the target 3D clothing image is a 3D clothing image that matches the pose of the model in the target model image;

[0037] The image extraction module is used to extract the clothing area image from the target model image;

[0038] The key point detection module is used to generate a key point image based on the target 3D clothing image;

[0039] The image generation module is used to generate a high-quality clothing image corresponding to the clothing to be launched based on the clothing area image and the key point image using a preset clothing image generation model.

[0040] In addition, to achieve the above objectives, this application also proposes a clothing image generation device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the clothing image generation method as described above.

[0041] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the clothing image generation method described above.

[0042] This application provides a method for generating clothing images. The method discloses obtaining a target model image of the clothing to be launched and a corresponding target 3D clothing image, wherein the target 3D clothing image is a 3D clothing image conforming to the model's pose in the target model image; extracting a clothing region image from the target model image; generating a key point image based on the target 3D clothing image; and generating a high-quality clothing image corresponding to the clothing to be launched based on the clothing region image and the key point image using a preset clothing image generation model. Compared to existing technologies where the production of clothing product images relies on post-processing and is limited by factors such as model pose and shooting environment, this application solves the technical problem of existing technologies where the production of clothing product images relies on post-processing, resulting in low efficiency and low quality. This is because the method can extract a clothing region image from the model image of the clothing to be launched and generate a high-quality clothing image based on the clothing region image and the key point image using a preset clothing image generation model. Attached Figure Description

[0043] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating an embodiment of the clothing image generation method of this application.

[0046] Figure 2 This is a flowchart illustrating the method for generating clothing images in this application.

[0047] Figure 3 This is a flowchart illustrating Embodiment 2 of the clothing image generation method of this application;

[0048] Figure 4 This is a flowchart illustrating Embodiment 3 of the clothing image generation method of this application;

[0049] Figure 5 This is a structural diagram of the preset clothing image generation model in the clothing image generation method of this application;

[0050] Figure 6 This is a schematic diagram of the module structure of the clothing image generation device according to an embodiment of this application;

[0051] Figure 7 This is a schematic diagram of the device structure of the hardware operating environment involved in the clothing image generation method in this application embodiment.

[0052] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0053] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0054] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0055] The main solution of this application embodiment is: to obtain the target model image of the garment to be launched and the target 3D garment image corresponding to the target model image, wherein the target 3D garment image is a 3D garment image that conforms to the model pose in the target model image; to extract the garment region image from the target model image; to generate a key point image based on the target 3D garment image; and to generate a high-quality garment image corresponding to the garment to be launched based on the garment region image and the key point image using a preset garment image generation model.

[0056] Because the production of clothing product images in the current technology relies on post-processing, it is subject to various limitations such as model pose and shooting environment, resulting in low efficiency and low quality in clothing image generation.

[0057] This application provides a solution that can extract clothing area images from model images of clothing to be launched online, and generate high-quality clothing images of the clothing to be launched online based on the clothing area images and key point images using a preset clothing image generation model. This solves the technical problem in the prior art that the production of clothing product images relies on post-processing, which is inefficient and of low quality.

[0058] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone; or an electronic device, clothing image generation device, or clothing image generation system including a clothing image generation device, capable of performing the above functions. The following description uses a clothing image generation system (hereinafter referred to as the system) as an example to illustrate this embodiment and the subsequent embodiments.

[0059] Based on this, embodiments of this application provide a method for generating clothing images, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the clothing image generation method of this application.

[0060] In this embodiment, the clothing image generation method includes steps S10 to S40:

[0061] Step S10: Obtain the target model image of the clothing to be uploaded and the target 3D clothing image corresponding to the target model image. The target 3D clothing image is a 3D clothing image that matches the pose of the model in the target model image.

[0062] It should be understood that the aforementioned clothing items to be launched can be any clothing products that will soon be displayed and sold on e-commerce platforms. These clothing items may have already been photographed with models trying them on, but further processing is required to generate more attractive three-dimensional images.

[0063] It should be noted that the aforementioned target model images can be original images of models wearing clothing to be launched online. These images can usually be taken from different angles to show the effect of the clothing on the model.

[0064] It should be noted that the aforementioned target 3D clothing image can be a clothing image with a sense of three-dimensionality and depth information that conforms to the pose of the model in the target model image. This clothing image can be between two-dimensional (2D) and three-dimensional (3D) and can display the three-dimensional effect of the clothing. Among them, the model pose can be the body posture and movement of the model in the target model image. In this embodiment, the model pose can be described by human body key points, which may include, but are not limited to, the positional information of parts such as the head, shoulders, elbows, wrists, waist, knees, and ankles.

[0065] In practical applications, 3D models of clothing can be created in advance using 3D modeling software (such as Blender, Maya, etc.) and rendered to generate 2.5D images with a stereoscopic effect. These 2.5D images are then stored in the e-commerce platform's database. Subsequently, when it is necessary to generate clothing images for clothing to be launched online, the target model image of a model wearing the clothing can be obtained from the e-commerce platform's database, along with a target stereoscopic clothing image with a 2.5D effect and the same modal pose as the target model image.

[0066] Step S20: Extract the clothing area image from the target model image.

[0067] It is understood that the aforementioned clothing area image can be an image extracted from the target model image that contains only the clothing portion. In this embodiment, the system can use image segmentation technology to separate the clothing area on the model in the target model image from the background and other non-clothing parts to obtain a clothing area image for subsequent processing.

[0068] Further, step S20 includes: inputting the target model image into a preset semantic segmentation model, wherein the preset semantic segmentation model is provided with a feature extraction layer; performing pixel classification on the target model image through the feature extraction layer; determining the clothing region in the target model image based on the classification result; and cropping the clothing region in the target model image to obtain a clothing region image.

[0069] It should be noted that the aforementioned preset semantic segmentation model can be a deep learning model used to classify each pixel in an image into different semantic categories. The semantic segmentation model in this embodiment may include, but is not limited to, U-Net, DeepLab, Mask R-CNN, etc. These models can extract image features through convolutional neural networks (CNN) and assign each pixel to a specific category (such as "clothing", "background", etc.).

[0070] It should be understood that the aforementioned feature extraction layer can be a module in a semantic segmentation model used to extract features from the input image, such as a convolutional layer and a pooling layer. In this embodiment, the feature extraction layer in the preset semantic segmentation model can progressively extract high-level features of the image through convolution operations and non-linear activation functions (such as ReLU), providing a foundation for subsequent pixel classification.

[0071] It is understandable that pixel classification of a target model image refers to the process of assigning each pixel in the target model image to a predefined category. In semantic segmentation, a pre-defined semantic segmentation model can learn the features of the target model image, predict the probability that each pixel belongs to a certain category, and classify the pixels into the category with the highest probability to obtain a classification result. In this embodiment, the classification result can be a label map of the same size as the input image, where the value of each pixel represents its category; for example, a pixel value of 1 may represent "clothing," and a pixel value of 0 may represent "background."

[0072] It is also understood that the aforementioned clothing area can be the clothing area in the target model image, that is, the area in the target model image classified as "clothing". In this embodiment, by using a preset semantic segmentation model, the clothing area on the model in the target model image can be separated from the background and other non-clothing parts to obtain the clothing area image.

[0073] In practical applications, the system can use the target model image as input to a pre-trained semantic segmentation model. Upon receiving the target model image, the semantic segmentation model processes it through its feature extraction layer, extracting image features, including edges, textures, and shapes, providing a foundation for subsequent pixel classification. Then, the semantic segmentation model can classify and predict the category of each pixel in the target model image based on the features extracted by the feature extraction layer, outputting a probability map representing the probability of each pixel belonging to different categories. Based on the probability map, each pixel in the target model image is assigned to the category with the highest probability, generating a classification result. Subsequently, the system generates a segmentation mask based on the classification result (pixels in the clothing area have a value of 1, while pixels in the background and other non-clothing areas have a value of 0), and uses the segmentation mask to extract the clothing area from the target model image, forming a separate clothing area image.

[0074] Step S30: Generate a key point image based on the target 3D clothing image.

[0075] It should be noted that the aforementioned key point image can be an image containing key parts of the human body (such as head, shoulders, elbows, knees, etc.) in the target 3D clothing image. The posture of the human body in the target 3D clothing image can be represented by a simple line drawing or heat map, that is, the posture of the model in the target model image.

[0076] In this embodiment, the system can use a pre-trained human keypoint detection model (such as OpenPose, Hourglass Networks, etc.) to process the target stereoscopic clothing image. Specifically, the human keypoint detection model can extract features from the target stereoscopic clothing image through a convolutional neural network and locate the key points of the human body in the target stereoscopic clothing image, such as the head, shoulders, elbows, waist, knees, etc. Then, based on the detected key point locations, a keypoint image is generated.

[0077] Step S40: Generate a high-quality clothing image corresponding to the clothing to be launched based on the clothing area image and the key point image using a preset clothing image generation model.

[0078] It should be noted that the aforementioned preset clothing image generation model can be a deep learning model used to generate high-quality clothing images with a three-dimensional feel. In this embodiment, the preset clothing image generation model may include, but is not limited to, diffusion models, which can generate high-quality clothing images by learning from a large amount of image data.

[0079] It should be understood that the aforementioned high-quality clothing images can be images with three-dimensionality and depth information corresponding to the clothing to be launched online. These high-quality clothing images retain the texture and color information of the clothing to be launched online, and also have effects such as shadows, perspective, and partial three-dimensional structure. By using them in e-commerce scenarios, it can help users better understand the style, details, and wearing effect of clothing, thereby enhancing the user experience.

[0080] In practical applications, refer to Figure 2 , Figure 2 This is a flowchart illustrating the process of generating a clothing image in the clothing image generation method of this application. Figure 2 As shown, the system extracts the clothing region image x_s from the target model image and generates a keypoint image x_p to represent the model's pose in the target model image y_t. The clothing region image x_s and the keypoint image x_p are then input into a preset clothing image generation model. The preset clothing image generation model encodes the clothing region image x_s and the keypoint image x_p to extract texture features from the clothing region image x_s and pose features from the keypoint image x_p. Then, it fuses the texture and pose features using a cross-attention mechanism or other feature fusion techniques to ensure that the generated clothing image conforms to both the texture details of the clothing and the geometric structure of the target pose. During the clothing image generation process, the model can perform multiple iterative denoising processes on the image, gradually generating high-quality clothing images. In each iteration, the model can combine the current noisy image, pose conditions, and texture features to predict and remove noise, gradually generating a clearer, higher-quality clothing image y_t-1.

[0081] This embodiment provides a method for generating clothing images. The method discloses obtaining a target model image of the clothing to be launched and a corresponding target 3D clothing image, wherein the target 3D clothing image is a 3D clothing image conforming to the model's pose in the target model image; extracting a clothing region image from the target model image; generating a key point image based on the target 3D clothing image; and generating a high-quality clothing image corresponding to the clothing to be launched based on the clothing region image and the key point image using a preset clothing image generation model. Compared to existing technologies where the production of clothing product images relies on post-processing and is limited by factors such as model pose and shooting environment, this embodiment can extract a clothing region image from the model image of the clothing to be launched and generate a high-quality clothing image of the clothing to be launched based on the clothing region image and the key point image using a preset clothing image generation model. This solves the technical problem of existing technologies where the production of clothing product images relies on post-processing, resulting in low efficiency and low quality.

[0082] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 , Figure 3 This is a flowchart illustrating Embodiment 2 of the clothing image generation method of this application.

[0083] In this embodiment, step S30 includes steps S301 to S302:

[0084] Step S301: Perform key point detection on the target 3D clothing image using a key point detection model to determine the human body key points in the target 3D clothing image.

[0085] It should be noted that the aforementioned keypoint detection model can be a deep learning model used to detect the location of key points on the human body in an image. This model can be based on a convolutional neural network and can identify key parts of the human body from the input image. Accordingly, the aforementioned human body keypoints can be the location information of key parts of the human body in the target 3D clothing image, typically including the head, shoulders, elbows, wrists, waist, knees, ankles, etc. The location information of these keypoints can be used to describe the posture and movement of the human body in the target 3D clothing image.

[0086] It should also be noted that the key point detection model can accurately locate key points of the human body in the target 3D clothing image, such as the head, shoulders, elbows, wrists, waist, knees, and ankles. In this embodiment, the model performs accurate key point detection, which can accurately estimate the posture and movement of the human body, ensuring that the generated clothing image conforms to the actual posture of the model, thereby improving the realism and naturalness of the generated clothing image.

[0087] Step S302: Generate a key point image based on the human body key points.

[0088] In practical applications, after acquiring the target 3D clothing image, the system first normalizes the image, adjusting its size and pixel value range to meet the input requirements of the keypoint detection model. Then, the system inputs the preprocessed 3D clothing image into the keypoint detection model. Upon receiving the image, the model extracts image features using a convolutional neural network and generates a heatmap of keypoints. Subsequently, the system extracts the location information of the keypoints from the heatmap, determines the coordinates of each keypoint, and generates a keypoint image based on the detected human body keypoint locations. In this embodiment, the keypoint image can be represented in the form of a simple line drawing, where each keypoint is represented by a dot or a small circle. Alternatively, the keypoint image can be represented as a heatmap, where the location of each keypoint is represented by a Gaussian distributed heatmap.

[0089] Further, step S302 includes: generating an initial human skeleton map based on the human key points using a preset image drawing method; constructing a skeleton map model by using the human key points in the initial human skeleton map as nodes and the connection relationships between each human key point as edges; and generating a key point image by associating the skeleton map model with key points through a preset graph convolutional network.

[0090] It should be noted that the aforementioned preset image drawing method can be an image drawing method for drawing a human skeleton diagram, such as a line drawing method or a heat map drawing method. The line drawing method can refer to using simple lines and dots to represent the key points and connections of the human body, where key points can be represented by small dots, and connections can be represented by straight lines or curves. The heat map drawing method can use heat maps to represent the positions of key points, where each key point corresponds to a heat map channel, and the Gaussian distribution in the heat map represents the position of the key point. In this embodiment, the system can use the line drawing method to draw the human skeleton in the target three-dimensional clothing image to generate a human skeleton diagram.

[0091] It should be understood that the aforementioned initial human skeleton diagram can be a skeleton diagram initially generated based on the location of key points of the human body. This initial skeleton diagram represents the posture of the human body in the form of a simple line drawing and usually includes key points and connecting lines between key points.

[0092] It should be noted that the aforementioned skeleton graph model can be a graph model representing the geometric structure of human posture. The aforementioned preset graph convolutional network can be a graph-based deep learning model, which can be used to process the skeleton graph model to perform posture correction on the image. In this embodiment, the graph convolutional network can optimize the position and connection relationships of key points by learning the relationships between nodes in the skeleton graph model, thereby generating high-quality key point images.

[0093] In practical applications, the system first uses a keypoint detection model (such as OpenPose) to extract the positions of human keypoints from the target 3D clothing image. Based on the extracted keypoint positions, an initial human skeleton image is generated using a simplified drawing method. This includes drawing keypoints (such as the head, shoulders, and elbows) and connecting lines (such as arms and legs), which represent the connections between keypoints. Then, the system performs pose correction on the initial human skeleton image to generate a keypoint image. Specifically, the system first defines each keypoint in the initial human skeleton image as a node and the connections between keypoints as edges, constructing a skeleton image model for subsequent graph convolutional network processing. Afterward, the system inputs the constructed skeleton image model into a pre-defined graph convolutional network. Upon receiving the skeleton image model, the pre-defined graph convolutional network learns the relationships between nodes in the model to optimize the positions and connections of keypoints. It then updates node features through graph convolution operations to ensure that the geometric relationships between keypoints conform to human anatomy. Finally, based on the optimized keypoint positions and connections, the final keypoint image is generated. Key point images can be represented in the form of line drawings or heat maps.

[0094] This embodiment discloses a method for detecting key points in a target 3D clothing image using a key point detection model to determine key human body points in the target 3D clothing image; and generating a key point image based on the key human body points. Since this embodiment can accurately locate the key human body points in the target 3D clothing image using a key point detection model, it is possible to accurately estimate the posture and movement of the human body, ensuring that the generated key point image conforms to the actual posture of the model, thereby improving the realism of the subsequently generated clothing image.

[0095] Based on the first and / or second embodiments of this application, in the third embodiment of this application, the content that is the same as or similar to the above embodiments can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 4 , Figure 4 This is a flowchart illustrating Embodiment 3 of the clothing image generation method of this application.

[0096] In this embodiment, step S40 includes steps S401 to S403:

[0097] Step S401: Input the clothing area image and the key point image into a preset clothing image generation model, which includes a feature extraction module and a texture diffusion module.

[0098] It should be noted that the aforementioned preset clothing image generation model can be a deep learning model used to generate high-quality clothing images, such as a diffusion model, which can generate high-quality images by progressively removing noise. In this embodiment, the preset clothing image generation model typically combines techniques such as feature extraction and texture diffusion to generate clothing images with a sense of three-dimensionality and high-quality details.

[0099] It should be understood that the aforementioned feature extraction module and texture diffusion module are components of a pre-defined clothing image generation model. The feature extraction module extracts features from the input image, and can be a convolutional neural network or a U-Net (encoder-decoder structure). In this embodiment, the image features extracted by the feature extraction module may include, but are not limited to, edge, texture, and color information, providing a foundation for subsequent image generation. The texture diffusion module fuses the extracted multi-scale texture features with the pose information of the keypoint image, and generates a high-quality clothing image through a diffusion process. In this embodiment, the texture diffusion module can be based on a cross-attention mechanism or other feature fusion techniques to ensure that the generated clothing image retains both the texture details of the clothing and conforms to the geometric structure of the modal pose.

[0100] Step S402: Extract the multi-scale image texture features corresponding to the clothing area image through the feature extraction module.

[0101] It should be noted that the aforementioned multi-scale image texture features can be multi-scale features extracted from the clothing area image. These features can include texture, edge, and color information at different levels. This embodiment, by extracting multi-scale features from the clothing area image, can provide rich details for the generation of clothing images, thereby helping to improve the quality of the generated clothing images.

[0102] Furthermore, the feature extraction module includes a texture encoder; step S402 includes: performing multi-scale feature extraction on the clothing area image through the texture encoder to obtain image texture features at different scales in the clothing area image; stacking the image texture features at each scale to form multi-scale image texture features corresponding to the clothing area image.

[0103] Understandably, the texture encoder described above can be a component used to extract multi-scale features from an input image, which is typically based on a convolutional neural network (CNN).

[0104] It is also understood that the aforementioned image texture features at different scales can refer to multi-scale features extracted from clothing area images. These features can include texture, edge, and color information at different levels. In this embodiment, the texture encoder can progressively extract the texture, edge, and color information of the input image through multiple convolutional and pooling layers. This feature extraction process can be performed at multiple scales to capture the texture information of the image from detail to overall texture.

[0105] In practical applications, after the target model image is input into a preset clothing image generation model, the texture encoder in the preset clothing image generation model can extract features of the image at different levels through multiple convolutional layers and pooling layers, where the feature map at each level captures texture information at different scales. Then, the texture encoder can stack the feature maps of different scales according to the channel direction to obtain the multi-scale image texture features corresponding to the clothing area image.

[0106] Step S403: Generate a high-quality clothing image corresponding to the clothing to be launched based on the multi-scale image texture features and the key point image through the texture diffusion module.

[0107] Furthermore, the texture diffusion module includes a cross-attention submodule; step S403 includes: extracting features from the keypoint image to generate a pose feature map; inputting the multi-scale image texture features as keys and key values, and the pose feature map as a query, into the cross-attention submodule; generating an attention weight map based on the keys, key values, and query through the cross-attention submodule; generating an enhanced image texture feature map based on the attention weight map and the pose feature map; and generating a high-quality clothing image corresponding to the clothing to be launched based on the enhanced image texture feature map.

[0108] It should be understood that the aforementioned pose feature map can be a feature map of the geometric structure representing the human pose extracted from the keypoint image. It can be extracted by a convolutional neural network and contains the position and pose information of the keypoints.

[0109] It should be noted that in the cross-attention mechanism, the key and key value are feature vectors extracted from multi-scale image texture features. The key is used to calculate the attention weights, and the key value is used to generate the final feature representation. The query is a feature vector extracted from the pose feature map, used to calculate similarity with the key to generate attention weights.

[0110] It should also be noted that the aforementioned cross-attention submodule can be a module that fuses pose feature maps with multi-scale image texture features through an attention mechanism. In this embodiment, the cross-attention submodule can generate an attention weight map by calculating the similarity between the query and the key, and use the key value to generate an enhanced image texture feature map.

[0111] It should be noted that the attention weight map mentioned above can be a weight map used to represent the degree of attention paid to multi-scale image texture features at each location in the pose feature map. It can be used to weighted sum the key values ​​to generate an enhanced image texture feature map. The enhanced image texture feature map can be a feature map that fuses the pose feature map and multi-scale image texture features, which can preserve the texture details of the clothing while conforming to the geometric structure of the modal pose.

[0112] In the specific implementation, refer to Figure 5 , Figure 5 This is a structural diagram of the preset clothing image generation model in the clothing image generation method of this application. (See diagram for example.) Figure 5 As shown, the system first uses a human keypoint detection model to extract keypoints from the target 3D clothing image y_t, and generates a keypoint image x_p based on the extracted keypoints. Simultaneously, the system uses a semantic segmentation model (such as U-Net) to analyze the input original model image, accurately segmenting the clothing regions (such as tops, pants, and dresses), and cropping the segmented clothing regions to form independent clothing region images x_s. In this embodiment, the preset clothing image generation model can adopt a dual-branch encoder design, including a noise prediction module, a texture encoder, and a cross-attention-based texture diffusion block (which internally includes a feature extraction module and a cross-attention submodule). The noise prediction module can be a U-Net-based backbone network responsible for the core denoising task; the texture encoder can be an independent encoder network used to process the clothing region image x_s to extract rich, multi-scale texture features from x_s; and the cross-attention-based texture diffusion block can efficiently and accurately inject the texture information of x_s into the generation process. In this embodiment, the texture diffusion block can embed the cross-attention mechanism module into various levels of the noise prediction module.

[0113] In this embodiment, after obtaining the clothing region image x_s, the system can input the clothing region image x_s into the texture encoder to obtain multi-scale texture features Fs. In each decoding layer (upsampling layer) of the noise prediction module, these texture features Fs can serve as keys and values, while the feature map of the current decoding layer of the noise prediction module serves as a query. Through the interaction of the query and key, the model can calculate an attention feature map, which clearly defines which part of the x_s texture should be "focused" on at each position of the current layer of the noise prediction module. Then, this weight is used to perform a weighted summation of the values ​​(texture features) and fused back into the features of the noise prediction module. During inference, the model can utilize the trained knowledge to generate the final high-quality clothing image starting from random noise. The sampling process can start with random noise that conforms to a Gaussian distribution. The process is a multi-step iterative denoising process. At each step, the noise prediction module can combine the current noisy image and pose condition x_p, and predict the noise of the current step and remove it through texture features Fs injected by cross-attention, thereby obtaining a clearer image. This process is then repeated until the end, and finally a high-quality clothing image is obtained. The texture of x_s can be preserved in the high-quality clothing image, but it is presented in the pose of x_p.

[0114] In this embodiment, a method is disclosed to input a clothing region image and a key point image into a preset clothing image generation model. The preset clothing image generation model includes a feature extraction module and a texture diffusion module. The feature extraction module extracts multi-scale image texture features corresponding to the clothing region image. The texture diffusion module generates a high-quality clothing image corresponding to the clothing to be uploaded based on the multi-scale image texture features and the key point image. Since this embodiment can extract multi-scale image texture features corresponding to the clothing region image through the feature extraction module, it can provide rich texture details for the generation of clothing images, which is beneficial to improving the quality of the generated clothing images.

[0115] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the clothing image generation method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0116] This application also provides a clothing image generation device, please refer to... Figure 6 The clothing image generation device includes:

[0117] Image acquisition module 10 is used to acquire a target model image of the clothing to be uploaded and a target 3D clothing image corresponding to the target model image, wherein the target 3D clothing image is a 3D clothing image that conforms to the model's pose in the target model image;

[0118] Image extraction module 20 is used to extract clothing area images from the target model image;

[0119] Key point detection module 30 is used to generate key point images based on the target 3D clothing image;

[0120] The image generation module 40 is used to generate a high-quality clothing image corresponding to the clothing to be launched based on the clothing area image and the key point image using a preset clothing image generation model.

[0121] The clothing image generation apparatus provided in this application, employing the clothing image generation method described in the above embodiments, can solve the technical problem in the prior art where the production of clothing product images relies on post-processing, resulting in low efficiency and poor quality. Compared with the prior art, the beneficial effects of the clothing image generation apparatus provided in this application are the same as those of the clothing image generation method provided in the above embodiments, and other technical features in the clothing image generation apparatus are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0122] This application provides a clothing image generation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the clothing image generation method in Embodiment 1 above.

[0123] The following is for reference. Figure 7 This document illustrates a structural schematic diagram of a clothing image generation device suitable for implementing embodiments of this application. The clothing image generation device in these embodiments may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 7 The clothing image generation device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0124] like Figure 7As shown, the garment image generation device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the garment image generation device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. The communication device 1009 allows the garment image generation device to communicate wirelessly or wiredly with other devices to exchange data. Although garment image generation devices with various systems are shown in the figures, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0125] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0126] The clothing image generation device provided in this application, employing the clothing image generation method in the above embodiments, can solve the technical problem of clothing image generation. Compared with the prior art, the beneficial effects of the clothing image generation device provided in this application are the same as those of the clothing image generation method provided in the above embodiments, and other technical features in this clothing image generation device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0127] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0128] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0129] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the clothing image generation method in the above embodiments.

[0130] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0131] The aforementioned computer-readable storage medium may be included in the garment image generating device; or it may exist independently and not be assembled into the garment image generating device.

[0132] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the garment image generation device, the garment image generation device causes the garment image generation device to: acquire a target model image of the garment to be uploaded and a target stereoscopic garment image corresponding to the target model image, wherein the target stereoscopic garment image is a stereoscopic garment image that conforms to the pose of the model in the target model image; extract a garment region image from the target model image; generate a key point image based on the target stereoscopic garment image; and generate a high-quality garment image corresponding to the garment to be uploaded based on the garment region image and the key point image using a preset garment image generation model.

[0133] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0134] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0135] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0136] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described clothing image generation method. This solves the technical problem in the prior art where the creation of clothing product images relies on post-processing, resulting in low efficiency and poor quality. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the clothing image generation method provided in the above embodiments, and will not be repeated here.

[0137] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for generating clothing images, characterized in that, The method includes: Obtain the target model image of the clothing to be launched and the target 3D clothing image corresponding to the target model image, wherein the target 3D clothing image is a 3D clothing image that matches the model's pose in the target model image; Extract the clothing area image from the target model image; Generate key point images based on the target 3D clothing image; A high-quality clothing image corresponding to the clothing to be launched is generated based on the clothing area image and the key point image using a preset clothing image generation model. The step of generating a high-quality clothing image corresponding to the clothing to be uploaded based on the clothing region image and the key point image using a preset clothing image generation model includes: The clothing region image and the key point image are input into a preset clothing image generation model. The preset clothing image generation model is equipped with a feature extraction module and a texture diffusion module. The texture diffusion module is equipped with a cross attention submodule. The feature extraction module extracts multi-scale image texture features corresponding to the clothing area image; Feature extraction is performed on the key point image to generate a pose feature map; The multi-scale image texture features are used as keys and key values, and the pose feature map is used as a query, which are then input into the cross-attention submodule. The cross-attention submodule generates an attention weight graph based on the key, the key value, and the query. An enhanced image texture feature map is generated based on the attention weight map and the pose feature map; A high-quality clothing image corresponding to the clothing to be launched is generated based on the enhanced image texture feature map.

2. The method as described in claim 1, characterized in that, The step of generating a key point image based on the target 3D clothing image includes: The key points of the human body in the target 3D clothing image are determined by using a key point detection model to detect key points in the target 3D clothing image. A key point image is generated based on the aforementioned human body key points.

3. The method as described in claim 2, characterized in that, The step of generating a key point image based on the human body key points includes: An initial human skeleton diagram is generated based on the key points of the human body using a preset image drawing method. The human skeleton model is constructed by taking the key points of the human body in the initial human skeleton diagram as nodes and the connection relationships between the key points as edges. The skeleton graph model is associated with key points using a pre-defined graph convolutional network to generate key point images.

4. The method as described in claim 1, characterized in that, The step of extracting the clothing area image from the target model image includes: The target model image is input into a preset semantic segmentation model, which includes a feature extraction layer. The target model image is classified by pixels through the feature extraction layer; The clothing area in the target model image is determined based on the classification results; The clothing area in the target model image is cropped to obtain a clothing area image.

5. The method as described in claim 1, characterized in that, The feature extraction module includes a texture encoder; the step of extracting multi-scale image texture features corresponding to the clothing area image through the feature extraction module includes: The texture encoder is used to extract multi-scale features from the clothing area image to obtain image texture features at different scales in the clothing area image. The image texture features at each scale are stacked to form multi-scale image texture features corresponding to the clothing area image.

6. A clothing image generation device, characterized in that, The device includes: The image acquisition module is used to acquire a target model image of the clothing to be uploaded and a target 3D clothing image corresponding to the target model image, wherein the target 3D clothing image is a 3D clothing image that matches the pose of the model in the target model image; The image extraction module is used to extract the clothing area image from the target model image; The key point detection module is used to generate a key point image based on the target 3D clothing image; The image generation module is used to generate a high-quality clothing image corresponding to the clothing to be launched based on the clothing area image and the key point image using a preset clothing image generation model; The image generation module is further configured to input the clothing region image and the key point image into a preset clothing image generation model, wherein the preset clothing image generation model includes a feature extraction module and a texture diffusion module, and the texture diffusion module includes a cross-attention submodule; the feature extraction module extracts multi-scale image texture features corresponding to the clothing region image; the key point image is used to extract features to generate a pose feature map; the multi-scale image texture features are used as keys and key values, and the pose feature map is used as a query, and input into the cross-attention submodule; the cross-attention submodule generates an attention weight map based on the keys, key values, and query; an enhanced image texture feature map is generated based on the attention weight map and the pose feature map; and a high-quality clothing image corresponding to the clothing to be launched is generated based on the enhanced image texture feature map.

7. A clothing image generation device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the clothing image generation method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the clothing image generation method as described in any one of claims 1 to 5.

Images