Design evaluation system, design evaluation method, and program

Abstract

This design evaluation system comprises: a feature quantification unit that, on the basis of a first trained model obtained by learning a first correspondence relationship between a material image indicating the design of a material and a feature amount obtained by quantifying an appearance feature of the design, quantifies the appearance feature of the design of a material indicated by a material image inputted as an evaluation target and generates feature amount information indicating the feature amount of the evaluation target; and a sensibility quantification unit that, on the basis of a second trained model obtained by learning a second correspondence relationship between the feature amount and a sensibility amount obtained by quantifying a sensibility word representing the appearance of the design, quantifies the sensibility word corresponding to the feature amount of the evaluation target indicated by the inputted feature amount information and generates sensibility amount information indicating the sensibility amount of the evaluation target.

Description

Design evaluation system, design evaluation method, and program

[0001] The present invention relates to a design evaluation system, a design evaluation method, and a program. This application claims priority to Japanese Patent Application No. 2024-082008 filed on May 20, 2024, and Japanese Patent Application No. 2025-042242 filed on March 17, 2025, the contents of which are incorporated herein by reference.

[0002] Conventionally, the process of creating designs for building materials and the like has been as follows: a business receives requests from a customer, creates a design, proposes the design to the customer, and receives evaluation from the customer.

[0003] In relation to this, various technologies capable of automatically generating designs have been proposed. For example, Patent Document 1 listed below discloses a technology in which image features corresponding to the human sense of an original image, such as its atmosphere or taste, are quantified, and a computer automatically generates other content that has a similar feel to the original image based on the image features.

[0004] Japanese Patent No. 7062890

[0005] When proposing a design, it is sometimes desirable to evaluate the design from an emotional perspective and convey its features through emotional expressions in order to clearly communicate its features.However, because design evaluation relies on personal sensibilities, it has been difficult to achieve an objective evaluation.

[0006] In view of the above-mentioned problems, an object of the present invention is to provide a design evaluation system, a design evaluation method, and a program that are capable of objectively evaluating a design.

[0007] In order to solve the above-mentioned problems, a design evaluation system according to one aspect of the present invention includes a feature quantification unit that quantifies appearance features of a material design indicated by a material image input as an evaluation target based on a first trained model that has learned a first correspondence relationship between a material image indicating a design of the material and a feature amount that quantifies appearance features of the design, and generates feature amount information that indicates the feature amount of the evaluation target; and an affective amount quantification unit that quantifies affective words corresponding to the feature amounts of the evaluation target indicated by the input feature amount information based on a second trained model that has learned a second correspondence relationship between the feature amount and affective amounts that quantify affective words that describe the appearance of the design, and generates affective amount information that indicates the affective amount of the evaluation target, The affective word is a verbalization of what a person who has viewed the material image feels about the appearance of the design, the affective quantity is a numerical representation of the degree to which the affective word applies to the appearance of the design, the first trained model is a model that has learned about the first correspondence using a first dataset in which the feature quantity, represented by a numerical value evaluated by a person, is tagged as first label information for the material image, and the second trained model is a model that has learned about the second correspondence using a second dataset in which the affective quantity, represented by a numerical value evaluated by a person, is tagged as second label information for the feature quantity information of the material image.

[0008] According to one aspect of the present invention, there is provided a computer-executable design evaluation method, comprising: a feature quantification step of quantifying visual features of a material design indicated by a material image input as an evaluation target, based on a first trained model that has learned a first correspondence relationship between a material image indicating a material design and a feature amount that quantifies visual features of the design, to generate feature amount information that indicates the feature amount of the evaluation target; and an affective quantification step of quantifying affective words corresponding to the feature amounts of the evaluation target indicated by the input feature amount information, based on a second trained model that has learned a second correspondence relationship between the feature amount and affective amounts that quantify affective words that describe the appearance of the design, to generate affective amount information that indicates the affective amount of the evaluation target, wherein the feature amount is a measure of the visual appearance of the design. The affective word is a verbalization of what a person who has viewed the material image feels about the appearance of the design, the affective quantity is a numerical representation of the degree to which the affective word applies to the appearance of the design, the first trained model is a model that has learned about the first correspondence using a first dataset in which the feature quantity, represented by a numerical value evaluated by a person, is tagged as first label information for the material image, and the second trained model is a model that has learned about the second correspondence using a second dataset in which the affective quantity, represented by a numerical value evaluated by a person, is tagged as second label information for the feature quantity information of the material image.

[0009] A program according to one aspect of the present invention causes a computer to function as: a feature quantification unit that quantifies appearance features of a material design indicated by a material image input as an evaluation target, based on a first trained model that has learned a first correspondence relationship between a material image indicating a design of the material and a feature amount that quantifies appearance features of the design, and generates feature amount information that indicates the feature amount of the evaluation target; and an affective amount quantification unit that quantifies affective words corresponding to the feature amounts of the evaluation target, indicated by the input feature amount information, based on a second trained model that has learned a second correspondence relationship between the feature amount and affective amounts that quantify affective words that describe the appearance of the design, and generates affective amount information that indicates the affective amount of the evaluation target. The affective words are formed by verbalizing the feelings of people who have viewed the material image about the appearance of the design, and the affective quantities are formed by numerically indicating the degree to which the affective words apply to the appearance of the design. The first trained model is a model that has learned about the first correspondence using a first dataset in which the feature quantities, represented by numerical values ​​evaluated by humans, are tagged as first label information for the material image, and the second trained model is a model that has learned about the second correspondence using a second dataset in which the affective quantities, represented by numerical values ​​evaluated by humans, are tagged as second label information for the feature quantity information of the material image.

[0010] According to the present invention, a design can be objectively evaluated.

[0011] FIG. 1 is a diagram showing an overview of a building material design evaluation service according to the present embodiment; FIG. 2 is a block diagram showing an example of the configuration of a design evaluation system according to the present embodiment; FIG. 3 is a diagram showing an overview of design learning according to the present embodiment; FIG. 4 is a block diagram showing an example of the functional configuration of a design learning device according to the present embodiment; FIG. 5 is a block diagram showing an example of the functional configuration of a design evaluation device according to the present embodiment; FIG. 6 is a flowchart showing an example of the flow of a design learning process according to the present embodiment; FIG. 7 is a flowchart showing an example of the flow of a design evaluation process according to the present embodiment.

[0012] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In this embodiment, a design evaluation system will be described that evaluates the design of a material based on an image showing the design of the material (hereinafter also referred to as a "material image"). The material is, for example, a material such as wood, stone, or fabric (cloth such as cloth or woven fabric). The design to be evaluated is, for example, the appearance of the design of the material. The appearance of the design of the material is, for example, a pattern or design on the surface of the material. In this case, the material image is an image showing the pattern or design on the surface of the material. As an example, if the material is wood, the material image is, for example, an image showing the appearance of wood with a wood grain pattern on its surface. The design evaluation system evaluates the appearance of the design of the material shown in the material image based on the material image input by the user, and outputs information indicating the design evaluation result (hereinafter also referred to as "evaluation information") to the user.

[0013] The following describes this embodiment, taking as an example an example a case where the design evaluation system is applied to a building material design evaluation service. Building materials are materials and components used for the interior and exterior of buildings. Examples of building materials include flooring, decorative paper, wallpaper, and interior decorative materials. Decorative sheets are used for the surface decoration of building materials. A variety of designs can be printed on the surface of the decorative sheets. The above-mentioned material images use images that show designs that can be applied to the design of building materials. Therefore, a user can apply the design of the material to the design of the building material by printing the material image on a decorative sheet and attaching the decorative sheet to the surface of the building material.

[0014] <1. Overview of building material design evaluation service> An overview of the building material design evaluation service according to this embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an overview of the building material design evaluation service according to this embodiment.

[0015] The building material design evaluation service SA shown in Fig. 1 is a service that evaluates the design of building materials handled by a user (e.g., a business operator) and provides evaluation information indicating the evaluation results to the user. The design evaluation involves feature evaluation and sensory evaluation.

[0016] In the feature evaluation, the visual characteristics of the material design shown in the material image are evaluated. The visual characteristics to be evaluated include, for example, layout, balance between straight grain and cross grain, distribution of luster, impression of knots (including leaf nodes), shading (modulation), finish, vessel color, impression of grain width, amount of grain width change, character, paint finish, pattern direction, hardness, coarseness, texture expression, and texture feel. The result of the feature evaluation is indicated by information (hereinafter also referred to as "feature amount") that quantifies the visual characteristics of the design. The feature amount indicates the degree to which the visual characteristics of the design are expressed in a numerical value in the material image. The feature evaluation is performed based on a correspondence relationship (first correspondence relationship) between the material image showing the appearance of the material design and the feature amount that quantifies the visual characteristics of the design. In this embodiment, the feature evaluation is performed using a feature quantification model (first trained model) that has learned (by machine learning) the correspondence relationship between the material image and the feature amount. The feature quantification model receives a material image to be evaluated as input, quantifies the visual features of the design of the material shown in the material image, and generates and outputs information indicating the features of the evaluation target (hereinafter also referred to as "feature information"). The feature quantification model, for example, divides the input material image into small pieces and classifies the features of each divided image using a deep learning model. In this way, the feature quantification model quantifies the visual features of the design of the input material image.

[0017] In the affective evaluation, affective words that describe the appearance of the design of the material shown in the material image are evaluated. The affective words are verbalized impressions of a person who views the material image regarding the appearance of the design of the material shown in the material image. Examples of affective words that are the subject of evaluation include gentle, elegant, tasteful, profound, stylish, gorgeous, refined, vibrant, timeless, refreshing, soothing, comfortable, calming, high-quality, and natural. The result of the affective evaluation is indicated by information (hereinafter also referred to as "affective amount") in which affective words for the appearance of the design are quantified. The affective amount indicates the degree to which the affective words apply to the appearance of the design as a number. The affective evaluation is performed based on a correspondence relationship (second correspondence relationship) between the feature amount of the appearance of the design shown in the material image and the affective amount of the appearance of the design shown in the material image. In this embodiment, the affective evaluation is performed by an affective quantification model (second trained model) that has learned (by machine learning) the correspondence between feature amounts and affective amounts. The affective quantification model receives as input a material image to be evaluated and feature amount information, quantifies affective words corresponding to the feature amounts of the evaluation target indicated by the feature amount information, and generates and outputs information indicating the affective amounts of the evaluation target (hereinafter also referred to as "affective amount information").

[0018] As shown in Fig. 1, a user of the building material design evaluation service SA first inputs a material image MG1 to be evaluated into the feature digitization model MD1 (step S11). The feature digitization model MD1 performs feature evaluation on the input material image MG1 and generates feature quantity information. The material image MG1 is, for example, an image showing a wood grain design as shown in Fig. 1.

[0019] Next, the user inputs the feature amount information generated by the feature digitization model MD1 through the feature evaluation into the affective digitization model MD2 (step S12). The affective digitization model MD2 performs affective evaluation based on the input feature amount information and generates affective amount information.

[0020] The affective quantification model MD2 outputs the generated affective quantity information to the user as evaluation information (step S13). The user uses the affective quantity information to propose a design to be applied to the building material to the customer (step S14). The affective quantities indicated by the affective quantity information are numerical values ​​of affective words objectively evaluated by the affective quantification model MD2. Therefore, by using affective quantity information indicating an objective evaluation, the user can make more persuasive proposals to the customer than when making proposals using the user's own subjective evaluation. Furthermore, by using affective quantity information, the user can make easier-to-understand proposals using affective words than when making proposals that simply explain the features of the design, which can also lead to an increase in the number of customers acquired.

[0021] 2. Configuration of the Design Evaluation System> The building material design evaluation service SA according to this embodiment has been described above. Next, the configuration of the design evaluation system according to this embodiment will be described with reference to Figs. 2 and 3. Fig. 2 is a block diagram showing an example of the configuration of the design evaluation system according to this embodiment. The design evaluation system 1 shown in Fig. 2 is a system for operating the building material design evaluation service SA outlined with reference to Fig. 1.

[0022] 2, the design evaluation system 1 includes an administrator terminal 10, a user terminal 20, a design learning device 30, and a design evaluation device 40. The network NW may be configured to transmit and receive information using, for example, a local area network (LAN), a wide area network (WAN), a telephone network (such as a mobile phone network or a landline telephone network), a regional Internet Protocol (IP) network, or the Internet.

[0023] (1) Administrator Terminal 10 The administrator terminal 10 is a terminal operated by an administrator (business operator) to manage the building material design evaluation service SA. The administrator terminal 10 is, for example, a smartphone, a tablet terminal, a PC (Personal Computer), etc. The administrator terminal 10 is connected to the design learning device 30 and the design evaluation device 40 via the network NW so as to be able to communicate with them.

[0024] In communication with the design learning device 30, the administrator terminal 10 transmits a plurality of material images, label information, etc. prepared for learning, and receives a feature quantification model MD1, a sensitivity quantification model MD2, etc. The administrator operates the administrator terminal 10 to transmit the material images and label information for learning to the design learning device 30.

[0025] The learning material images are, for example, image data (hereinafter also referred to as "scanned image data") obtained by scanning pattern manuscript data prepared for building material design, or image data (hereinafter also referred to as "design image data") showing a design created by a designer or the like for building material design.

[0026] The label information is tagged for learning purposes based on the perspective of applying the material design to the design of building materials. The label information includes label information (first label information) prepared for learning characteristics and label information (second label information) prepared for learning sensibility.

[0027] Label information prepared for feature learning (hereinafter also referred to as "feature label information") is information indicating visual features of the design shown in the training material image. Feature label information includes information tagged to the training material image by a human and information tagged to the training material image by image recognition processing. Information tagged to the training material image by a human is, for example, a feature indicated by a numerical value evaluated by a human. Information tagged to the training material image by image recognition processing is, for example, a feature obtained by quantifying features captured by image recognition of the training material image. The person who evaluates the features is, for example, a person (expert) with knowledge of the design of the material. The expert is, for example, a designer. In evaluating the features, the expert, such as a designer, may, for example, actually look at the material image, capture the visual features of the design shown in the material image, and respond numerically to the degree to which the features are expressed. Alternatively, a human, such as a designer, may check whether each design feature, which is prepared in advance as an evaluation item, is expressed in the design of the material image, and respond numerically to the degree to which the features are expressed.

[0028] Label information prepared for sensibility learning (hereinafter also referred to as "sensibility label information") is information that indicates the affective amount of the appearance of the design shown in the learning material image. The affective label information is information tagged by a human. The information tagged by a human is an affective amount indicated by a numerical value evaluated by a human. The person who evaluates the affective amount is, for example, a monitor. The monitor is an ordinary person who answers a questionnaire for evaluating the affective amount. The monitors are not particularly limited, and it is desirable to cover a wide range of people regardless of their background, age, gender, etc. In the questionnaire, for example, the monitor is presented with a material image to be evaluated and multiple affective words, and is asked to respond to the degree to which each affective word applies to the appearance of the design shown in the material image. The response to the degree to which the affective word applies may be any method, such as by selecting the most appropriate number from multiple options or by freely entering a number.

[0029] The feature digitization model MD1 and the sensibility digitization model MD2 are models generated by the design learning device 30 based on learning material images and label information transmitted from the administrator terminal 10.

[0030] In communication with the design evaluation device 40, the administrator terminal 10 transmits a feature quantification model MD1 and a sensibility quantification model MD2. The feature quantification model MD1 and the sensibility quantification model MD2 are models that the administrator terminal 10 receives from the design learning device 30.

[0031] On the administrator terminal 10, various UIs (User Interfaces) are displayed by an application (hereinafter also referred to as the "administrator app") that allows the administrator to manage the building material design evaluation service SA. The administrator can manage the building material design evaluation service SA by operating the UI displayed on the administrator terminal 10 by the administrator app. Note that the functions of the administrator app may be provided by installing the administrator app on the administrator terminal 10 (i.e., a native app), or may be provided by a web system (i.e., a web app). In the case of a web app, the administrator app is managed by a server, and its functions are provided via a web browser.

[0032] (2) User Terminal 20 The user terminal 20 is a terminal operated by a user to use the building material design evaluation service SA. The user terminal 20 is, for example, a smartphone, a tablet terminal, a PC, etc. The user terminal 20 is connected to the design evaluation device 40 via the network NW so as to be able to communicate with the design evaluation device 40.

[0033] In communication with the design evaluation device 40, the user terminal 20 transmits material images and receives evaluation information. The user operates the user terminal 20 to input the material image MG1 to be evaluated and transmit it to the design evaluation device 40. The evaluation information is affective amount information generated by the design evaluation device 40 based on the material image transmitted from the user terminal 20, or information for display generated by processing the affective amount information.

[0034] Various UIs are displayed on the user terminal 20 by an application (hereinafter also referred to as a "user app") that allows the user to use the building material design evaluation service SA. The user can use the building material design evaluation service SA by operating the UI displayed on the user terminal 20 by the user app. The functions of the user app may be provided by installing the user app on the user terminal 20 (i.e., a native app), or may be provided by a web system (i.e., a web app). In the case of a web app, the user app is managed by a server, and its functions are provided via a web browser.

[0035] (3) Design Learning Device 30 The design learning device 30 is a device that generates the feature quantification model MD1 and the sensibility quantification model MD2. The design learning device 30 is configured, for example, by one or more PCs or server devices (e.g., cloud servers). The design learning device 30 is connected to the administrator terminal 10 via the network NW so as to be able to communicate with the administrator terminal 10.

[0036] In communication with the administrator terminal 10, the design learning device 30 receives a plurality of material images, label information, etc. prepared for learning, and transmits a feature quantification model MD1, a sensibility quantification model MD2, etc. The design learning device 30 creates a learning dataset based on the learning material images and label information received from the administrator terminal 10, and generates the feature quantification model MD1 and the sensibility quantification model MD2 using the learning dataset.

[0037] Here, an overview of design learning according to this embodiment will be described with reference to Fig. 3. Fig. 3 is a diagram showing an overview of design learning according to this embodiment. In design learning, learning is performed on the correspondence relationship between material images and feature amounts (first correspondence relationship) and the correspondence relationship between feature amounts of material images and affective amounts (second correspondence relationship).

[0038] As shown in Fig. 3, in the building material design evaluation service SA, an administrator prepares learning material images MG2 in advance. The learning material images MG2 are, for example, scanned image data or design image data. In the example shown in Fig. 3, a material image MG2-1 showing a wood grain design and a material image MG2-2 showing a different wood grain design from the material image MG2-1 are shown.

[0039] First, the administrator operates the administrator terminal 10 to tag feature quantities for the learning material images MG2 (step S21). To do this, the administrator inputs feature label information into the administrator terminal 10 and transmits it to the design learning device 30. The input feature label information is, for example, feature quantities indicated by numerical values ​​evaluated by a designer. Upon receiving the feature label information, the design learning device 30 creates a dataset (hereinafter also referred to as a "feature learning dataset") in which the corresponding learning material images MG2 are tagged with the feature label information. For example, as shown in FIG. 3, the design learning device 30 tags the appearance feature quantities of the design of the material shown in material image MG2-1 as label information for material image MG2-1. Furthermore, the design learning device 30 tags the appearance feature quantities of the design of the material shown in material image MG2-2 as label information for material image MG2-2. As a result, the design learning device 30 creates a feature learning dataset DS1 in which each material image MG2 is tagged with feature label information.

[0040] The design learning device 30 uses the created feature learning dataset DS1 to learn about the features of the material images (the correspondence between the material images and the feature amounts) and generates a feature digitization model MD1 (step S22).

[0041] Next, the administrator operates the administrator terminal 10 to tag the learning material image MG2 with affective quantities (step S23). To do this, the administrator inputs affective label information into the administrator terminal 10 and transmits it to the design learning device 30. The affective label information input here is, for example, affective quantities obtained through a survey of monitors. When the design learning device 30 receives the affective label information, it creates a dataset (hereinafter also referred to as a "affective learning dataset") in which the corresponding feature learning dataset DS1 is further tagged with affective label information. For example, as shown in FIG. 3, the design learning device 30 tags the affective quantities of the design of the material represented by material image MG2-1 as label information for the feature learning dataset DS1 of material image MG2-1. Furthermore, it tags the affective quantities of the design of the material represented by material image MG2-2 as label information for the feature learning dataset DS1 of material image MG2-2. As a result, the design learning device 30 creates a sensibility learning dataset DS2 in which each feature learning dataset DS1 is tagged with sensibility label information.

[0042] The design learning device 30 uses the created sensory learning dataset DS2 to learn about the sensory perception of the material image (the correspondence between the feature amounts of the material image and the sensory perception amounts) and generates a sensory perception quantification model MD2 (step S24).

[0043] (4) Design Evaluation Device 40 The design evaluation device 40 is a device that evaluates the design of the material image MG1 to be evaluated. The design evaluation device 40 is configured, for example, by one or more PCs or server devices (e.g., cloud servers). The design evaluation device 40 is communicably connected to the administrator terminal 10 and the user terminal 20 via the network NW.

[0044] In communication with the administrator terminal 10, the design evaluation device 40 receives the feature quantification model MD1 and the affective quantification model MD2. The design evaluation device 40 uses the feature quantification model MD1 and the affective quantification model MD2 received from the administrator terminal 10 to evaluate the design of the material image MG1 to be evaluated.

[0045] In communication with the user terminal 20, the design evaluation device 40 receives material images and transmits evaluation information. The design evaluation device 40 inputs the material images received from the user terminal 20 into a feature quantification model MD1, and inputs the feature amount information output from the feature quantification model MD1 into a sensitivity quantification model MD2, thereby generating evaluation information.

[0046] <3. Functional Configuration of Design Learning Device 30> The configuration of the design evaluation system 1 according to this embodiment has been described above. Next, the functional configuration of the design learning device 30 according to this embodiment will be described with reference to Fig. 4. Fig. 4 is a block diagram showing an example of the functional configuration of the design learning device 30 according to this embodiment. As shown in Fig. 4, the design learning device 30 includes a communication unit 310, a storage unit 320, and a control unit 330.

[0047] (1) Communication Unit 310 The communication unit 310 has a function of transmitting and receiving various information. The communication unit 310 is communicably connected to the administrator terminal 10 via the network NW, and transmits and receives various information. In communication with the administrator terminal 10, the communication unit 310 receives a plurality of material images prepared for learning, label information, etc., and transmits a feature quantification model MD1, a sensibility quantification model MD2, etc.

[0048] (2) Storage Unit 320 The storage unit 320 has a function of storing various information. The storage unit 320 is configured by a storage medium provided as hardware in the design learning device 30, such as a hard disk drive (HDD), a solid state drive (SSD), a flash memory, an electrically erasable programmable read-only memory (EEPROM), a random access read / write memory (RAM), a read-only memory (ROM), or any combination of these storage media. As shown in FIG. 4 , the storage unit 320 includes a material image storage unit 321, a learning dataset storage unit 322, and a model storage unit 323.

[0049] (2-1) Material Image Storage Unit 321 The material image storage unit 321 has a function of storing material images. The material image storage unit 321 stores, for example, learning material images MG2 that are prepared in advance by the administrator and received by the communication unit 310 from the administrator terminal 10. The material image storage unit 321 stores a plurality of material images with different combinations of material and design appearance characteristics.

[0050] (2-2) Training Dataset Storage Unit 322 The training dataset storage unit 322 has a function of storing a training dataset. The training dataset storage unit 322 stores, for example, a training dataset created by the data processing unit 332 described below. The training dataset storage unit 322 stores, as a training dataset, a plurality of data tagged with label information for each material image stored in the material image storage unit 321.

[0051] (2-3) Model Storage Unit 323 The model storage unit 323 has a function of storing various models. As shown in Fig. 4, the model storage unit 323 stores a feature digitization model MD1 and a sensibility digitization model MD2.

[0052] (3) Control Unit 330 The control unit 330 has a function of controlling the overall operation of the design learning device 30. The control unit 330 is realized, for example, by causing a central processing unit (CPU) or a graphics processing unit (GPU) provided as hardware in the design learning device 30 to execute a program. As shown in FIG. 4 , the control unit 330 includes a data acquisition unit 331, a data processing unit 332, a learning unit 333, and an output processing unit 334.

[0053] (3-1) Data Acquisition Unit 331 The data acquisition unit 331 has a function of acquiring various data. For example, the data acquisition unit 331 acquires the learning material image MG2 and label information that the communication unit 310 receives from the administrator terminal 10. The label information acquired by the data acquisition unit 331 includes feature label information or affective label information.

[0054] (3-2) Data Processing Unit 332 The data processing unit 332 has the function of performing various types of data processing. For example, the data processing unit 332 creates a learning dataset based on the learning material images MG2 and label information acquired by the data acquisition unit 331. When the data acquisition unit 331 acquires the learning material images MG2 and feature label information, the data processing unit 332 creates a feature learning dataset DS1 in which the learning material images MG2 are tagged with the feature label information. When the data acquisition unit 331 acquires affective label information, the data processing unit 332 creates a affective learning dataset DS2 in which the feature learning dataset DS1 is further tagged with affective label information.

[0055] (3-3) Learning Unit 333 The learning unit 333 has a function of generating a trained model through machine learning. For example, the learning unit 333 uses a training dataset created by the data processing unit 332 to learn various correspondence relationships.

[0056] When the training dataset is the feature training dataset DS1, the training unit 333 uses the feature training dataset DS1 to train on the visual features of the design indicated by the material images. In training on the features, the training unit 333 trains on the correspondence between the material images indicated by the feature training dataset DS1 and the feature quantities. Through this training, when a material image is input, the training unit 333 generates a feature quantification model MD1 that can digitize the visual features of the design indicated by the material image and generate and output feature quantity information indicating the feature quantities of the evaluation target. In other words, the feature quantification model MD1 is a model that has trained on the correspondence between material images and feature quantities using the feature training dataset DS1 (first dataset) in which feature quantities indicated by numerical values ​​evaluated by humans are tagged as feature label information (first label information) for the material images.

[0057] Note that the feature label information according to this embodiment is information tagged for learning purposes based on the perspective of applying the design of a material to the design of a building material. Furthermore, the feature quantification model MD1 used to quantify the features is a trained model trained using a feature learning dataset DS1 that includes the feature label information. Therefore, the feature quantity information generated by the feature quantification model MD1 is information generated based on the perspective of applying the design of a material to the design of a building material.

[0058] In addition, the feature learning dataset DS1 used by the learning unit 333 to learn features includes a dataset tagged with feature label information by a person with knowledge of material design (e.g., a designer), and a dataset tagged with feature label information by image recognition of material images.

[0059] When the training dataset is the affective learning dataset DS2, the learning unit 333 uses the affective learning dataset DS2 to learn about the affect of the appearance of the design represented by the material image. In the affective learning, the learning unit 333 learns the correspondence between the feature quantities of the material image represented by the affective learning dataset DS2 and the affective quantities. Through this learning, the learning unit 333 generates an affective quantification model MD2 that, when a material image and feature quantity information are input, can quantify affective words for the appearance of the design represented by the material image and generate and output affective quantity information indicating the affective quantities of the evaluation target. In other words, the affective quantification model MD2 is a model that has learned about the correspondence between the feature quantities of the material image and the affective quantities using the affective learning dataset DS2 (second dataset) in which affective quantities indicated by numerical values ​​evaluated by humans are tagged with affective label information (second label information) for the feature quantity information of the material image.

[0060] The affective label information according to this embodiment is tagged information for learning purposes based on the perspective of applying the design of a material to the design of a building material. The affective quantification model MD2 used to quantify the affect is a trained model trained using the affective learning dataset DS2 that includes the affective label information. Therefore, the affective quantity information generated by the affective quantification model MD2 is information generated based on the perspective of applying the design of a material to the design of a building material.

[0061] (3-4) Output Processing Unit 334 The output processing unit 334 has a function of performing processing related to the output of various information. For example, the output processing unit 334 transmits the feature digitization model MD1 and the sensibility digitization model MD2 generated by the learning unit 333 from the communication unit 310 to the administrator terminal 10.

[0062] 4. Functional Configuration of the Design Evaluation Device 40 The functional configuration of the design learning device 30 according to this embodiment has been described above. Next, the functional configuration of the design evaluation device 40 according to this embodiment will be described with reference to FIG. 5. FIG. 5 is a block diagram showing an example of the functional configuration of the design evaluation device 40 according to this embodiment. As shown in FIG. 5, the design evaluation device 40 includes a communication unit 410, a storage unit 420, and a control unit 430.

[0063] (1) Communication Unit 410 The communication unit 410 has a function of transmitting and receiving various information. The communication unit 410 is communicably connected to the administrator terminal 10 and the user terminal 20 via the network NW, and transmits and receives various information to and from each terminal. In communication with the administrator terminal 10, the communication unit 410 receives the feature quantification model MD1 and the affective quantification model MD2. In communication with the user terminal 20, the communication unit 410 receives material images and transmits evaluation information.

[0064] (2) Storage Unit 420 The storage unit 420 has a function of storing various types of information. The storage unit 420 is configured by a storage medium provided as hardware in the design evaluation device 40, such as an HDD, SSD, flash memory, EEPROM, RAM, ROM, or any combination of these storage media. As shown in FIG. 5 , the storage unit 420 includes a model storage unit 421 and a material image storage unit 422.

[0065] (2-1) Model Storage Unit 421 The model storage unit 421 has a function of storing various models. As shown in Fig. 5, the model storage unit 421 stores a feature quantification model MD1 and a sensibility quantification model MD2. The feature quantification model MD1 and the sensibility quantification model MD2 are models generated by the design learning device 30 and received by the communication unit 410 from the administrator terminal 10.

[0066] (2-2) Material Image Storage Unit 422 The material image storage unit 422 has a function of storing material images. The material image storage unit 422 stores, for example, material images MG1 to be evaluated that are prepared by the user and received by the communication unit 410 from the user terminal 20.

[0067] (3) Control Unit 430 The control unit 430 has a function of controlling the overall operation of the design evaluation device 40. The control unit 430 is realized, for example, by causing a CPU or GPU provided as hardware in the design evaluation device 40 to execute a program. As shown in FIG. 5 , the control unit 430 includes an image acquisition unit 431, a design evaluation unit 432, and an output processing unit 433.

[0068] (3-1) Image Acquisition Unit 431 The image acquisition unit 431 has a function of acquiring a material image MG1 to be evaluated. The image acquisition unit 431 acquires the material image MG1 received by the communication unit 410 from the user terminal 20 as the evaluation target.

[0069] (3-2) Design Evaluation Unit 432 The design evaluation unit 432 has a function of evaluating the design represented by the material image MG1 to be evaluated. As shown in FIG. 5, the design evaluation unit 432 includes a feature digitization unit 4321 and an affective digitization unit 4322.

[0070] (3-2-1) Feature digitization unit 4321 The feature digitization unit 4321 has a function of evaluating the design shown in the material image MG1 to be evaluated from the perspective of its features. The feature digitization unit 4321 evaluates the visual features of the design of the material shown in the material image MG1 to be evaluated acquired by the image acquisition unit 431, based on the correspondence between the training material image MG2 and the feature amounts obtained by quantifying the visual features of the design shown in the material image MG2. In evaluating the visual features, the feature digitization unit 4321 digitizes the visual features of the design of the material shown in the material image MG1 to be evaluated, and generates feature amount information indicating the feature amounts of the evaluation target.

[0071] The feature digitization unit 4321 generates feature amount information by inputting the material image MG1 to be evaluated into the feature digitization model MD1 stored in the model storage unit 421. By using the feature digitization model MD1, the feature digitization unit 4321 can objectively evaluate the features of the evaluation target.

[0072] (3-2-2) Sensitivity Quantification Unit 4322 The sensitivity quantification unit 4322 has a function of evaluating the design represented by the material image MG1 to be evaluated from a sensitivity perspective. The sensitivity quantification unit 4322 evaluates the sensitivity to the appearance of the design of the material represented by the material image MG1 to be evaluated acquired by the image acquisition unit 431, based on the correspondence between the feature amounts of the training material image MG2 and the sensitivity amounts obtained by quantifying sensitivity words that express human sensitivity to the appearance of the design represented by the material image MG2. In evaluating sensitivity, the sensitivity quantification unit 4322 quantifies sensitivity words that correspond to the feature amounts of the appearance of the design of the material represented by the material image MG1 to be evaluated, and generates sensitivity amount information that indicates the sensitivity amount to the appearance of the evaluation target.

[0073] The sensibility digitization unit 4322 generates sensibility amount information by inputting feature amount information of the material image MG1 to be evaluated into the sensibility digitization model MD2 stored in the model storage unit 421. The sensibility digitization unit 4322 can objectively evaluate the sensibility of the evaluation target by using the sensibility digitization model MD2.

[0074] (3-3) Output Processing Unit 433 The output processing unit 433 has a function of performing processing related to the output of various information. For example, the output processing unit 433 transmits the affective amount information generated by the affective digitization unit 4322 as evaluation information from the communication unit 410 to the user terminal 20 and displays it on the user terminal 20.

[0075] 5. Processing Flow The functional configuration of the design evaluation device 40 according to this embodiment has been described above. Next, the processing flow according to this embodiment will be described with reference to FIGS.

[0076] (1) Design Learning Processing The flow of the design learning processing according to this embodiment will be described with reference to Fig. 6. Fig. 6 is a flowchart showing an example of the flow of the design learning processing according to this embodiment.

[0077] 6, first, the data acquisition unit 331 of the design learning device 30 acquires the learning material image MG2 (step S101). Specifically, the data acquisition unit 331 acquires the learning material image MG2 that is sent by the administrator from the administrator terminal 10 to the design learning device 30 and received by the communication unit 310.

[0078] The data acquisition unit 331 also acquires feature label information corresponding to the acquired learning material image MG2 (step S102). Specifically, the data acquisition unit 331 acquires the feature label information that is sent by the administrator from the administrator terminal 10 to the design learning device 30 and received by the communication unit 310.

[0079] Next, the data processing unit 332 of the design learning device 30 creates a feature learning dataset DS1 (step S103) based on the learning material images MG2 and the feature label information acquired by the data acquisition unit 331. Specifically, the data processing unit 332 creates the feature learning dataset DS1 by tagging each of the learning material images MG2 with feature label information indicating the corresponding feature amount.

[0080] Next, the learning unit 333 of the design learning device 30 uses the feature learning dataset DS1 created by the data processing unit 332 to learn the visual features of the design represented by the material image (step S104). Specifically, the learning unit 333 uses the feature learning dataset DS1 to learn the correspondence between the learning material image MG2 and the visual features of the design. Through this learning, the learning unit 333 generates a feature quantification model MD1 (step S105). The output processing unit 334 links the feature quantification model MD1 created by the learning unit 333 to the design evaluation device 40 via the administrator terminal 10.

[0081] Next, the data acquisition unit 331 acquires affective label information corresponding to the acquired learning material image MG2 (step S106). Specifically, the data acquisition unit 331 acquires affective label information that is sent by the administrator from the administrator terminal 10 to the design learning device 30 and received by the communication unit 310.

[0082] Next, the data processing unit 332 of the design learning device 30 creates a sensibility learning dataset DS2 based on the feature learning dataset DS1 created by the data processing unit 332 and the sensibility label information acquired by the data acquisition unit 331 (step S107). Specifically, the data processing unit 332 creates the sensibility learning dataset DS2 by tagging each of the feature learning datasets DS1 with sensibility label information indicating the corresponding sensibility amount.

[0083] Next, the learning unit 333 of the design learning device 30 uses the sensitivity learning dataset DS2 created by the data processing unit 332 to learn the sensitivity to the appearance of the design represented by the material image (step S108). Specifically, the learning unit 333 uses the sensitivity learning dataset DS2 to learn the correspondence between the visual feature quantities and the sensitivity quantities of the design of the training material image MG2. Through this learning, the learning unit 333 generates a sensitivity quantification model MD2 (step S109). The output processing unit 334 links the sensitivity quantification model MD2 created by the learning unit 333 to the design evaluation device 40 via the administrator terminal 10.

[0084] (2) Design Evaluation Processing The flow of the design evaluation processing according to this embodiment will be described with reference to Fig. 7. Fig. 7 is a flowchart showing an example of the flow of the design evaluation processing according to this embodiment.

[0085] 7, first, the image acquisition unit 431 of the design evaluation device 40 acquires a material image MG1 to be evaluated (step S201). Specifically, the image acquisition unit 431 acquires the material image MG1 that is input by the user to the user terminal 20, transmitted from the user terminal 20 to the design evaluation device 40, and received by the communication unit 410.

[0086] Next, the feature digitization unit 4321 of the design evaluation device 40 evaluates the visual features of the design indicated by the material image to be evaluated (step S202). Specifically, the feature digitization unit 4321 inputs the material image acquired by the image acquisition unit 431 into the feature digitization model MD1 stored in the model storage unit 421, thereby generating feature amount information.

[0087] Next, the affective digitization unit 4322 of the design evaluation device 40 evaluates the affective sensation for the appearance of the design represented by the material image to be evaluated (step S203). Specifically, the affective digitization unit 4322 generates affective amount information by inputting the feature amount information generated by the feature digitization unit 4321 for the material image acquired by the image acquisition unit 431 into the affective digitization model MD2 stored in the model storage unit 421.

[0088] Next, the output processing unit 433 of the design evaluation device 40 outputs the evaluation information (step S204). Specifically, the output processing unit 433 transmits the affective amount information generated by the affective digitization unit 4322 as evaluation information from the communication unit 410 to the user terminal 20, and displays it on the user terminal 20.

[0089] The processing flow according to this embodiment has been described above. As described above, the design evaluation system 1 according to this embodiment includes: a feature digitization unit 4321 that digitizes the appearance features of the design of a material indicated by a material image input as an evaluation target, based on a feature digitization model (first trained model) that has learned a first correspondence relationship between a material image indicating the design of the material and a feature amount obtained by quantifying the appearance features of the design, and generates feature amount information that indicates the feature amount of the evaluation target; and an affective digitization unit 4322 that digitizes affective words corresponding to the feature amounts of the evaluation target indicated by the input feature amount information, based on an affective digitization model (second trained model) that has learned a second correspondence relationship between the feature amount and an affective amount obtained by quantifying affective words that describe the appearance of the design, and generates affective amount information that indicates the affective amount of the evaluation target. The affective words are formed by verbalizing the feelings of people who have viewed the material images about the appearance of the design, and the affective quantities are formed by numerically indicating the degree to which the affective words apply to the appearance of the design. The feature quantification model is a model that has learned about a first correspondence using a feature learning dataset DS1 (first dataset) in which feature quantities indicated by numerical values ​​evaluated by people are tagged as feature label information (first label information) for material images, and the affective quantification model is a model that has learned about a second correspondence using a affective learning dataset DS2 (second dataset) in which affective quantities indicated by numerical values ​​evaluated by people are tagged as affective label information (second label information) for feature quantity information of material images.

[0090] With this configuration, the design evaluation system 1 according to this embodiment can evaluate designs without relying on personal sensibilities, making it possible to objectively evaluate designs.

[0091] 6. Modifications The above describes the embodiments. Next, modifications of the above-described embodiments will be described. Note that each modification described below may be applied to the embodiment alone or in combination with the embodiment. Furthermore, each modification may be applied in place of the configuration described in the embodiment, or may be applied in addition to the configuration described in the embodiment.

[0092] In the above-described embodiment, an example has been described in which the design learning function and the design evaluation function are respectively implemented by different devices (the design learning device 30 and the design evaluation device 40), but the present invention is not limited to such an example. For example, the design learning function and the design evaluation function may be implemented by a single device having both functions.

[0093] In the above-described embodiment, an example has been described in which the learning material image MG2 and label information are transmitted from the administrator terminal 10 to the design learning device 30, and the learning dataset DS is created in the design learning device 30. However, the present invention is not limited to this example. For example, the learning dataset DS may be created in advance by the administrator in the administrator terminal 10, and transmitted from the administrator terminal 10 to the design learning device 30.

[0094] In the above-described embodiment, an example has been described in which the feature quantification model MD1 and the sensitivities quantification model MD2 generated by the design learning device 30 are registered in the design evaluation device 40 via the administrator terminal 10, but the present invention is not limited to such an example. For example, the design learning device 30 and the design evaluation device 40 may be able to communicate with each other via the network NW, and the feature quantification model MD1 and the sensitivities quantification model MD2 may be transmitted directly from the design learning device 30 to the design evaluation device 40.

[0095] In the above-described embodiment, an example in which affective amount information is output as evaluation information has been described, but the present invention is not limited to such an example. For example, only feature amount information may be output as evaluation information, or information in which feature amount information and affective amount information are combined may be output as evaluation information.

[0096] The above describes a modified embodiment of the present invention. Note that some or all of the design evaluation system 1, administrator terminal 10, user terminal 20, design learning device 30, and design evaluation device 40 in the above-described embodiment may be implemented by a computer. In this case, a program for implementing these functions may be recorded on a computer-readable recording medium, and the program may be loaded into a computer system and executed. Note that the term "computer system" here includes hardware such as an OS and peripheral devices. Furthermore, the term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into a computer system. Furthermore, the term "computer-readable recording medium" may also include media that dynamically store programs for a short period of time, such as communication lines used when transmitting programs via networks such as the Internet or telephone lines, and media that store programs for a fixed period of time, such as volatile memory within the computer systems that serve as servers or clients in such cases. Furthermore, the program may be a program for realizing some of the above-described functions, or may be a program that can realize the above-described functions in combination with a program already recorded in a computer system, or may be a program that is realized using a programmable logic device such as an FPGA (Field Programmable Gate Array).

[0097] The embodiments of the present invention have been described in detail above with reference to the drawings, but the specific configuration is not limited to that described above, and various design changes can be made within the scope of the gist of the present invention.

[0098] 1...design evaluation system, 10...administrator terminal, 20...user terminal, 30...design learning device, 40...design evaluation device, 310...communication unit, 320...storage unit, 321...material image storage unit, 322...learning dataset storage unit, 323...model storage unit, 330...control unit, 331...data acquisition unit, 332...data processing unit, 333...learning unit, 334...output processing unit, 410...communication unit, 420...storage unit, 421...model storage unit, 422...material image storage unit, 430...control unit, 431...image acquisition unit, 432...design evaluation unit, 4321...feature quantification unit, 4322...sensibility quantification unit, 433...output processing unit, MD1...feature quantification model, MD2...sensibility quantification model, NW...network, SA...building material design evaluation service

Claims

1. A feature quantification unit that quantifies the visual characteristics of a material design shown in a material image input as an evaluation target based on a first trained model that has learned a first correspondence relationship between a material image showing the design of the material and a feature amount that quantifies the visual characteristics of the design, and generates feature amount information that indicates the feature amount of the evaluation target; and an affective quantification unit that quantifies the affective words corresponding to the feature amounts of the evaluation target shown in the input feature amount information based on a second trained model that has learned a second correspondence relationship between the feature amount and an affective amount that quantifies an affective word that expresses the appearance of the design, and generates affective amount information that indicates the affective amount of the evaluation target, wherein the feature amount indicates, in a numerical value, the degree to which the visual characteristics of the design are expressed in the material image, and the affective word is a verbalization of what a person who views the material image feels about the appearance of the design, and the affective amount indicates, in a numerical value, the degree to which the affective word applies to the appearance of the design, A design evaluation system, wherein the first trained model is a model that has learned about the first correspondence using a first dataset in which the feature quantities indicated by numerical values ​​evaluated by humans are tagged as first label information for the material images, and the second trained model is a model that has learned about the second correspondence using a second dataset in which the affective quantities indicated by numerical values ​​evaluated by humans are tagged as second label information for the feature quantity information of the material images.

2. The design evaluation system according to claim 1, wherein the first dataset includes a dataset in which the material images are tagged with the first label information by image recognition.

3. The design evaluation system according to claim 1, wherein the feature information and the affective information are information generated based on a perspective when the design of the material is applied to the design of a building material.

4. A feature quantification process for quantifying the visual characteristics of a material design shown in a material image input as an evaluation target based on a first trained model that has learned a first correspondence relationship between a material image showing the design of the material and a feature amount that quantifies the visual characteristics of the design, and generating feature amount information that indicates the feature amount of the evaluation target; and an affective quantification process for quantifying the affective words corresponding to the feature amount of the evaluation target indicated by the input feature amount information based on a second trained model that has learned a second correspondence relationship between the feature amount and an affective amount that quantifies an affective word that expresses the appearance of the design, and generating affective amount information that indicates the affective amount of the evaluation target, wherein the feature amount indicates, in a numerical value, the degree to which the visual characteristics of the design are expressed in the material image, the affective word is a verbalization of what a person who has viewed the material image felt about the appearance of the design, and the affective amount indicates, in a numerical value, the degree to which the affective word applies to the appearance of the design, A computer-implemented design evaluation method, wherein the first trained model is a model that has learned about the first correspondence using a first dataset in which the feature quantities indicated by numerical values ​​evaluated by humans are tagged as first label information for the material images, and the second trained model is a model that has learned about the second correspondence using a second dataset in which the affective quantities indicated by numerical values ​​evaluated by humans are tagged as second label information for the feature quantity information of the material images.

5. A computer is caused to function as: a feature quantification unit that quantifies the appearance characteristics of a material design shown in a material image input as an evaluation target, based on a first trained model that has learned a first correspondence relationship between a material image showing the design of the material and a feature amount that quantifies the appearance characteristics of the design, and generates feature amount information that indicates the feature amount of the evaluation target; and an affective quantification unit that quantifies the affective words corresponding to the feature amounts of the evaluation target, based on a second trained model that has learned a second correspondence relationship between the feature amount and an affective amount that quantifies an affective word that expresses the appearance of the design, and generates affective amount information that indicates the affective amount of the evaluation target, wherein the feature amount indicates, in a numerical value, the degree to which the appearance characteristics of the design are expressed in the material image, and the affective word is a verbalization of what a person who has viewed the material image felt about the appearance of the design, and the affective amount indicates, in a numerical value, the degree to which the affective word applies to the appearance of the design, The first trained model is a model that has learned about the first correspondence using a first dataset in which the feature quantities indicated by numerical values ​​evaluated by humans are tagged as first label information for the material images, and the second trained model is a model that has learned about the second correspondence using a second dataset in which the affective quantities indicated by numerical values ​​evaluated by humans are tagged as second label information for the feature quantity information of the material images.

Images