Object recommendation method, apparatus, device, and computer storage medium

Abstract

Embodiments of the present application provide an object recommendation method, device and equipment and a computer storage medium. The object recommendation method comprises: obtaining a reference image, the reference image comprising a target object; performing a search in a matching image library based on the reference image to obtain an associated matching image, the associated matching image comprising at least an associated object and a matching object corresponding to the associated object, the similarity between the associated object and the target object being greater than or equal to a preset threshold; performing a search in a recommended object library based on the associated matching image to obtain a plurality of matching object images, the recommended object library comprising subject images of a plurality of recommendable objects; using a multi-image matching model to analyze and match the associated matching image and the plurality of matching object images to determine a recommended matching object image; and performing an object recommendation operation based on the recommended matching object image. The embodiments can automatically and efficiently perform the object recommendation operation while reducing the cost required for the object recommendation operation.

Description

Technical Field

[0001] This application relates to the field of information processing technology, and in particular to an object recommendation method, apparatus, device, and computer storage medium. Background Technology

[0002] With the rapid development of internet technology, e-commerce platforms are being used more and more widely. Currently, the apparel industry on e-commerce platforms is undergoing a significant sales paradigm upgrade, shifting from a traffic-centric model focused on promoting individual items to a "scenario-based outfit combination" recommendation model aimed at enhancing user value. For example, when a user browses a top, the platform not only recommends similar tops but also aims to recommend a complete outfit consisting of matching pants, shoes, and accessories. This model better meets users' styling needs and significantly improves conversion rates.

[0003] Currently, recommendations are often made by human stylists who configure "scenario-based outfit combinations," which is not only inefficient and costly, but also limited in style and fails to meet the personalized needs of a large number of users. Summary of the Invention

[0004] This application provides an object recommendation method, apparatus, device, and computer storage medium that can automatically and efficiently perform object recommendation operations. This not only ensures the quality and efficiency of object recommendation but also reduces the cost of object recommendation operations and meets the personalized needs of different users.

[0005] In a first aspect, embodiments of the present invention provide an object recommendation method, including: Obtain a reference image, wherein the reference image includes the target object; Based on the reference image, a search is performed in the matching image library to obtain related matching images. The related matching images include at least: a related object and a matching object corresponding to the related object. The similarity between the related object and the target object is greater than or equal to a preset threshold. Based on the associated matching images, a search is performed in the recommendation object library to obtain multiple matching object images, wherein the recommendation object library includes the main images of multiple recommendable objects; The associated matching image and the multiple matching object images are analyzed and matched using a multi-image matching model to determine the recommended matching object image; Perform object recommendation operations based on the recommended matching object images.

[0006] Secondly, embodiments of the present invention provide an object recommendation device, comprising: The first acquisition module is used to acquire a reference image, wherein the reference image includes the target object; The first retrieval module is used to search in the matching image library based on the reference image to obtain related matching images, wherein the related matching images include at least: a related object and a matching object corresponding to the related object, and the similarity between the related object and the target object is greater than or equal to a preset threshold. The first retrieval module is further configured to perform a retrieval in the recommendation object library based on the associated matching images to obtain multiple matching object images, wherein the recommendation object library includes the main images of multiple recommendable objects; The first processing module is used to analyze and match the associated matching image and the multiple matching object images using a multi-image matching model to determine the recommended matching object image; The first processing module is further configured to perform an object recommendation operation based on the recommended matching object image.

[0007] Thirdly, embodiments of the present invention provide a clothing recommendation method, including: Obtain a reference image, wherein the reference image includes the target clothing; Based on the reference image, a search is performed in the matching image library to obtain related matching images. The related matching images include at least: related clothing and matching clothing corresponding to the related clothing. The similarity between the related clothing and the target clothing is greater than or equal to a preset threshold. Based on the associated matching images, a search is performed in the recommendation object library to obtain multiple matching clothing images, wherein the recommendation object library includes multiple main images of recommendable clothing; The associated matching image and the multiple matching clothing images are analyzed and matched using a multi-image matching model to determine the recommended matching clothing image; The clothing recommendation operation is performed based on the recommended clothing images.

[0008] Fourthly, embodiments of the present invention provide a clothing recommendation device, comprising: The second acquisition module is used to acquire a reference image, wherein the reference image includes the target clothing; The second retrieval module is used to search the matching image library based on the reference image to obtain related matching images. The related matching images include at least: related clothing and matching clothing corresponding to the related clothing. The similarity between the related clothing and the target clothing is greater than or equal to a preset threshold. The second retrieval module is further configured to perform a retrieval in the recommendation object library based on the associated matching images to obtain multiple matching clothing images, wherein the recommendation object library includes multiple main images of recommendable clothing; The second processing module is used to analyze and match the associated matching images and the multiple matching clothing images using a multi-image matching model to determine the recommended matching clothing images; The second processing module is also used to perform clothing recommendation operations based on the recommended clothing images.

[0009] Fifthly, embodiments of the present invention provide an electronic device, including: a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions, when executed by the processor, implement the methods in the first or second aspect described above.

[0010] In a sixth aspect, embodiments of the present invention provide a computer storage medium for storing a computer program, which, when executed by a computer, implements the methods described in the first or second aspect above.

[0011] In a seventh aspect, embodiments of the present invention provide a computer program product, comprising: a computer-readable storage medium storing computer instructions, which, when executed by one or more processors, cause one or more processors to perform the steps of the methods in the first or second aspect described above.

[0012] The object recommendation method, apparatus, device, and computer storage medium provided in this embodiment acquire a reference image, search a matching image library based on the reference image to obtain associated matching images, then search a recommendation object library based on the associated matching images to obtain multiple matching object images, and use a multi-image matching model to analyze and match the associated matching images and the multiple matching object images to determine recommended matching object images; then perform object recommendation operations based on the recommended matching object images, thereby enabling automatic and efficient object recommendation operations. This not only ensures the quality and efficiency of object recommendation operations but also reduces the time and labor costs required for object recommendation operations. Furthermore, it allows for personalized object recommendation operations for different reference images, thereby meeting the personalized matching needs of different users and further improving the practicality of the object recommendation method. Attached Figure Description

[0013] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A schematic diagram illustrating a scenario of an object recommendation method provided for an exemplary embodiment of this application; Figure 2 A flowchart illustrating an object recommendation method provided for an exemplary embodiment of this application; Figure 3 A flowchart illustrating another object recommendation method provided for an exemplary embodiment of this application; Figure 4 A schematic diagram illustrating the process of retrieving multiple matching object images from a recommendation object library based on the associated matching images, as provided in an exemplary embodiment of this application; Figure 5 This is a schematic diagram illustrating a process for analyzing and matching the associated matching image and the plurality of matching object images using a multi-image matching model to determine the recommended matching object image, provided as an exemplary embodiment of this application. Figure 6 A flowchart illustrating an exemplary embodiment of this application for a clothing recommendation method; Figure 7 A schematic diagram of the structure of an object recommendation device provided for an exemplary embodiment of this application; Figure 8 A schematic diagram of the structure of an electronic device provided in an exemplary embodiment of this application; Figure 9 A schematic diagram of the structure of an apparel recommendation device provided for an exemplary embodiment of this application; Figure 10 A schematic diagram of the structure of an electronic device provided for an exemplary embodiment of this application. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] It should be noted that, in the case of user information involved in the embodiments of this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0016] The various models involved in this application (including but not limited to language models or large models) comply with relevant laws and standards. Furthermore, the technical solutions provided in the embodiments of this application can employ deep learning models with relatively large parameter scales. The large model is merely an example, and the embodiments of this application do not limit the number of model parameters supported by the deep learning model used, aiming to meet actual needs. The deep learning models involved in the embodiments of this application can be artificial intelligence-based language models (LM) or multimodal models (MM).

[0017] Additionally, it should be noted that when user interaction operations or triggering operations are involved in the embodiments of this application, these operations include, but are not limited to, various interaction methods such as touch operations, gesture operations, voice operations, head movement operations, and eye movement operations. Touch operations include, but are not limited to, click operations, double-click operations, long-press operations, swipe operations, pinch operations, or mouse hover operations. Swipe operations include, but are not limited to, straight-line swipes and curved-line swipes.

[0018] To facilitate understanding of the object recommendation method, apparatus, device, and computer storage medium provided in the embodiments of this application, the relevant technologies are briefly described below: Currently, with the rapid development of e-commerce technology, the apparel industry is undergoing a significant sales paradigm upgrade. This upgrade is shifting from a past model focused on driving traffic and promoting individual items to a "scenario-based outfit combination" sales model aimed at increasing user value and average order value. For example, when a user browses a top, the platform not only recommends similar tops but also aims to recommend a complete outfit consisting of matching pants, shoes, and accessories. This model not only better meets users' styling needs but also significantly improves conversion rates and average order value.

[0019] Currently, e-commerce platforms rely on "scenario-based outfit combinations" configured by human stylists to recommend products. This approach is not only inefficient and costly, but also limited in style options, failing to meet the personalized needs of a massive user base. This presents a significant challenge to achieving scalable, scenario-based outfit recommendation solutions. Therefore, how to leverage artificial intelligence (AI) technology to achieve automated, high-quality, and scalable clothing outfit recommendations remains a crucial technical challenge for the service industry.

[0020] To address the aforementioned technical issues, related technology 1 provides an end-to-end clothing matching method using a Large Language Model (LLM). However, this method has the following drawbacks: due to the lack of in-depth fashion expertise, LLM has limited ability to understand complex visual information such as style, cut, material, and color, which is difficult to accurately describe in text, resulting in generated matching suggestions that often lack professionalism and practicality.

[0021] Related technology 2 provides a method to directly understand reference images and generate matching images using a multi-modal large language model (MLLM). However, this method has the following drawbacks: although existing MLLMs can understand image content, they cannot perform global, fine-grained comparisons and matches within a product library of millions or even hundreds of millions of stock keeping units (SKUs), resulting in recommendations that cannot be translated into specific purchasable products.

[0022] Related technology 3 provides a method for single-item matching and recommendation based on image search. Specifically, it first finds high-quality outfit images, and then uses image search technology to find the same or similar SKUs for each item in the high-quality outfit images. However, this method has the following drawbacks: 1) The products in the outfit images are affected by factors such as shooting light, model posture, clothing wrinkles, and post-processing filters, resulting in significant differences from the white background / model images of the product SKUs; 2) Subtle differences in color (such as off-white and pure white) and style (such as V-neck and round neck) between products mean that the accuracy of existing MLLM direct visual matching is far from meeting the needs of the scenario.

[0023] To address the aforementioned technical problems, embodiments of this application provide an object recommendation method, apparatus, device, and computer storage medium, as detailed in the appendix. Figure 1As shown, the execution entity of this object recommendation method can be the object recommendation device 200, which can be implemented as a local server, a cloud server, or an edge server. When the object recommendation device 200 is implemented as a cloud server, the object recommendation method can be executed in the cloud. Several computing nodes (cloud servers) can be deployed in the cloud, each with computing and storage resources. In the cloud, multiple computing nodes can be organized to provide a certain service; of course, a single computing node can also provide one or more services. The cloud can provide this service by providing an external service interface, which users call to use the corresponding service. Service interfaces include Software Development Kits (SDKs) and Application Programming Interfaces (APIs).

[0024] The object recommendation device 200 is communicatively connected to the client 100, which is used by the user to trigger object recommendation operations. The client 100 can be any computing device with a certain information interaction capability; specifically, it can be a mobile phone, a personal computer (PC), a tablet computer, a settings application, etc. Furthermore, the basic structure of the client 100 may include at least one processor. The number of processors depends on the client's configuration and type. The client 100 may also include memory, which can be volatile, such as Random Access Memory (RAM), or non-volatile, such as Read-Only Memory (ROM), flash memory, etc., or both types. The memory typically stores the operating system (OS), one or more applications, and may also store program data. In addition to the processing unit and memory, the client 100 also includes some basic configurations, such as a network interface card (NIC) chip, an I / O bus, a display component, and some peripheral devices. Optionally, some peripheral devices may include, for example, a keyboard, a mouse, a stylus, a printer, etc. Other peripheral devices are well known in the art and will not be described in detail here.

[0025] Object recommendation device 200 refers to a device capable of performing object recommendation operations in a network virtual environment, typically referring to a device that utilizes a network for information planning and object recommendation operations. Object recommendation device 200 can be an object recommendation model used to implement object recommendation operations. Physically, object recommendation device 200 can be any device capable of providing computing services and performing corresponding object recommendation operations, such as processors, servers, etc. The composition of object recommendation device 200 mainly includes a processor, hard disk, memory, system bus, etc., and its architecture is similar to that of a general-purpose computer.

[0026] In this embodiment described above, the object recommendation device 200 and the client 100 establish a network connection, which can be a wireless or wired network connection. If the object recommendation device 200 and the client 100 establish a communication connection, the mobile network standard can be any one of 2G (Global System for Mobile Communications GSM), 2.5G (General Packet Radio Service GPRS), 3G (Wideband Code Division Multiple Access (WCDMA), Time Division Synchronous Code Division Multiple Access (TD-SCDMA), 4G (Long Term Evolution LTE), 4G+ (Enhanced Long Term Evolution LTE+), Global Microwave Access Interoperability (WiMax), 5G, 6G, etc.

[0027] In this embodiment, the client 100 is used by a user to generate a reference image for triggering an object recommendation operation. The reference image includes a target object, which can be any of the following: clothing, home furnishings, apparel, etc. To enable the object recommendation operation, the reference image can be sent to the object recommendation device 200, allowing the device to perform a corresponding object recommendation operation based on the reference image.

[0028] The object recommendation device 200 is used to acquire reference images sent by the client 100. Since the reference images include the main information or visual information of the target object, in order to implement the recommendation of other matching objects in the application scenario of searching for the target object, a retrieval operation can be performed in the matching image library based on the reference images. The matching image library includes multiple high-quality matching images, and each matching image includes at least two matched objects. Different application scenarios may correspond to different types of objects. For example, in the application scenario of clothing, the matching image may include a matched shirt and jeans; or, in the application scenario of home furnishing, the matching image may include a matched sofa and coffee table.

[0029] After searching the matching image library using the reference image, you can obtain the associated matching images corresponding to the reference image. There can be one or more associated matching images. Each associated matching image includes at least: the associated object and the matching object corresponding to the associated object. The similarity between the associated object and the target object is greater than or equal to a preset threshold, that is, the similarity between the associated object and the target object is high. For example, if the target object is "pants A", the associated object can be "pants A'" or "pants A"; if the target object is "sofa B", the associated object can be "sofa B'".

[0030] To enable object recommendation based on a user-specified target object, after obtaining the associated matching image, a search can be performed in the recommendation object database based on the associated matching image to obtain multiple matching object images. The recommendation object database includes main images of multiple recommendable objects. Then, a multi-image matching model is used to analyze and match the associated matching image and the multiple matching object images to obtain recommended matching object images corresponding to the reference image. The number of recommended matching object images can be one or more, and these images can include matching objects suitable for recommendation. Object recommendation can then be performed based on these recommended matching object images. Specifically, the recommended matching object images can be directly sent to client 100 for display, or a recommended matching image can be generated based on the recommended matching object image and the target object, and then sent to client 100 for display. Thus, when a user searches for information based on the reference image, not only can relevant information about the target object be obtained and displayed, but also relevant information about the recommended matching objects and the matching effect between the recommended matching objects and the target object can be obtained and displayed, thereby achieving automatic and efficient object recommendation.

[0031] In this embodiment, the method can automatically and efficiently recommend matching objects in application scenarios where target objects are searched. This not only ensures the quality and efficiency of object recommendation operations, but also reduces the time and manpower costs required for object recommendation operations. At the same time, it can perform personalized object recommendation operations for different reference images, thereby meeting the personalized needs of different users and further improving the practicality of the object recommendation method.

[0032] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0033] Figure 2 A flowchart illustrating an object recommendation method provided as an exemplary embodiment of this application; see attached diagram. Figure 2As shown, this embodiment provides an object recommendation method. The execution subject of this method is an object recommendation device, which can be implemented as software or a combination of software and hardware. When the object recommendation device is implemented as hardware, it can be various electronic devices capable of performing object recommendation operations. In some instances, the object recommendation device can be implemented as an application client, server, cloud server, etc. When the object recommendation device is implemented as software, it can be installed in the electronic devices exemplified above. Specifically, the object recommendation method provided in this embodiment can include: Step S201: Obtain a reference image, which includes the target object.

[0034] Step S202: Search the matching image library based on the reference image to obtain related matching images. The related matching images include at least: the related object and the matching object corresponding to the related object. The similarity between the related object and the target object is greater than or equal to a preset threshold.

[0035] Step S203: Search the recommendation object library based on the associated matching images to obtain multiple matching object images, wherein the recommendation object library includes the main images of multiple recommendable objects.

[0036] Step S204: Use a multi-image matching model to analyze and match the associated matching image and multiple matching object images to determine the recommended matching object image.

[0037] Step S205: Perform object recommendation operation based on the recommended matching object image.

[0038] The specific implementation methods and principles of each of the above steps are explained in detail below: Step S201: Obtain a reference image, which includes the target object.

[0039] In object search applications, to ensure accuracy and reliability while better meeting potential object matching needs, reference images can be obtained. These reference images can include target objects, which can be objects of interest to the user. In different application scenarios, target objects can correspond to different types of objects. For example, in clothing applications, target objects can be clothing (e.g., jeans, shirts, T-shirts, etc.), accessories (e.g., handkerchiefs, gloves, ties, jewelry, cuffs, etc.), or shoes and hats, etc.; in home applications, target objects can be sofas, coffee tables, TV cabinets, etc.

[0040] The reference image can be obtained through image selection operations in an image search scenario. In this case, obtaining the reference image may include: displaying a human-computer interaction interface, which includes multiple displayed images; obtaining the selection operation input by the user for any of the displayed images (including: click operation, double-click operation, long-term viewing operation, etc.); and obtaining the reference image in response to the selection operation input for any of the displayed images. This effectively ensures the accuracy and reliability of obtaining the reference image.

[0041] In other instances, reference images can be obtained not only through image selection operations but also through text-based generalized search operations in text search scenarios. In this case, obtaining a reference image may include: obtaining the search text entered by the user; performing a generalized search operation based on the search text to obtain multiple images that match the search text; and obtaining the reference image in response to the user's image selection operation on any one of the multiple images. This also ensures the accuracy and reliability of obtaining the reference image.

[0042] Step S202: Search the matching image library based on the reference image to obtain related matching images. The related matching images include at least: the related object and the matching object corresponding to the related object. The similarity between the related object and the target object is greater than or equal to a preset threshold.

[0043] After obtaining the reference image, a search operation can be performed in the matching image library based on the reference image to obtain related matching images. The matching image library can include multiple high-quality matching images, and the obtained related matching images can be at least a portion of these multiple matching images; that is, the number of obtained related matching images can be one or more. For each related matching image, it includes at least a related object and a matching object corresponding to the related object. The similarity between the related object and the target object is greater than or equal to a preset threshold, meaning the similarity between the related object and the target object is high. In some specific scenarios, the related object can be the target object.

[0044] For example, if the target object in the reference image is "jeans A", the resulting associated matching image may include "jeans A1" and "T-shirt B"; or, the associated matching image may also include "jeans A" and "shirt C", etc. Alternatively, if the target object in the reference image is "sofa L", the resulting associated matching image may include "sofa L1" and "coffee table M"; or, the associated matching image may also include "sofa L" and "TV cabinet N", etc.

[0045] Furthermore, for associated matching images, they can be obtained directly through a search operation based on the similarity between a reference image and matching images in a matching image library. In this case, retrieving associated matching images from the matching image library based on the reference image can include: obtaining multiple matching images from the matching image library; determining the image similarity between the reference image and each matching image; and obtaining the associated matching image based on the image similarity between the reference image and each matching image. When there is only one associated matching image, the multiple matching images can be sorted based on the image similarity between the reference image and each matching image, and the matching image with the highest image similarity is determined as the associated matching image. Alternatively, when there are multiple associated matching images, the image similarity between the reference image and each matching image can be analyzed and compared with a preset similarity threshold. If the image similarity is greater than or equal to the preset similarity threshold, the matching image corresponding to that image similarity is determined as the associated matching image. This effectively ensures the accuracy and reliability of obtaining associated matching images.

[0046] In other instances, associated matching images can be obtained not only through retrieval operations based on the similarity between the reference image and matching images in the matching image library, but also through a mixed image-text retrieval operation in the matching image library by combining text description information and the reference image. In this case, retrieving associated matching images in the matching image library based on the reference image may include: determining the text description information corresponding to the reference image; and performing a mixed image-text retrieval in the matching image library based on the text description information and the reference image to obtain associated matching images.

[0047] In order to accurately obtain the associated matching images, the text description information corresponding to the reference image can be determined first. This text description information can be the visual description information of the target object obtained by recognizing the reference image. For example, when the target object in the reference image is "shirt", the text description information corresponding to the reference image obtained by recognizing the reference image can be "a blue shirt"; or, when the target object in the reference image is "pants", the text description information corresponding to the reference image obtained by recognizing the reference image can be "white pants", and so on.

[0048] In some instances, textual description information can be obtained by manually annotating a reference image, or by analyzing and processing the reference image using a deep learning model. In such cases, determining the textual description information corresponding to the reference image can include: identifying a pre-trained deep learning model for extracting image description information; inputting the reference image into the deep learning model for recognition; and obtaining the textual description information output by the deep learning model. This effectively ensures the degree of automation and intelligence in acquiring textual description information.

[0049] After obtaining the text description information, a mixed text-image search can be performed in the matching image library based on the text description information and reference images to obtain related matching images. In some examples, performing a mixed text-image search in the matching image library based on text description information and reference images to obtain related matching images may include: performing a search operation in the matching image library based on the reference images to obtain a first set of search images; performing a search operation in the matching image library based on the text description information to obtain a second set of search images; and determining related matching images based on the first set of search images and the second set of search images. Specifically, the intersection of the first set of search images and the second set of search images can be determined as the related matching images, thus effectively ensuring the accuracy and reliability of obtaining related matching images.

[0050] In other instances, associated matching images can be determined not only by the intersection of the image sets obtained from two searches, but also by vectorizing the text description information and the reference image. In this case, obtaining associated matching images by performing a mixed text-image search in the matching image library based on the text description information and the reference image may include: determining the image vector corresponding to the reference image and the text vector corresponding to the text description information; performing a fusion operation on the image vector and the text vector to obtain fusion features. Specifically, the fusion features can be obtained by concatenating the image vector and the text vector, or by performing a weighted summation operation on the image vector and the text vector; and then performing a search operation in the matching image library based on the fusion features to obtain associated matching images. This also ensures the accuracy and reliability of obtaining associated matching images.

[0051] Step S203: Search the recommendation object library based on the associated matching images to obtain multiple matching object images, wherein the recommendation object library includes the main images of multiple recommendable objects.

[0052] Since the associated pairing image includes the pairing object and related objects similar to the target object, the pairing objects included in the associated pairing image may correspond to actual object products, or they may not correspond to actual object products. To avoid invalid object recommendation operations for objects that do not correspond to actual object products, after obtaining the associated pairing image, a search operation can be performed in the recommendation object library based on the associated pairing image, thereby obtaining multiple pairing object images. The recommendation object library can include the main images of multiple recommendable objects (i.e., objects that correspond to actual object products). The obtained multiple pairing object images can be at least a part of multiple main images, and any two pairing object images can correspond to different pairing objects, and any two pairing object images can correspond to the same or different object types.

[0053] Example 1: When there is only one associated pairing image, a search operation can be performed directly in the recommendation object library based on the associated pairing image. Specifically, the pairing object region in the associated pairing image is determined; based on the pairing object region, a search and matching operation is performed in the recommendation object library to obtain multiple pairing object images. Among them, the similarity between any pairing object image and the pairing object region is greater than or equal to a preset threshold. This effectively ensures the accuracy and reliability of obtaining multiple pairing object images.

[0054] Example 2: When there are multiple associated matching images, such as associated matching image A, associated matching image B, and associated matching image C, a retrieval operation can be performed in the recommendation object library based on associated matching image A. Specifically, the matching object region 'a' in associated matching image A is determined; based on the matching object region 'a', a retrieval and matching operation is performed in the recommendation object library to obtain multiple matching object images 'aa'. These multiple matching object images 'aa' can form a set AA, where the similarity between any matching object image 'aa' and the matching object region 'a' is greater than or equal to a preset threshold. For associated matching images B and C, a similar retrieval operation can be performed as described above for associated matching image A, obtaining the set 'BB' corresponding to associated matching image B and the set 'CC' corresponding to associated matching image C, respectively. Then, the complete set consisting of cluster AA, set BB, and set CC can be used to determine the multiple matching object images, thus effectively ensuring the accuracy and reliability of obtaining multiple matching object images.

[0055] Step S204: Use a multi-image matching model to analyze and match the associated matching image and multiple matching object images to determine the recommended matching object image.

[0056] The associated pairing image includes the associated object and the corresponding pairing object. That is, the associated pairing image can present the pairing effect between the associated object and the pairing object. The pairing object image is the main image corresponding to the recommended object. That is, the associated pairing image and the pairing object image can present the relevant attribute information of the pairing object from different dimensions. Therefore, in order to accurately determine the recommended pairing object image, a multi-image matching model can be used to analyze and match the associated pairing image and multiple pairing object images. Specifically, the associated pairing image and multiple pairing object images are input into the multi-image matching model for analysis and matching to obtain the recommended pairing object image output by the multi-image matching model. In this way, the recommended pairing object image is stably determined.

[0057] Furthermore, the multi-graph matching model involved in this application embodiment can be a Large Language Model (LLM) based on artificial intelligence. This application embodiment does not limit the number of model parameters supported by the model, aiming to meet actual needs. If the model has relatively more parameters, the model size will be relatively larger, and the model performance will be relatively better. Of course, more time and resources will be consumed during inference or training. If the model has relatively fewer parameters, the model size will be relatively smaller. While meeting performance requirements, the model is more lightweight, and consumes relatively less time and resources during inference or training. This multi-graph matching model can be a deep learning model used to process and generate natural language text or multimodal data. It can be implemented based on a neural network architecture and can be pre-trained on large amounts of data. In an optional implementation, the multi-graph matching model can include an encoder, a decoder, a self-attention layer, and a feed-forward neural network, etc. The encoder is mainly used to convert input data (usually in sequence form) into vector representation. This process can capture the semantic features of the input data. The decoder is responsible for converting the intermediate representation generated by the encoder into output data (usually in sequence form). The self-attention layer is a mechanism that allows the model to pay attention to other positions in the sequence to better encode the current position information. The feedforward neural network can perform nonlinear transformations on the output of the self-attention layer to enhance the model's expressive power. All parts work together, enabling the model built on them to perform well in various complex processing tasks, such as natural language processing, computer vision, speech recognition, machine translation, text summarization, and intelligent question answering.

[0058] Step S205: Perform object recommendation operation based on the recommended matching object image.

[0059] After obtaining the recommended pairing object image, object recommendation operations can be performed based on the recommended pairing object image. In some instances, object recommendation operations can be implemented by displaying the recommended pairing object image. In this case, object recommendation operations based on the recommended pairing object image may include: displaying the recommended pairing object image in response to an object search operation based on a reference image.

[0060] Specifically, when a user performs an object search based on a reference image, not only can images of the target object corresponding to the reference image be displayed, but also recommended matching object images can be directly shown. The recommended matching object images are displayed below or after the images of the target object. For example, the images of the target object can be displayed at the top of the search results page, while the recommended matching object images can be displayed at the bottom. This allows users to not only directly view the images of the target object when performing an object search based on a reference image, but also to view recommended matching object images that have a pairing relationship with the target object, thus better meeting the user's pairing needs.

[0061] In other instances, object recommendation operations can be implemented not only by displaying recommended pairing object images, but also by displaying images based on recommended pairing objects. In this case, object recommendation operations based on recommended pairing object images can include: generating recommended pairing images based on reference images and recommended pairing object images; and displaying recommended pairing images in response to object search operations based on reference images.

[0062] Specifically, when a user performs an object search based on a reference image, not only can the relevant images of the target object corresponding to the reference image be displayed, but also recommended matching object images can be directly shown. In order to enable users to more intuitively understand the matching effect between the target object and the recommended matching object, after obtaining the recommended matching object image, the reference image and the recommended matching object image can be analyzed and processed to generate a recommended matching image. In some instances, the recommended matching image can be obtained by analyzing and processing the reference image and the recommended matching object image through an image generation model or an image stitching model.

[0063] Since the recommended pairing image includes both the target object and the recommended pairing object, when a user performs an object search operation based on the reference image, displaying the recommended pairing image after the object search operation allows the user to intuitively understand the pairing effect between the target object and the recommended pairing object. This can better meet the user's pairing needs and help improve the conversion rate of the target object and the recommended pairing object.

[0064] The object recommendation method provided in this embodiment obtains a reference image, searches a matching image library based on the reference image to obtain related matching images, and then searches a recommendation object library based on the related matching images to obtain multiple matching object images. A multi-image matching model is then used to analyze and match the related matching images and the multiple matching object images to determine the recommended matching object images. Finally, object recommendation is performed based on the recommended matching object images, thus enabling automatic and efficient object recommendation. This not only ensures the quality and efficiency of object recommendation but also reduces the time and manpower costs required for object recommendation. Furthermore, it allows for personalized object recommendation based on different reference images, thereby meeting the personalized matching needs of different users and further improving the practicality of the object recommendation method.

[0065] Figure 3 A flowchart illustrating another object recommendation method provided as an exemplary embodiment of this application; based on the above embodiments, refer to the appendix... Figure 3 As shown, in order to accurately perform the matching object recommendation operation, before searching the matching image library based on the reference image, it is necessary to construct a matching image library to determine the associated matching images. At this time, the method in this embodiment may also include: Step S301: Obtain the data source and quality parameters corresponding to the publicly available pairing chart. The quality parameters include at least one of the following: content quality parameters and effect quality parameters.

[0066] The image library can include multiple matching images used to determine associated matching images. To ensure the accuracy and reliability of determining associated matching images, the multiple matching images in the image library need to be of high quality. Specifically, the data source and quality parameters corresponding to publicly available matching images can be obtained first. Publicly available matching images can refer to publicly available service matching, interior design matching, or other types of visual matching images. The data source corresponding to the publicly available matching images can include at least one of the following: social media, e-commerce platforms, trending figures, and professionals. The quality parameters corresponding to the publicly available matching images can include at least one of the following: content quality parameters and effect quality parameters. Content quality parameters can include at least one of the following: relevance between matching objects, completeness of objects, rationality between image objects and scenes, matching rationality, aesthetic level, etc. Effect quality parameters can include at least one of the following: user behavior indicators: click-through rate, conversion rate, dwell time, interaction rate, negative feedback rate, completion rate, etc.

[0067] Quality parameters can include not only content quality parameters and effect quality parameters, but also image quality parameters, which can include at least one of the following: resolution, sharpness, sharpness, noise level, dynamic range, color accuracy, white balance, etc.

[0068] In addition, the data source and quality parameters corresponding to the publicly available pairing chart can be obtained through human-computer interaction. In this case, obtaining the data source and quality parameters corresponding to the publicly available pairing chart can include: displaying the data configuration interface; obtaining the data configuration operation entered by the user in the data configuration interface; and obtaining the data source and quality parameters corresponding to the publicly available pairing chart based on the data configuration operation. This effectively ensures the accuracy and reliability of obtaining the data source and quality parameters corresponding to the publicly available pairing chart.

[0069] Step S302: Based on the data source and quality parameters, the publicly available matching images are filtered to obtain multiple filtered matching images.

[0070] Since different data sources often correspond to public matching images of different qualities, and different quality parameters often correspond to public matching images of different qualities, in order to ensure the quality of matching images in the matching image library, after obtaining the data source and quality parameters corresponding to the public matching images, the public matching images can be filtered based on the data source and quality parameters, thereby obtaining multiple filtered matching images, which can be public matching images of higher quality.

[0071] In some instances, multiple filtered pairing images can be obtained through image filtering operations using a preset source set and preset parameter thresholds. In this case, filtering public pairing images based on data source and quality parameters to obtain multiple filtered pairing images may include: identifying whether the data source is within the preset source set; if the data source is not within the preset source set, the public pairing image corresponding to that data source can be determined as not a filtered pairing image; if the data source is within the preset source set, the public pairing image corresponding to that data source can be determined as a process pairing image; analyzing and comparing the quality parameters corresponding to the process pairing image with preset parameter thresholds; if the quality parameters corresponding to the process pairing image are less than the preset parameter thresholds, the process pairing image corresponding to those quality parameters is determined as not a filtered pairing image; if the quality parameters corresponding to the process pairing image are greater than or equal to the preset parameter thresholds, the process pairing image corresponding to those quality parameters is determined as a filtered pairing image. This effectively ensures the accuracy and reliability of determining the filtered pairing images.

[0072] In other instances, multiple filtered pairing images can be obtained by filtering public pairing images using a pre-trained image filtering model. In this case, filtering public pairing images based on data sources and quality parameters to obtain multiple filtered pairing images may include: determining a pre-trained network filtering model; inputting public pairing images, their corresponding data sources, and quality parameters into the network filtering model to perform image filtering operations, thereby obtaining multiple filtered pairing images. These multiple filtered pairing images can be high-quality public pairing images, thus ensuring the flexibility and reliability of determining the filtered pairing images.

[0073] Step S303: Generate a matching image library based on multiple filtered matching images.

[0074] After obtaining multiple filtered matching images, these images can be analyzed and processed to generate a matching image library. In some cases, multiple filtered matching images can be directly combined to form a matching image library, which effectively ensures the accuracy and reliability of the generated matching image library.

[0075] In other instances, the outfit matching library can be generated not only directly from multiple filtered outfit matching images, but also from the outfit matching evaluation information corresponding to each of the multiple filtered outfit matching images. In this case, generating the outfit matching library based on multiple filtered outfit matching images may include: determining the outfit matching evaluation information corresponding to each of the multiple filtered outfit matching images; selecting multiple reference outfit matching images from the multiple filtered outfit matching images based on the outfit matching evaluation information, wherein the value corresponding to the outfit matching evaluation information of the reference outfit matching images is greater than or equal to a preset threshold; and generating the outfit matching library based on the multiple reference outfit matching images.

[0076] After obtaining multiple filtered outfit images, these images can be analyzed to determine their respective evaluation information. This evaluation information typically refers to subjective or objective assessments of the outfit images in terms of aesthetics, harmony, practicality, and stylistic consistency. This evaluation information can be a rating or score, such as excellent, good, average, fair, or poor, or scores like 60, 70, 80, 90, or 95. In some cases, the evaluation information can be obtained through annotation by professionals. In this case, determining the evaluation information for each filtered outfit image can include: displaying a user interface; obtaining the evaluation configuration settings entered by the user in the user interface; and determining the evaluation information for each outfit image based on the user's input. This effectively ensures the accuracy and reliability of the evaluation information determination.

[0077] In other instances, the matching evaluation information can be determined based on a pre-trained aesthetic evaluation model. In this case, determining the matching evaluation information corresponding to each of the multiple filtered matching images can include: determining the pre-trained aesthetic evaluation model; inputting the multiple filtered matching images into the aesthetic evaluation model for analysis and processing, and obtaining the matching evaluation information corresponding to each of the multiple filtered matching images output by the aesthetic evaluation model. This also ensures the flexibility and reliability of determining the matching evaluation information.

[0078] After determining the matching evaluation information corresponding to each of the multiple filtered matching images, multiple reference matching images can be selected from the multiple filtered matching images based on the matching evaluation information. Among them, the value corresponding to the matching evaluation information of the reference matching image is greater than or equal to a preset threshold, that is, the reference matching image can be a high-quality filtered matching image. After obtaining multiple reference matching images, a matching image library can be generated based on the multiple reference matching images. Specifically, multiple reference matching images can be directly composed of matching image libraries, which effectively ensures the flexibility and reliability of determining the matching image library.

[0079] In this embodiment, the data source and quality parameters corresponding to the publicly available matching images are obtained, and then the publicly available matching images are filtered based on the data source and quality parameters to obtain multiple filtered matching images. A matching image library is then generated based on the multiple filtered matching images, so that the obtained matching image library includes high-quality matching images. In this way, when performing object recommendation operations based on the matching image library, the accuracy and reliability of the object recommendation operation can be effectively guaranteed.

[0080] Figure 4 This is a schematic diagram illustrating a process for retrieving multiple matching object images from a recommendation object library based on associated matching images, as provided in an exemplary embodiment of this application; based on the above embodiment, refer to the appendix... Figure 4 As shown, for multiple matching object images, they can be determined not only by the similarity between the matching object region of the associated matching image and any matching object image in the recommendation object library, but also by the matching object description of the matching object in the associated matching image. In this case, searching the recommendation object library based on the associated matching image to obtain multiple matching object images can include: Step S401: Determine the vector representation information of the associated pairing image and the description of the pairing objects in the associated pairing image.

[0081] Among them, the vector representation information of the associated matching image can refer to the numerical representation information obtained by mapping the image to a low-dimensional or high-dimensional continuous vector space, which can semantically represent the content information of the image. In order to obtain multiple matching object images, after obtaining the associated matching image, the associated matching image can be analyzed and processed to determine the vector representation information of the associated matching image. In some instances, the vector representation information can be determined by analyzing and processing the associated matching image based on a pre-trained image processing model or image processing algorithm.

[0082] Furthermore, to ensure the accuracy and reliability of identifying multiple matching object images, the associated matching images can be analyzed to determine the matching object descriptions within them. These descriptions can be object description information obtained through a combination of computer vision and natural language processing techniques. The matching object descriptions can include at least one of the following: object category (e.g., dog, car, person), attribute information (including color, size, posture, etc.), positional relationship (e.g., dog to the left of the chair), behavior / state ("running," "sitting," "holding a cup"), etc. In some instances, the matching object descriptions can be determined by analyzing the associated matching images using pre-trained image recognition models or computer vision algorithms.

[0083] Step S402: Based on vector representation information, perform search and matching in the recommendation object library to obtain multiple recall object images.

[0084] Since vector representation information can represent the content information of associated paired images, in order to accurately obtain multiple paired object images, a search and matching operation can be performed in the recommendation object library based on the vector representation information, thereby obtaining multiple recalled object images. Among them, multiple recalled object images can be images in the recommendation object library that match the vector representation information. Here, "matching" can mean that the similarity between the image in the recommendation object library and the vector representation information is greater than or equal to a preset threshold.

[0085] Step S403: Filter multiple recall object images based on the matching object description to determine multiple matching object images.

[0086] After obtaining the matching object description, multiple recall object images can be filtered based on the matching object description to determine multiple matching object images. The multiple matching objects can be recall object images that match the matching object description among the multiple recall object images.

[0087] In some instances, the filtering operation for multiple recall object images can be determined by a pre-trained image filtering model. In this case, filtering multiple recall object images based on the matching object description and determining multiple matching object images may include: determining the pre-trained image filtering model; inputting the matching object description and multiple recall object images into the image filtering model for processing, and obtaining multiple matching object images output by the image filtering model. This effectively ensures the accuracy and reliability of determining multiple matching object images.

[0088] In other instances, the filtering operation of multiple recalled object images can be determined not only by a pre-trained image filtering model, but also based on the recalled object descriptions and object attributes corresponding to each of the multiple recalled object images. In this case, filtering multiple recalled object images based on object description information to determine multiple matching object images may include: determining the recalled object description information corresponding to each of the multiple recalled object images; filtering multiple recalled object images based on object description information and recalled object description information to determine multiple matching object images.

[0089] In order to perform image filtering based on multiple recalled object images, the recall object description information corresponding to each of the multiple recalled object images can be determined. The recall object description information is object description information obtained by combining computer vision and natural language processing technology. The recall object description information may include at least one of the following: object category (e.g., dog, car, person), attribute information (including color, size, posture, function, material, etc.), positional relationship (e.g., dog is on the left of the chair), behavior / state ("running", "sitting", "holding a cup"), etc.

[0090] After obtaining the recall object description information corresponding to each of the multiple recalled object images, the multiple recalled object images can be filtered based on the object description information and the recall object description information to determine multiple matching object images. In some instances, multiple matching object images can be determined based on a pre-trained image filtering model, or multiple matching object images can be determined based on the similarity between the object description information and the recall object description information. In this case, filtering multiple recalled object images based on the object description information and the recall object description information to determine multiple matching object images can include: determining the similarity between the object description information and the recall description information; if the similarity is greater than or equal to a preset threshold, it indicates that the matching degree between the recalled object image and the matching object image is high, and thus the recalled object image corresponding to the recall object description information can be determined as a matching object image. Conversely, if the similarity is less than the preset threshold, it indicates that the matching degree between the recalled object image and the matching object image is low, and thus the recalled object image corresponding to the recall object description information can be determined as not a matching object image. This effectively achieves the accuracy and reliability of determining the matching object images.

[0091] In this embodiment, by determining the vector representation information of the associated matching images and the description of the matching objects in the associated matching images, and then searching and matching in the recommendation object library based on the vector representation information, multiple recall object images are obtained. The multiple recall object images are then filtered based on the matching object descriptions to determine multiple matching object images. This effectively ensures the quality and efficiency of determining multiple matching object images, and facilitates the accuracy and reliability of object recommendation operations based on multiple matching object images.

[0092] Figure 5 This application provides a schematic diagram illustrating a process for analyzing and matching associated matching images and multiple matching object images using a multi-image matching model to determine recommended matching object images, based on any of the above embodiments; (Refer to the appendix for details). Figure 5 As shown, for the recommended pairing object image, it can not only be determined by directly analyzing and matching the associated pairing image and multiple pairing object images based on a multi-image matching model, but also by combining the object identification boxes of the pairing objects in the associated pairing image. In this case, using a multi-image matching model to analyze and match the associated pairing image and multiple pairing object images to determine the recommended pairing object image can include: Step S501: Determine the object identification box of the matching object in the associated matching image.

[0093] The associated matching image can include the matching object and the background area. The background area contains information unrelated to the matching object. Therefore, in order to accurately determine the recommended matching object image, after obtaining the associated matching image, the associated matching image can be analyzed and processed to determine the object identification box of the matching object in the associated matching image. The object identification box can be a rectangular area used to surround a target object in the image to clarify the position of the target object in the image.

[0094] In some instances, object bounding boxes can be obtained by analyzing and processing the associated matching images based on a pre-trained object detection model. In this case, determining the object bounding boxes of the matching objects in the associated matching images can include: inputting the associated matching images into the pre-trained object detection model for analysis and processing to obtain the object bounding boxes of the matching objects in the associated matching images. This effectively ensures the accuracy and reliability of determining the object bounding boxes of the matching objects in the associated matching images.

[0095] Step S502: Based on the object identifier box, determine the reference object image corresponding to the matching object.

[0096] Since the object identification box can clearly indicate the position of the target object in the image, after obtaining the object identification box, a reference object image corresponding to the matching object can be determined based on the object identification box. In some instances, the reference object image can be the image area corresponding to the object identification box, or the reference object image can be a combination area image between the image area corresponding to the object identification box and the preset border area.

[0097] Step S503: Use a multi-image matching model to analyze and match the reference object image and multiple matching object images to determine the recommended matching object image. The similarity between the recommended matching object image and the reference object image is greater than the similarity between other matching object images and the reference object image.

[0098] Since the reference image is the image region where the matching object is located, it does not include the background information in the associated matching images. Therefore, after obtaining the reference image, a multi-image matching model can be used to analyze and match the reference image and multiple matching object images. This allows for the stable determination of the recommended matching object image output by the multi-image matching model. The similarity between the recommended matching object image and the reference image is greater than the similarity between other matching object images and the reference image. In other words, the recommended matching object image is the matching object image with the highest similarity to the reference image among multiple matching object images. This effectively ensures the flexibility and reliability of determining the recommended matching object image.

[0099] In this embodiment, object identification boxes of matching objects in associated matching images are determined, and then reference object images corresponding to the matching objects are determined based on the object identification boxes. A multi-image matching model is used to analyze and match the reference object images and multiple matching object images to determine recommended matching object images. Then, object recommendation operations can be performed based on the recommended matching object images, thereby effectively realizing automatic and efficient object recommendation operations. This not only ensures the quality and efficiency of object recommendation but also reduces the cost required for object recommendation operations, while meeting the personalized needs of different users.

[0100] Figure 6 A flowchart illustrating an exemplary embodiment of this application provides a method for recommending clothing; see attached diagram. Figure 6 As shown, this embodiment provides a clothing recommendation method. The execution subject of this method is a clothing recommendation device, which can be implemented as software or a combination of software and hardware. When the clothing recommendation device is implemented as hardware, it can be various electronic devices capable of performing clothing recommendation operations. In some instances, the clothing recommendation device can be implemented as an application client, server, cloud server, etc. When the clothing recommendation device is implemented as software, it can be installed in the electronic devices exemplified above. Specifically, the clothing recommendation method provided in this embodiment can include: Step S601: Obtain a reference image, which includes the target clothing.

[0101] Step S602: Search the matching image library based on the reference image to obtain related matching images. The related matching images include at least: related clothing and matching clothing corresponding to the related clothing. The similarity between the related clothing and the target clothing is greater than or equal to a preset threshold.

[0102] Step S603: Search the recommendation object library based on the associated matching images to obtain multiple matching clothing images, wherein the recommendation object library includes multiple main images of recommendable clothing.

[0103] Step S604: Use a multi-image matching model to analyze and match the associated matching image and multiple matching clothing images to determine the recommended matching clothing image.

[0104] Step S605: Perform clothing recommendation operation based on the recommended clothing images.

[0105] In this embodiment, the specific implementation method and effect of steps S601-S605 are the same as those described above. Figure 2 The specific implementation methods and effects of steps S201-S205 in the corresponding embodiments are similar. Please refer to the above description for details, which will not be repeated here.

[0106] Specifically, when the target clothing item is an item of interest or a garment of interest, in order to address the problem of "low efficiency and difficulty in scaling up manual clothing matching" in the clothing industry within e-commerce applications, the clothing recommendation method implemented in this embodiment can be implemented as an automated "reference-retrieval-matching" matching recommendation chain. This clothing recommendation method may include the following steps: Step 1: Build a fashion knowledge base and conduct inspiration searches (Reference - Search).

[0107] To accurately recommend clothing items, a clothing knowledge base can be built first. This knowledge base can contain 1.3 million high-quality, verified (identifying excellent outfit combinations) clothing images. Each clothing image can include at least two outfits or videos. The data sources for these images can be verified sources, such as fashion bloggers or reputable merchants; or, the images can be high-quality or influential (e.g., high likes, high views, or high sales volume of the corresponding clothing) outfits from fashion bloggers or reputable merchants.

[0108] For the outfit images included in the outfit knowledge base, they may include object identification boxes for locating clothing objects or clothing items, and may also include object description information corresponding to the clothing objects or clothing items in the outfit images.

[0109] After constructing the outfit knowledge base, a mixed image and text search operation can be performed based on it. Specifically, when it is necessary to obtain an image of a specific clothing item that a user is interested in (e.g., a particular blue shirt), the text description information corresponding to the clothing item image can be determined first. Then, using the clothing item image and text description information as input, a mixed image and text search operation is performed in the outfit knowledge base to find related outfit images containing clothing items most similar to the clothing item image in terms of style, color, and design (e.g., a high-quality outfit image B containing item A' (A' is similar to A)). The related outfit images must include at least: the related clothing item (item A') and the corresponding matching clothing item (high-quality outfit image B). The similarity between the related clothing item (item A') and the clothing item in the clothing item image (item A) is greater than or equal to a preset threshold. Furthermore, there can be one or more related outfit images. The obtained related outfit images can be used as "inspiration images" to achieve automated outfit recommendation operations. Step 2: Text pre-screening (matching-coarse screening) based on Large Language Model (LLM).

[0110] After obtaining related outfit images through inspiration retrieval based on the outfit knowledge base, you can first search the recommended clothing library (including: main images of multiple recommended clothing items) based on the related outfit images to obtain matching clothing images corresponding to the related clothing (e.g.: white trousers paired with a blue shirt). When the recommended clothing library is a product library in an e-commerce platform, since the product library contains a massive number of clothing products, directly performing image matching would result in an excessively large candidate range, high computational cost, and inaccuracy.

[0111] To avoid the aforementioned high search costs, after obtaining the associated outfit images, a search can be performed in the product database based on these images to obtain multiple images of matching clothing (e.g., a small set of candidate SKUs for other items in outfit image B, such as pants C). The product database can include multiple product images. To accurately obtain multiple images of matching clothing, recommended clothing descriptions corresponding to each product image in the database can be generated first. Specifically, an efficient Multi-modal Large Language Model (MLLM) can be used offline to generate descriptive text (captions) for the product images in the database, such as "white high-waisted wide-leg trousers".

[0112] In some instances, the MLLM model can directly analyze and process product images to output a description of the clothing object; alternatively, it can determine the clothing product details page or product attribute information corresponding to the product image. Then, the clothing product details page or product attribute information, along with the product image, can be input into the MLLM model for processing to obtain the clothing object description output by the MLLM model. This clothing object description can include not only visual feature information but also information such as the material of the clothing and the suitable season.

[0113] Furthermore, to accurately identify multiple matching clothing images, after obtaining the associated matching images, multiple matching clothing images can be determined based on the associated matching images. Specifically, the powerful text understanding and reasoning capabilities of the LLM model can be used to perform a fast filtering operation on the product library based on the matching clothing images. At this time, the LLM model can obtain the trigger instruction, such as: "Please find all options belonging to the 'pants' category, 'white', and suitable for 'spring and summer' wear from the candidate SKUs in the following product library." Then, by analyzing the clothing object description "Caption" corresponding to each product image, the LLM model can quickly filter out obviously inconsistent options (such as filtering out tops, skirts, winter pants, non-white pants, etc.) and obtain multiple matching clothing images that match the description information. This can drastically reduce the number of candidate clothing items in the product library from tens of thousands or even hundreds of thousands to less than a few dozen, which greatly reduces the complexity of subsequent fine screening.

[0114] Step 3: Fine screening based on multi-graph matching model (matching-fine screening).

[0115] After step 2 above, a small set of candidate products is obtained, which includes multiple images of matching outfits. These candidate outfit images meet the requirements in terms of macro attributes such as category and color, but there are subtle differences in style, fit, and details. Therefore, in order to improve the quality and effectiveness of the clothing recommendation operation, a more sophisticated and computationally more expensive multi-image matching model can be used. This multi-image matching model can comprehensively compare multiple images, including the target item image in the "associated matching images" and the main clothing images in the multiple matching outfit images. Specifically, it can compare deep visual features such as texture, cropping, and design details (e.g., button and pocket styles), and calculate the precise matching score between each candidate outfit in each matching outfit image and the target item. This precise matching score is used to identify the similarity between the candidate outfit and the target item. Then, the candidate outfit image with the highest matching score can be selected as the recommended matching outfit image. Clothing recommendation operations can then be performed based on the recommended matching outfit image. For example, the matching images between the recommended matching outfit image and the outfit of interest (e.g., item A+) can be output. The system matches the SKUs of pants and makes clothing recommendations based on the matching images.

[0116] The clothing recommendation method provided in this embodiment utilizes a constructed clothing knowledge base to determine related clothing images as inspiration sources, and then performs clothing recommendation operations based on these related images. This effectively compensates for the lack of creativity in the fashion field of general large models. Secondly, it determines recommended clothing images through a multi-stage matching process of "LLM text pre-screening + multi-image model fine screening," specifically realizing an automated clothing matching generation chain of "reference-retrieval-matching." This solution does not rely on large models to create "out of thin air," but rather uses the "reference-retrieval-matching" steps to generate recommended clothing images from a market-proven, high-quality... Drawing inspiration from a knowledge base of fashion styling tips provides AI with a "fashion expert knowledge base," ensuring the professionalism and style of the styling solutions. This cleverly avoids the shortcomings of the model's own lack of creativity and domain knowledge, effectively solving the technical challenge of low accuracy in matching fine-grained products with a massive amount of goods. It increases the matching accuracy and recall rate of individual items to over 85%, enabling the large-scale generation of high-quality styling solutions and possessing extremely high commercial value. When this application solution is applied to e-commerce scenarios, it can upgrade the sales paradigm from "single items" to "scenario-based outfit combinations," further improving the practicality of the solution.

[0117] Furthermore, in some of the processes described in the above embodiments and accompanying drawings, multiple operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or they may be executed in parallel. The operation numbers, such as 11, 12, etc., are merely used to distinguish different operations and do not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0118] Figure 7 A schematic diagram of an object recommendation device provided for an exemplary embodiment of this application; see attached drawing. Figure 7 As shown, this embodiment provides an object recommendation device, which is used to perform the above-described... Figure 2 The object recommendation method shown, specifically, the object recommendation device may include: The first acquisition module 11 is used to acquire a reference image, which includes the target object; The first retrieval module 12 is used to search in the matching image library based on the reference image to obtain related matching images. The related matching images include at least: the related object and the matching object corresponding to the related object. The similarity between the related object and the target object is greater than or equal to a preset threshold. The first retrieval module 12 is also used to perform retrieval in the recommendation object library based on the associated matching images to obtain multiple matching object images, wherein the recommendation object library includes the main images of multiple recommendable objects; The first processing module 13 is used to analyze and match the associated matching image and multiple matching object images using a multi-image matching model to determine the recommended matching object image; The first processing module 13 is also used to perform object recommendation operations based on the recommended matching object image.

[0119] The object recommendation device in this embodiment can also perform the above-described... Figures 1-5 The description of the embodiments shown is for reference only, and will not be elaborated upon here.

[0120] like Figure 8 As shown, this embodiment provides an electronic device for performing the above-described... Figure 2 The object recommendation method shown includes an electronic device that may include a memory 24 and a processor 25.

[0121] Memory 24 is used to store computer programs and can be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device, data structures, contact data, phone book data, messages, pictures, videos, etc.

[0122] The processor 25, coupled to the memory 24, executes a computer program in the memory 24 for: acquiring a reference image, which includes a target object; searching a matching image library based on the reference image to obtain associated matching images, wherein the associated matching images include at least: an associated object and a matching object corresponding to the associated object, and the similarity between the associated object and the target object is greater than or equal to a preset threshold; searching a recommendation object library based on the associated matching images to obtain multiple matching object images, wherein the recommendation object library includes main images of multiple recommendable objects; analyzing and matching the associated matching images and multiple matching object images using a multi-image matching model to determine recommended matching object images; and performing object recommendation operations based on the recommended matching object images.

[0123] Furthermore, such as Figure 8 As shown, the electronic device also includes other components such as a communication component 26, a display 27, a power supply component 28, and an audio component 29. Figure 8 The diagram only shows some components and does not mean that the electronic device includes only these components. Figure 8 The components shown. Additionally... Figure 8The components within the center frame are optional, not mandatory, and their specific requirements depend on the product form of the work node. In this embodiment, the work node can be a terminal device such as a desktop computer, laptop computer, smartphone, or IoT device, or a server-side device such as a conventional server, cloud server, or server array. If the work node in this embodiment is implemented as a terminal device such as a desktop computer, laptop computer, or smartphone, it may include... Figure 8 The components within the center frame; if the working node in this embodiment is implemented as a server-side device such as a conventional server, cloud server, or server array, it may not include... Figure 8 The component within the center frame.

[0124] Figure 9 A schematic diagram of the structure of an apparel recommendation device provided for an exemplary embodiment of this application; see attached drawing. Figure 9 As shown, this embodiment provides a clothing recommendation device, which is used to perform the above-described... Figure 6 The clothing recommendation method shown may specifically include: The second acquisition module 31 is used to acquire a reference image, which includes the target clothing; The second retrieval module 32 is used to search in the matching image library based on the reference image to obtain related matching images. The related matching images include at least: related clothing and matching clothing corresponding to the related clothing. The similarity between the related clothing and the target clothing is greater than or equal to a preset threshold. The second retrieval module 32 is also used to search in the recommendation object library based on the associated matching images to obtain multiple matching clothing images, wherein the recommendation object library includes multiple main images of recommendable clothing; The second processing module 33 is used to analyze and match associated matching images and multiple matching clothing images using a multi-image matching model to determine recommended matching clothing images; The second processing module 33 is also used to perform clothing recommendation operations based on recommended clothing images.

[0125] The clothing recommendation device in this embodiment can also perform the above-described functions. Figure 6 The description of the embodiments shown is for reference only, and will not be elaborated upon here.

[0126] like Figure 10 As shown, this embodiment provides an electronic device for performing the above-described... Figure 6 The object recommendation method shown includes an electronic device that may include a memory 44 and a processor 45.

[0127] Memory 44 is used to store computer programs and can be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device, data structures, contact data, phone book data, messages, pictures, videos, etc.

[0128] The processor 45, coupled to the memory 44, executes a computer program in the memory 44 for: acquiring a reference image, which includes the target clothing; searching a matching image library based on the reference image to obtain related matching images, wherein the related matching images include at least: the related clothing and matching clothing corresponding to the related clothing, and the similarity between the related clothing and the target clothing is greater than or equal to a preset threshold; searching a recommendation object library based on the related matching images to obtain multiple matching clothing images, wherein the recommendation object library includes multiple main images of recommendable clothing; analyzing and matching the related matching images and multiple matching clothing images using a multi-image matching model to determine recommended matching clothing images; and performing clothing recommendation operations based on the recommended matching clothing images.

[0129] Furthermore, such as Figure 10 As shown, the electronic device also includes other components such as a communication component 46, a display 47, a power supply component 48, and an audio component 49. Figure 10 The diagram only shows some components and does not mean that the electronic device includes only these components. Figure 10 The components shown. Additionally... Figure 10 The components within the center frame are optional, not mandatory, and their specific requirements depend on the product form of the work node. In this embodiment, the work node can be a terminal device such as a desktop computer, laptop computer, smartphone, or IoT device, or a server-side device such as a conventional server, cloud server, or server array. If the work node in this embodiment is implemented as a terminal device such as a desktop computer, laptop computer, or smartphone, it may include... Figure 10 The components within the center frame; if the working node in this embodiment is implemented as a server-side device such as a conventional server, cloud server, or server array, it may not include... Figure 10 The component within the center frame.

[0130] The aforementioned memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0131] The aforementioned communication component is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as 2G, 3G, 4G / LTE, 5G, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel.

[0132] The aforementioned display includes a screen, which may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a Touch Panel, the screen can be implemented as a touchscreen to receive input signals from the user. The Touch Panel includes one or more touch sensors to sense touches, swipes, and gestures on the Touch Panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation.

[0133] The aforementioned power supply components provide power to various components within the device in which they reside. These power supply components may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device in which they reside.

[0134] The aforementioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals can be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0135] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to implement the steps in the above-described method embodiments. The computer-readable storage medium includes volatile or non-volatile components, or a combination thereof, and can be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technologies, CD-ROM, Digital Video Disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium. Accordingly, this application also provides a computer program product, which includes a computer program or instructions. When the computer program or instructions are executed by a processor, the processor is able to implement the steps in the above method embodiments. It should be understood that each step or combination of steps in the above method flow can be implemented by the computer program or instructions. In addition, these computer programs or instructions can be applied to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device, so that the processor of the general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device can be implemented as a means to implement the corresponding functions in the above method embodiments.

[0136] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0137] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An object recommendation method, characterized in that, include: Obtain a reference image, wherein the reference image includes the target object; Based on the reference image, a search is performed in the matching image library to obtain related matching images. The related matching images include at least: a related object and a matching object corresponding to the related object. The similarity between the related object and the target object is greater than or equal to a preset threshold. Based on the associated matching images, a search is performed in the recommendation object library to obtain multiple matching object images, wherein the recommendation object library includes the main images of multiple recommendable objects; The associated matching image and the multiple matching object images are analyzed and matched using a multi-image matching model to determine the recommended matching object image; Perform object recommendation operations based on the recommended matching object images.

2. The method according to claim 1, characterized in that, Before performing a search in the matching image library based on the reference image, the method further includes: Obtain the data source and quality parameters corresponding to the publicly available pairing charts. The quality parameters include at least one of the following: content quality parameters and effect quality parameters. Based on the data source and the quality parameters, the publicly available matching images are filtered to obtain multiple filtered matching images; Based on the multiple filtered matching images, the matching image library is generated.

3. The method according to claim 2, characterized in that, Based on the multiple filtered matching images, the matching image library is generated, including: Determine the matching evaluation information corresponding to each of the multiple filtered matching images; Based on the matching evaluation information, multiple reference matching images are selected from the multiple filtered matching images, wherein the value corresponding to the matching evaluation information of the reference matching images is greater than or equal to a preset threshold. The matching image library is generated based on the multiple reference matching images.

4. The method according to claim 1, characterized in that, Based on the reference image, a search is performed in the matching image library to obtain related matching images, including: Determine the text description information corresponding to the reference image; Based on the text description information and the reference image, a mixed text and image search is performed in the matching image library to obtain the associated matching image.

5. The method according to claim 1, characterized in that, Based on the associated matching images, a search is performed in the recommendation object library to obtain multiple matching object images, including: Determine the vector representation information of the associated matching image and the matching object description of the matching objects in the associated matching image; Based on the vector representation information, a search and matching process is performed in the recommendation object library to obtain multiple recall object images; Based on the description of the matching objects, the multiple recall object images are filtered to determine multiple matching object images.

6. The method according to claim 5, characterized in that, Based on the object description information, the multiple recalled object images are filtered to determine multiple matching object images, including: Determine the recall object description information corresponding to each of the multiple recall object images; Based on the object description information and the recall object description information, the multiple recall object images are filtered to determine multiple matching object images.

7. The method according to claim 6, characterized in that, Based on the object description information and the recalled object description information, the multiple recalled object images are filtered to determine multiple matching object images, including: Determine the similarity between the object description information and the recalled object description information; If the similarity is greater than or equal to a preset threshold, the recall object image corresponding to the recall object description information is determined as the matching object image.

8. The method according to any one of claims 1-7, characterized in that, The associated matching image and the multiple matching object images are analyzed and matched using a multi-image matching model to determine the recommended matching object image, including: Determine the object identification box of the matching object in the associated matching image; Based on the object identification box, a reference object image corresponding to the matching object is determined; A multi-image matching model is used to analyze and match the reference object image and the multiple matching object images to determine a recommended matching object image. The similarity between the recommended matching object image and the reference object image is greater than the similarity between the other matching object images and the reference object image.

9. The method according to any one of claims 1-7, characterized in that, The object recommendation operation based on the recommended matching object image includes: In response to an object search operation based on the reference image, the recommended matching object image is displayed.

10. The method according to any one of claims 1-7, characterized in that, The object recommendation operation based on the recommended matching object image includes: Based on the reference image and the recommended pairing object image, a recommended pairing image is generated; In response to an object search operation based on the reference image, the recommended matching image is displayed.

11. A method for recommending clothing, characterized in that, include: Obtain a reference image, wherein the reference image includes the target clothing; Based on the reference image, a search is performed in the matching image library to obtain related matching images. The related matching images include at least: related clothing and matching clothing corresponding to the related clothing. The similarity between the related clothing and the target clothing is greater than or equal to a preset threshold. Based on the associated matching images, a search is performed in the recommendation object library to obtain multiple matching clothing images, wherein the recommendation object library includes multiple main images of recommendable clothing; The associated matching image and the multiple matching clothing images are analyzed and matched using a multi-image matching model to determine the recommended matching clothing image; The clothing recommendation operation is performed based on the recommended clothing images.

12. An object recommendation device, characterized in that, include: The first acquisition module is used to acquire a reference image, wherein the reference image includes the target object; The first retrieval module is used to search in the matching image library based on the reference image to obtain related matching images, wherein the related matching images include at least: a related object and a matching object corresponding to the related object, and the similarity between the related object and the target object is greater than or equal to a preset threshold. The first retrieval module is further configured to perform a retrieval in the recommendation object library based on the associated matching images to obtain multiple matching object images, wherein the recommendation object library includes the main images of multiple recommendable objects; The first processing module is used to analyze and match the associated matching image and the multiple matching object images using a multi-image matching model to determine the recommended matching object image; The first processing module is further configured to perform an object recommendation operation based on the recommended matching object image.

13. A clothing recommendation device, characterized in that, include: The second acquisition module is used to acquire a reference image, wherein the reference image includes the target clothing; The second retrieval module is used to search the matching image library based on the reference image to obtain related matching images. The related matching images include at least: related clothing and matching clothing corresponding to the related clothing. The similarity between the related clothing and the target clothing is greater than or equal to a preset threshold. The second retrieval module is further configured to perform a retrieval in the recommendation object library based on the associated matching images to obtain multiple matching clothing images, wherein the recommendation object library includes multiple main images of recommendable clothing; The second processing module is used to analyze and match the associated matching images and the multiple matching clothing images using a multi-image matching model to determine the recommended matching clothing images; The second processing module is also used to perform clothing recommendation operations based on the recommended clothing images.

14. An electronic device, characterized in that, include: A memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions, when executed by the processor, implement the method of any one of claims 1-11.

15. A computer storage medium, characterized in that, Used to store a computer program that, when executed by a computer, implements the method of any one of claims 1-11.

16. A computer program product, characterized in that, include: A computer program, when executed by a processor of an electronic device, causes the processor to perform the steps of the method of any one of claims 1-11.

Images