KR102814832B1 - Wholesale and retail integrated management platform and strategic product sourcing system with generative ai based product image creation function - Google Patents

Abstract

According to an embodiment of the present invention, a wholesale and retail integrated management platform and a strategic product sourcing system that operate in conjunction with a plurality of wholesaler terminals, retailer terminals, and producer systems are provided, including a product registration module that receives product information from a plurality of wholesaler terminals and stores the product information in a database; an order list registration module that provides product information, inventory information, and receipt information stored in a database to a plurality of retailer terminals, and receives orders received from consumers from the retailer terminals as order list information; a wholesaler link module that connects a retailer that has registered the order list with a wholesaler that supplies the corresponding product; an order list sorting/merging module that sorts and merges order lists connected to wholesalers, aggregates quantities, and generates a sorted order list; and an out-of-stock/delayed-in-stock product list extraction module that extracts and stores out-of-stock and delayed-in-stock products from the sorted order list, wherein the wholesaler terminal receives an order list from which out-of-stock and delayed-in-stock products are excluded from the sorted order list.

Description

{WHOLESALE AND RETAIL INTEGRATED MANAGEMENT PLATFORM AND STRATEGIC PRODUCT SOURCING SYSTEM WITH GENERATIVE AI BASED PRODUCT IMAGE CREATION FUNCTION}

The present invention relates to a wholesale and retail integrated management platform and a strategic product sourcing system having a product image generation function based on generative AI, and more specifically, to a system that integrates and manages a plurality of wholesale and retail systems, facilitates payment through automation, analyzes market trends to source optimal products, and generates product images based on generative AI when new products are launched.

Typically, transactions between wholesalers who supply goods and retailers who sell the goods supplied by wholesalers to consumers are conducted through the retailer visiting multiple wholesalers to look for new products, and the retailer who receives orders from consumers enters the customers' orders by hand or creates an Excel file, tallying up the quantities by transaction partner, product name, color, and size, and making phone calls, faxes, or in person during the wholesaler's business hours.

In cases like the above, there is a problem that the retailer has to visit the wholesaler in person, and especially in the case of clothing items, since most wholesalers work late at night or in the early morning, the retailer's operator has to extend his/her working hours or assign a separate person in charge to visit the wholesaler after the retailer's sales (business) hours.

In addition, wholesalers have to compare and confirm orders received from retailers with actual product inventory before shipping the products, and manually write receipts for the shipped products and send them out. This takes a lot of time, and errors can occur during settlement due to errors in writing unit prices or product names, and there is a problem in that the details of the actual shipped products cannot be confirmed.

Meanwhile, products such as clothing have a very short seasonal life cycle and styles change quickly. It takes a lot of time and money to produce sample products before the start of the season and then create product images through photo studios and model hiring. If the schedule is delayed, there is a problem of missing the sales point.

Therefore, technologies are needed to enable this process to be carried out efficiently.

Publication of Patent Publication No. 10-2008-0009039 (Published on January 24, 2008) Patent Publication No. 10-2592596 (Published on October 23, 2023) Publication of Patent Publication No. 10-2023-0169673 (Published on December 18, 2023)

The purpose of the present invention is to solve the problems of the prior art described above.

The purpose of the present invention is to enable wholesalers, retailers, and manufacturers to input or output product sales information within a single system, thereby ensuring speed and stability in mass product transactions.

The purpose of the present invention is to systematically analyze a product database accumulated in a system to provide an optimal product sourcing reference point that matches the trends, seasons, and age-specific conditions of the target market.

The purpose of the present invention is to enable existing wholesale and retail businesses to produce various models (faces and body types) and backgrounds based on images taken of fashion items being worn on mannequins, hangers, or the like rather than actual models, or flat photos of fashion items, through a generative AI model, and to automatically generate new AI product images, thereby easily producing product description images and encouraging consumers to desire to purchase.

The purposes of the present invention are not limited to those mentioned above, and other purposes not mentioned will be clearly understood from the description below.

According to one embodiment of the present invention for achieving the above-described purpose, a wholesale and retail integrated management platform and strategic product sourcing system that operate in conjunction with a plurality of wholesaler terminals, retailer terminals, and producer systems, the platform includes a product registration module that receives product information from a plurality of wholesaler terminals and stores it in a database; an order list registration module that provides product information, inventory information, and receipt information stored in a database to a plurality of retailer terminals, and receives orders received from consumers from the retailer terminals as order list information; a wholesaler link module that connects a retailer that has registered the order list with a wholesaler that supplies the corresponding product; an order list sorting/merging module that sorts and merges order lists connected to wholesalers to aggregate quantities and generates a sorted order list; and an out-of-stock/delayed-in-stock product list extraction module that extracts and stores out-of-stock and delayed-in-stock products from the sorted order list, wherein the wholesaler terminal receives an order list from which out-of-stock and delayed-in-stock products are excluded from the sorted order list.

The above product registration module includes a first image receiving unit that receives a first wearing image including a fashion item from a wholesale terminal; a model generating unit that estimates a human model based on analysis of the fashion item; and a second image generating unit that generates a second wearing image in which the fashion item is fitted on the human model estimated by the model generating unit, wherein the first wearing image is an image of the fashion item being worn on a mannequin, hanger, or the like rather than an actual model, or a flat image of the fashion item. A wholesale and retail integrated management platform and a strategic product sourcing system are provided.

The above-mentioned wholesale and retail integrated management platform and strategic product sourcing system may further include a trend analysis module that combines information on fashion styles desired by consumers with current fashion trends, derives optimal options by season and age, and suggests them to wholesalers and retailers.

According to an embodiment of the present invention, wholesalers, retailers, and manufacturers can input or output product sales information within a single system, thereby achieving the effect of ensuring speed and stability in mass product transactions.

According to an embodiment of the present invention, by systematically analyzing a product database accumulated in a system, it is possible to provide an optimal product sourcing reference point that matches the trend, season, and age-specific conditions of the target market.

According to an embodiment of the present invention, existing wholesale and retail businesses can easily produce product description images by producing various models (faces and body types) and backgrounds based on images of fashion items being worn on mannequins, hangers, or the like rather than actual models, or flat images of fashion items, and automatically generating new AI product images through a generative AI model, thereby encouraging consumers to desire to purchase.

In addition, new AI product images can be created as local customized product images with model body types, faces, and backgrounds specialized for each country, which can help boost overseas sales.

Figure 1 is a conceptual diagram showing the connection of a wholesale and retail integrated management platform and a strategic product sourcing system according to an embodiment of the present invention.
Figure 2 is a configuration diagram showing the configuration of an integrated management platform according to one embodiment of the present invention.
FIG. 3 is a block diagram illustrating a detailed configuration of a trend analysis module of an integrated management platform according to the embodiment of FIG. 2.
Figure 4 is a flowchart showing the process of a method for automatic ordering and sharing inventory data between wholesalers and retailers according to the present invention.
FIG. 5 is a block diagram for explaining a product information image generation operation of a product registration module according to one embodiment of the present invention.
Figure 6 is an example of a first wearing image, which is an image of a mannequin wearing a fashion item.
Figure 7 is another example of the first wearing image as a model image wearing a fashion item.
FIGS. 7 and 8 are examples showing second wearing images generated according to an embodiment of the present invention.

The detailed description of the present invention set forth below refers to the accompanying drawings which illustrate specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that the various embodiments of the present invention, while different from one another, are not necessarily mutually exclusive. For example, specific shapes, structures, and features described herein may be implemented in other embodiments without departing from the spirit and scope of the invention. It should also be understood that the positions or arrangements of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. Accordingly, the following detailed description is not intended to be limiting, and the scope of the invention is defined only by the appended claims, along with the full scope of equivalents to which such claims are entitled, if any. Like reference numerals in the drawings designate the same or similar functionality throughout the several aspects.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings so that a person skilled in the art to which the present invention pertains can easily practice the present invention.

FIG. 1 is a conceptual diagram showing the connection of a wholesale and retail integrated management platform and a strategic product sourcing system (hereinafter referred to as “management system”) according to an embodiment of the present invention, and is described with reference to FIG. 1.

The integrated management platform (100) is connected to a number of wholesale terminals (200), a number of retail terminals (300), a number of producer terminals (400), a financial institution system (500), and a delivery system (600) via the Internet.

The integrated management platform (100) records information on products sold through the Internet and transmits the information to consumers who need the products, thereby enabling product transactions. In addition, it uses the financial institution system (500) to pay for the purchase of products and requests delivery of paid products to the delivery system (600).

The integrated management platform (100) registers product information, product features and advantages, product inventory information (quantity information of a product that a wholesaler has when entering product information), and out-of-stock information received from multiple wholesaler terminals (200), and a retailer accesses the integrated management platform (100) using a retailer terminal (300) to search the registered product information, selects only the products to be sold, copies the product information, synchronizes the product name that the retailer wants to use with the product name used by the wholesaler, and registers the product to be sold in various shopping malls (open markets, etc.) with the synchronized product name to start selling online.

The inventory information and out-of-stock information of the above product is the quantity information that the wholesaler has when entering the product information. When the wholesaler registers the quantity of the product in the integrated management platform (100), the information is automatically updated and stored according to the retailer's order.

The wholesale terminal (200) is a terminal of a wholesaler that collects and distributes products such as clothing and raw materials in bulk from producers or manufacturers, and is connected to the integrated management platform (100) via the Internet.

The retail terminal (300) is a terminal of a retailer that purchases small quantities of products such as clothing from wholesalers and resells them to a large number of consumers, and is connected to the integrated management platform (100) via the Internet.

When an order list is registered, the integrated management platform (100) links the name of the wholesaler who supplies the product to the product in the order list, sorts and merges the order list linked to the name of the wholesaler by wholesaler, product name, color, and size, and automatically aggregates the order quantity to create and save a sorted order list.

Products that are out of stock or have delayed delivery are automatically extracted from the above-mentioned sorted order list, an out of stock/delayed delivery list is created and sent to the retailer to notify them, and an order list of products actually ordered is created by excluding out of stock/delayed delivery products from the above-mentioned sorted order list, and orders are placed in bulk and automatically for each wholesaler.

In addition, the information on products ordered through the above order list is used to update the data on the stock quantity of products currently registered in the integrated management platform (100). That is, the stock quantity value is reduced by the quantity of products included in the order list to store the change in stock quantity.

In addition, the integrated management platform (100) automatically generates a receipt for the price of the product that the retailer must pay to the wholesaler based on the order list transmitted to the wholesaler. The receipt includes product information such as wholesaler name/product name/color/size and is generated and stored.

The wholesaler ships products ordered from retailers based on the order list automatically received from the integrated management platform (100).

In addition, since retailers can view product information registered in the integrated management platform (100) anytime and anywhere, they have the advantage of efficiently conducting customer C/S for consumers who purchased products from a shopping mall (open market, etc.) using product information such as product inventory quantity and arrival date.

The producer terminal (400) is a terminal of a producer who directly produces and ships clothing products, and is connected to the integrated management platform (100) via the Internet.

The financial institution system (500) is a computer system of a bank, credit card company, etc., and provides a system for payment of sales costs between wholesalers, retailers, manufacturers, and distributors.

In order to pay for sales costs, various methods can be used, such as direct remittance, account transfer, pre-accumulation of accumulated points, and credit card payment. In the present invention, a typical e-commerce payment method can be applied.

The delivery system (600) is a computer system of a cargo transportation company, courier company, etc., and enables delivery of products traded through the integrated management platform (100) to recipients.

Figure 2 is a configuration diagram showing the configuration of an integrated management platform (100), and is described with reference to Figure 2.

The integrated management platform (100) may include a web server (102), a product registration module (104), a product synchronization module (106), an order list registration module (108), an order list sorting/merging module (110), an out-of-stock/delayed delivery product list extraction module (114), a receipt generation module (116), and a data storage module (118).

The web server (102) allows a wholesale terminal (200) or a retail terminal (300) to access the integrated management platform (100) via the Internet and exchange various data using a web browser or a dedicated application.

In addition, various processing modules included in the integrated management platform (100) are controlled to process data necessary for product transactions and product delivery.

The product registration module (104) registers products received from multiple wholesalers and stores them in the product database (110).

According to one embodiment, the product registration module (104) receives product (e.g. clothing) information from a wholesaler, receives a photo of the product, creates an image of the product being used (e.g. clothing being worn), and forms the image as product information. This will be described in detail later.

The product synchronization module (106) synchronizes the product name used in the wholesaler registered by the product registration module (104) with the name used in the retailer.

The order list registration module (108) registers an order list received from a retailer.

The wholesaler link module (105) for each product in the order list connects the products in the registered order list with the wholesaler that supplies the products.

The order list sorting/merging module (110) sorts and merges the order list connected to the above wholesaler by wholesaler/product name/color/size and automatically aggregates the quantity to create a sorted order list.

The trend analysis module (112) combines information on the fashion style desired by consumers with current fashion trends, derives optimal options by season and age, and suggests them to wholesalers and retailers.

The out-of-stock/delayed delivery product list extraction module (114) extracts products that are out-of-stock or delayed in delivery from the above-mentioned sorted order list and creates an out-of-stock/delayed delivery list.

The receipt generation module (116) automatically creates a receipt for products supplied by a wholesaler to a retailer.

The data storage module (118) stores data on the above-mentioned out-of-stock/delayed delivery product list, order list registered from retailers, purchase order list, receipt information, and product inventory quantity.

The above data storage module (118) includes a wholesaler information database (120), a wholesaler information database (122), a producer information database (124), a payment information database (126), and a delivery information database (128).

The wholesale information database (120) receives and stores information necessary for business transactions with wholesalers, such as the name, address, and products handled by the wholesalers registered in the integrated management platform (100), from the product database (110).

The wholesale information database (122), like the wholesale information database (120), stores information on retailers.

The producer information database (124) stores information such as the name, address, and product items of the producer of the product being sold.

In order to determine whether a person attempting to register as a member of the integrated management platform (100) is a wholesaler, retailer, or producer, information on the business registration certificate is used.

In other words, members whose business type is listed as wholesale on their business registration certificate are registered as wholesalers, and members whose business type is listed as retailers are registered as retailers. Since this is determined based on the information of the registered business operator, not the user's choice, it allows for more accurate classification of sellers.

In addition, a person who registers as a member to use the integrated management platform (100) is identified as a general member using his or her resident registration number.

The payment information database (126) receives and stores information on payment of costs incurred in product transactions between wholesalers, retailers, and manufacturers from the payment information management module (108).

The delivery information database (128) receives and stores product delivery information from the delivery information management module (110).

FIG. 3 is a block diagram illustrating a detailed configuration of a trend analysis module of an integrated management platform according to the embodiment of FIG. 2.

Referring to FIG. 3, a trend analysis module (112) may be included. The trend analysis module (112) analyzes big data online and provides an analysis report on the latest fashion trends. The trend analysis module (112) may be implemented to include a market information collection machine (1121), a market trend analysis machine (1122), and a market trend report generation machine (1123). The components of the trend analysis module (112) may be configured as software on a single server, but may also be implemented by being distributed to multiple servers that are physically separated and connected by a network.

The market information collection machine (1121) collects information such as images, texts, and hashtags (hereinafter referred to as "market information") through fashion-related web pages published in online fashion malls selling shopping products such as Naver Shopping Mall and 11th Street, including power bloggers including social media such as Instagram and Facebook related to fashion products, homepages of influential celebrities in the fashion industry, and blogs of influencers. At this time, the market information is not limited to information collected in the online market, and may include all information related to market demand. When collecting market information, the market information collection machine (1121) separates and extracts non-visual information such as fashion image information and text.

Market information may include information about the influence of each web page (number of visitors, sales, etc.). Fashion image information or non-visual information collected from a web page may be weighted and reflected by the influence value of the web page when analyzing fashion trends. For example, fashion image information on Instagram with 100,000 followers may be reflected with a higher weight when analyzing fashion trends compared to fashion image information on Instagram with 1,000 followers.

The market trend analysis machine (1122) analyzes non-visual information such as fashion image information and text describing the fashion image information, hash codes, buyer reviews, and visitor responses. Based on the analyzed results, fashion images are given classification codes for each fashion product, which are similar to each fashion product.

For example, when an image of a person wearing a dress shirt and glasses is collected from a recent post of a Facebook user, the market trend analysis machine (1122) can recognize an object called 'dress shirt' from an image of clothes separated from the image, and detect objective fashion elements such as 'long sleeves', 'white', 'with collar', 'blue collar', and 'slim fit'. When object recognition and detection of fashion elements are completed in this way, a detailed product classification such as 'white long-sleeved dress shirt with a blue collar and slim fit' can be made. In addition, non-visual information such as hash codes and replies can be analyzed to find non-visual product elements such as '40s' and 'male'.

In this way, when the analysis of fashion image information and non-visual information is completed, only then can a similar product classification of 'a slim-fit white long-sleeved dress shirt for men in their 40s with a blue collar' be created and a fashion product classification code be assigned to distinguish it. This fashion product classification code is a unique code of the trend analysis module (112), and the separation code system may differ by era or product market (country, region, etc.), and the fashion product classification code may be added or modified according to market changes.

The market trend analysis machine (1122) can be implemented to consist of an image analysis engine, a non-visual information processing engine, and a fashion product classification code generation engine. The image analysis engine can be implemented using object detection and multi-element (Multi-Label Attribute Classification) detection technology among deep learning visual recognition technologies. Object recognition technologies include Faster R-CNN, YOLO, and SSD, which have recently been utilized. The multi-element detection technology utilizes CNN technology, and it is necessary to design a system that classifies the features of clothing to suit the Korean situation and an optimized learning function for multi-element detection.

The non-visual information processing engine uses natural language analysis technology. That is, it analyzes the relationship between words or sentences based on a language dictionary. In the case of language classification models, there are Naive Bayes, Linea SVM, BERT, GPTs, etc. All of these methods operate based on data processed by a morphological analyzer. The embodiment of the present invention uses the morphological analysis and entity recognition technology of 'Exobrain' when processing natural language data and builds a fashion natural language dictionary. 'Exobrain' is a Korean AI platform implemented through a national project led by the Electronics and Telecommunications Research Institute. By using 'Exobrain', fashion-related chatbots, SNS analysis, and open market data mining can be easily realized.

The fashion product classification code generation engine provides detailed and precise classification codes for fashion products according to trends, unlike the abstract existing classifications. This makes it easier to analyze fashion product trends based on big data, and it is also easy to calculate the correlation between each fashion product.

The market trend report generation machine (1123) generates demand forecast information for a certain fashion product from similar product mining information. Demand forecast information for fashion products can be provided in a very detailed product unit, such as a fashion product classification code, but can also be provided in a more abstract and broad product unit.

Meanwhile, the market trend report generation machine (1123) receives actual sales information from online or offline cooperative retailers, compares and analyzes the difference between the demand forecast information for each fashion product and the actual sales, and updates (upgrades) the demand forecast model.

Market trend reports can be generated in regular cycles, for example, every two weeks or a month. That is, demand forecast information using the demand forecast model in the current cycle is generated using the results of similar product mining performed up to the previous cycle. The results of similar product mining performed in the current cycle can be used when generating demand forecast information for the next cycle.

FIG. 4 is a flowchart showing the process of a method for automatic ordering and sharing inventory data between wholesalers and retailers according to the present invention, and the sequential operation process is explained with reference to FIG. 4.

A number of wholesalers access the integrated management platform (100) to register the wholesalers' products (S201), the retailers view the products registered by the wholesalers, select the products they wish to sell, and copy the product information (S202), and if the retailers wish to use a different name for the selected product than the name designated by the wholesaler, they set the desired name and synchronize it with the wholesalers' product name (S203).

The integrated management platform (100) registers the order list received from the retailer (S204), connects the products in the registered order list with the wholesaler that supplies the products (S205), and automatically aggregates the quantity and creates a sorted order list by sorting and merging the order list with the wholesaler information connected by wholesaler/product name/color/size.

Products that are out of stock or whose arrival is delayed are extracted from the above-mentioned sorted order list, and an out of stock/delayed arrival list, which is a list of the extracted out of stock or whose arrival is delayed, and an order list, which is a list of products that are out of stock or whose arrival is delayed, are created (S207). The out of stock/delayed arrival list is automatically sent to retailers to notify them, and the order list is automatically used to place orders for each sorted wholesaler (S208).

In addition, based on the above order list, the product inventory data is updated using information on products ordered, and changes in inventory are stored (S209), and a receipt for payment for products supplied by a wholesaler to a retailer is automatically generated based on the above order list (S210).

Below, the process of creating an image as product information using the product registration module (104) described above will be described.

FIG. 5 is a block diagram for explaining a product information image generation operation of a product registration module according to one embodiment of the present invention.

Referring to FIG. 5, the product registration module (104) may be configured to include a first image receiving unit (104a), a model generating unit (104b), and a second image generating unit (104c).

The first image receiving unit (104a) can receive a first wearing image equipped with a fashion item from the wholesale terminal (200). In the present embodiments, the fashion item can be clothing, hats, shoes, etc. that can be worn on a user's body area, and can have various styles and colors for each product. The first wearing image can be, for example, an image of a mannequin equipped with the fashion item, an image of a mannequin equipped with the fashion item, a hanger, or a hanger, an image of an actual model equipped with the fashion item, or a flat image of the fashion item.

Figure 6 is an example of an image of a mannequin wearing a fashion item.

As shown in Figure 6, mannequin images inevitably have some body areas missing or unnatural, and therefore have limitations in stimulating users' purchasing power for fashion items compared to actual model images.

The unnaturalness of these mannequin images causes users to have a lower purchasing power for fashion items, which in turn leads to a decline in sales of fashion items.

In addition, if the first wearing image is an actual model image, it has the advantage of being more natural than a mannequin image. However, it takes a lot of time and money to recruit an actual model for shooting for each fashion item, and even if the body area of the actual model is synthesized onto the fashion item for virtual fitting, there are many cases where it is difficult to form empathy with the user because the age, race, etc. of the user looking at the virtually fitted wearing image are not considered at all. In addition, in this case, the image of the actual model often does not match the style, color, etc. of the fashion item, causing the user to feel unnatural. Fig. 7 is a drawing showing an example of a case where the first wearing image is an actual model image.

Accordingly, in the present invention, a virtual human model expected to maximize sales of a fashion item is estimated from the first wearing image, and then the virtual human model is synthesized into a body area within the first wearing image to generate a second wearing image.

The model generation unit (104b) analyzes the first wearing image to determine the body area within the first wearing image, and analyzes the type, style, and color of the fashion item to estimate a virtual human model that is expected to maximize sales of the fashion item. In the present embodiments, the human model is a virtual model implemented in the form of an actual person, and can be dynamically changed according to the type, style, and color of the fashion item.

First, the model generation unit (104b) can analyze the first wearing image to determine a body area within the first wearing image. Here, the body area within the first wearing image means a body area that is exposed outside of the fashion item within the first wearing image.

The body area in the first wearing image can be, for example, the face area, the hand area, the arm area, the calf area, etc.

The model generation unit (104b) can determine a body region in the first wearing image based on the learning data stored in the database. As an example, the model generation unit (104b) can compare the learning data stored in the database with the first wearing image, select learning data whose similarity with the first wearing image is higher than a reference value, and determine the body region in the first wearing image from the body region in the selected learning data. At this time, the model generation unit (104b) can determine the body region in the first wearing image using a widely known image classification technique or image analysis technique, such as SVM (Support vector machine).

Next, the model generation unit (104b) can analyze the type, style, and color of the fashion item to estimate a virtual human model in which sales of the fashion item are expected to be maximized the most. To this end, the model generation unit (104b) can classify the fashion item into one of a plurality of pre-classified categories according to the style and color of the fashion item. Here, the type of the fashion item can be, for example, a men's jacket, men's pants, a women's one-piece dress, a women's blouse, sneakers, shoes, a casual hat, etc. In addition, the style of the fashion item can be, for example, casual, street, vintage, feminine, dandy, classic, manish, elegance, ethnic, modern, natural, romantic, sporty, newtro, hiphop, maximum, punk, etc. In addition, the color of the fashion item can be, for example, white, black, blue, red, etc. As an example, the model generation unit (104b) can classify the fashion item into categories such as top/outdoor/men's suit/dandy/black, etc. As another example, the model generation unit (104b) can classify the fashion item into categories such as top/outdoor/women's blouse/feminine/white, etc.

Thereafter, the model generation unit (104b) can select learning data corresponding to the above-mentioned classified category from among the learning data stored in the database. The learning data stored in the database are images of wear that have been inserted into past advertising pages, and may be classified into one of a plurality of categories. In addition, the database may store the time each learning data was inserted into the advertising page and the number of user clicks that occurred during that time.

The model generation unit (104b) can estimate a human model based on learning data corresponding to the above-mentioned classified categories, in which the number of clicks on the advertisement page exceeds a reference value relative to the time inserted into the advertisement page. Since it is expected that many purchases have occurred by users in the case of learning data corresponding to the above-mentioned classified categories in which the number of clicks on the advertisement page exceeds a reference value relative to the time inserted into the advertisement page, the model generation unit (104b) can estimate a human model optimized for the sale of fashion items based on learning data corresponding to the above-mentioned classified categories in which the number of clicks on the advertisement page exceeds a reference value relative to the time inserted into the advertisement page.

Specifically, the model generation unit (104b) can learn attribute information including at least one of a facial shape, a body part feature, a skin tone, an expression, and a hairstyle of learning data corresponding to the classified category, in which the number of clicks on an advertisement page exceeds a reference value relative to the time inserted into the advertisement page. The model generation unit (104b) can estimate a human model that fits a body area in the first wearing image based on the learned attribute information. As an example, if the attribute information of the first learning data in which the number of clicks on an advertisement page exceeds a reference value relative to the time inserted into the advertisement page is a Korean woman in her early 20s with an oval face shape, white skin tone, smiling expression, and long black hair, the model generation unit (104b) can generate in real time a human model of the Korean woman that fits the body area in the first wearing image while corresponding to the learned attribute information. As another example, if the attribute information of the second learning data, which has generated a reference value or more in number of clicks on the advertising page compared to the time inserted in the advertising page, is a Korean woman in her early 20s with a round face shape, white skin tone, expressionless face, and long black hair, the model generation unit (104b) can generate in real time a human model of the Korean woman that matches the body area in the first wearing image while corresponding to the learned attribute information. In this way, there may be a plurality of learning data for which the reference value or more in number of clicks on the advertising page compared to the time inserted in the advertising page. At this time, the model generation unit (104b) can change the expression of the human model into one of a plurality of pre-classified expression categories according to the type, style, and color of the fashion item. The expression categories may be, for example, pure, chic, soft, etc. Since the facial expression of the human model varies depending on the type, style, and color of the fashion item, the model generation unit (104b) can estimate the facial expression of the human model that matches the type, style, and color of the fashion item based on the facial expressions in the learning data for which the number of clicks on the advertisement page exceeds a reference value compared to the time inserted into the advertisement page. As an example, the model generation unit (104b) can change the facial expression of the human model to an innocent form if the fashion item is an elegant white one-piece dress for women, and can change the facial expression of the human model to a chic form if the fashion item is a men's suit.

Meanwhile, if the age group information among the attribute information of each learning data matches the target age group information calculated in proportion to the price of the fashion item included in the first wearing image, a relatively high weight (e.g. a weight exceeding 1) can be applied to the learning data.

Accordingly, in the case of expensive products, even if the number of clicks on the advertising page is low compared to the time inserted on the advertising page, if the age group is high, the probability of estimating a human model according to the attribute increases.

In addition, the model generation unit (104b) can estimate the human model by correcting each attribute information using a Generative Adversarial Network (GAN) model and a Diffusion model. For example, the model generation unit (104b) can correct each attribute information corresponding to multiple learning data in which the number of clicks on the advertising page exceeds a reference value compared to the time inserted into the advertising page, and compare the return on ad spend (ROAS) of the advertising page including the learning data with the highest similarity to the corrected attribute information, and then estimate the human model using the attribute information when the return on ad spend was the highest. In the case of the GAN model, attribute information of an area that users feel is unnatural (e.g., eyes, nose, mouth, ears, skin, hair, etc.) can be individually learned, and then each attribute information can be corrected to have a naturalness exceeding a reference value. At this time, if only the GAN model is used, a problem may occur in which the print or texture of the fashion item appears blurry, so the model generation unit (104) can newly draw the blurry or empty part using the Diffusion model. In the case of the Diffusion model, the blurry or empty part can be automatically generated and corrected based on a large number of reference images. In this case, a more natural human model can be automatically estimated while the advertising revenue is expected to be the highest.

In addition, the model generation unit (104b) can generate a human model by changing a first pose included in the first wearing image to a second pose different from the first pose, or by changing the shooting angle of the fashion item from the first shooting angle (e.g., side) to the second shooting angle (e.g., rear) using a generative AI model such as a GAN model. That is, the model generation unit (104b) generates a human model based on the first wearing image, but can change the pose or shooting angle in the process of generating the human model using the generative AI model. At this time, the model generation unit (104b) can determine a pose or shooting angle that can maximize the sales of the fashion item according to the type, style, and color of the fashion item, and determine the pose or shooting angle of the human model according to the determined pose or shooting angle. The model generation unit (104b) can determine the pose and shooting angle of a human model optimized for the sale of fashion items based on learning data corresponding to the classified categories, in the same manner as described above, in which the number of clicks on the advertising page exceeds a reference value compared to the time inserted in the advertising page.

The pose can be selected from among model poses in the learning data that are identical to or have a similarity level higher than a standard value with the type of fashion item, style, and color included in the first wearing image, and one pose from among the poses in the first wearing image.

Since the pose in the first wearing image may be the result of the wholesaler setting it as the pose that best suits the fashion item, a weight exceeding 1 may be applied to poses among the model poses of the learning data that are identical to the pose in the first wearing image or have a similarity level higher than a standard value.

As a result of applying weights in the above manner among the model poses of the learning data that are identical to or have a similarity level higher than a standard value with respect to the specific fashion item type, style, and color, and the poses in the first wearing image, the pose with the largest number can be determined as the final pose.

In addition, the model generation unit (104b) can estimate the human model based on the location (country, etc. information) and race of the target consumer received from the wholesale terminal (200). To this end, the information of the target consumer can include at least one of nationality, gender, age, and race. The model generation unit (104b) estimates the human model based on learning data corresponding to the classified category, in which the number of clicks on the advertisement page relative to the time inserted therein exceeds a reference value. In the process of estimating the human model, if there are multiple learning data in which the number of clicks on the advertisement page relative to the time inserted therein exceeds a reference value, the model generation unit (104b) can estimate the human model based on learning data having a high similarity to the information of the target consumer.

For example, if a Korean woman is the target consumer, the model generation unit (104b) can generate a human model of a Korean woman, and if an American woman is the target consumer, it can generate a human model of an American woman.

The database is a storage where learning data required for estimating and generating a human model is stored. As described above, the database may store a plurality of wearing images that have been inserted into past advertising pages as learning data. The learning data may be classified into one of a plurality of categories. In addition, the database may store the time at which each learning data was inserted into the advertising page and the number of user clicks that occurred during that time. In addition, the database may store a plurality of background images corresponding to each learning data.

The second image generation unit (104c) synthesizes the human model estimated by the model generation unit (104b) into the body area in the first wearing image to generate a second wearing image.

The algorithm for driving artificial intelligence described above can be driven by MLOps (Machine Learning Operations). This is a machine learning operation for self-learning through automation, and generative AI can also learn for natural product presentation based on MLOps.

FIGS. 8 and 9 are examples showing second wearing images generated according to an embodiment of the present invention.

Referring to FIG. 8, the second image generating unit (104c) can generate a second wearing image synthesized from the image shown in FIG. 6, which represents a Korean woman in her early 20s with an oval face shape and long brown hair.

At this time, the second image generation unit (104c) can replace not only the human model but also the background image in the first wearing image with a new background image. The second image generation unit (104c) can select a background image with the highest degree of compatibility with the human model among the background images stored in the database, and replace the background image in the first wearing image with the selected background image. The degree of compatibility between each attribute information of the human model and the background image can be stored in a numerical form in the database. For example, the degree of compatibility can increase as the frequency of existence of learning data having specific attribute information and background data having specific attribute information increases.

Meanwhile, referring to FIG. 9, the second image generation unit (104c) can generate a second wearing image synthesized with a human model representing a white woman from the image shown in FIG. 7. In addition, in the model, accessories (e.g., earrings) can be changed to something else, or the hairstyle can be changed. This can vary depending on the fashion style identified by the model generation unit (104b), for example. For example, in the case of an elegant style, the model generation unit (104b) can estimate that the human model's hair is long and change the size of the accessory to a small one.

Additionally, according to another embodiment, depending on the color of the identified fashion item, a smaller sized accessory may be applied to the human model for an item with lower brightness.

The above description of the present invention is for illustrative purposes, and those skilled in the art will understand that the present invention can be easily modified into other specific forms without changing the technical idea or essential characteristics of the present invention. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive. For example, each component described as a single component may be implemented in a distributed manner, and likewise, components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the claims described below, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention.

100: Integrated Management Platform
200: Wholesale Terminal
300: Retail Terminal
400: Producer Terminal
500: Financial Institution System
600: Delivery System

Claims (3)

As a wholesale and retail integrated management platform and strategic product sourcing system that operates in conjunction with multiple wholesale terminals, retail terminals, and producer systems,
A product registration module that receives product information from multiple wholesaler terminals and stores it in a database;
An order list registration module that provides product information, inventory information, and receipt information stored in a database to a number of retail terminals and receives orders received from consumers from the retail terminals as order list information;
A wholesaler link module that connects retailers who have registered the above order list with wholesalers who supply the corresponding products;
Order list sorting/merging module that generates a sorted order list by sorting and merging the linked order list of wholesalers and counting the quantities; and
Includes an out-of-stock/delayed-delivery product list extraction module that extracts and stores out-of-stock and delayed-delivery products from the above-mentioned sorted order list.
The above wholesale terminal receives an order list excluding out-of-stock and delayed delivery products from the sorted order list.
The above product registration module,
A first image receiving unit receiving a first wearing image including a fashion item from a wholesale terminal;
A model generation unit that estimates a human model based on analysis of the above fashion items;
Including a second image generation unit that generates a second wearing image in which the fashion item is equipped on the human model estimated by the model generation unit;
The above first wearing image is an image of a fashion item being worn on a mannequin, not an actual model.
The above model generation unit learns attribute information of learning data among the previously stored learning data in which the number of clicks on the advertising page is greater than a reference value compared to the time inserted into the advertising page, and estimates a human model suitable for a body area in the first wearing image based on the learned attribute information, and if the age group information among the attribute information of each learning data matches the target age group information calculated to be proportional to the price of the fashion item included in the first wearing image, a first weight exceeding 1 is applied to the learning data, and a second weight exceeding 1 is applied to learning data having a pose that is identical to or has a similarity greater than a reference value with respect to the pose in the first wearing image among the model poses of the learning data.
The second image generation unit applies accessories to a human model, and the lower the brightness of the fashion item, the smaller the size of the accessories is applied. This is a wholesale and retail integrated management platform and strategic product sourcing system. delete In the first paragraph,
A wholesale and retail integrated management platform and strategic product sourcing system that further includes a trend analysis module that combines information on the fashion styles desired by consumers with current fashion trends, derives optimal options by season and age, and suggests them to wholesalers and retailers.