Electronic device, method, and non-transitory computer-readable storage medium for searching for three-dimensional object by using embedding vector

Abstract

This electronic device may comprise a memory storing instructions, and at least one processor. The instructions, when executed by the at least one processor, may cause the electronic device to: identify an event for determining, from among a plurality of three-dimensional (3D) objects, a 3D object corresponding to specified text; obtain similarities between a first embedding vector corresponding to the specified text and second embedding vectors obtained on the basis of sets of images respectively corresponding to the plurality of 3D objects; and on the basis of obtaining, from among the similarities, a similarity exceeding a threshold similarity, determine a 3D object corresponding to the similarity, as a 3D object corresponding to the specified text.

Description

Electronic device, method, and non-transient computer-readable storage medium for retrieving 3D objects using embedding vectors

[0001] The present disclosure relates to an electronic device, a method, and a non-transient computer-readable storage medium for retrieving 3D objects using embedding vectors.

[0002] An electronic device may be used to search for specified data within a database. For example, the electronic device may scan the database to identify whether the specified data is included within the database. When scanning the database, the electronic device may output a data cell containing the specified data. For example, the electronic device may transmit the payload to an external electronic device via a communication circuit. The external electronic device may output the data cell based on receiving the data cell.

[0003] The information described above may be provided as related art for the purpose of aiding understanding of the present disclosure.

[0004] No claim or determination is made as to whether any of the foregoing can be applied as prior art related to the present disclosure.

[0005] An electronic device is described. The electronic device may include a memory comprising one or more storage media for storing instructions. The electronic device may include at least one processor comprising a processing circuit. The instructions may cause the electronic device to identify an event for determining a three-dimensional object corresponding to a specified text among a plurality of three-dimensional objects when executed individually or collectively by the at least one processor. The instructions may cause the electronic device to obtain similarities between a first embedding vector corresponding to the specified text and second embedding vectors obtained based on sets of images corresponding to each of the plurality of three-dimensional objects when executed individually or collectively by the at least one processor. The instructions may cause the electronic device to determine the three-dimensional object corresponding to the similarity as the three-dimensional object corresponding to the specified text, based on obtaining a similarity among the similarities that exceeds a threshold similarity.

[0006] A method is provided. The method may be executed within an electronic device. The method may include an operation of identifying an event for determining a 3D (three-dimensional) object corresponding to a specified text among a plurality of 3D objects. The method may include an operation of obtaining similarities between a first embedding vector corresponding to the specified text and second embedding vectors obtained based on sets of images corresponding to each of the plurality of 3D objects. The method may include an operation of determining a 3D object corresponding to a similarity as the 3D object corresponding to the specified text, based on obtaining a similarity among the similarities that exceeds a threshold similarity.

[0007] A non-transient computer-readable storage medium is provided. The non-transient computer-readable storage medium may store one or more programs. The one or more programs may include instructions that cause the electronic device to identify an event for determining a three-dimensional object corresponding to a specified text among a plurality of three-dimensional objects when executed by the electronic device. The one or more programs may include instructions that cause the electronic device to obtain similarities between a first embedding vector corresponding to the specified text and second embedding vectors obtained based on sets of images corresponding to each of the plurality of three-dimensional objects when executed by the electronic device. The one or more programs may include instructions that cause the electronic device to determine the three-dimensional object corresponding to the similarity as the three-dimensional object corresponding to the specified text when executed by the electronic device, based on obtaining a similarity among the similarities that exceeds a threshold similarity.

[0008] Figure 1 illustrates an example of an electronic device that transmits data about a 3D (three-dimensional) object.

[0009] Figure 2 is a simplified block diagram of an exemplary electronic device.

[0010] Figure 3 is a flowchart illustrating the operation of an electronic device that determines a 3D object using similarities.

[0011] FIG. 4 illustrates an exemplary operation of an electronic device that obtains similarities using embedding vectors.

[0012] FIG. 5 illustrates an exemplary operation of an electronic device for acquiring embedding vectors based on images of 3D objects.

[0013] FIG. 6 illustrates an exemplary operation of an electronic device for acquiring an embedding vector based on text obtained through an image of a 3D object.

[0014] FIG. 7 illustrates an exemplary operation of an electronic device that determines a 3D object using the Reciprocal Rank Fusion (RRF) technique.

[0015] FIG. 8 illustrates examples of operations performed within an electronic device and an external electronic device.

[0016] Figure 1 illustrates an example of an electronic device that transmits data about a 3D (three-dimensional) object.

[0017] Referring to FIG. 1, an electronic device (100) may be used to search for 3D objects or to transmit data representing 3D objects. For example, the electronic device (100) may be referred to as a search server or a server for searching. For example, the 3D object may be referred to as a 3D image or a 3D asset. For example, the electronic device (100) may include a communication circuit (e.g., the communication circuit (205) of FIG. 2). For example, the electronic device (100) may receive a signal representing a specified text (e.g., the specified text (405) of FIG. 4) from an external electronic device (110) through the communication circuit. For example, the external electronic device (110) may transmit the signal to the electronic device (100) based on receiving user input representing the specified text. For example, the user input may be received based on a software application for a search engine.

[0018] Based on receiving the signal, the electronic device (100) can search for a 3D object corresponding to the specified text within a database included in the electronic device (100). For example, to search for a 3D object corresponding to the specified text, the electronic device (100) can identify texts (e.g., title, label, description, etc.) tagged on each of the plurality of 3D objects included in the database. For example, the electronic device (100) can identify whether the 3D object corresponding to the specified text is included among the plurality of 3D objects based on identifying similarities between the specified text and the tagged texts. For example, the electronic device (100) can identify the 3D object corresponding to the specified text among the plurality of 3D objects based on identifying similarities between the specified text and the tagged texts. For example, searching for or determining a 3D object based on similarity between the texts above may be referred to as a lexical search or a search based on lexical analysis.

[0019] The result of the lexical search may depend on the text tagged to the 3D object. For example, if text that is less relevant to the 3D object is tagged, it may be difficult to identify the 3D object that is highly relevant to the specified text (e.g., the specified text (405) in FIG. 4) among multiple 3D objects. For example, if text that correctly describes the 3D object is not tagged to the 3D object, it may be difficult to identify the 3D object that is highly relevant to the specified text among multiple 3D objects. For example, the electronic device (100) may be required to determine the 3D object that is highly relevant to the specified text among multiple 3D objects, regardless of the text tagged to each of the 3D objects.

[0020] For example, the electronic device (100) may be required to perform a vector-based search to identify or determine, among a plurality of 3D objects, a 3D object that is highly relevant to the specified text, regardless of the texts tagged to each of the plurality of 3D objects. For example, the vector-based search may be referred to as a vector search. For example, the vector search may be performed by calculating the similarity between embedding vectors. For example, the electronic device (100) may perform the vector search by obtaining the similarity between the embedding vector for the 3D object and the embedding vector for the specified text. For example, the electronic device (100) may obtain an embedding vector (e.g., the embedding vector (420) of FIG. 4) using a plurality of images obtained through the 3D object. For example, the plurality of images may be obtained based on the specified coordinates of the 3D object. For example, the plurality of images may be described as 2D images of the 3D object based on the specified coordinates. For example, the electronic device (100) can determine or output a 3D object corresponding to the specified text regardless of the text tagged on the 3D object by performing a vector search using an embedding vector obtained using the plurality of images.

[0021] For example, the electronic device (100) may include hardware components used to perform or execute the above operations. The hardware components are described and illustrated with reference to FIG. 2.

[0022] Figure 2 is a simplified block diagram of an exemplary electronic device.

[0023] Referring to FIG. 2, the electronic device (100) may include at least one processor (207), a communication circuit (205), and a memory (206).

[0024] At least one processor (207) may include a hardware component for processing data using instructions stored in memory (206). The hardware component for processing data may include a central processing unit (e.g., including a processing circuit). The hardware component for processing data may include a neural processing unit (e.g., including a processing circuit).

[0025] At least one processor (207) may include one or more cores. For example, at least one processor (207) may have the structure of a multi-core processor such as a dual core, a quad core, or a hexa core.

[0026] Memory (206) may include a hardware component for storing data and / or instructions that are input to and / or output from at least one processor (207). Memory (206) may include, for example, volatile memory such as RAM (random-access memory) and / or non-volatile memory such as ROM (read-only memory). Volatile memory may include, for example, at least one of DRAM (dynamic RAM), SRAM (static RAM), cache RAM, and PSRAM (pseudo SRAM). Non-volatile memory may include, for example, at least one of PROM (programmable ROM), EPROM (erasable PROM), EEPROM (electrically erasable PROM), flash memory, hard disk, compact disk, and EMMC (embedded multimedia card).

[0027] The communication circuit (205) may include hardware components to support the transmission and / or reception of signals between the electronic device (100) and an external electronic device. The communication circuit (205) may include, for example, at least one of a modem, an antenna, and an O / E (optic / electronic) converter. The communication circuit (205) may support the transmission and / or reception of signals based on various types of protocols such as Ethernet, LAN (local area network), WAN (wide area network), WiFi (wireless fidelity), Bluetooth, BLE (Bluetooth low energy), Zigbee, LTE (long term evolution), and 5G NR (new radio).

[0028] At least one processor (207) can identify an event for determining a 3D object (e.g., 3D object (410) of FIG. 4) corresponding to a specified text (e.g., specified text (405) of FIG. 4) among a plurality of 3D objects. For example, at least one processor (207) can receive a signal representing the specified text from an external electronic device (110) through a communication circuit (205). For example, the event may include the reception of the signal. For example, at least one processor (207) can obtain similarities between a first embedding vector (e.g., embedding vector (415) of FIG. 4) corresponding to the specified text, and second embedding vectors obtained based on sets of images corresponding to each of the plurality of 3D objects. For example, at least one processor (207) can determine a 3D object corresponding to the similarity as a 3D object corresponding to the specified text based on obtaining a similarity that exceeds a threshold similarity among the similarities.

[0029] FIG. 3 is a flowchart illustrating the operation of an electronic device for determining a 3D object using similarities. This method may be executed by the electronic device (100) illustrated in FIG. 2 or at least one processor (207) of the electronic device (100).

[0030] Referring to FIG. 3, in operation 310, the electronic device (100) can identify an event for determining a 3D object (e.g., 3D object (410) of FIG. 4) corresponding to a specified text (e.g., specified text (405) of FIG. 4) among a plurality of 3D objects. For example, an external electronic device (110) can receive user input representing the specified text. For example, based on receiving the user input, the external electronic device (110) can transmit a signal representing the specified text to the electronic device (100) through a communication circuit included in the external electronic device (110). For example, based on receiving the signal, the electronic device (100) can determine or search for a 3D object corresponding to the specified text among a plurality of 3D objects. For example, the event may include receiving the signal.

[0031] In operation 320, the electronic device (100) can obtain similarities between a first embedding vector corresponding to a specified text (e.g., specified text (405) of FIG. 4) and second embedding vectors obtained based on sets of images corresponding to each of the plurality of 3D objects. For example, the first embedding vector may be described as an embedding vector representing the specified text. For example, the electronic device (100) can convert the specified text into the first embedding vector.

[0032] According to one embodiment, an electronic device (100) can obtain the first embedding vector representing the specified text by providing the specified text to the first trained model. For example, the first trained model may be trained to output an embedding vector using the text. For example, the first trained model may include a model trained through a machine learning technique (or a deep learning technique).

[0033] The electronic device (100) can obtain second embedding vectors corresponding to each of a plurality of 3D objects stored in a database. For example, the electronic device (100) can obtain second embedding vectors corresponding to each of a plurality of 3D objects before identifying an event to determine a 3D object corresponding to the specified text (e.g., the 3D object (410) of FIG. 4). For example, each second embedding vector can be obtained based on a plurality of images obtained through each 3D object (e.g., the 3D object (410) of FIG. 4). For example, each second embedding vector can be obtained by synthesizing embedding vectors corresponding to each of the plurality of images obtained through each 3D object (e.g., the embedding vector (532), embedding vector (534), embedding vector (536), and embedding vector (538) of FIG. 5). For example, the acquisition of each of the above second embedding vectors will be described later with reference to FIG. 5.

[0034] In operation 330, the electronic device (100) may determine a 3D object corresponding to a similarity (e.g., 3D object (410) of FIG. 4) that exceeds a threshold similarity among the similarities between the first embedding vector and the second embedding vector, and determine that the 3D object corresponding to said similarity is a 3D object corresponding to a specified text (e.g., specified text (405) of FIG. 4). For example, the electronic device (100) may determine that the 3D object corresponding to a similarity exceeding the threshold similarity corresponds to said specified text. For example, the electronic device (100) may transmit data representing said determined 3D object to an external electronic device (110) through a communication circuit (205). For example, the electronic device (100) may transmit a signal causing said determined 3D object to be displayed to an external electronic device (110) through a communication circuit (205). For example, the electronic device (100) may transmit a signal to an external electronic device (110) via a communication circuit (205) that causes a part of the determined 3D object (e.g., a representative image of the 3D object) to be displayed. For example, the electronic device (100) may determine that a 3D object corresponding to a similarity exceeding a threshold similarity corresponds to a specified text (405).

[0035] The electronic device (100) can obtain similarities between embedding vectors obtained through a specified text (e.g., specified text (405)) (e.g., embedding vector (415) of FIG. 4) and embedding vectors obtained through a 3D object (e.g., 3D object (410) of FIG. 4). For example, the acquisition of the embedding vector and the acquisition of similarities are described and illustrated in more detail with reference to FIG. 4.

[0036] FIG. 4 illustrates an exemplary operation of an electronic device that obtains similarities using embedding vectors.

[0037] Referring to FIG. 4, the electronic device (100) may obtain embedding vectors for each of a plurality of 3D objects before identifying an event to determine a 3D object (410) corresponding to a specified text (405). For example, the electronic device (100) may obtain, identify, or determine a similarity (425) between an embedding vector (415) corresponding to the specified text (405) and an embedding vector (420) for a 3D object (410) included in the plurality of 3D objects, based on identifying the event. For example, the electronic device (100) may obtain, identify, or determine a similarity (435) between an embedding vector (415) corresponding to the specified text (405) and an embedding vector (430) for a 3D object (410) included in the plurality of 3D objects, based on identifying the event. For example, the embedding vector (420) can be described as an embedding vector obtained using multiple images obtained through the 3D object (410). For example, the embedding vector (430) can be described as an embedding vector obtained using text obtained through the 3D object (410). For example, the embedding vector (420) can be referred to as an image embedding vector. For example, the embedding vector (430) can be referred to as a text embedding vector.

[0038] For example, the electronic device (100) may store data representing image embedding vectors and text embedding vectors for each of a plurality of 3D objects. For example, the electronic device (100) may obtain similarities (e.g., similarities including similarity (425)) between the image embedding vectors for the plurality of 3D objects and the embedding vector (415) corresponding to the specified text (405) based on identifying an event for determining or searching for a 3D object corresponding to a specified text (405). For example, the electronic device (100) may obtain other similarities (e.g., similarities including similarity (435)) between the text embedding vectors for the plurality of 3D objects and the embedding vector (430) corresponding to the specified text (405) based on identifying an event for determining or searching for a 3D object corresponding to a specified text (405). For example, the electronic device (100) may determine a 3D object corresponding to a similarity exceeding a threshold similarity among the similarities as a 3D object corresponding to a specified text (405). However, it is not limited thereto. For example, the electronic device (100) may determine a 3D object corresponding to another similarity exceeding a threshold similarity among the other similarities as a 3D object corresponding to a specified text (405).

[0039] The electronic device (100) can search for or determine a 3D object corresponding to a specified text (405) by using image embedding vectors corresponding to each of a plurality of 3D objects. For example, the electronic device (100) can obtain similarities between image embedding vectors obtained based on an embedding vector (415) corresponding to the specified text (405) and sets of images corresponding to each of the plurality of 3D objects. For example, an image embedding vector corresponding to each 3D object included in the plurality of 3D objects may correspond to a set of images (e.g., images (522), (524), (526), ​​and (528) of FIG. 5). For example, the electronic device (100) can obtain an embedding vector (420) corresponding to said set by using a set of images. For example, the electronic device (100) can obtain embedding vectors (e.g., image embedding vectors) including an embedding vector (420) by using a set of images corresponding to a plurality of 3D objects.

[0040] According to one embodiment, at least one of image embedding vectors for a plurality of 3D objects and text embedding vectors for a plurality of 3D objects may be obtained based on crawling. For example, an electronic device (100) may obtain one or more 3D objects from web pages using a software application for crawling. For example, the electronic device (100) may obtain at least one of one or more image embedding vectors for the one or more 3D objects obtained and one or more text embedding vectors for the one or more 3D objects obtained. For example, the electronic device (100) may tag the at least one obtained one to the one or more 3D objects obtained. For example, the electronic device (100) may tag or map data representing the at least one obtained one to each of the one or more 3D objects obtained. For example, the electronic device (100) can store a 3D object obtained through crawling in memory (206) along with an image embedding vector corresponding to the 3D object and a text embedding vector corresponding to the 3D object. For example, the electronic device (100) can create or manage a database (or dataset) containing the image embedding vector and / or text embedding vector through the storage. For example, the electronic device (100) can obtain data for a 3D object having at least one of the extensions GLB, FBX, and OBJ based on the crawling.

[0041] According to one embodiment, the electronic device (100) can obtain a 3D object (410) corresponding to a specified text (405) among a plurality of 3D objects by performing a Reciprocal Rank Fusion (RRF) technique based on the similarities and other similarities. The operation of obtaining a 3D object (410) corresponding to a specified text (405) based on the RRF technique will be described later with reference to FIG. 7.

[0042] The electronic device (100) can acquire multiple images through a 3D object (410). For example, the electronic device (100) can acquire embedding vectors for each of the multiple images. For example, the electronic device (100) can acquire an image embedding vector (e.g., embedding vector (420)) by synthesizing the embedding vectors. The acquisition of an image embedding vector based on the embedding vectors for each of the multiple images is described and illustrated in more detail with reference to FIG. 5.

[0043] FIG. 5 illustrates an exemplary operation of an electronic device for acquiring embedding vectors based on images of 3D objects.

[0044] Referring to FIG. 5, the electronic device (100) can obtain an embedding vector (420) using a 3D object (510). For example, the 3D object (510) may be an example of the 3D object (410) of FIG. 4. For example, the electronic device (100) can obtain a plurality of images (522, 524, 526, 528) using the 3D object (510). For example, the electronic device (100) can obtain a plurality of images (522, 524, 526, 528) using the 3D object (510) based on preset coordinates. For example, the preset coordinates may be coordinates set according to the World Coordinate System (WCS). However, it is not limited thereto. For example, the above-mentioned preset coordinates may be coordinates set according to an Object Coordinate System (OCS). For example, the above-mentioned preset coordinates may be coordinates set according to a View Coordinate System. For example, the above-mentioned View Coordinate System may be described as a coordinate system for the viewpoint of a camera or an observer.

[0045] The above-mentioned preset coordinates may include a first coordinate, a second coordinate, a third coordinate, and a fourth coordinate. For example, image (522) may be described as an image based on the first coordinate of a 3D object (510). For example, image (524) may be described as an image based on the second coordinate of a 3D object (510). For example, image (526) may be described as an image based on the third coordinate of a 3D object (510). For example, image (528) may be described as an image based on the fourth coordinate of a 3D object (510). For example, a plurality of images (522, 524, 526, 528) may be referred to as multi-view images.

[0046] The electronic device (100) can obtain embedding vectors (532, 534, 536, 538) using a plurality of images (522, 524, 526, 528). For example, the embedding vector (532) may correspond to the image (522). For example, the embedding vector (534) may correspond to the image (524). For example, the embedding vector (536) may correspond to the image (526). For example, the embedding vector (538) may correspond to the image (528).

[0047] According to one embodiment, when an electronic device (100) obtains a plurality of embedding vectors using a plurality of images (522, 524, 526, 528), a second trained model may be used. For example, the second trained model may be described as a model trained to output embedding vectors using images. For example, the second trained model may include a model trained through a machine learning technique (or a deep learning technique).

[0048] The electronic device (100) can obtain an embedding vector (420) by using embedding vectors (532, 534, 536, 538). For example, the electronic device (100) can obtain an embedding vector (420) by synthesizing the embedding vectors (532, 534, 536, 538). For example, the embedding vector (420) can be obtained by synthesizing embedding vectors obtained based on images corresponding to the coordinates of a 3D object (410). For example, the electronic device (100) may be required to obtain an embedding vector corresponding to a 3D object (410) in order to perform a vector search. For example, obtaining an embedding vector (420) corresponding to a 3D object (410) based on the images above may be easier than obtaining an embedding vector for a 3D object (410). For example, converting a 3D object (410) into an embedding vector (directly) may be difficult due to the complexity of the data. For example, converting a 3D object (410) into an embedding vector (directly) may be more expensive than obtaining an embedding vector (420) based on the images above. For example, converting a 3D object (410) into an embedding vector (directly) may result in low compatibility between different embedding vectors because the representation of the 3D object varies. For example, the electronic device (100) can obtain an embedding vector (420) corresponding to a 3D object (410) through images based on multiple coordinates of the 3D object (410). For example, the electronic device (100) can obtain a fast and highly compatible embedding vector at low cost through the images.

[0049] Referring to FIG. 5, the order (or method) in which an embedding vector (420) is obtained based on four images (522, 524, 526, 528) of a 3D object (510) is illustrated, but the embodiment is not limited thereto. For example, each of the four images may be described as an image corresponding to the front, rear, left side, and right side of the 3D object (510). For example, an electronic device (100) may obtain an embedding vector (420) using four images (522, 524, 526, 528) of a 3D object (510) and one or more additional images. For example, the one or more additional images may include at least one of an image corresponding to a top view, an image corresponding to a bottom view, and a perspective view. For example, the electronic device (100) may be required to obtain an embedding vector (420) using a predetermined number of images (e.g., 4 to 6). For example, the more images used to obtain the embedding vector (420), the higher the resources and / or costs required to obtain the embedding vector (420).

[0050] For example, the electronic device (100) may use a text-based embedding vector (430) obtained through a 3D object (410) to increase the accuracy of the embedding vector (420) based on multiple images obtained through a 3D object (410). For example, the acquisition of the text-based embedding vector (430) is described and illustrated in more detail with reference to FIG. 6.

[0051] FIG. 6 illustrates an exemplary operation of an electronic device for acquiring an embedding vector based on text obtained through an image of a 3D object.

[0052] Referring to FIG. 6, an electronic device (100) may obtain an image (522) from a 3D object (510) to obtain an embedding vector (430). For example, the electronic device (100) may obtain text (620) that represents or describes the image (522) using the image (522). For example, the text (620) may be referred to as a caption. For example, the electronic device (100) may obtain text (620) that represents or describes the image (522) by providing the image (522) to a model (610). For example, the electronic device (100) may be required to obtain an image (522) that represents the features of the 3D object (510) in order to obtain the text (620). For example, an image (522) obtained from a 3D object (510) and used to obtain text (620) can (sufficiently) represent the features of the 3D object (510). For example, an image (522) representing the features of the 3D object can be obtained through the 3D object (510) according to predetermined coordinates. For example, the image (522) may be an image corresponding to the front of the 3D object (510). However, it is not limited thereto. For example, an image representing the features of the 3D object may be a perspective view of the 3D object (510).

[0053] For example, the model (610) may be described as a model trained to output text using an image. For example, the model (610) may include a model trained through machine learning techniques (or deep learning techniques). For example, the model (610) may include a Vision-Language Model (VLM). For example, the model (610) may include a Large Multimodal Model (LMM). For example, the model (610) may be trained to describe or explain a provided image. For example, the model (610) may be used to analyze visual objects contained in the image (522) by performing object recognition and to output text (620) that describes or expresses said visual objects. For example, the electronic device (100) may obtain text (620) that describes or describes the image (522) by performing object recognition on the image (522).

[0054] According to one embodiment, the first trained model used to convert a specified text (405) into an embedding vector (415) may be different from the second trained model used to convert an image into an embedding vector. However, it is not limited thereto. For example, the first trained model may be the same as the second trained model.

[0055] According to one embodiment, the first trained model may be different from the model (610). However, it is not limited thereto. For example, the first trained model may be the same as the model (610).

[0056] According to one embodiment, the second trained model may be different from the model (610). However, it is not limited thereto. For example, the second trained model may be the same as the model (610).

[0057] For example, the electronic device (100) can obtain an embedding vector (430) using text (620). For example, the embedding vector (430) may correspond to text (620) obtained through an image obtained based on preset coordinates of a 3D object (410). For example, the electronic device (100) can obtain an embedding vector (430) corresponding to text (620) using text (620) representing an image obtained based on preset coordinates of a 3D object (410).

[0058] According to one embodiment, the electronic device (100) may use a third trained model when obtaining an embedding vector (430) using text (620). For example, the third trained model may be described as a model trained to generate an embedding vector using text. For example, the third trained model may be the same as the first trained model used to obtain an embedding vector (415) corresponding to a specified text (405). However, it is not limited thereto. For example, the third trained model may be different from the first trained model.

[0059] For example, the electronic device (100) can determine a 3D object (410) corresponding to a specified text (405) from among a plurality of 3D objects by using an embedding vector (420) (or image embedding vector) and an embedding vector (430) (or text embedding vector). For example, the electronic device (100) can determine a 3D object (410) corresponding to a specified text (405) by obtaining similarities between an embedding vector (415) corresponding to a specified text (405) and image embedding vectors for a plurality of 3D objects. For example, the electronic device (100) can determine a 3D object (410) corresponding to a specified text (405) by obtaining other similarities between an embedding vector (415) corresponding to a specified text (405) and text embedding vectors for a plurality of 3D objects. For example, the operation of determining a 3D object (410) corresponding to a designated text (405) based on the similarities and other similarities is described and illustrated in more detail with reference to FIG. 7.

[0060] FIG. 7 illustrates an exemplary operation of an electronic device that determines a 3D object using the Reciprocal Rank Fusion (RRF) technique.

[0061] Referring to FIG. 7, the table (710) may indicate the order of 3D objects determined according to an embedding vector (420) obtained based on a plurality of images (e.g., image (522), image (524), image (526), ​​image (528) of FIG. 5) obtained through a 3D object (410). For example, the plurality of 3D objects may include a first 3D object, a second 3D object, a third 3D object, and a fourth 3D object. For example, the electronic device (100) may obtain or calculate similarities between an embedding vector (415) corresponding to a specified text (405) and image embedding vectors of the plurality of 3D objects (e.g., image embedding vector of the first 3D object, image embedding vector of the second 3D object, image embedding vector of the third 3D object, and image embedding vector of the fourth 3D object). For example, the table (710) may be described as a table arranged in order of highest similarity among the plurality of 3D objects according to the similarities. For example, according to the similarities, the similarity corresponding to the second 3D object may be higher than the similarity corresponding to the fourth 3D object, the similarity corresponding to the third 3D object may be higher than the similarity corresponding to the second 3D object, and the similarity corresponding to the first 3D object may be higher than the similarity corresponding to the third 3D object. For example, the electronic device (100) may transmit data regarding a 3D object corresponding to a similarity exceeding a threshold similarity to an external electronic device (110) through a communication circuit (205). For example, the threshold similarity may be lower than the similarity corresponding to the second 3D object and higher than the similarity corresponding to the fourth 3D object. However, it is not limited thereto. For example, the above threshold similarity may be lower than the similarity corresponding to the first 3D object and higher than the similarity corresponding to the third 3D object.

[0062] Table (720) may indicate the order of 3D objects determined according to an embedding vector (430) obtained based on text (e.g., text (620)) obtained through a 3D object (410). For example, an electronic device (100) may obtain or calculate other similarities between an embedding vector (430) corresponding to a specified text (405) and text embedding vectors for the 3D objects (e.g., text embedding vector of a first 3D object, text embedding vector of a second 3D object, text embedding vector of a third 3D object, and text embedding vector of a fourth 3D object). For example, table (720) may be described as a table arranged in order of different similarities among the plurality of 3D objects according to the other similarities. For example, according to the other similarities mentioned above, the other similarity corresponding to the fourth 3D object may be higher than the other similarity corresponding to the third 3D object, the other similarity corresponding to the second 3D object may be higher than the other similarity corresponding to the fourth 3D object, and the other similarity corresponding to the first 3D object may be higher than the other similarity corresponding to the second 3D object. For example, the electronic device (100) may transmit data for a 3D object corresponding to another similarity that exceeds another threshold similarity to an external electronic device (110) through a communication circuit (205). For example, the other threshold similarity may be lower than the other similarity corresponding to the fourth 3D object and higher than the other similarity corresponding to the third 3D object. However, it is not limited thereto. For example, the other threshold similarity may be lower than the other similarity corresponding to the first 3D object and higher than the other similarity corresponding to the second 3D object.

[0063] Table (730) can be described as a table obtained by combining Table (710) and Table (720). For example, an electronic device (100) can obtain a table having a single order for a plurality of 3D objects by using an RRF technique. For example, the RRF technique can be described as a technique for obtaining a single result by using a set of results. For example, the RRF technique can be performed based on the reciprocal of the order (or rank) that appears by each result. For example, the electronic device (100) can obtain a single table (730) by using a table (710) related to image embedding vectors and a table (720) related to text embedding vectors through the RRF technique. For example, the electronic device (100) can obtain an order (or rank) of a plurality of 3D objects by using the similarities related to the image embedding vector and other similarities related to the text embedding vector through the RRF technique. For example, according to the order obtained through the RRF technique, the first 3D object may have a higher priority than the second 3D object, the second 3D object may have a higher priority than the third 3D object, and the third 3D object may have a higher priority than the fourth 3D object. For example, determining the priority for each of the plurality of 3D objects through the RRF technique makes it easier to determine the 3D object corresponding to the specified text (405) than determining the priority related to the image embedding vector or the priority related to the text embedding vector. For example, determining or searching for a 3D object corresponding to a specified text (405) among multiple 3D objects using the RRF technique can produce a more accurate result (or a result preferred by the user) than determining a 3D object corresponding to a specified text (405) using one of an image embedding vector and a text embedding vector.

[0064] According to one embodiment, the electronic device (100) can obtain values ​​for each of a plurality of 3D objects based on performing an RRF technique on similarities for image embedding vectors and other similarities for text embedding vectors. For example, the electronic device (100) can determine that the 3D object corresponding to the value exceeding a threshold value among the values ​​is the 3D object corresponding to the specified text (405). For example, the electronic device (100) can determine that the 3D object corresponding to the value exceeding the threshold value among the values ​​is the object corresponding to the specified text (405). For example, the electronic device (100) can determine that the 3D object corresponding to the value exceeding the threshold value corresponds to the specified text (405).

[0065] According to one embodiment, the electronic device (100) may assign a weight to at least one of the similarities associated with an image embedding vector and other similarities associated with a text embedding vector. For example, the electronic device (100) may determine a 3D object (410) corresponding to a specified text (405) among a plurality of 3D objects based on the assigned weight. For example, the electronic device (100) may obtain modified similarities using the similarities and other similarities based on the assigned weight. For example, the electronic device (100) may determine a modified similarity among the modified similarities that exceeds a threshold modified similarity. For example, the electronic device (100) may determine a 3D object corresponding to a modified similarity exceeding a threshold modified similarity as a 3D object corresponding to a specified text (405) among a plurality of 3D objects. For example, the electronic device (100) can determine that a 3D object corresponding to a modification similarity exceeding a threshold modification similarity corresponds to a specified text (405).

[0066] According to one embodiment, the electronic device (100) can obtain modified embedding vectors by synthesizing image embedding vectors for a plurality of 3D objects and text embedding vectors for a plurality of 3D objects. For example, the electronic device (100) can obtain modified embedding vectors by synthesizing an embedding vector (420) and an embedding vector (430). For example, the electronic device (100) can obtain similarities between the modified embedding vectors and an embedding vector (415) corresponding to a specified text (405). For example, the electronic device (100) can determine that a 3D object corresponding to a similarity exceeding a threshold similarity among the similarities corresponds to the specified text (405).

[0067] According to one embodiment, the modified embedding vectors can be obtained using a trained model. For example, the trained model may be described as a model trained to output a single embedding vector using the provided embedding vectors. For example, an electronic device (100) can obtain the modified embedding vector by providing the trained model with an image embedding vector for a 3D object (410) (e.g., embedding vector (420)) and a text embedding vector for a 3D object (410) (e.g., embedding vector (430)).

[0068] The electronic device (100) can transmit data representing a 3D object to an external electronic device (110) via a communication circuit (205) based on determining or identifying a 3D object corresponding to a specified text (405) among a plurality of 3D objects. For example, the external electronic device (110) can output or display content regarding the 3D object based on receiving the data. For example, the output of content regarding the 3D object is described and illustrated in more detail with reference to FIG. 8.

[0069] FIG. 8 illustrates examples of operations performed within an electronic device and an external electronic device.

[0070] Referring to FIG. 8, in operation 810, an external electronic device (110) may transmit a first signal representing a specified text (405) to an electronic device (100) through a communication circuit included in the external electronic device (110). For example, the external electronic device (110) may receive user input to search for or determine a 3D object corresponding to the specified text (405) among a plurality of 3D objects. For example, the electronic device (100) may transmit a first signal corresponding to the specified text (405) represented by the user input to the electronic device (100). For example, the electronic device (100) may receive the first signal from the external electronic device (110) through a communication circuit (205).

[0071] In operation 820, the electronic device (100) may obtain similarities between a first embedding vector and second embedding vectors corresponding to a specified text (405) based on receiving the first signal. For example, each second embedding vector may be described as an embedding vector corresponding to a plurality of images obtained through a 3D object (410). For example, each second embedding vector may include an embedding vector (420). For example, the second embedding vectors may be referred to as image embedding vectors. For example, operation 820 may correspond to operation 320 of FIG. 3.

[0072] In operation 830, the electronic device (100) may determine that a 3D object (410) corresponding to a similarity exceeding a threshold similarity among the similarities is a 3D object corresponding to a specified text (405). For example, the electronic device (100) may determine that a 3D object (410) corresponding to a similarity exceeding a threshold similarity corresponds to a specified text (405). For example, the electronic device (100) may determine that a 3D object corresponding to a similarity exceeding a threshold similarity corresponds to a specified text (405) based on identifying the similarity exceeding the threshold similarity. For example, operation 830 may correspond to operation 330 of FIG. 3.

[0073] In operation 840, the electronic device (100) may transmit a second signal (or data) representing a 3D object corresponding to a specified text (405) to an external electronic device (110) via a communication circuit (205) based on the above determination. For example, the electronic device (100) may transmit a second signal to an external electronic device (110) via a communication circuit (205) that causes content for the 3D object to be displayed or output. For example, the external electronic device (110) may receive the second signal from the electronic device (100) via a communication circuit included in the external electronic device (110).

[0074] In operation 850, the electronic device (100) may display content for the determined 3D object based on receiving the second signal. For example, the content may include an image of the determined 3D object (e.g., an image or thumbnail image corresponding to preset coordinates of the 3D object). For example, the content may be described as content for indicating that the determined 3D object has been acquired. For example, the external electronic device (110) may output or display the content through a display included in the external electronic device (110). For example, a user of the external electronic device (110) may recognize that a 3D object corresponding to a specified text (405) has been found or determined based on the display of the content.

[0075] According to one embodiment, the electronic device (100) may provide or display content for a 3D object corresponding to a specified text (405). For example, an operation in which the content is displayed through an external electronic device (110) is described in detail, but the embodiment is not limited thereto. For example, the electronic device (100) may acquire or identify the specified text (405) based on receiving user input for searching for a 3D object. For example, the electronic device (100) may perform operations to determine a 3D object corresponding to the specified text (405) among a plurality of 3D objects based on identifying an event of acquiring the specified text (405). For example, the operations may include operations 320 and 330 of FIG. 3. For example, the electronic device (100) may output or provide data corresponding to the 3D object based on determining a 3D object corresponding to the specified text (405) among a plurality of 3D objects. For example, the electronic device (100) may further include a display (not shown). For example, the electronic device (100) may display a 3D object corresponding to a specified text (405) through the display based on the above determination. For example, the electronic device (100) may display an image (e.g., a thumbnail image) of a 3D object corresponding to a specified text (405) through the display using the above data.

[0076] An electronic device as described above may include a memory for storing instructions. The electronic device may include at least one processor. The instructions may cause the electronic device to identify an event for determining a three-dimensional object corresponding to a specified text among a plurality of three-dimensional objects when executed individually or collectively by the at least one processor. The instructions may cause the electronic device to obtain similarities between a first embedding vector corresponding to the specified text and second embedding vectors obtained based on sets of images corresponding to each of the plurality of three-dimensional objects when executed individually or collectively by the at least one processor. The instructions may cause the electronic device to determine a three-dimensional object corresponding to the similarity as the three-dimensional object corresponding to the specified text when executed individually or collectively by the at least one processor, based on obtaining a similarity among the similarities that exceeds a threshold similarity.

[0077] According to one embodiment, each of the images may be an image corresponding to each of the preset coordinates of each 3D object.

[0078] According to one embodiment, each of the second embedding vectors can be obtained by synthesizing the third embedding vectors corresponding to the images.

[0079] According to one embodiment, each of the second embedding vectors can be obtained by synthesizing a fourth embedding vector obtained using text and a third embedding vector obtained using the images of each set. The text can represent different images obtained through each 3D object.

[0080] According to one embodiment, the instructions may cause the electronic device to obtain different similarities between the first embedding vector and the fourth embedding vectors corresponding to texts representing different images obtained through each of the plurality of 3D objects, when executed individually or collectively by the at least one processor. The instructions may cause the electronic device to obtain values ​​for each of the plurality of 3D objects based on performing a Reciprocal Rank Fusion (RRF) technique on the similarities and the other similarities, when executed individually or collectively by the at least one processor. The instructions may cause the electronic device to determine the 3D object corresponding to the value exceeding a threshold among the values ​​as the 3D object corresponding to the specified text, when executed individually or collectively by the at least one processor.

[0081] According to one embodiment, the instructions may cause the electronic device to provide data representing the 3D object based on the determination of the 3D object corresponding to the specified text when executed individually or collectively by the at least one processor.

[0082] According to one embodiment, the event may include obtaining the specified text based on user input for searching for a 3D object.

[0083] According to one embodiment, the first embedding vector can be obtained using a first model trained to generate an embedding vector using text. The second embedding vectors can be obtained using a second model trained to generate an embedding vector using an image.

[0084] A method performed by an electronic device as described above may include an operation of identifying an event for determining a 3D (three-dimensional) object corresponding to a specified text among a plurality of 3D objects. The method may include an operation of obtaining similarities between a first embedding vector corresponding to the specified text and second embedding vectors obtained based on sets of images corresponding to each of the plurality of 3D objects. The method may include an operation of determining a 3D object corresponding to a similarity as the 3D object corresponding to the specified text, based on obtaining a similarity among the similarities that exceeds a threshold similarity.

[0085] According to one embodiment, each of the images may be an image corresponding to each of the preset coordinates of each 3D object.

[0086] According to one embodiment, each of the second embedding vectors can be obtained by synthesizing the third embedding vectors corresponding to the images.

[0087] According to one embodiment, each of the second embedding vectors can be obtained by synthesizing a fourth embedding vector obtained using text and a third embedding vector obtained using the images of each set. The text can represent different images obtained through each 3D object.

[0088] According to one embodiment, the method may include an operation of obtaining different similarities between the first embedding vector and the fourth embedding vectors corresponding to texts representing different images obtained through each of the plurality of 3D objects. The method may include an operation of obtaining values ​​for each of the plurality of 3D objects based on performing a Reciprocal Rank Fusion (RRF) technique on the similarities and the other similarities. The method may include an operation of determining the 3D object corresponding to a value exceeding a threshold among the values ​​as the 3D object corresponding to the designated text.

[0089] According to one embodiment, the method may include an operation of providing data representing the 3D object based on the determination of the 3D object corresponding to the specified text.

[0090] According to one embodiment, the event may include obtaining the specified text based on user input for searching for a 3D object.

[0091] According to one embodiment, the first embedding vector can be obtained using a first model trained to generate an embedding vector using text. The second embedding vectors can be obtained using a second model trained to generate an embedding vector using an image.

[0092] In a computer-readable storage medium storing one or more programs as described above, the one or more programs may include instructions that cause the electronic device to identify an event for determining a three-dimensional object corresponding to a specified text among a plurality of three-dimensional objects when executed by the electronic device. The one or more programs may include instructions that cause the electronic device to obtain similarities between a first embedding vector corresponding to the specified text and second embedding vectors obtained based on sets of images corresponding to each of the plurality of three-dimensional objects when executed by the electronic device. The one or more programs may include instructions that cause the electronic device to determine the three-dimensional object corresponding to the similarity as the three-dimensional object corresponding to the specified text, based on obtaining a similarity among the similarities that exceeds a threshold similarity.

[0093] According to one embodiment, each of the images may be an image corresponding to each of the preset coordinates of each 3D object.

[0094] According to one embodiment, each of the second embedding vectors can be obtained by synthesizing the third embedding vectors corresponding to the images.

[0095] According to one embodiment, each of the second embedding vectors can be obtained by synthesizing a fourth embedding vector obtained using text and a third embedding vector obtained using the images of each set. The text can represent different images obtained through each 3D object.

[0096] According to one embodiment, the one or more programs may include instructions that cause the electronic device to obtain different similarities between the first embedding vector and the fourth embedding vectors corresponding to texts representing different images obtained through each of the plurality of 3D objects when executed by the electronic device. The one or more programs may include instructions that cause the electronic device to obtain values ​​for each of the plurality of 3D objects based on performing a Reciprocal Rank Fusion (RRF) technique on the similarities and the other similarities when executed by the electronic device. The one or more programs may include instructions that cause the electronic device to determine the 3D object corresponding to the value exceeding a threshold among the values ​​as the 3D object corresponding to the designated text when executed by the electronic device.

[0097] According to one embodiment, the one or more programs may include instructions that cause the electronic device to provide data representing the 3D object based on the determination of the 3D object corresponding to the specified text when executed by the electronic device.

[0098] According to one embodiment, the event may include obtaining the specified text based on user input for searching for a 3D object.

[0099] According to one embodiment, the first embedding vector can be obtained using a first model trained to generate an embedding vector using text. The second embedding vectors can be obtained using a second model trained to generate an embedding vector using an image.

[0100] The device described above may be implemented as a hardware component, a software component, and / or a combination of a hardware component and a software component. For example, the device and components described in the embodiments may be implemented using one or more general-purpose or special-purpose computers, such as a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing unit may execute an operating system (OS) and one or more software applications executed on said operating system. Additionally, the processing unit may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing unit may be described as being used as a single unit, but those skilled in the art will understand that the processing unit may include multiple processing elements and / or multiple types of processing elements. For example, the processing unit may include multiple processors or one processor and one controller. In addition, other processing configurations, such as parallel processors, are also possible.

[0101] Software may include computer programs, code, instructions, or a combination of one or more of these, and may configure a processing unit to operate as desired or instruct the processing unit independently or collectively. Software and / or data may be embodied in any type of machine, component, physical device, computer storage medium, or device so as to be interpreted by the processing unit or to provide instructions or data to the processing unit. Software may be distributed over networked computer systems and may be stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

[0102] The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. In this case, the medium may continuously store a computer-executable program, or temporarily store it for execution or download. Additionally, the medium may be various recording or storage means in the form of a single or several combined hardware, and may not be limited to a medium directly connected to a computer system but may exist distributed over a network. Examples of media may include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and media configured to store program instructions, including ROM, RAM, and flash memory. Additionally, other examples of media may include recording or storage media managed by app stores that distribute applications or sites and servers that supply or distribute various other software.

[0103] Although the embodiments have been described above with reference to limited examples and drawings, those skilled in the art can make various modifications and variations from the description above. For example, suitable results may be achieved even if the described techniques are performed in a different order than described, and / or the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

[0104] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below.

Claims

1. In an electronic device, Memory comprising one or more storage media for storing instructions; and It includes at least one processor comprising processing circuitry, and When the above instructions are executed individually or collectively by the at least one processor, Identify an event to determine a 3D (three-dimensional) object corresponding to specified text among multiple 3D objects, and Similarities between a first embedding vector corresponding to the specified text and second embedding vectors obtained based on sets of images corresponding to each of the plurality of 3D objects are obtained, and Based on obtaining a similarity that exceeds a threshold similarity among the above similarities, the 3D object corresponding to the similarity is determined as the 3D object corresponding to the specified text. causing the above electronic device, Electronic device.

2. In claim 1, each of the images is, An image corresponding to each of the preset coordinates of each 3D object, Electronic device.

3. In claim 1, each of the second embedding vectors is, Obtained by synthesizing third embedding vectors corresponding to the above images, Electronic device.

4. In claim 1, each of the second embedding vectors is, A fourth embedding vector obtained using text and a third embedding vector obtained using the images of each set are obtained by synthesizing, and The above text is, Representing different images obtained through each 3D object, Electronic device.

5. In Claim 1, When the above instructions are executed individually or collectively by the at least one processor, Obtaining different similarities between fourth embedding vectors corresponding to texts representing different images obtained through each of the first embedding vector and the plurality of 3D objects, and Based on performing the RRF (Reciprocal Rank Fusion) technique on the above similarities and the other similarities, values ​​for each of the plurality of 3D objects are obtained, and To determine the 3D object corresponding to the value exceeding the threshold among the above values ​​as the 3D object corresponding to the above specified text, causing the above electronic device, Electronic device.

6. In Claim 1, When the above instructions are executed individually or collectively by the at least one processor, Based on the determination of the 3D object corresponding to the specified text, to provide data representing the 3D object, causing the above electronic device, Electronic device.

7. In claim 1, the above event is, Acquiring the aforementioned specified text based on user input for searching 3D objects, Electronic device.

8. In claim 1, the first embedding vector is, Obtained using a first model trained to generate embedding vectors using text, and The above second embedding vectors are, Obtained using a second model trained to generate embedding vectors using images, Electronic device.

9. In a non-transient computer-readable storage medium storing one or more programs, said one or more programs, said one or more programs are, When executed by an electronic device, Identify an event to determine a 3D (three-dimensional) object corresponding to specified text among multiple 3D objects, and Similarities between a first embedding vector corresponding to the specified text and second embedding vectors obtained based on sets of images corresponding to each of the plurality of 3D objects are obtained, and Based on obtaining a similarity that exceeds a threshold similarity among the above similarities, the 3D object corresponding to the similarity is determined as the 3D object corresponding to the specified text. Including instructions that cause the above electronic device, Non-transient computer-readable storage media.

10. In claim 9, each of the images is, An image corresponding to each of the preset coordinates of each 3D object, Non-transient computer-readable storage media.

11. In claim 9, each of the second embedding vectors is, Obtained by synthesizing third embedding vectors corresponding to the above images, Non-transient computer-readable storage media.

12. In claim 9, each of the second embedding vectors is, A fourth embedding vector obtained using text and a third embedding vector obtained using the images of each set are obtained by synthesizing, and The above text is, Representing different images obtained through each 3D object, Non-transient computer-readable storage media.

13. In Claim 9, When the above one or more programs are executed by the electronic device, Obtaining different similarities between fourth embedding vectors corresponding to texts representing different images obtained through each of the first embedding vector and the plurality of 3D objects, and Based on performing the RRF (Reciprocal Rank Fusion) technique on the above similarities and the other similarities, values ​​for each of the plurality of 3D objects are obtained, and To determine the 3D object corresponding to the value exceeding the threshold among the above values ​​as the 3D object corresponding to the above specified text, Including instructions that cause the above electronic device, Non-transient computer-readable storage media.

14. In Claim 9, When the above one or more programs are executed by the electronic device, Based on the determination of the 3D object corresponding to the specified text, to provide data representing the 3D object, Including instructions that cause the above electronic device, Non-transient computer-readable storage media.

15. In claim 9, the above event is, Acquiring the aforementioned specified text based on user input for searching 3D objects, Non-transient computer-readable storage media.

16. In claim 9, the first embedding vector is, Obtained using a first model trained to generate embedding vectors using text, and The above second embedding vectors are, Obtained using a second model trained to generate embedding vectors using images, Non-transient computer-readable storage media.

17. In a method executed within an electronic device, An operation to identify an event for determining a 3D (three-dimensional) object corresponding to specified text among multiple 3D objects, and The operation of obtaining similarities between a first embedding vector corresponding to the specified text and second embedding vectors obtained based on sets of images corresponding to each of the plurality of 3D objects, and Based on obtaining a similarity that exceeds a threshold similarity among the similarities, the operation of determining a 3D object corresponding to the similarity as a 3D object corresponding to the specified text, method.

18. In claim 17, each of the images is, An image corresponding to each of the preset coordinates of each 3D object, method.

19. In claim 17, each of the second embedding vectors is, Obtained by synthesizing third embedding vectors corresponding to the above images, method.

20. In claim 17, each of the second embedding vectors is, A fourth embedding vector obtained using text and a third embedding vector obtained using the images of each set are obtained by synthesizing, and The above text is, Representing different images obtained through each 3D object, method.

Images