Image / Text-Based Design Generation Device and Method

Abstract

To provide a device and a method for extracting and generating fashionable design based on an image and a text.SOLUTION: A design generation device 200 includes: a parsing unit configured to parse image data and text data; a learning unit for learning a design generation model using artificial intelligence based on parsed image data and text data; and a design generation unit for generating a new design of a specific item or an element of design using the design generation model learned by the learning unit.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to an apparatus and method for extracting and generating popular designs based on images and texts.

Background Art

[0002] The speed of change in design trends is extremely fast. In particular, with the emergence of global SPA brands and the increasing activity of SNS, the speed of change in trends has become even faster, and trendy items tend to be embraced as the same design globally.

[0003] Thus, in order to match the fast speed of trend change and the supply by global designs, designers need repeated data collection and assistance in early design work, and fashion merchandisers need to separately collect and analyze the latest trends and emerging brand information. Also, designers must thoroughly investigate the rapidly changing fashion trends and create designs that match them. However, in a hurry, although they must create three designs within seven days, in reality, they can only complete one new design after drawing 200 - 300 pieces.

[0004] Also, in the case of fashion merchandisers, it is inefficient because it takes a lot of cost, labor, and time to update fashion trends and generate new designs, such as visiting collections to investigate trends and brands, going to major fashion cities like New York / Paris, and having to create order lists and look books.

Summary of the Invention

Problems to be Solved by the Invention

[0005] As one embodiment of the present invention, the present invention provides a design generation device including: a parsing unit that parses image data and text data; a learning unit that uses artificial intelligence to train a design generation model based on the parsed image data and text data; and a design generation unit that generates a new design or design elements of a specific item using the design generation model trained by the learning unit.

[0006] As another embodiment of the present invention, the present invention provides a design generation method including: a step of using artificial intelligence to train a design generation model based on the parsed image data and text data; and a step of generating a new design or new design elements of a specific item using the trained design generation model.

[0007] As another embodiment of the present invention, the present invention provides a program recorded on a recording medium capable of generating a new design based on an image and text, the program including: a step of using artificial intelligence to train a design generation model based on the parsed image data and text data; and a step of generating a new design or new design elements of a specific item using the trained design generation model.

Advantages of the Invention

[0008] According to various embodiments of the present invention, a new design can be generated by a more time- and cost-efficient means due to the rapidly changing design trends.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Embodiments for Carrying Out the Invention

[0010] Hereinafter, the present disclosure will be described in detail with reference to the drawings. At this time, the same components in each drawing are denoted by the same reference numerals as much as possible. In addition, detailed descriptions of already known functions and / or configurations are omitted. The content disclosed below focuses on the parts necessary for understanding the operations according to various embodiments, and descriptions of elements that may obscure the gist of the description are omitted.

[0011] In this document, expressions such as "A or B", "at least one of A or / and B", or "one or more of A or / and B" may all include all possible combinations of the listed items. For example, "A or B", "at least one of A and B", or "at least one of A or B" can all refer to cases including (1) including at least one A, (2) including at least one B, or (3) including both at least one A and at least one B.

[0012] Expressions such as "first", "second", "the first", or "the second" used in various embodiments can modify various components without regard to order and / or importance, and do not limit the components. For example, without departing from the scope of the rights of the present invention, the first component may be named the second component, and similarly, the second component may also be named the first component.

[0013] Also, some components in the figures may be exaggerated, omitted, or shown schematically. The size of each component does not fully reflect the actual size, and thus, the content described herein is not limited by the relative sizes and intervals of the components depicted in each figure.

[0014] FIG. 1 is a diagram showing the overall configuration according to an embodiment of the present invention. The design generation device 200 parses images that are exposed in real time on the Internet, such as websites, SNSs, blogs, etc., and text data related to the images. The images may include all images that can be extracted on the Internet. Specifically, for example, the images may include images containing all items related to fashion, images containing all items related to accessories, game content-related images (all images required for games, such as game backgrounds, characters, clothes, items, buildings, maps, etc.), movie content-related images (all images required for movies, such as movie backgrounds, characters, special effects, buildings, maps, etc.), brand logo images, interior-related images (all images required for interiors, such as furniture images, furniture arrangements, wallpapers, raw silk, etc.), various tableware images, etc. Also, the images include all images that can be visually identified, such as landscape photos and paintings. Also, the images may include animal images such as birds and puppies.

[0015] In addition, when the design generation device 200 parses images on a website such as a shopping mall, it can also parse data on the sales volume and sales amount corresponding to the images. Also, the text related to the images may be, for example, when selling T-shirts in a shopping mall, text such as "T-shirt", "tee", or "top" described in the category or product name of the image. Also, the text data may include information on the product specifications of the corresponding product, such as materials and textures. Also, product names included in the text data may include accessories such as "earrings" and "necklaces", interior terms such as "wallpaper", "raw silk", and "fabric", and furniture terms.

[0016] Also, the text data may include content design terms. The content may include all of games, movies, manga, etc. Also, the text data may include all terms related to the images to be parsed.

[0017] In addition, the design generation device 200 can receive by parsing in real time image data uploaded to SNS such as Instagram or Facebook for a certain period of time. The latest trends can be understood through images uploaded to SNS such as Instagram or Facebook. Also, the design generation device 200 can parse the image data and text data recorded in the design generation device 200 itself.

[0018] The design generation device 200 that parses images and text data related to the images that are exposed in real time on the Internet such as websites, SNS, blogs, etc., or are in the design generation device 200 generates new designs or design elements for specific items based on the parsed images and text. Here, the design elements may include at least one of all elements that can constitute a design, such as patterns, forms, hues, etc. The design elements may be, for example, elements that constitute clothing, footwear, etc., and the design elements may be, for example, elements that constitute interior-related items (all items related to furniture, wallpaper interior). Also, the design elements may be elements that constitute game content (backgrounds of games, characters, clothes, items, buildings, maps, etc.). Also, the design elements may be elements that constitute movie content (for example, backgrounds of movies, characters, special effects, buildings, maps, etc.). Also, the design elements may be elements that constitute brand logos, etc. Also, the design elements can constitute all elements that can be generated in photo images. For example, design elements that can constitute landscape images, animal images, human images, etc. can be cited as examples.

[0019] Specifically, the design generation device 200 extracts items using the parsed image and text data. In item extraction, in addition to the parsed image and text data, an item extraction model (object detection) is used. For example, items may include various items such as footwear, bags, dresses, neckties, shirts, hats, cardigans, coats, etc. Items may include all accessories, all items related to interiors, components of movie content, components of game content, brand logos, etc.

[0020] For example, if there is a vector value corresponding to footwear in the image, the "footwear" part of the image can be recognized to extract a footwear item.

[0021] Also, text data regarding the parsed image may also be used to extract items within the image. Also, multiple items can be extracted from one image. For example, if there is a man in an image wearing a beanie hat, a turtleneck sweater over a pea coat, jeans, and sneakers, the items extracted from the image may be "beanie hat", "turtleneck sweater", "pea coat", "jeans", and "sneakers".

[0022] In addition, the design generation device 200 extracts design elements for each of the extracted items. As described above, the design elements may be elements included in all images, and will be described in more detail below based on clothing. The design elements include all external elements that make up an item, such as hue, sleeve length, fit, pattern, print, etc. For example, referring to FIG. 3, the design elements of a knee-length white flare one-piece without sleeves with an overall floral print include elements such as the floral print, no sleeves, white, knee length, and flare one-piece. Further, the design generation device 200 can extract design elements, such as patterns, hues, shapes, etc., not only using the design elements for each item, but also using elements such as the overall mood, shape, hue, etc. in the images in which the items have not been extracted among the collected images.

[0023] The design generation device 200 can extract the top designs or design elements from the extracted designs or design elements. Specifically, the top designs or design elements can be determined based on the exposure frequency of specific designs or design elements on images exposed on SNS or websites being equal to or higher than a reference value. Also, the top designs or design elements can be determined based on the sales volume and sales amount of items containing specific designs or design elements on a specific website being equal to or higher than a reference value. Furthermore, the top designs or design elements can be determined based on the preference degree for the image exposed on the Internet such as a specific website or SNS (judgment based on all data that can estimate the preference degree for the image, such as the number of clicks, feedback numbers, share numbers, etc.). Also, for the design elements themselves, design elements equal to or higher than a certain reference value can be extracted considering all of the exposure frequency, sales volume, preference degree, etc. For example, referring to FIG. 3, if the design elements in the extracted design are "white", "sleeveless", "floral pattern", and "flare one-piece", each element may be a top design element. Also, when determining whether a specific design or design element is a top design or design element, it can be determined whether it is a design or design element included in an item worn by a famous person such as a celebrity, and a higher weight value can be given to such a design or design element for judgment. Also, for example, when the top design is furniture, the design elements can include all elements forming the furniture such as "raw wood", "leather", "presence or absence of armrests", etc. Thus, the top design can be any design (game content, movie content, brand logo, etc.) that can be included in the image, and in this case, the same top design extraction process may be applied.

[0024] The design generation device 200 extracts a higher-level design or elements of a design and records them in a recording unit. The design generation device 200 extracts and updates elements of the higher-level design in real time according to rapid changes in the design, categorizes and records the elements of the higher-level design for a large number of items. Also, when there are no items (such as a general image), only the elements of the design can be recorded.

[0025] In addition, the design generation device 200 can generate and train a design generation model by reflecting elements of the higher-level design among the image data and text data parsed using artificial intelligence, the elements of the design / elements of the design for the extracted specific item, and the elements of the design for the specific item extracted by the design extraction unit. Furthermore, the design generation device 200 trains the design generation model by additionally considering the generated design for a specific item or elements of the design, and also evaluates the generated design for a specific item or the elements of the design, and trains the design generation model by additionally considering the results. The evaluation of the design or elements of the design will be described later.

[0026] After that, the design generation device 200 generates a new design (higher-level design or elements of the design) using the design generation model that is trained in real time by the learning unit.

[0027] To generate a new design, GAN (generative adversarial networks) that can specialize in categories and perform random generation, VAE (Variational Autoencoder) + GAN that can randomly transform a specific design and generate a new design in a form with a high degree of similarity to the characteristics of the specific design, genetic algorithm + GAN that can recognize the elements of a specific design like human genetic traits (such as facial features), create various crosses, and repeatedly transfer new and more marketable variations across generations based on the feedback for them, conditional GAN for design changes, and style transfer technology that extracts inspiration for creating a new design from data such as various images and transforms the style while maintaining the external appearance of the existing design may be used.

[0028] Machine learning can be broadly classified into three types. The first is supervised learning, the second is reinforcement learning, and the last is unsupervised learning.

[0029] Here, GAN belongs to unsupervised learning and generates new images as a model in which a Generator (G) that generates images and a Discriminator (D) that evaluates the images generated by the Generator oppose each other and improve each other's performance. Specifically, D tries to determine that only the original data is real, and G generates fake data so that D cannot distinguish it as fake, improving the performance of both models. Eventually, D can no longer distinguish between real and fake data.

[0030] As one embodiment of generating a design, there is a method of changing the vector value of the elements of a design to generate elements of a new design. The change of the vector value may be input as a set value and may be performed by a learned design generation model. Specifically, for example, referring to FIG. 5, if a floral one-piece is in vogue, the vector value of the existing pattern in the floral one-piece extracted from the collected images is changed, and a design of a new floral one-piece can be generated.

[0031] Also, as another embodiment of generating elements of a new design, there is a method of synthesizing different elements among the elements of a design to generate a new design for a specific item. Specifically, different elements among the elements of each popular design can be merged to generate a new design. For example, referring to E in FIG. 2, by synthesizing the "puff sleeves" of knit, the "ultra violet color" of a shirt, and the "long one-piece" among the elements of the design, a design of a long one-piece with puff sleeves and in ultra violet color can be generated. Also, such different elements may be automatically and randomly merged, or may be merged according to the input set value (for the elements of the design to be merged). Different elements of a design are applicable even when the items are different. For example, the pattern of clothing may be applied to furniture, or the shape of an accessory may be applied to a game item.

[0032] Also, as another embodiment of generating elements of a new design, there is a method of randomly changing and synthesizing the elements of a design to generate elements of a new design. For example, referring to FIG. 7, by extracting the pattern for a flower from a photo of a flower and extracting the sunset color and mood from a photo of a setting sun, a pattern with a floral pattern and in sunset color may be generated.

[0033] In addition, the design or design elements generated in this way may be reflected when the design generation model is learned again to learn the design generation model, and the newly learned design generation model may be used when generating new designs or design elements again.

[0034] Next, the design generation device 200 evaluates the generated new design or design elements according to certain criteria. At this time, the design generation device 200 can rank the new designs or design elements extracted in consideration of various elements. For ranking the new designs or design elements, generally, the exposure frequency of a specific design or design element on the image exposed on an SNS or website, the sales volume and sales amount of items including the specific design or design element, and the preference for the image exposed on the Internet such as a specific website or SNS (judgment based on all data that can estimate the preference for the image such as the number of clicks, the number of feedbacks, and the number of shares) can be used for judgment. Also, when judging the exposure frequency or determining the ranking, a weight value is given to the design or design element exposed in the SNS or image of a celebrity or famous person, and among the same designs or design elements, if the preference, sales volume, or sales amount for the design or design element sold by a specific brand or specific website is higher, a weight value can be given to the design or design element (hereinafter, "judgment of weight value"). The design generation device 200 ranks the new designs or new design elements generated in this way.

[0035] In addition, in this way, the design generation model can be learned using the information ranking the new designs or new design elements.

[0036] FIG. 2 is a diagram showing the configuration of the design generation device 200 according to an embodiment of the present invention.

[0037] The design generation device 200 includes an input unit 201, a parsing unit 202, an extraction unit 203, a recording unit 204, an upper-level design extraction unit 206, a learning unit 207, a design generation unit 208, a display unit 210, and a transmission unit 212.

[0038] It can receive text and labels via the input unit 201 and reflect the changes in vector values.

[0039] The parsing unit 202 parses images and text data related to the images in real time from an external website, blog, or SNS server via a network, or from the recording unit 204. It also parses sales volume data and sales amount data related to specific images from the server of a specific website or the design generation device itself. The text related to the image may be, for example, text such as "T-shirt", "tea", or "top" described in the category or product name of the image when selling T-shirts in a shopping mall.

[0040] The extraction unit 203 extracts items using the parsed images and text data. In item extraction, in addition to the parsed images and text data, an item extraction model (object detection) is used. For example, items may include various items such as footwear, bags, dresses, neckties, shirts, hats, cardigans, coats, etc. At this time, if there is a vector value corresponding to footwear in the image, the "footwear" part in the image can be recognized and the footwear item can be extracted.

[0041] In addition, the extraction unit 203 extracts design elements for each of the extracted items. The design elements include all external elements that make up the item, such as hue, sleeve length, fit, pattern, print, etc. Furthermore, the extraction unit 203 can extract multiple items from one image. For example, referring to FIG. 4, if there is an image of a woman holding a cat, wearing sunglasses, hot pants, a black T-shirt, and straw sandals, the items extracted are sunglasses, hot pants, T-shirt, and footwear (straw sandals). Here, the cat is not extracted as an item.

[0042] In addition, the extraction unit 203 extracts design elements for each of the extracted items. The design elements include all external elements that make up the item, such as hue, sleeve length, fit, pattern, print, material, etc. For example, referring to FIG. 3, the design elements of a knee-length white flare one-piece without sleeves with an overall floral print include elements such as floral print, no sleeves, white, knee length, and flare one-piece.

[0043] In addition, the extraction unit 203 can extract design elements, such as pattern, hue, shape, etc., using not only the design elements for each item but also elements such as the overall mood, shape, and hue in the images where items have not been extracted among the collected images.

[0044] The recording unit 204 records the item extraction model. The item extraction model can determine which item the vector value in the image corresponds to. In addition, the recording unit 204 categorizes and records the extracted items. Furthermore, the recording unit 204 categorizes and records the top-level designs for each item extracted by the top-level design extraction unit 206.

[0045] The upper design extraction unit 206 determines what the upper design or design elements are among the extracted designs. Specifically, the upper design or design elements can be determined based on the case where the exposure frequency of a specific design is equal to or higher than a reference value in terms of the image exposed on an SNS or website, etc. Also, in a specific website, the upper design or design elements can be determined based on the case where the sales volume and sales amount of items including a specific design or design elements are equal to or higher than a reference value. Furthermore, the upper design or design elements can be determined based on the preference degree for the image exposed on the Internet such as a specific website or SNS (judgment based on all data that can estimate the preference degree for the image, such as the number of clicks, the number of feedbacks, the number of shares, etc.). For example, referring to FIG. 3, if the design elements in the extracted design are "white", "sleeveless", "floral pattern", and "flare one-piece", each element may be an upper design. Also, when determining the upper design or design elements, by determining whether a famous person such as a celebrity has worn it, a predetermined weight value can be given to such a design or design elements for judgment.

[0046] The learning unit 207 can generate and learn a design generation model by reflecting the image data and text data parsed using artificial intelligence, the design elements / design elements for the extracted specific item, and the elements of the upper design among the design elements for the specific item extracted by the design extraction unit. Also, the learning unit 207 learns the design generation model by additionally considering the generated design for the specific item or the design elements, and also evaluates the generated design for the specific item or the design elements, and learns the design generation model by additionally considering the result.

[0047] The design generation unit 208 generates a new design using a design generation model that is learned in real time by the learning unit. To generate a new design in the design generation unit 208, GAN (generative adversarial networks) that enables specialization of categories and random generation, VAE+GAN that enables random deformation of a specific design, genetic algorithm+GAN, conditional GAN for design change, and style transfer technology may be used.

[0048] Specifically, the design generation unit 208 uses the following exemplary methods to generate a new design or design elements.

[0049] As one embodiment of generating a design, there is a method of changing the vector value of an element of a popular design to generate a new design element. The change of the vector value may be input as a set value and may be automatically performed by the learned design generation model. Specifically, for example, referring to FIG. 5, if a floral one-piece is popular, the vector value of the existing floral pattern in the floral one-piece extracted from the collected images is changed to generate a new design of a floral one-piece.

[0050] Also, as another embodiment of generating a new design element, there is a method of synthesizing different elements among the design elements to generate a new design for a specific item. Specifically, different elements among the elements of popular designs can be combined to generate a new design. For example, referring to FIG. 6, by combining the "puff sleeves" of knitwear, the "ultra violet color" of a shirt, and the "long one-piece" among the elements of popular designs, a design of a long one-piece with puff sleeves in ultra violet color can be generated. Also, such different elements may be automatically randomly combined or may be combined according to the input set values (for the design elements to be combined).

[0051] Also, as another embodiment of generating new design elements, there is a method of generating new design elements using a design generation model based on the design elements.

[0052] For example, referring to FIG. 7, a new pattern may be generated by extracting a pattern for a flower from a photograph of the flower and extracting the sunset color and mood from a photograph of a sunset.

[0053] Also, the design or design elements generated in this way may be reflected when learning the design generation model again to learn the design generation model, and the newly learned design generation model may be used when generating new designs or design elements again. Further, such a learned design generation model may include individual models depending on the types of its input values and output values.

[0054] Next, the design generation unit 208 evaluates the generated new design or design elements according to certain criteria. Specifically, the design generation unit 208 ranks the new design or design elements extracted in consideration of various elements.

[0055] To rank a new design or design element, generally, it can be determined based on the exposure frequency of a specific design or design element on an image exposed on an SNS or website, etc., the sales volume and sales amount of items including the specific design or design element, and the preference for the image exposed on the Internet such as a specific website or SNS (all data that can estimate the preference for the image, such as the number of clicks, the number of feedbacks, and the number of shares). Also, when judging the exposure frequency or determining the ranking, a weight value is given to the design or design element exposed on the SNS or image of a celebrity or famous person. Among the top design elements, if the preference, sales volume, or sales amount for the design sold under a specific brand or specific website is higher, a weight value can be given to the design element (hereinafter, "judgment of weight value"). The design generation device 200 can rank the new design or design element generated in this way. Also, in this way, the design generation model can be learned using the information in which the new design or new design element is ranked.

[0056] The display unit 210 can display the design generated by the design generation unit 208.

[0057] When proceeding with manufacturing for the generated design, the transmission unit 212 can send the design to be manufactured and the quantity of manufacturing to the factory side.

[0058] FIG. 8 is a diagram showing the configuration of the extraction unit 203 according to an embodiment of the present invention.

[0059] The extraction unit 203 includes an item extraction unit 300 and a design extraction unit 302.

[0060] The item extraction unit 300 extracts items using the received image and text data. In item extraction, in addition to the received image and text data, the recorded item extraction model is also used. Here, items include various items such as footwear, bags, dresses, neckties, shirts, hats, cardigans, coats, etc. If there is a vector value corresponding to footwear in the image, the "footwear" part in the image can be recognized and the footwear item can be extracted. Also, to extract an item from the received image, text related to the image may be used as a reference. Furthermore, the item extraction unit 300 can extract multiple items from one image. For example, referring to FIG. 4, if there is an image of a woman holding a cat, wearing sunglasses, hot pants and a black T-shirt, and straw sandals, the extracted items are sunglasses, hot pants, T-shirt, footwear (straw sandals). Here, the cat is not extracted as an item.

[0061] The design extraction unit 302 extracts the design and design elements of the extracted items. Also, the design extraction unit 302 can extract design elements through the parsed image, text, etc. Design elements include all external elements that make up items such as hue, sleeve length, fit, pattern, print, etc. Also, design elements can include hue, mood, material, pattern, etc. in a general image even if they do not make up an item.

[0062] FIG. 9 is a diagram showing the configuration of the design generation unit 208 according to an embodiment of the present invention.

[0063] The design generation unit 208 includes a design reception unit 400, a design change unit 402, a design synthesis unit 404, and an evaluation unit 406.

[0064] The design generation unit 208 generates a new design using a design generation model that is learned in real time in the learning unit.

[0065] To generate a new design, GAN (generative adversarial networks) that can specialize in categories and perform random generation, VAE+GAN that can randomly transform a specific design, genetic algorithm+GAN, conditional GAN for design modification, and style transfer technology may be used.

[0066] Machine learning can be broadly classified into three categories. The first is supervised learning, the second is reinforcement learning, and the last is unsupervised learning.

[0067] Here, GAN belongs to unsupervised learning. A new image is generated as a model in which a Generator (G) that generates an image and a Discriminator (D) that evaluates the image generated by the Generator oppose each other to improve each other's performance. Specifically, D tries to judge only the original data as true, and G generates fake data so that D cannot distinguish it as fake, improving the performance of both models. Eventually, D can no longer distinguish between true and fake data.

[0068] Also, cGAN differs from the existing GAN in that it can generate what is desired instead of randomly generating.

[0069] The design receiving unit 400 receives all elements to generate a new design. For example, it can receive a design generation model and elements of the design. The design change unit 402 generates a new design by changing the vector values of the design elements using the design elements extracted for each item. Specifically, for example, if there is a shirt with a floral pattern, it is possible to change the vector value of the floral pattern on the shirt extracted from the collected images to generate a new design for the shirt with a floral pattern. Also, the design change unit 402 can generate new design elements by changing the vector values for the design elements extracted from an item and the design elements extracted from an image without the item using a design generation model. The change of the vector value may be performed automatically or may be performed upon receiving a set value.

[0070] Using the design elements extracted from the collected items, the design composition unit 404 synthesizes different elements among the trendy design elements for each collected item to generate a new design for the item. Also, the design change unit and the design composition unit can be used in combination to generate a new design.

[0071] Specifically, referring to FIG. 6, for example, among the elements of trendy designs, the “puff sleeves” of a knit, the “ultra violet color” of a shirt, and the “long one-piece” are synthesized, and a design of a long one-piece with puff sleeves and in ultra violet color can be generated. Also, different elements of the same item can be synthesized. For example, the “cuffed sleeves” of a blouse, the “floral color” of the blouse, and the “mint color” of the blouse, which are elements of trendy designs, may be combined to generate one blouse design.

[0072] Also, the design composition unit 404 can generate new design elements by synthesizing the design elements extracted from an item and the design elements extracted from an image without the item.

[0073] In addition, the design synthesis unit 404 can also synthesize the design elements extracted from the item and the design elements extracted from the image that does not contain the item to generate a new design for a specific item.

[0074] The evaluation unit 406 evaluates the newly generated design or design elements by the design change unit 402 or the design synthesis unit 404 according to certain criteria. Specifically, the evaluation unit 406 can rank the newly generated design or design elements extracted by considering various elements.

[0075] To rank the new design or design elements, generally, the exposure frequency of a specific design or design elements on the image exposed on SNS or websites, the sales volume and sales amount of the items including the specific design or design elements, and the preference for the image (including the design or design elements similar to the new design or design elements) exposed on the Internet such as a specific website or SNS (judgment based on all data that can estimate the preference for the image, such as the number of clicks, the number of feedbacks, and the number of shares) can be considered. Also, when judging the exposure frequency or determining the ranking, a weight value can be given to the design or design elements exposed on the SNS or image of a celebrity or famous person. If the preference, sales volume, or sales amount for the design sold by a specific brand or specific website among the top design elements is even higher, a weight value can be given to the design elements (hereinafter, "judgment of weight value"). The evaluation unit 406 can rank the new design generated in this way.

[0076] For example, if the generated design is a short cardigan in burgundy color, the burgundy color can be expressed in various ways for each product. That is, although it is the same burgundy color, different tones of burgundy color can be extracted little by little. Therefore, in order to select the burgundy color for various types of burgundy short cardigans, even for the same burgundy color, a more preferred color can be extracted to derive the final burgundy short cardigan. Here, in order to extract a more preferred color although it is the same burgundy color, it can be determined based on elements that can determine the preference degree between colors, such as whether it is a color used by a specific brand with a high preference degree or a color used in an image with a high exposure degree.

[0077] Figure 10 is a flowchart of a method for extracting popular designs according to an embodiment of the present invention.

[0078] The parsing unit 202 parses the website, SNS, images and texts related to the images that are exposed in real time on the blog received via the parsing unit 202 (S500). In addition, the parsing unit 202 parses the images, texts related to the images, and texts recorded in the design generation device.

[0079] The item extraction unit 300 extracts items based on the collected images and texts (S502).

[0080] Specifically, the item extraction unit 300 extracts items using the parsed image and text data. In item extraction, in addition to the parsed image and text data, a recorded item extraction model is used. Here, items include various items such as footwear, bags, one-piece dresses, neckties, shirts, hats, cardigans, coats, etc. If there is a vector value recognized as a shirt in the image, the "shirt" part of the image can be recognized and the shirt item can be extracted. Also, to extract an item from the received image, text related to the image may be used as a reference. Further, the item extraction unit 300 can extract a plurality of items from one image.

[0081] If an item is extracted by the item extraction unit 300, the design extraction unit 302 extracts design elements from the item (S504). In item extraction, in addition to the parsed image and text data, an item extraction model (objection detection) is used. For example, items may include various items such as footwear, bags, one-piece dresses, neckties, shirts, hats, cardigans, coats, etc. At this time, if there is a vector value corresponding to footwear in the image, the "footwear" part of the image can be recognized and the footwear item can be extracted.

[0082] Specifically, the design extraction unit 302 extracts design elements of the extracted item (S504). Design elements include all external elements that make up an item, such as hue, sleeve length, fit, pattern, print, etc. Also, design elements can include materials, mood, etc. Further, the design extraction unit 302 can use not only the design elements for each item, but also elements such as the overall mood, shape, hue, etc. in the images where items have not been extracted among the collected images to extract design elements, such as pattern, hue, shape, etc.

[0083] Next, the upper design is determined from the designs or design elements extracted by the upper design extraction unit 206 and the design extraction unit 302 (S506).

[0084] Specifically, a popular design or design element can be determined based on the case where the exposure frequency of a specific design or design element on an image exposed on an SNS or website is equal to or higher than a reference value. Also, the upper design or design element can be determined based on the case where the sales volume and sales amount of items including a specific design or design element on a specific website are equal to or higher than a reference value. Further, the upper design or design element can be determined based on the degree of preference for the image exposed on the Internet such as a specific website or SNS (judgment based on all data capable of estimating the degree of preference for the image, such as the number of clicks, the number of feedbacks, and the number of shares). Also, when determining whether a specific design or design element is an upper design or design element, it can be determined whether a famous person such as a celebrity has worn it, and a higher weight value can be given to such a design or design element for judgment.

[0085] Next, a design generation model that generates new designs or design elements using artificial intelligence is learned (S508).

[0086] The learning unit 207 can generate and learn a design generation model by reflecting the image data and text data parsed using artificial intelligence, the design elements / design elements for the extracted specific item, and the upper design elements among the design elements for the specific item extracted by the design extraction unit. Also, the learning unit 207 learns the design generation model by additionally considering the generated design for the specific item or the design element, and learns the design generation model by additionally considering the result of evaluating the generated design for the specific item or the design element. Next, a new design is generated using the learned design generation model (S510).

[0087] Specifically, the design generation unit 208 uses the following exemplary methods to generate a new design or design elements.

[0088] As one embodiment of generating a design, there is a method of generating new design elements by changing the vector values of elements of a popular design. The change of the vector value may be input as a set value and may be performed by the learned design generation model. Specifically, for example, referring to FIG. 5, if a floral one-piece is popular, the vector value of the existing floral pattern in the floral one-piece extracted from the collected images is changed to generate a new design of a floral one-piece.

[0089] Also, as another embodiment of generating new design elements, there is a method of generating a new design for a specific item by combining different elements among the design elements. Specifically, different elements among the elements of a popular design can be combined to generate a new design. For example, referring to FIG. 6, among the elements of a popular design, the "puff sleeves" of a knit, the "ultra-violet color" of a shirt, and the "long one-piece" are combined to generate a design of a long one-piece with puff sleeves in ultra-violet color. Also, such different elements may be automatically randomly combined, or may be combined according to the input set value (for the design elements to be combined).

[0090] Also, as another embodiment of generating new design elements, there is a method of randomly changing and combining design elements to generate new design elements.

[0091] For example, referring to FIG. 7, a pattern for a flower is extracted from a photo of a flower, and the sunset color and mood are extracted from a photo of a sunset, whereby a pattern in sunset color with a floral pattern can be generated.

[0092] Also, the design or design elements thus generated may be reflected when the design generation model is learned again to learn the design generation model, and the newly learned design generation model may be used when generating a new design or design elements again.

[0093] Although not shown in the figure, the design generation unit evaluates and ranks the design generated after generating the design. Specifically, the design generation unit 208 ranks new designs or design elements extracted in consideration of various elements.

[0094] To rank new designs or design elements, generally, it can be determined based on the exposure frequency of a specific design in the image exposed on an SNS or website, the sales volume and sales amount of items including the specific design, and the preference for the image exposed on the Internet such as a specific website or SNS (all data that can estimate the preference for the image such as the number of clicks, the number of feedbacks, the number of shares, etc.). Also, when judging the exposure frequency or determining the ranking, a weight value is given to the design or design elements exposed in the SNS or image of a celebrity or famous person, and if the preference, sales volume, or sales amount for the design sold under a specific brand or specific website among the upper design elements is even higher, a weight value can be given to the design elements (hereinafter, "judgment of weight value"). The design generation device 200 can rank the new designs or design elements thus generated. Also in this way, the design generation model can be learned using the information in which the new design or new design elements are ranked.

[0095] FIG. 11 is a flowchart of a design change method according to an embodiment of the present invention.

[0096] Next, the design change unit 402 generates a new design or design elements by changing the vector values of the design elements using the learned design generation model (S602). The vector values may be changed by the set values received via the input unit or may be automatically changed.

[0097] Also, the design or design elements generated in this way may be reflected when the design generation model is learned again to learn the design generation model, and the newly learned design generation model may be used when generating a new design or design elements again.

[0098] The generated design or design elements are evaluated and ranked (S604).

[0099] To rank new designs or design elements, generally, based on the exposure frequency of a specific design in the image exposed on SNS or websites, the sales volume and sales amount of items including a specific design or design elements, and the preference for the image exposed on the Internet such as a specific website or SNS (all data that can estimate the preference for the image such as the number of clicks, the number of feedbacks, and the number of shares). Also, when judging the exposure frequency or determining the rank, a weight value is given to the design or design elements exposed in the SNS or image of a celebrity or famous person, and if the preference, sales volume, or sales amount for the design sold under a specific brand or specific website among the upper-ranked design elements is even higher, a weight value can be given to the design elements (hereinafter, "judgment of weight value"). The design generation device 200 can rank the new designs or design elements generated in this way. Also in this way, the design generation model can be learned using the information in which the new design or new design elements are ranked.

[0100] FIG. 12 is a flowchart of a design generation method according to an embodiment of the present invention.

[0101] Using the design generation model, the design synthesis unit 404 generates a new design using different elements among the collected trendy design elements for each item (S702).

[0102] The generated new design may be limited to the design for a specific item. However, the items from which different elements are extracted may be different items.

[0103] The generated design may be reflected when the design generation model is learned again to learn the design generation model, and the newly learned design generation model may be used when generating a new design or design elements again.

[0104] The generated design is evaluated and ranked (S704).

[0105] To rank a new design, generally, it can be determined based on the exposure frequency of a specific design on an image exposed on an SNS or website, the sales volume and sales amount of items including the specific design, and the preference for the image exposed on the Internet such as a specific website or SNS (all data that can estimate the preference for the image such as the number of clicks, feedbacks, and shares). Also, when judging the exposure frequency or determining the rank, a weight value is given to the design exposed on the SNS or image of a celebrity or famous person, and if the preference, sales volume, or sales amount for the design element sold on a specific brand or specific website among the design elements included in the generated design is higher, a weight value can be given to the design element (hereinafter, "judgment of weight value"). The design generation device 200 can rank the newly generated design in this way. Also in this way, the design generation model can be learned using the information in which the new design is ranked.

[0106] Figure 13 is a flowchart of a design generation method according to an embodiment of the present invention.

[0107] Generate elements of a new design using a design generation model (S706).

[0108] Generate elements of a new design using not only the elements of the design for a specific item but also the elements of the design extracted based on an image from which the item is not extracted. The elements of the new design may include patterns, hues, fits, etc.

[0109] Evaluate and rank the generated design elements (S708).

[0110] To rank the elements of the new design, generally, it can be determined based on the exposure frequency of a specific design on an image exposed on an SNS or website, the sales volume and sales amount of items including the elements of the specific design, and the preference for the image exposed on the Internet such as a specific website or SNS (all data that can estimate the preference for the image, such as the number of clicks, the number of feedbacks, and the number of shares). Also, when judging the exposure frequency or determining the ranking, a weight value can be given to the elements of the design exposed in the SNS or image of a celebrity or famous person. Similarly, if the preference, sales volume, or sales amount for the design sold by a specific brand or specific website among the elements of the design is higher, a weight value can be given to the elements of the design (hereinafter, "judgment of weight value"). The design generation device 200 can rank the new design or the elements of the design generated in this way. Also, in this way, the design generation model can be learned using the information in which the new design or the elements of the new design are ranked.

[0111] Referring to FIG. 14, a new design generated via the design change section 402 is shown. When a shirt with various prints is in vogue, a shirt with a new print is designed by changing the existing print elements.

[0112] Referring to FIG. 15, an example is shown in which elements of different designs are combined to generate a new design.

[0113] FIG. 16 is a diagram showing the hardware configuration in which a program for executing the method according to an embodiment of the present invention is recorded.

[0114] The server 1001 corresponds to a design generation device according to an embodiment of the present invention.

[0115] The server 1001 includes a memory 1000, a processor 1002, an input section 1004, a display section 1006, and a communication section 1008.

[0116] The memory 1000 corresponds to the recording section 204 and records an item extraction model. The item extraction model can determine which item the vector value in the image corresponds to.

[0117] Also, the memory 1000 categorizes and records the extracted items. Further, the memory 1000 categorizes and records the design elements for each item extracted by the fashion design extraction section 206. Also, the memory 1000 can record a model for ranking the generated new designs.

[0118] The processor 1002 generates a new design using the collected images and text. It corresponds to the extraction section 203 and the design generation section 208 of the design generation device.

[0119] The processor 1002 uses the item extraction model recorded in the memory 1000 by using the collected images and text, extracts items using the images and text, and then extracts the design or design elements for each item. Alternatively, the processor 1002 can extract the design and design elements using the images and text from which items are not extracted without extracting the items.

[0120] In addition, the processor 1002 extracts the upper-level design or upper-level design elements from the extracted design or design elements. Specifically, the upper-level design or design elements can be determined based on the case where the exposure frequency of a specific design or design element on an image exposed on an SNS or a website is equal to or higher than a reference value. Also, the upper-level design or design elements can be determined based on the case where the sales volume and sales amount of items including a specific design or design element on a specific website are equal to or higher than a reference value. Furthermore, the upper-level design or design elements can be determined based on the preference for the image exposed on the Internet such as a specific website or SNS (judgment based on all data that can estimate the preference for the image, such as the number of clicks, the number of feedbacks, and the number of shares). Also, when determining whether a specific design or design element is an upper-level design or design element, by determining whether a famous person such as a celebrity has worn it, a weight value can be further given to such a design for judgment.

[0121] In addition, the processor 1002 learns the design generation model.

[0122] The processor 1002 can generate and learn a design generation model by reflecting the image data and text data parsed using artificial intelligence, the design elements / design elements for the extracted specific items, and the upper-level design elements among the design elements for the specific items extracted by the design extraction unit.

[0123] Further, the processor 1002 causes the design generation model to learn by additionally considering the generated design for a specific item or elements of the design, and also performs an evaluation on the generated design for a specific item or elements of the design, and causes the design generation model to learn by additionally considering the result.

[0124] Next, the processor 1002 generates a new design using the upper design and the design generation model.

[0125] As one embodiment of generating a design, there is a method of generating elements of a new design by changing vector values of elements of a popular design. The change of the vector value may be input as a set value and may be performed by the learned design generation model. Specifically, for example, referring to FIG. 5, if a floral one-piece is popular, the vector value of the existing floral pattern in the floral one-piece extracted from the collected images is changed to generate a design of a new floral one-piece.

[0126] Also, as another embodiment of generating elements of a new design, there is a method of generating a new design for a specific item by combining different elements among the elements of the design. Specifically, different elements among the elements of a popular design can be merged to generate a new design. For example, referring to FIG. 6, among the elements of a popular design, the "puff sleeve" of a knit, the "ultra violet color" of a shirt, and the "long one-piece" are combined to generate a design of a long one-piece with puff sleeves in ultra violet color. Also, such different elements may be automatically and randomly combined, or may be combined according to set values input (for the elements of the design to be combined).

[0127] Also, as another embodiment of generating elements of a new design, there is a method of randomly changing and combining elements of a design to generate elements of a new design.

[0128] In addition, the processor 1002 evaluates and ranks the generated design or design elements. To rank a new design, generally, based on the exposure frequency of a specific design or design elements on an image exposed on an SNS or website, the sales volume and sales amount of items including the specific design, and the preference for the image exposed on the Internet such as a specific website or SNS (judgment based on all data that can estimate the preference for the image, such as the number of clicks, the number of feedbacks, and the number of shares). Also, when judging the exposure frequency or determining the ranking, a weight value is given to the design or design elements exposed in the SNS or image of a celebrity or famous person, and if the preference, sales volume, or sales amount for the design sold under a specific brand or specific website among the design elements is higher, a weight value can be given to the design elements (hereinafter, "judgment of weight value"). The design generation device 200 ranks the newly generated design or design elements in this way and derives only the designs above a certain rank as the final result design.

[0129] The result ranked in this way is reflected when learning the design generation model.

[0130] The display unit 1006 can display the generated design.

[0131] The communication unit 1008 can receive image data or text data, or when proceeding with the manufacture of the generated design, send the design to be manufactured and the quantity of manufacture to the factory side.

[0132] In addition, an apparatus or system according to various embodiments may include at least one or more of the foregoing components, some may be omitted, or may further include additional other components. And the embodiments described in this document are presented for the purpose of explaining and understanding the disclosed technical content, and do not limit the scope of the present invention. Therefore, the scope of this document should be construed as including all modifications or various other embodiments based on the technical idea of the present invention.

Claims

1. A parsing unit that parses image data and text data in real time online; A learning unit that uses artificial intelligence to learn a design generation model based on the parsed image data and text data; and A design generation unit that generates a new design or design elements for a specific item using the design generation model learned based on the data parsed in real time by the learning unit without a separate input from the user; An extraction unit that extracts at least one of an item, a design, and design elements from the image data and text data; A recording unit that categorizes and records the extracted item, the design, and the design elements; comprising The learning unit and the design generation unit perform learning of the design generation model and generation of a design or design elements using at least one of GAN, VAE+GAN, GA+GAN, cGAN, and style transfer, The design generation unit receives a vector value corresponding to the item, the design, or the design element categorized by the recording unit, and includes a design change unit that reflects the received vector value in the input and reflects it in the generation of a new design or design elements of the specific item, a design generation device.

2. The design generation device according to claim 1, wherein the design generation unit further includes a design synthesis unit that generates a new design in real time using the design generation model learned in real time.

3. The design generation device according to claim 1, further including an evaluation unit that evaluates the frequency of exposure, sales volume, preference, and a preset weighting value of the newly generated design or new design elements of the specific item to at least one of a website, a blog, and an SNS.

4. The design generation device according to claim 3, wherein the evaluation result of the evaluation unit is reflected when the learning unit learns the design generation model.

5. The design generation unit according to claim 1 generates a new design or design element of the specific item based on a design or design element extracted from image data and text data that does not include the design or design element of the specific item.

Images