A furniture information processing method, device, equipment and medium

Abstract

The application relates to a furniture information processing method and device, equipment and medium, and relates to the technical field of indoor design. The method comprises the following steps: acquiring three-dimensional model data of target furniture, performing feature extraction processing on the three-dimensional model data to obtain implicit feature information and attribute feature information, performing feature fusion processing on the implicit feature information and the attribute feature information to obtain target feature information, determining candidate furniture information similar to the target furniture according to the target feature information, and determining target similar furniture corresponding to the target furniture based on the candidate furniture information, so that the most similar furniture to the target furniture can be acquired, and the problem that the prior art cannot find a specific furniture similar set from a large number of furniture is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of interior design, and in particular to a furniture information processing method and device, equipment and medium. BACKGROUND

[0002] At present, when a user selects furniture using an indoor home design platform, the user usually sets furniture one by one for different positions in an indoor space, which slows down the progress of indoor design.

[0003] The existing solution is that after a user selects a desired furniture, the indoor home design platform automatically selects furniture of the same category from a database to match other furniture in the space according to the category of the furniture selected by the user. However, for a mature furniture manufacturer or design platform, a large number of furniture of the same category may exist (for example, there may be more than 1000 furniture of the same category), and the attributes of different furniture of the same category are different, and different users have different definitions of similar furniture. The selected furniture from the database cannot meet the definition of similar furniture of different users. It can be seen that the existing technology cannot find similar furniture of a specified furniture from a large number of furniture in the database. SUMMARY

[0004] The present application provides a furniture information processing method, device, equipment and medium to solve the problem that the existing technology cannot find a similar set of a specified furniture from a large number of furniture.

[0005] In a first aspect, the present application provides a furniture information processing method, comprising:

[0006] obtaining three-dimensional model data of a target furniture;

[0007] performing feature extraction processing according to the three-dimensional model data to obtain implicit feature information and attribute feature information;

[0008] performing feature fusion processing according to the implicit feature information and the attribute feature information to obtain target feature information;

[0009] determining candidate furniture information similar to the target furniture according to the target feature information;

[0010] determining a target similar furniture corresponding to the target furniture based on the candidate furniture information.

[0011] Optionally, the obtaining of the three-dimensional model data of the target furniture comprises:

[0012] obtaining user input information;

[0013] extracting furniture identification from the user input information;

[0014] Based on the furniture identification, the three-dimensional model data is extracted from a preset furniture database.

[0015] Optionally, the feature extraction processing according to the three-dimensional model data obtains implicit feature information and attribute feature information, including:

[0016] Based on the three-dimensional model data, point cloud format data and rendering graph data of the target furniture are determined;

[0017] The point cloud feature vector and the rendering graph feature vector are obtained by respectively performing encoding processing on the point cloud format data and the rendering graph data;

[0018] The point cloud feature vector and the rendering graph feature vector are determined as the implicit feature information;

[0019] The attribute feature information is obtained by performing classification processing based on the rendering graph data.

[0020] Optionally, the point cloud format data and the rendering graph data of the target furniture are determined based on the three-dimensional model data, including:

[0021] The point cloud format data is obtained by performing format conversion processing based on the three-dimensional model data;

[0022] The rendering graph data is obtained by performing rendering processing based on the three-dimensional model data, or the three-dimensional model data corresponding to the preset rendering graph data is extracted from the furniture database.

[0023] Optionally, the point cloud feature vector and the rendering graph feature vector are obtained by respectively performing encoding processing on the point cloud format data and the rendering graph data, including:

[0024] The point cloud format data is input into a preset first encoder for encoding processing, and the point cloud feature vector output by the first encoder is obtained;

[0025] The rendering graph data is input into a preset second encoder for encoding processing, and the rendering graph feature vector output by the second encoder is obtained.

[0026] Optionally, the target feature information is obtained by performing feature fusion processing according to the implicit feature information and the attribute feature information, including:

[0027] Feature weight information is obtained, and the feature weight information includes attribute weight parameters corresponding to the attribute feature information and implicit weight parameters corresponding to the implicit feature information;

[0028] The attribute feature information and the implicit feature information are spliced and weighted based on the attribute weight parameter and the implicit weight parameter, to obtain the target feature information.

[0029] Optionally, the attribute feature information is obtained by performing classification processing on the rendering graph data, including: inputting the rendering graph data into a preset material classification model to perform material classification processing, to obtain a material feature vector; inputting the rendering graph data into a preset category classification model to perform category classification processing, to obtain a category feature vector; and taking the material feature vector and the category feature vector as the attribute feature information.

[0030] The candidate furniture information similar to the target furniture is determined according to the target feature information, including: searching a first similar vector matching the target feature information from a preset similar vector retrieval library; and determining furniture information corresponding to the first similar vector as the candidate furniture information.

[0031] In a second aspect, the present application provides a furniture information processing apparatus, including:

[0032] The acquisition module is configured to acquire three-dimensional model data of a target furniture.

[0033] The feature extraction processing module is configured to perform feature extraction processing on the three-dimensional model data, to obtain implicit feature information and attribute feature information.

[0034] The feature fusion processing module is configured to perform feature fusion processing on the implicit feature information and the attribute feature information, to obtain target feature information.

[0035] The candidate furniture information determination module is configured to determine candidate furniture information similar to the target furniture according to the target feature information.

[0036] The target similar furniture determination module is configured to determine a target similar furniture corresponding to the target furniture based on the candidate furniture information.

[0037] In a third aspect, the present application provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete communication with each other through the communication bus.

[0038] The memory is configured to store a computer program.

[0039] The processor is configured to execute the program stored on the memory, to implement the steps of the furniture information processing method according to any one of the embodiments of the first aspect.

[0040] In a fourth aspect, the present application provides a computer readable storage medium, having stored thereon a computer program, which, when executed by a processor, implements the steps of the furniture information processing method according to any one of the embodiments of the first aspect.

[0041] To sum up, the embodiments of the present application obtain the three-dimensional model data of the target furniture, perform feature extraction processing according to the three-dimensional model data to obtain implicit feature information and attribute feature information, perform feature fusion processing according to the implicit feature information and the attribute feature information to obtain target feature information, determine candidate furniture information similar to the target furniture according to the target feature information, and determine the target similar furniture corresponding to the target furniture based on the candidate furniture information, so that the most similar furniture to the target furniture can be obtained, and the problem that the prior art cannot find a specific furniture similar set from a large number of furniture is solved. BRIEF DESCRIPTION OF DRAWINGS

[0042] The accompanying drawings, which are incorporated herein and constitute part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required to be used in the embodiments or the prior art description will be briefly introduced as follows. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without creative labor.

[0044] Figure 1 A flowchart of a furniture information processing method provided by an embodiment of the present application;

[0045] Figure 2 A step flowchart of a furniture information processing method provided by an optional embodiment of the present application;

[0046] Figure 3 A similar furniture determination flowchart provided by an optional embodiment of the present application;

[0047] Figure 4 A model training flowchart of similar furniture provided by an optional embodiment of the present application;

[0048] Figure 5 A structural block diagram of a furniture information processing device provided by an embodiment of the present application;

[0049] Figure 6 A structural diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0050] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] To facilitate understanding of the embodiments of this application, further explanations and descriptions will be provided below in conjunction with the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of this application.

[0052] Figure 1 This is a flowchart illustrating a furniture information processing method provided in an embodiment of this application. Figure 1 As shown, the furniture information processing method provided in this application embodiment may specifically include the following steps:

[0053] Step 110: Obtain the 3D model data of the target furniture.

[0054] Specifically, the target furniture can be furniture selected by the user, that is, the target furniture can be furniture specified by the user. When the user needs to find a set of similar furniture from a large number of furniture, the specified furniture can be used as the target furniture. This application embodiment does not limit this. The target furniture can have corresponding three-dimensional model data, which can include the three-dimensional (3D) model file of the target furniture. This application embodiment does not limit this. Specifically, this application embodiment can use furniture specified by the user as the target furniture, and then obtain the three-dimensional model file of the target furniture as three-dimensional model data.

[0055] In a specific implementation, the embodiments of this application can pre-build a furniture database, which can store the three-dimensional model data corresponding to each piece of furniture. After the target furniture is determined, the three-dimensional model data corresponding to the target furniture can be extracted from the furniture database.

[0056] Step 120: Perform feature extraction processing based on the three-dimensional model data to obtain latent feature information and attribute feature information.

[0057] Specifically, the implicit feature information can include a point cloud feature vector and a rendering graph feature vector, and the embodiments of the present application do not limit this; the attribute feature information can include a material feature vector and a category feature vector, and the embodiments of the present application do not limit this. Specifically, the embodiments of the present application can perform feature extraction processing according to the three-dimensional model data to obtain a point cloud feature vector, a rendering graph feature vector, a material feature vector, and a category feature vector, respectively, and then the point cloud feature vector and the rendering graph feature vector can be taken as the implicit feature information, and the material feature vector and the category feature vector can be taken as the attribute feature information, and subsequent feature fusion processing can be performed based on the implicit feature information and the attribute feature information, that is, step 130 is executed.

[0058] In a specific implementation, after the embodiments of the present application obtain the three-dimensional model data of the target furniture, the three-dimensional model data can be converted into a point cloud format to obtain a three-dimensional model in a point cloud format, and the three-dimensional model data can be visualized and rendered to obtain a rendering graph. The three-dimensional model in the point cloud format and the rendering graph can include objective attributes of the furniture, such as furniture color, material, furniture three-dimensional model structure, length of each edge of the furniture, radian of an angle, degree of reflection under light at each position, and transparency of the material, etc. When performing feature extraction, the three-dimensional model in the point cloud format can be subjected to feature extraction to obtain a point cloud feature vector, and the rendering graph can be subjected to feature extraction to obtain a rendering graph feature vector, a material feature vector, and a category feature vector. Then, the point cloud feature vector and the rendering graph feature vector can be taken as the implicit feature information, so that the implicit feature information can more comprehensively consider various feature attributes of the furniture and cover the definition of similarity by different users; the material feature vector and the category feature vector can be taken as the attribute feature information, so that the attribute feature information can combine a plurality of different classification feature vectors to express most attributes of the target furniture. Subsequently, feature fusion can be performed on the implicit feature information and the attribute feature information, so that when determining similar furniture, the accuracy of the similar furniture can be improved, and thus the user experience can be improved.

[0059] Step 130: performing feature fusion processing according to the implicit feature information and the attribute feature information to obtain target feature information.

[0060] Specifically, the target feature information can include a target feature vector, and the embodiments of the present application do not limit this. Specifically, the embodiments of the present application can splice and weight each feature vector included in the implicit feature information and the attribute feature information to obtain a target feature vector as the target feature information. By combining the implicit feature information and the attribute feature information, different users' definitions of similarity can be more comprehensively and more possibly covered, and by retaining original attribute features such as material and category of the furniture through the attribute feature information, the accuracy of obtaining similar furniture can be effectively improved, and the furniture most similar to the user-specified furniture can be obtained.

[0061] In step 140, candidate furniture information similar to the target furniture is determined according to the target feature information.

[0062] Specifically, the furniture database in the embodiment of the present application can contain target furniture and non-target furniture, wherein the non-target furniture can contain candidate furniture, and each non-target furniture can have corresponding furniture feature information. After the target feature information is determined, the target feature information can be matched with each non-target furniture corresponding to the furniture feature vector to obtain the similarity between the target feature information and each furniture feature vector, and then the non-target furniture similar to the target furniture can be determined as the candidate furniture information according to the similarity.

[0063] In a specific implementation, the embodiment of the present application can perform feature extraction on the three-dimensional model data corresponding to each furniture contained in the furniture database to obtain the furniture feature vector corresponding to each furniture as the furniture feature information stored in the feature vector database. After the target feature information corresponding to the target furniture is determined, the target feature information can be matched with the furniture feature information stored in the feature vector database to determine the candidate furniture information similar to the target furniture.

[0064] In step 150, the target similar furniture corresponding to the target furniture is determined based on the candidate furniture information.

[0065] Specifically, after the candidate furniture information is determined, the target similar furniture corresponding to the target furniture can be determined based on the candidate furniture information, thereby solving the problem that the prior art cannot filter the similar furniture required by the user from a large number of furniture.

[0066] For example, the candidate furniture information can be sorted in descending order of similarity to obtain a sorting result, and the similarity sorting manner is not limited in the example. Then, the candidate furniture information with the highest similarity can be selected as the target similar furniture according to the sorting result. Of course, one or more candidate furniture information can be selected as the target similar furniture in descending order of similarity according to the actual needs of the user, and the example is not limited in this regard.

[0067] As can be seen, the embodiment of the present application obtains the three-dimensional model data of the target furniture, performs feature extraction processing on the three-dimensional model data to obtain implicit feature information and attribute feature information, then performs feature fusion processing on the implicit feature information and the attribute feature information to obtain target feature information, and then determines the candidate furniture information similar to the target furniture according to the target feature information, to determine the target similar furniture corresponding to the target furniture based on the candidate furniture information, thereby being able to obtain the most similar furniture to the target furniture and solving the problem that the prior art cannot find a similar set of specific furniture from a large number of furniture.

[0068] Referring to Figure 2 , a step flow diagram of a furniture information processing method provided by an optional embodiment of the present application is shown, which can specifically include the following steps:

[0069] Step 210, obtaining three-dimensional model data of a target furniture.

[0070] In an optional embodiment, the three-dimensional model data of the target furniture is obtained, which can specifically include: obtaining user input information; extracting furniture identification from the user input information; and extracting the three-dimensional model data from a preset furniture database based on the furniture identification. Specifically, the furniture identification can include an identity document, which is not limited by the embodiments of the present application.

[0071] As an example, referring to Figure 3 , the furniture ID can be extracted from the user input information, then the target furniture can be determined according to the furniture ID, and the three-dimensional model corresponding to the target furniture (i.e. three-dimensional model data) can be extracted from the furniture database.

[0072] Step 220, determining point cloud format data and rendering graph data of the target furniture based on the three-dimensional model data.

[0073] Specifically, the point cloud format data can include point cloud format model data, which is not limited by the embodiments of the present application; and the rendering graph data can include a rendering graph corresponding to the three-dimensional model data, which is also not limited by the embodiments of the present application. Specifically, after obtaining the three-dimensional model data, the three-dimensional model data can be converted to obtain the point cloud format data, and the three-dimensional model data can be visualized to obtain the rendering graph data. The point cloud format data and the rendering graph data completely cover the objective attributes of the furniture itself, can more comprehensively cover the similar definitions of different users, and improve the accuracy of similar furniture.

[0074] In a specific implementation, the point cloud data in the embodiments of the present application can contain information of each structure of the furniture. Since the point cloud format data is generated by sampling from a 3D model without rendering, part of the furniture detail information may be lost in the point cloud format data, such as the furniture surface reflection degree and texture details, etc. The rendering graph data can contain the furniture surface reflection degree and texture details, etc. missing in the point cloud format data. The rendering graph data can be a projection of a view of the furniture and can not contain information of the back structure of the furniture, etc. Therefore, when obtaining the point cloud format data of the 3D model, the embodiments of the present application can obtain the rendering graph data corresponding to the 3D model data, combine the point cloud format data and the rendering graph data to make up for the defects of the point cloud format data or the rendering graph data, and perform feature extraction by combining the point cloud format data and the rendering graph data, so as to improve the furniture feature extraction effect, and make the point cloud feature vector and the rendering graph feature vector obtained by feature extraction cover the input attributes of the furniture and improve the similar furniture judgment ability.

[0075] In an optional embodiment, the embodiments of the present application determine the point cloud format data and the rendering graph data of the target furniture based on the 3D model data, which can specifically include: performing format conversion processing based on the 3D model data to obtain the point cloud format data; performing rendering processing based on the 3D model data to obtain the rendering graph data; or extracting the rendering graph data corresponding to the 3D model data from the furniture database.

[0076] As an example, referring to Figure 3 , the format of the obtained 3D model data can be a model file format (obj format), and the format of the 3D model data is not limited in this example. The 3D model data can be converted into a point cloud format by an open source library (such as Open3d) and saved to obtain point cloud format model data, which is used as point cloud format data. The 3D model data can be rendered, such as visual rendering in a computer by using a 3D rendering tool (such as 3Dmax), to visualize the structure and color of the 3D model data, to obtain a rendering graph, which is used as rendering graph data.

[0077] In a specific implementation, the furniture database in the embodiment of the present application can pre-render the stored three-dimensional model data to obtain rendering image data corresponding to the three-dimensional model data, so as to serve as preset rendering image data corresponding to the three-dimensional model data, and can store the rendering image data. Before rendering the three-dimensional model data, it can be determined whether the rendering image data corresponding to the three-dimensional model data already exists in the furniture database. If it is determined that the rendering image data already exists in the furniture database, the preset rendering image data corresponding to the three-dimensional model data can be directly extracted from the furniture database, so as to improve the efficiency. If it is determined that the rendering image data corresponding to the three-dimensional model data does not exist in the furniture database, the three-dimensional model data can be rendered to obtain the rendering image data.

[0078] In step 230, the point cloud feature vector and the rendering image feature vector are obtained by respectively performing encoding processing on the point cloud format data and the rendering image data.

[0079] Specifically, after obtaining the point cloud format data and the rendering image data, the point cloud format data can be encoded to obtain a point cloud feature vector corresponding to the point cloud format data, and the rendering image data can be encoded to obtain a rendering image feature vector corresponding to the rendering image data, so as to extract multiple features by combining the input of the point cloud format data and the input of the rendering image data, thereby more comprehensively covering the input attributes, and improving the accuracy of the most similar furniture obtained subsequently.

[0080] In a specific implementation, the embodiment of the present application can pre-construct two autoencoders. The two autoencoders can be a kind of neural network model, for example, the two autoencoders can be a first encoder and a second encoder, and the embodiment of the present application does not limit this. The autoencoder can compress the input vector into a feature vector with smaller dimension and shorter length. For example, the input vector can include three-dimensional model point cloud (i.e., point cloud format data) and two-dimensional image (i.e., rendering image data), etc. The autoencoder can also satisfy that it generates a same-dimension vector with high similarity to the input vector in the decoding stage, so that the feature vector obtained by the autoencoder can be equivalent to the input. Specifically, the point cloud format data can be input into the first encoder, the point cloud format data is encoded by the first encoder to obtain a point cloud feature vector, and the rendering image data can be input into the second encoder, the rendering image data is encoded by the second encoder to obtain a rendering image feature vector.

[0081] Optionally, the above encoding processing on the point cloud format data and the rendering image data to obtain the point cloud feature vector and the rendering image feature vector can specifically include the following sub-steps:

[0082] Sub-step 2301, input the point cloud format data into a preset first encoder for encoding processing to obtain the point cloud feature vector output by the first encoder.

[0083] Specifically, the first encoder can include a first autoencoder, and the embodiments of the present application do not limit this. Specifically, the embodiments of the present application can input the point cloud format data as an input vector into the pre-trained first encoder, and the first encoder can encode and compress the input point cloud format data to obtain a feature vector with smaller dimensions and shorter length as a point cloud feature vector.

[0084] As an example, referring to Figure 3 After the three-dimensional model data is format-converted to obtain the point cloud format data, the point cloud format data can be input into the pre-trained autoencoder 1AE1 (i.e., the first encoder) to obtain the feature vector 1 (i.e., the point cloud feature vector).

[0085] Sub-step 2302, input the rendering graph data into a preset second encoder for encoding processing to obtain the rendering graph feature vector output by the second encoder.

[0086] Specifically, the second encoder can include a second autoencoder, and the embodiments of the present application do not limit this. Specifically, the embodiments of the present application can input the rendering graph data as an input vector into the pre-trained second encoder, and the second encoder can encode and compress the input rendering graph data to obtain a feature vector with smaller dimensions and shorter length as a rendering graph feature vector.

[0087] As an example, referring to Figure 3 The three-dimensional model data can be rendered to obtain the rendering graph data, and then the rendering graph data can be input into the pre-trained autoencoder 2AE2 (i.e., the second encoder) to obtain the feature vector 2 (i.e., the rendering graph feature vector).

[0088] In a specific implementation, the embodiments of the present application can pre-train two autoencoders to obtain the trained first encoder and the second encoder. For example, referring to Figure 4 Two autoencoders can be pre-constructed, which can include the autoencoder 1AE1 and the autoencoder 2AE2, and then the three-dimensional model and the rendering graph corresponding to each furniture can be extracted from the preset furniture database, the point cloud format data is obtained based on the three-dimensional model, the point cloud format data is input into the autoencoder 1AE1 for model training to obtain the first encoder, and the rendering graph (i.e., the rendering graph data) can be input into the autoencoder 2AE2 for model training to obtain the second encoder, and the present example does not limit this.

[0089] Step 240, the point cloud feature vector and the rendering map feature vector are determined as the implicit feature information.

[0090] Specifically, after obtaining the point cloud feature vector and the rendering map feature vector, the point cloud feature vector and the rendering map feature vector can be determined as the implicit feature information.

[0091] Step 250, classification processing is performed based on the rendering map data to obtain the attribute feature information.

[0092] Specifically, the attribute feature information can be obtained by performing feature extraction based on the attribute label data and the rendering map data. Specifically, a material feature vector can be obtained by performing feature extraction based on the material attribute label included in the attribute label data and the rendering map data, and a category feature vector can be obtained by performing feature extraction based on the category attribute label included in the attribute label data and the rendering map data. Subsequently, the category feature vector and the material feature vector can be determined as the attribute feature information. In the process of performing feature extraction based on the attribute label data, the rendering map data is added for feature extraction, so that the information such as the degree of reflection and the texture details of the furniture surface included in the rendering map is used to improve the furniture feature extraction effect.

[0093] In an optional embodiment, the attribute feature information is obtained by performing classification processing based on the rendering map data. Specifically, the attribute feature information can be obtained by performing material classification processing on the rendering map data by inputting the rendering map data into a pre-trained material classification model, performing category classification processing on the rendering map data by inputting the rendering map data into a pre-trained category classification model, and determining the material feature vector and the category feature vector as the attribute feature information.

[0094] As an example, referring to Figure 3 The rendering map and the material attribute label in the attribute label data can be input into the pre-trained material classification model Ml, the input rendering map and the material attribute label can be subjected to material classification processing by the material classification model Ml, and a feature vector 3 (i.e., a material feature vector) can be obtained. The rendering map and the category attribute label in the attribute label data can be input into the pre-trained category classification model M2, the input rendering map and the category attribute label can be subjected to category classification processing by the category classification model M2, and a feature vector 4 (i.e., a category feature vector) can be obtained. Subsequently, the category feature vector and the material feature vector can be determined as the attribute feature information. By combining the characteristics of the furniture itself and outputting the feature vector by the neural network model, the ability to judge similar furniture can be improved.

[0095] In a specific implementation, the embodiment of the present application can pre-define two classification models through a convolutional neural network model, which can be MobileNetV3 convolutional neural network model M1 and MobileNetV3 convolutional neural network model M2. Subsequently, the convolutional neural network model M1 and the convolutional neural network model M2 can be model trained, and the trained convolutional neural network model M1 is used as the material classification model M1, and the trained convolutional neural network model M2 is used as the category classification model M2.

[0096] As an example, referring to Figure 4 The attribute label corresponding to each piece of furniture can be obtained from the furniture database, which can include a material attribute label and a category attribute label, and the present example is not limited thereto. Specifically, the embodiment of the present application can use the rendering image as a training set and the material attribute label and the category attribute label corresponding to the rendering image as a control set. When model training is performed, such as model training of the material classification model M1, the rendering image can be input into the material classification model M1 for model training to obtain a material classification prediction result output by the material classification model M1. The material classification prediction result is compared with the material attribute label to determine whether the model converges. For example, when the accuracy of the material classification prediction result output by the material classification model M1 reaches 90% or above, it is determined that the model converges, and the trained material classification model M1 is obtained.

[0097] Similarly, when the category classification model M2 is model trained, the rendering image can be input into the category classification model M2 for model training to obtain a category classification prediction result output by the category classification model M2. The category classification prediction result is compared with the material attribute label to determine whether the model converges. For example, when the accuracy of the category classification prediction result output by the category classification model M2 reaches 90% or above, it is determined that the model converges, and the trained category classification model M2 is obtained.

[0098] In actual processing, after obtaining the trained material classification model M1 and the category classification model M2, the rendering image data can be input into the material classification model M1 and the category classification model M2 for feature extraction during actual use, so that the feature vector of the furniture can be extracted when the attribute label corresponding to the furniture stored in the furniture database is empty or the attribute label is incorrect, realizing attribute-free furniture query. The attribute label feature is further converted and refined into a digital feature vector (i.e., a material feature vector and a category feature vector) through deep learning, so as to retain the original attribute label feature of the furniture, and then express most of the attributes of the furniture through the digital feature vector.

[0099] In actual processing, after the automatic encoders AE1 and AE2, the material classification model M1, and the category classification model M2 are trained, the point cloud format data and the rendering graph data corresponding to all furniture in the furniture database can be input into the corresponding models to extract the feature vectors corresponding to each attribute. Subsequently, the dimensions of the feature vectors of each furniture can be converted into one dimension, and after being given a certain weight, the feature vectors are connected and combined, and stored in a similar vector retrieval library (Faiss), so that after the target feature information is determined, the vector closest to the target feature information is searched from the Faiss based on the target feature information, and the furniture corresponding to the closest vector found can be used as the target similar furniture, so as to find similar furniture or similar set of specific furniture in a large number of furniture.

[0100] In step 260, feature weight information is obtained, which includes attribute weight parameters corresponding to the attribute feature information and hidden weight parameters corresponding to the hidden feature information.

[0101] Specifically, the feature weight information can be set by the user, and of course, in the case where the user does not set it, the feature weight information can also be preset, and the embodiments of the present application do not limit this. The feature weight information can include attribute weight parameters corresponding to the attribute feature information and hidden weight parameters corresponding to the hidden feature information. The hidden weight parameters can include point cloud weight corresponding to the point cloud feature vector and rendering graph weight corresponding to the rendering graph feature vector. The attribute weight parameters can include material attribute weight corresponding to the material feature vector and category attribute weight corresponding to the category feature vector, and the embodiments of the present application do not limit this. Specifically, in the embodiments of the present application, the user can set the corresponding feature weight according to the demand of the attribute feature information and the hidden feature information. For example, in the case where the user pays more attention to the material, a higher weight can be set for the material feature vector; in the case where the user pays more attention to the category, a higher weight can also be set for the category feature vector.

[0102] In step 270, based on the attribute weight parameters and the hidden weight parameters, the attribute feature information and the hidden feature information are spliced and weighted to obtain the target feature information.

[0103] Specifically, the embodiments of the present application can splice and weight the attribute feature information and the hidden feature information based on the attribute weight parameters and the hidden weight parameters to obtain the target feature information. For example, referring to Figure 3The feature weight information can be determined according to the user input weight, the feature weight information is used to determine the weights corresponding to the feature vector 1 (i.e., the point cloud feature vector), the feature vector 2 (i.e., the rendering graph feature vector), the feature vector 3 (i.e., the material feature vector), and the feature vector 4 (i.e., the category feature vector), different weights are given to the feature vector 1, the feature vector 2, the feature vector 3, and the feature vector 4, and then the feature vector 1, the feature vector 2, the feature vector 3, and the feature vector 4 can be spliced and weighted in combination with the weights to obtain a target feature vector as the target feature information.

[0104] In step 280, candidate furniture information similar to the target furniture is determined according to the target feature information.

[0105] In an optional embodiment, the embodiment of the present application determines the candidate furniture information similar to the target furniture according to the target feature information, which can specifically include: searching a first similar vector matching the target feature information from a preset similar vector retrieval library; and determining the furniture information corresponding to the first similar vector as the candidate furniture information.

[0106] As an example, referring to Figure 3 After the target feature information is determined, a feature vector matching the target feature information can be searched in the similar vector retrieval library (Faiss), the similarity between the target feature vector and each feature vector in the Faiss is calculated by using the Euclidean distance algorithm, then each feature vector in the similar vector retrieval library can be sorted in descending order of similarity to obtain a sorting result, the furniture ID corresponding to each feature vector in the sorting result is returned to the user, and the furniture information corresponding to the furniture ID can be used as the candidate furniture information.

[0107] In step 290, the target similar furniture corresponding to the target furniture is determined based on the candidate furniture information.

[0108] Specifically, the embodiment of the present application can use the furniture information with the highest similarity to the target feature information in the candidate furniture information as the target similar furniture, so that the furniture most similar to the user-specified furniture can be obtained from a large number of furniture, and the problem that the prior art cannot find a similar set from a large number of furniture and a specific furniture is solved.

[0109] In a specific implementation, after the sorting result of the furniture similarity is obtained, one or more furniture with the highest similarity in the sorting result can be selected according to the user demand and returned to the user to facilitate the user to make a selection.

[0110] In summary, the embodiment of the present application obtains the three-dimensional model data of the target furniture, determines the point cloud format data and the rendering graph data of the target furniture based on the three-dimensional model data, respectively encodes the point cloud format data and the rendering graph data to obtain the point cloud feature vector and the rendering graph feature vector, determines the point cloud feature vector and the rendering graph feature vector as the implicit feature information, classifies the rendering graph data to obtain the attribute feature information, then obtains the feature weight information, splices and weights the attribute feature information and the implicit feature information based on the attribute weight parameter and the implicit weight parameter to obtain the target feature information, and then determines the candidate furniture information similar to the target furniture based on the target feature information, and determines the target similar furniture corresponding to the target furniture based on the candidate furniture information. The three-dimensional point cloud model and the two-dimensional rendering graph are combined to extract multiple features, so that the input attribute is more comprehensive, and the attribute label feature of the furniture can be extracted through the classification processing of the rendering graph data. The attribute label feature is converted and refined into a digital feature vector, combined with the extracted multiple features, the furniture most similar to the user-specified furniture can be obtained from a large number of furniture, and the problem that the prior art cannot find a similar set of specific furniture from a large number of furniture is solved.

[0111] It should be noted that, for the method embodiment, in order to simply describe, it is expressed as a series of action combinations, but those skilled in the art should know that the embodiment of the present application is not limited by the described action sequence, because according to the embodiment of the present application, certain steps can be performed in other order or simultaneously.

[0112] As shown in Figure 5 the embodiment of the present application also provides a furniture information processing device 500, which comprises:

[0113] The acquisition module 510 is configured to acquire three-dimensional model data of a target furniture.

[0114] The feature extraction processing module 520 is configured to perform feature extraction processing on the three-dimensional model data to obtain implicit feature information and attribute feature information.

[0115] The feature fusion processing module 530 is configured to perform feature fusion processing on the implicit feature information and the attribute feature information to obtain target feature information.

[0116] The candidate furniture information determination module 540 is configured to determine candidate furniture information similar to the target furniture according to the target feature information.

[0117] The target similar furniture determination module 550 is configured to determine the target similar furniture corresponding to the target furniture based on the candidate furniture information.

[0118] Optionally, the acquisition module comprises:

[0119] a user input information obtaining sub-module, configured to obtain user input information;

[0120] a furniture identification extracting sub-module, configured to extract a furniture identification from the user input information;

[0121] an extracting sub-module, configured to extract the three-dimensional model data from a preset furniture database based on the furniture identification.

[0122] Optionally, the feature extraction processing module comprises:

[0123] a data determining sub-module, configured to determine point cloud format data and rendering graph data of the target furniture based on the three-dimensional model data;

[0124] an encoding processing sub-module, configured to perform encoding processing on the point cloud format data and the rendering graph data respectively to obtain a point cloud feature vector and a rendering graph feature vector;

[0125] an implicit feature information determining sub-module, configured to determine the point cloud feature vector and the rendering graph feature vector as the implicit feature information;

[0126] an attribute feature information determining sub-module, configured to perform classification processing based on the rendering graph data to obtain the attribute feature information.

[0127] Optionally, the data determining sub-module comprises:

[0128] a point cloud format data determining unit, configured to perform format conversion processing based on the three-dimensional model data to obtain the point cloud format data;

[0129] a rendering graph data determining unit, configured to perform rendering processing based on the three-dimensional model data to obtain the rendering graph data, or extract preset rendering graph data corresponding to the three-dimensional model data from the furniture database.

[0130] Optionally, the encoding processing sub-module comprises:

[0131] a point cloud feature vector determining module, configured to input the point cloud format data into a preset first encoder to perform encoding processing, so as to obtain the point cloud feature vector output by the first encoder;

[0132] a rendering graph feature vector determining module, configured to input the rendering graph data into a preset second encoder to perform encoding processing, so as to obtain the rendering graph feature vector output by the second encoder.

[0133] Optionally, the feature fusion processing module comprises:

[0134] The feature weight information acquisition submodule is configured to acquire feature weight information, which includes an attribute weight parameter corresponding to the attribute feature information and a latent weight parameter corresponding to the latent feature information.

[0135] The weighted splicing processing submodule is configured to perform weighted splicing processing on the attribute feature information and the latent feature information based on the attribute weight parameter and the latent weight parameter, to obtain the target feature information.

[0136] Optionally, the attribute feature information determination submodule includes:

[0137] The material classification processing unit is configured to input the rendered graph data into a preset material classification model to perform material classification processing, to obtain a material feature vector.

[0138] The category classification processing unit is configured to input the rendered graph data into a preset category classification model to perform category classification processing, to obtain a category feature vector.

[0139] The attribute feature information determination unit is configured to take the material feature vector and the category feature vector as the attribute feature information.

[0140] The candidate furniture information determination module is specifically configured to find a first similar vector matching the target feature information from a preset similar vector retrieval library; and determine furniture information corresponding to the first similar vector as candidate furniture information.

[0141] It should be noted that the furniture information processing apparatus provided in the embodiments of the present application can execute the furniture information processing method provided in any embodiment of the present application, and has the corresponding functions and beneficial effects of executing the furniture information processing method.

[0142] In a specific implementation, the above-mentioned furniture information processing apparatus can be integrated in a device, so that the device can perform feature extraction and feature fusion based on three-dimensional model data corresponding to a target furniture, to obtain target feature information, and then determine a target similar furniture similar to the target furniture based on the target feature information, as an electronic device, to realize searching for a similar furniture based on a target furniture specified by a user. The electronic device can be composed of two or more physical entities, or can be composed of one physical entity, for example, the electronic device can be a personal computer (PC), a computer, a server, etc., and the embodiments of the present application do not make a specific limitation in this regard.

[0143] As Figure 6As shown, the embodiment of the present application provides an electronic device, comprising a processor 111, a communication interface 112, a memory 113 and a communication bus 114, wherein the processor 111, the communication interface 112 and the memory 113 complete mutual communication through the communication bus 114; the memory 113 is used for storing a computer program; the processor 111 is used for executing the program stored on the memory 113, and the steps of the furniture information processing method provided by any one of the preceding method embodiments are implemented. Exemplarily, the steps of the furniture information processing method can comprise the following steps: acquiring three-dimensional model data of a target furniture; performing feature extraction processing according to the three-dimensional model data to obtain implicit feature information and attribute feature information; performing feature fusion processing according to the implicit feature information and the attribute feature information to obtain target feature information; determining candidate furniture information similar to the target furniture according to the target feature information; and determining a target similar furniture corresponding to the target furniture based on the candidate furniture information.

[0144] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the furniture information processing method provided by any one of the preceding method embodiments.

[0145] It should be noted that, in this document, the relationship terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or device including the element.

[0146] The above description is only a specific embodiment of the present application, which enables those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features applied herein.

Claims

1. A furniture information processing method characterized by comprising: The method comprises the following steps: obtaining three-dimensional model data of a target furniture; performing feature extraction processing on the three-dimensional model data to obtain implicit feature information and attribute feature information; performing feature fusion processing on the implicit feature information and the attribute feature information to obtain target feature information; determining candidate furniture information similar to the target furniture according to the target feature information; determining a target similar furniture corresponding to the target furniture based on the candidate furniture information; wherein the feature extraction processing on the three-dimensional model data to obtain the implicit feature information and the attribute feature information comprises: determining point cloud format data and rendering graph data of the target furniture based on the three-dimensional model data; performing encoding processing on the point cloud format data and the rendering graph data respectively to obtain point cloud feature vectors and rendering graph feature vectors; determining the point cloud feature vectors and the rendering graph feature vectors as the implicit feature information; performing classification processing on the rendering graph data to obtain the attribute feature information; the feature fusion processing on the implicit feature information and the attribute feature information to obtain the target feature information comprises: obtaining feature weight information, wherein the feature weight information comprises attribute weight parameters corresponding to the attribute feature information and implicit weight parameters corresponding to the implicit feature information; performing splicing and weighting processing on the attribute feature information and the implicit feature information based on the attribute weight parameters and the implicit weight parameters to obtain the target feature information; the classification processing on the rendering graph data to obtain the attribute feature information comprises: inputting the rendering graph data into a preset material classification model to perform material classification processing to obtain material feature vectors; inputting the rendering graph data into a preset category classification model to perform category classification processing to obtain category feature vectors; and taking the material feature vectors and the category feature vectors as the attribute feature information; the determination of the candidate furniture information similar to the target furniture according to the target feature information comprises: searching for a first similar vector matching the target feature information from a preset similar vector retrieval library; and determining furniture information corresponding to the first similar vector as the candidate furniture information.

2. The method of claim 1, wherein, The method comprises the following steps: obtaining three-dimensional model data of a target furniture; performing feature extraction processing on the three-dimensional model data to obtain implicit feature information and attribute feature information; performing feature fusion processing on the implicit feature information and the attribute feature information to obtain target feature information; 3. The method of claim 1, wherein, determining candidate furniture information similar to the target furniture according to the target feature information; determining a target similar furniture corresponding to the target furniture based on the candidate furniture information; wherein the feature extraction processing on the three-dimensional model data to obtain the implicit feature information and the attribute feature information comprises:

4. The method of claim 1, wherein, determining point cloud format data and rendering graph data of the target furniture based on the three-dimensional model data; performing encoding processing on the point cloud format data and the rendering graph data respectively to obtain point cloud feature vectors and rendering graph feature vectors; determining the point cloud feature vectors and the rendering graph feature vectors as the implicit feature information; performing classification processing on the rendering graph data to obtain the attribute feature information; the feature fusion processing on the implicit feature information and the attribute feature information to obtain the target feature information comprises: obtaining feature weight information, wherein the feature weight information comprises attribute weight parameters corresponding to the attribute feature information and implicit weight parameters corresponding to the implicit feature information; performing splicing and weighting processing on the attribute feature information and the implicit feature information based on the attribute weight parameters and the implicit weight parameters to obtain the target feature information; the classification processing on the rendering graph data to obtain the attribute feature information comprises: inputting the rendering graph data into a preset material classification model to perform material classification processing to obtain material feature vectors; inputting the rendering graph data into a preset category classification model to perform category classification processing to obtain category feature vectors; and taking the material feature vectors and the category feature vectors as the attribute feature information; the determination of the candidate furniture information similar to the target furniture according to the target feature information comprises: searching for a first similar vector matching the target feature information from a preset similar vector retrieval library; and determining furniture information corresponding to the first similar vector as the candidate furniture information. The method comprises the following steps: obtaining three-dimensional model data of a target furniture; performing feature extraction processing on the three-dimensional model data to obtain implicit feature information and attribute feature information; performing feature fusion processing on the implicit feature information and the attribute feature information to obtain target feature information; determining candidate furniture information similar to the target furniture according to the target feature information; determining a target similar furniture corresponding to the target furniture based on the candidate furniture information; wherein the feature extraction processing on the three-dimensional model data to obtain the implicit feature information and the attribute feature information comprises: determining point cloud format data and rendering graph data of the target furniture based on the three-dimensional model data; performing encoding processing on the point cloud format data and the rendering graph data respectively to obtain point cloud feature vectors and rendering graph feature vectors; determining the point cloud feature vectors and the rendering graph feature vectors as the implicit feature information; performing classification processing on the rendering graph data to obtain the attribute feature information; the feature fusion processing on the implicit feature information and the attribute feature information to obtain the target feature information comprises: obtaining feature weight information, wherein the feature weight information comprises attribute weight parameters corresponding to the attribute feature information and implicit weight parameters corresponding to the implicit feature information; performing splicing and weighting processing on the attribute feature information and the implicit feature information based on the attribute weight parameters and the implicit weight parameters to obtain the target feature information; the classification processing on the rendering graph data to obtain the attribute feature information comprises: inputting the rendering graph data into a preset material classification model to perform material classification processing to obtain material feature vectors; inputting the rendering graph data into a preset category classification model to perform category classification processing to obtain category feature vectors; and taking the material feature vectors and the category feature vectors as the attribute feature information; the determination of the candidate furniture information similar to the target furniture according to the target feature information comprises: searching for a first similar vector matching the target feature information from a preset similar vector retrieval library; and determining furniture information corresponding to the first similar vector as the candidate furniture information. the obtaining of the three-dimensional model data of the target furniture comprises: obtaining user input information; extracting furniture identification from the user input information; extracting the three-dimensional model data from a preset furniture database based on the furniture identification. the determination of the point cloud format data and the rendering graph data of the target furniture based on the three-dimensional model data comprises: performing format conversion processing on the three-dimensional model data to obtain the point cloud format data; performing rendering processing on the three-dimensional model data to obtain the rendering graph data; or extracting preset rendering graph data corresponding to the three-dimensional model data from the furniture database. the encoding processing on the point cloud format data and the rendering graph data respectively to obtain the point cloud feature vectors and the rendering graph feature vectors comprises: Input the point cloud format data into a preset first encoder for encoding processing to obtain the point cloud feature vector output by the first encoder; Input the rendering graph data into a preset second encoder for encoding processing to obtain the rendering graph feature vector output by the second encoder.

5. An information processing apparatus for furniture, characterized by comprising: Comprise: An acquisition module is used to acquire three-dimensional model data of a target furniture; A feature extraction processing module is used to perform feature extraction processing according to the three-dimensional model data to obtain implicit feature information and attribute feature information; A feature fusion processing module is used to perform feature fusion processing according to the implicit feature information and the attribute feature information to obtain target feature information; A candidate furniture information determination module is used to determine candidate furniture information similar to the target furniture according to the target feature information; A target similar furniture determination module is used to determine a target similar furniture corresponding to the target furniture based on the candidate furniture information; The feature extraction processing module comprises: A data determination sub-module is used to determine point cloud format data and rendering graph data of the target furniture based on the three-dimensional model data; An encoding processing sub-module is used to perform encoding processing on the point cloud format data and the rendering graph data respectively to obtain a point cloud feature vector and a rendering graph feature vector; An implicit feature information determination sub-module is used to determine the point cloud feature vector and the rendering graph feature vector as the implicit feature information; An attribute feature information determination sub-module is used to perform classification processing based on the rendering graph data to obtain the attribute feature information; The feature fusion processing module comprises: A feature weight information acquisition sub-module is used to acquire feature weight information, which contains attribute weight parameters corresponding to the attribute feature information and implicit weight parameters corresponding to the implicit feature information; A weighted splicing processing sub-module is used to perform splicing and weighting processing on the attribute feature information and the implicit feature information based on the attribute weight parameters and the implicit weight parameters to obtain the target feature information; The attribute feature information determination sub-module comprises: A material classification processing unit is used to input the rendering graph data into a preset material classification model for material classification processing to obtain a material feature vector; A category classification processing unit is used to input the rendering graph data into a preset category classification model for category classification processing to obtain a category feature vector; An attribute feature information determination unit is used to take the material feature vector and the category feature vector as the attribute feature information; The candidate furniture information determination module is specifically used to find a first similar vector matching the target feature information from a preset similar vector retrieval library; and determine furniture information corresponding to the first similar vector as candidate furniture information.

6. An electronic device, comprising: Comprise a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus; The memory is used to store a computer program; The processor is used to execute the program stored on the memory to realize the steps of the furniture information processing method in any one of claims 1-4.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the furniture information processing method according to any one of claims 1-4.

Images