WO2023230769A1

WO2023230769A1 - Cad model search method, cad model clustering and classification model generation method, apparatus and storage medium

Info

Publication number: WO2023230769A1
Application number: PCT/CN2022/096006
Authority: WO
Inventors: 邹文超; 王海峰
Original assignee: 西门子股份公司; 西门子（中国）有限公司
Priority date: 2022-05-30
Filing date: 2022-05-30
Publication date: 2023-12-07

Abstract

Provided are a CAD model search method, a CAD model clustering and classification model generation method, an apparatus, and a storage medium. The CAD model clustering and classification model generation method comprises: for each CAD model in a database, extracting geometric information thereof, performing structured representation on the geometric information, and converting the geometric information in a structured representation form into a vector of a predefined length to obtain a vector of the CAD model; clustering the vectors of all the CAD models in the database to obtain a plurality of CAD model clusters which do not exceed a set number size, so that when a CAD model search is carried out, a search range is positioned to a CAD model cluster which does not exceed the set number size; and for each CAD model cluster, performing training by using all the CAD models in the CAD model cluster as training samples so as to obtain a classification model corresponding to the CAD model cluster, so that CAD model searching in the CAD model cluster is performed by means of the corresponding classification model. The technical solution in the embodiments can improve the CAD model search efficiency.

Description

CAD model search, clustering and classification model generation method, device and storage medium

Technical field

The present application relates to the field of computers, and in particular to a CAD model clustering and classification model generation, as well as a CAD model search method, device and computer-readable storage medium.

Background of the invention

Currently, people have a great demand for CAD search engines, and they need to find similar CAD models in order to reuse their designs or the manufacturing processes behind them. However, how to improve the search efficiency of CAD search engines is currently a big challenge. Search efficiency is related to the time consumption of the search strategy. For example, the simplest implementation of a search engine looks like the following pseudocode:

```Pseudo Code Start#code start

Vars:Input Model＝IN,Model＝M,Database＝DB,Similarity＝S,Threshold＝T,Candidate List＝L,Display Number＝NUM#Input model IN, search for model M in database DB, similarity is S, The similarity threshold is T, the candidate list is L, and the display number is NUM

In the above implementation, the search time will increase as the number of CAD models in the database increases. If each iteration takes 1 second, the database size is 10 ⁴ models. Then the search time is 10 ⁴ seconds, which is about 3 hours. This is unacceptable to search engine users.

To this end, those skilled in the art are still working on finding other CAD model search solutions.

Contents of the invention

In view of this, embodiments of the present application provide, on the one hand, a CAD model clustering and classification model generation method and a CAD model search method, and on the other hand, a CAD model clustering and classification model generation device, a CAD model search device, The computer device and the computer-readable storage medium can improve the efficiency of CAD model search.

In order to solve the above technical problems, the technical solution of this application is implemented as follows:

A method for generating CAD model clustering and classification models, including: for each CAD model in the database, extracting the geometric information of the CAD model, performing a structured representation of the geometric information, and converting the structured expressed geometric information into Convert to a vector of predefined length to obtain the vector of the CAD model; cluster the vectors of all CAD models in the database to obtain multiple CAD model clusters not exceeding the set number in order to perform CAD model When searching, locate the search scope to a CAD model cluster that does not exceed the set number; for each CAD model cluster, use all CAD models within the CAD model cluster as training samples to train to obtain the corresponding CAD The classification model of model clustering is used so that when searching for CAD models, the corresponding classification model is used to search for CAD models within the CAD model clusters that do not exceed the set number.

A CAD model search method, including: receiving an input current CAD model; extracting geometric information of the current CAD model, performing a structured representation of the geometric information, and converting the structured represented geometric information into a predefined length vector to obtain the vector of the current CAD model; based on the vector of the current CAD model, search for a CAD model cluster sample to obtain a target CAD model cluster that is most similar to the current CAD model and does not exceed a set number. class; the CAD model clustering sample includes: a plurality of CAD model clusters not exceeding a set number obtained after clustering the vectors of all CAD models in a database; taking the current CAD model as an input The sample input is a classification model corresponding to the target CAD model cluster, and the likelihood of each CAD model in the current CAD model and the target CAD model cluster is obtained; the classification model is configured by clustering the target CAD model. All CAD models in the class are trained as training samples; the target CAD model is obtained according to the likelihood of each CAD model in the current CAD model and the leaf node CAD model cluster.

A CAD model clustering and classification model generation device, including: a first module, used for extracting the geometric information of the CAD model for each CAD model in the database, and performing a structured representation of the geometric information, and The geometric information of the structured representation is converted into a vector of predefined length to obtain the vector of the CAD model; the second module is used to cluster the vectors of all CAD models in the database to obtain multiple vectors that do not exceed the set CAD model clusters of a certain number and size, so that when searching for CAD models, the search scope is positioned to a CAD model cluster that does not exceed the set number and size; the third module is used to cluster each CAD model, using the All CAD models in the CAD model cluster are trained as training samples to obtain a classification model corresponding to the CAD model cluster, so that when performing a CAD model search, the corresponding classification model is used to cluster CAD models that do not exceed the set number. CAD model search within classes.

A CAD model search device, including: a fourth module for receiving an input current CAD model, extracting geometric information of the current CAD model, and performing structured representation of the geometric information, and converting the structured expressed geometric information into Convert to a vector of predefined length to obtain the vector of the current CAD model; the fifth module is used to search for a CAD model clustering sample based on the vector of the current CAD model to obtain the most similar to the current CAD model A target CAD model clustering that does not exceed a set number; the CAD model clustering sample includes: a plurality of CAD models that do not exceed a set number and are obtained after clustering the vectors of all CAD models in a database Clustering; the sixth module is used to input the current CAD model as an input sample into a classification model corresponding to the target CAD model cluster, and obtain the current CAD model and each CAD in the target CAD model cluster. The likelihood of the model; the classification model is obtained by training all CAD models in the target CAD model cluster as training samples; the seventh module is used to cluster the current CAD model and the leaf node CAD model according to The likelihood of each CAD model in the class is calculated to obtain the target CAD model.

A computer device, including at least one memory and at least one processor, wherein: the at least one memory is used to store a computer program; the at least one processor is used to call the computer program stored in the at least one memory to execute any of the above The CAD model clustering and classification model generation method described in one embodiment, or the CAD model search method described in any embodiment.

A computer-readable storage medium with a computer program stored thereon; the computer program can be executed by a processor and implement the CAD model clustering and classification model generation method described in any of the above embodiments, or any implementation The CAD model search method described in the method.

It can be seen from the above technical solution that in this application, each CAD model in the database is converted into a vector, and then all CAD models in the database are clustered based on the vector, and multiple CAD models that do not exceed the set number are obtained. CAD model clustering allows the search scope to be targeted to a CAD model cluster that does not exceed the set number when searching for CAD models, thereby narrowing the search space and improving search efficiency; through cluster training for each CAD model A classification model is used, and the corresponding classification model is used to search for CAD models within the CAD model clusters that do not exceed the set number, thereby further accelerating the search speed and improving the search efficiency.

Brief description of the drawings

In order to better understand the present application, the embodiments of the present application will be described in detail below with reference to the accompanying drawings, so that the above and other features and advantages of the present application will be clearer to those of ordinary skill in the art. In the accompanying drawings:

Figure 1 is an exemplary flow chart of a CAD model clustering and classification model generation method in an embodiment of the present application.

2A and 2B are schematic diagrams of geometric information of a CAD model in an example of the present invention.

3A and 3B are respectively schematic diagrams of geometric information of a structured representation of a CAD model in an example of the present invention.

Figure 4 is a schematic diagram of re-clustering of clustered CAD models in an example of this application.

Figure 5 is a schematic diagram of a hierarchically indexed CAD model clustering sample in an example of this application.

Figure 6 is an exemplary flow chart of a CAD model search method in an embodiment of the present application.

Figure 7 is a flow chart of a method for searching and locating CAD model clusters in an embodiment of the present application.

Figure 8 is an exemplary structural diagram of a CAD model clustering and classification model generation device in an embodiment of the present invention.

Figure 9 is an exemplary structural diagram of a CAD model search device in an embodiment of the present invention.

Figure 10 is an exemplary structural diagram of a computer device in an embodiment of the present application.

Among them, the reference signs are as follows:

标号label	含义meaning
101～105、601～605、701～704101～105, 601～605, 701～704	步骤step
21twenty one	通孔Through hole
22、2322, 23	圆台Round table
24twenty four	通槽 slot
4141	聚类的中心点center point of cluster
4242	下一层聚类Next level clustering
4343	下一层聚类的中心点The center point of the next layer of clustering

801801	第一单元The first unit
802802	第二单元 Second unit
803803	第三单元 Unit 3
901901	第四单元 Unit 4
902902	第五单元 Unit 5
903903	第六单元 Unit 6
904904	第七单元 Unit 7
10011001	存储器 memory
10021002	处理器 processor
10031003	总线bus

How to implement this application

In the embodiment of the present application, in order to improve the efficiency of CAD model search, it is considered to provide a search solution that first clusters and then searches based on the classification model corresponding to each cluster, so as to reduce the search space and speed up the search process.

In order to make the purpose, technical solutions and advantages of the present application more clear, the technical solutions of the present application will be described in detail below with reference to the accompanying drawings and examples.

Figure 1 is an exemplary flow chart of a CAD model clustering and classification model generation method in an embodiment of the present application. As shown in Figure 1, the method may include the following steps:

Step 101: Extract the geometric information of each CAD model in the database.

In this step, the geometric information may include design features and boundary representation (B-rep) information. Among them, design features are basic shape descriptions that define the appearance of the CAD model. Typical design features may include bosses, holes, pockets, grooves, and notches, etc. Although different CAD model suppliers may have different definitions of design features, in this application example they can be converted through standard CAD formats, that is, the design features of CAD models from different providers are first converted based on standard CAD formats. Transform to obtain uniformly defined design features. B-rep information consists of the topological components of the CAD model, which can include faces, edges, and vertices.

2A and 2B show geometric information of a CAD model in an example. Among them, Figure 2A is a boundary representation information map, and Figure 2B is a geometric feature map. Labels ① to ⑨ in Fig. 2A represent boundary representation information in the CAD model, that is, each face. Reference numeral 21 in Figure 2B represents a through hole, reference numeral 22 represents a circular cone 1, reference numeral 23 represents a circular cone 2, and reference numeral 24 represents a through groove.

Step 102: Structurally represent the extracted geometric information of each CAD model.

In this step, the geometric information extracted in step 101 is organized into an appropriate form for subsequent calculation. Since the extracted CAD geometric information contains semantics, it can be organized into a graph-like form, that is, a topological graph structure as shown in Figure 3A or a tree structure as shown in Figure 3B. The above-mentioned methods of using structured representations such as topological graphs or trees only have different embedding methods, but they can all be counted as graphic structured representations.

Step 103: Convert the structured representation of geometric information into a vector of predefined length to obtain the vector of each CAD model.

In this step, the geometric information of the structured representation is converted into a vector of predefined length, for example, the geometric information of the structured representation is embedded to obtain the corresponding embedding vector. For the structured representation of topological graphs, topological graph embedding methods such as random walk and Node2Vec can be used to convert them into embedding vectors. For tree structured representation, tree embedding methods such as tree kernels and attention selection vector (ACV) trees can be used to convert them into embedding vectors. In this embodiment, both the topological graph embedding method and the tree embedding method will output a normalized vector v∈R ^d , where d is the predefined vector size (dimension), R represents a real number, and R ^d represents a d-dimensional real number vector. .

Step 104: Cluster the vectors of all CAD models in the database to obtain multiple CAD model clusters that do not exceed the set number, so that when searching for CAD models, the search range is positioned to one that does not exceed the set number. Quantity-sized CAD model clustering.

In this step, vectors of a large number of CAD models can be clustered in a hierarchical manner to obtain CAD model clusters that do not exceed a set number, thereby reducing the search space to a single cluster that does not exceed a predefined size.

In specific implementation, the vectors of a large number of CAD models can be clustered according to the distance by calculating the distance between the vectors of each CAD model in a high-dimensional (for example, greater than three-dimensional) space. Clustering methods can use K-means, Gaussian mixture model (GMM), density-based clustering method with noise (DBSCAN, Density-Based Spatial Clustering of Applications with Noise), etc.

In order to divide clusters that exceed the predefined size into smaller clusters to obtain sub-clusters that do not exceed the predefined size, in this embodiment, a hierarchical recursive method can be used for clustering, as shown in the following pseudo code :

It can be seen that the process expressed by the above pseudocode is: use the vectors of all CAD models in the database as the current CAD models to be clustered; cluster the current CAD models to be clustered according to the principle of not exceeding the maximum number of clusters, and obtain The current number of clusters; for each cluster in the current number of clusters, determine whether the size of the cluster exceeds the set number. If so, use the cluster that exceeds the set number as the current number. The CAD model to be clustered is returned to perform the operation of clustering the current CAD model to be clustered according to the principle of not exceeding the maximum number of clusters; otherwise, a cluster having a size not exceeding the set number is obtained.

Figure 4 shows a schematic diagram of re-clustering the clustered CAD models in an example. As shown in Figure 4, for each cluster, there is a CAD model 41 embedded in the position closest to the cluster center in space, and the next layer of clusters 42 divided by each cluster is the same and also has a closest location. CAD model of cluster centers43. In this embodiment, in order to facilitate retrieval of clusters that do not exceed the set number, a knowledge graph can be set for all clusters involved in the clustering process according to their inclusion relationships, which is called a hierarchical index here. CAD model clustering sample. And in order to reduce the amount of retrieval, the CAD model clustering at each layer can use the CAD model closest to the cluster center as the representative of the clustering layer. As shown in Figure 5, for the input IN, the vector of the CAD model of the center point of the first iterative clustering can be used as the sample of the first layer, recorded as SP, and the vector of the CAD model of the center point of the second iterative clustering can be used as the sample of the first layer. The vector is used as the sample of the second layer, recorded as SSP. The vector of the CAD model of the center point of the third iterative clustering is used as the sample of the third layer, recorded as S ² SP. The vector of the center point of the fourth iterative clustering is CAD. The vector of the model is used as the fourth layer sample, which is recorded as S ³ SP, and so on, so that the hierarchically indexed CAD model clustering sample shown in Figure 5 can be obtained. It can be seen that the hierarchical index CAD model clustering sample is a tree structure, including CAD model clustering as leaf nodes and CAD model clustering of parent nodes at each level located above the leaf nodes. Each parent node CAD model The cluster includes each of its sub-node CAD model clusters, and the samples of each CAD model cluster are represented by the vector of the CAD model located at the center point within the cluster. Further, as shown in Figure 5, each leaf node CAD model cluster is further associated with the vector CLV of all the CAD models it contains.

When retrieving, you can cluster samples based on the hierarchically indexed CAD model tree, starting from the uppermost parent node and retrieving the leaf nodes in sequence until the leaf node is retrieved. When retrieving the parent node, you only need to change the current CAD model Just compare with a CAD model that represents the CAD model cluster of the parent node, which greatly reduces the number of comparisons.

Clustering methods can be approximately exponential. For example, if you set "Maximum number of clusters" to 3, clustering is completely completed every iteration, assuming 6 iterations are completed. Then, the final output cluster size will be divided by 3 ⁶ , which is 729. As shown in Figure 5, assuming each output cluster size is 500 CAD models, the traditional search method may need to compare a total of 500 × 7 = 3500 models. Using the clustering method, the search process is retrieved from top to bottom based on the hierarchically indexed CAD model tree clustering samples shown in Figure 5, for example, the retrieval path is shown along the bold double-dot dash line marked box in Figure 5 , the retrieval route is shown as the bold double-dotted line in Figure 5. In this embodiment, only 7+500=507 times need to be compared.

Step 105: For each cluster, use all CAD models in the cluster as training samples to train to obtain a classification model corresponding to the cluster. In order to perform CAD model search, the corresponding classification model is used to perform CAD model search within the CAD model clusters that do not exceed the set number.

In this step, all CAD models in the cluster can be used as training samples for classification model training. For example, for each CAD model cluster, each CAD model in the CAD model cluster can be used as an input training sample, and the likelihood between the CAD model and each CAD model in the CAD model cluster is as The training samples are output to train an intelligent network to obtain a classification model corresponding to the CAD model clustering. There are many implementation methods for specific training, some of which are briefly listed below:

The first method: for each CAD model in the cluster, use screenshots of different views of the CAD model as a training data set, use each screenshot as an input sample in turn, and compare the screenshot with each of the sub-clusters. The likelihood between CAD models is used as an output sample to train an intelligent network such as a convolutional neural network, and a classification model corresponding to the clustering of the CAD models is obtained.

The second type: for each CAD model in the cluster, use videos of the CAD model including different perspectives as input training samples, and calculate the likelihood between the CAD model and each CAD model in the CAD model cluster. Degrees are used as output training samples to train an intelligent network such as a convolutional neural network, and obtain a classification model corresponding to the CAD model clustering.

The third method: for each CAD model in the cluster, use the vector of the CAD model as the input training sample, and the likelihood between the CAD model and each CAD model in the CAD model cluster as the output training Samples, train an intelligent network such as a convolutional neural network, and obtain a classification model corresponding to the CAD model clustering

Among them, convolutional neural networks can use VGG, MobileNet, ResNet, etc. In addition, in order to improve the robustness of the classification model, more data augmentation (Data Augmentation) methods can be used to enrich the training data set. For example, you can take screenshots of existing data such as CAD models, add random background images, or rotate screenshots to obtain new training data.

Further, the classification model corresponding to each CAD model cluster that does not exceed the predefined size can be further associated with the leaf node CAD model cluster of the CAD model cluster sample of the aforementioned hierarchical index as shown in the dotted line part in Figure 5 superior.

Figure 6 is an exemplary flow chart of a CAD model search method in an embodiment of the present application. As shown in Figure 6, the method may include the following steps:

Step 601: Receive the input current CAD model.

Step 602: Extract the geometric information of the current CAD model, perform a structured representation of the geometric information, and convert the structured geometric information into a vector of a predefined length to obtain a vector of the current CAD model.

The specific implementation process of this step is similar to steps 101 to 103, and will not be described again here.

Step 603: Based on the vector of the current CAD model, search for a CAD model cluster sample to obtain a target CAD model cluster that is most similar to the current CAD model and does not exceed a set number. Wherein, the CAD model clustering samples include: a plurality of CAD model clusters that do not exceed a set number and are obtained by clustering vectors of all CAD models in a database.

As mentioned above, in one embodiment, the CAD model clustering sample may be a hierarchically indexed CAD model tree clustering sample. The hierarchically indexed CAD model tree clustering sample includes the CAD model clusters that do not exceed a set number as leaf nodes and the CAD model clusters of parent nodes at each level located above the leaf nodes. Each Each parent node CAD model cluster includes the CAD model clusters of each of its child nodes, and the samples of each CAD model cluster are represented by the vector of the CAD model located at the center point within the cluster. Furthermore, each leaf node CAD model cluster is further associated with its corresponding classification model and the vectors of all CAD models it contains.

Correspondingly, in this step, the CAD model tree clustering samples of the hierarchical index can be searched layer by layer to obtain a leaf node CAD model cluster that is most similar to the current CAD model. This leaf node CAD model cluster is Clustering of target CAD models. The specific process of searching and locating CAD model clusters can be described in Figure 7, including the following processing:

Step 701: Compare the vector of the current CAD model with the uppermost CAD model samples in the hierarchically indexed CAD model tree clustering samples in order to determine the most similar CAD model sample.

Step 702: Determine whether the most similar CAD model sample is the last layer CAD model sample. If so, perform step 704; otherwise, perform step 703.

Step 703: Search each CAD model sample below the most similar CAD model sample, and determine the most similar CAD model sample, and then return to step 702.

Step 704: Determine that the most similar CAD model sample corresponds to a CAD model cluster that does not exceed a set number.

Step 604: Use the current CAD model as an input sample to input a classification model corresponding to the target CAD model cluster, and obtain the likelihood of each CAD model in the current CAD model and the target CAD model cluster. The classification model is trained by using all CAD models in the target CAD model cluster as training samples.

In this step, during specific implementation, for different training methods of the aforementioned classification model, there can also be multiple implementation methods in this step, which are described separately below.

The first method: Obtain screenshots of the current CAD model from different perspectives, use the screenshots from different perspectives as an input sample to input a classification model corresponding to the target CAD model clustering, and obtain the screenshot and the target CAD The likelihood of each CAD model in model clustering. For the first type, you can further post-process the likelihoods corresponding to screenshots from different perspectives of the current CAD model, such as the mathematical average of the likelihoods or the standard deviation of the likelihoods, etc., to obtain the final likelihood. The following pseudocode shows how this is done in an example:

Second method: Obtain videos of the current CAD model including different perspectives, use the video as an input sample to input a classification model corresponding to the target CAD model clustering, and obtain the current CAD model and the target CAD model. The likelihood of each CAD model in the cluster.

The third method: use the vector of the current CAD model as an input sample to input the classification model corresponding to the target CAD model cluster, and obtain the likelihood of each CAD model in the current CAD model and the target CAD model cluster. Spend.

Step 605: Obtain a target CAD model based on the likelihood of each CAD model in the current CAD model and the leaf node CAD model cluster.

The above is a detailed description of the CAD model clustering and classification model generation method and the CAD model search method in the embodiment of the present invention. Next, the CAD model clustering and classification model generation device and the CAD model search device in the embodiment of the present invention are described in detail. Describe in detail. The device embodiments in the embodiments of the present invention can be used to implement the corresponding method embodiments in the embodiments of the present invention. For details that are not disclosed in detail in the device embodiments of the present invention, please refer to the corresponding descriptions in the method embodiments of the present invention.

Figure 8 is an exemplary structural diagram of a CAD model clustering and classification model generation device in an embodiment of the present invention. As shown in Figure 8, it includes: a first module 801, a second module 802 and a third module 803.

Among them, the first module 801 is used to extract the geometric information of each CAD model in the database, perform a structured representation of the geometric information, and convert the structured represented geometric information into a predefined length. Vector, get the vector of the CAD model.

The second module 802 is used to cluster the vectors of all CAD models in the database to obtain multiple CAD model clusters that do not exceed a set number.

During specific implementation, the second module 802 can use the vectors of all CAD models in the database as the current CAD models to be clustered; cluster the current CAD models to be clustered according to the principle of not exceeding the maximum number of clusters, and obtain the current number of clusters; for each cluster in the current number of clusters, determine whether the size of the cluster exceeds the set number. If so, use the cluster that exceeds the set number as the current number. Cluster the CAD model, and return to perform the operation of clustering the current CAD model to be clustered according to the principle of not exceeding the maximum number of clusters; otherwise, obtain a cluster whose size does not exceed the set number. In addition, the second module 802 further uses the inclusion relationship between the clusters generated in each iteration of the clustering process to construct a hierarchical index CAD model tree clustering sample; the hierarchical index CAD model tree clustering The sample includes CAD model clusters as leaf nodes and CAD model clusters of parent nodes at all levels located above the leaf nodes. Each parent node CAD model cluster includes CAD model clusters of its child nodes as leaf nodes. The CAD model cluster is a CAD model cluster that does not exceed a set number of sizes, and the samples of each CAD model cluster are represented by the vector of the CAD model located at the center point within the cluster.

The third module 803 is configured to use all CAD models in the CAD model cluster as training samples for each CAD model cluster to train to obtain a classification model corresponding to the CAD model cluster. During specific implementation, the third module 803 can cluster each CAD model and use screenshots of different perspectives of each CAD model in the CAD model cluster as input training samples, and the screenshots are related to the CAD model. The likelihood between each CAD model in the cluster is used as the output training sample, and the classification model corresponding to the CAD model cluster is trained; or, for each CAD model cluster, each CAD model cluster within the CAD model cluster is used. Videos of a CAD model including different perspectives are used as input training samples, the likelihood between the CAD model and each CAD model in the CAD model cluster is used as an output training sample, and the classification corresponding to the CAD model cluster is obtained through training. model; or, for each CAD model cluster, use the vector of each CAD model in the CAD model cluster as an input training sample, and the similarity between the CAD model and each CAD model in the CAD model cluster is The probability is used as the output training sample, and a classification model corresponding to the CAD model clustering is obtained by training.

Figure 9 is an exemplary structural diagram of a CAD model search device in an embodiment of the present invention. As shown in Figure 3, the device may include: a fourth module 901, a fifth module 902, a sixth module 903, and a seventh module 904.

Among them, the fourth module 901 is used to receive the input current CAD model, extract the geometric information of the current CAD model, perform a structured representation of the geometric information, and convert the structured representation of the geometric information into a vector of predefined length. , get the vector of the current CAD model.

The fifth module 902 is used to search for a CAD model cluster sample based on the vector of the current CAD model, and obtain a target CAD model cluster that is most similar to the current CAD model and does not exceed a set number; the CAD Model clustering samples include: multiple CAD model clusters that do not exceed a set number and are obtained by clustering the vectors of all CAD models in a database. In one embodiment, the CAD model clustering sample may be a hierarchically indexed CAD model tree clustering sample; the hierarchically indexed CAD model tree clustering sample includes as leaf nodes the no more than A set number of CAD model clusters and CAD model clusters of parent nodes at all levels located above the leaf node. Each parent node CAD model cluster includes CAD model clusters of its child nodes. Each CAD model The samples of the cluster are represented by the vector of the CAD model located at the center point within the cluster.

The sixth module 903 is used to input the current CAD model as an input sample into a classification model corresponding to the target CAD model cluster, and obtain the similarity between the current CAD model and each CAD model in the target CAD model cluster. probability; the classification model is obtained by training all CAD models in the target CAD model cluster as training samples. During specific implementation, the sixth module 903 may obtain screenshots of the current CAD model from different perspectives, and use the screenshots from different perspectives as an input sample to input a classification model corresponding to the target CAD model clustering, to obtain the screenshots. and the likelihood of each CAD model in the target CAD model cluster; or also obtain the video of the current CAD model including different perspectives, and use the video as an input sample to input the corresponding target CAD model cluster Classify models to obtain the likelihood of each CAD model in the current CAD model and the target CAD model cluster; or you can also use the vector of the current CAD model as an input sample to input the corresponding target CAD model cluster. The classification model is used to obtain the likelihood of each CAD model in the clustering of the current CAD model and the target CAD model.

The seventh module 904 is used to obtain a target CAD model based on the likelihood of each CAD model in the current CAD model and the leaf node CAD model clustering.

In fact, the CAD model clustering and classification model generation device and the CAD model search device provided by this embodiment of the present application can be implemented in various ways. For example, by using an application programming interface that conforms to specific rules, the CAD model clustering and classification model generation device and the CAD model search device can be compiled into a plug-in installed in the smart terminal, or can be packaged into an application program for users to download. and use.

The CAD model clustering and classification model generation method and CAD model search method provided by this implementation of the present application can be stored in various storage media in an instruction storage mode or an instruction set storage mode. These storage media include but are not limited to: floppy disk, optical disk, DVD, hard disk, flash memory, USB flash memory, CF card, SD card, MMC card, SM card, memory stick and xD card.

In addition, the CAD model clustering and classification model generation method and CAD model search method provided by this embodiment of the present application can also be applied to flash memory (Nand-flash)-based storage media, such as USB flash drives, CF cards, SD card, SDHC card, MMC card, SM card, memory stick and xD card.

It should be clear that the operating system operating in the computer can not only implement the program code read by the computer from the storage medium, but also implement part or all of the actual operations by using instructions based on the program code to implement the above embodiments. function of any embodiment.

For example, FIG. 10 is an exemplary structural diagram of a computer device in an embodiment of the present application. The device can be used to perform the method shown in Figure 1 or Figure 6, or to implement the device of Figure 8 or Figure 9. As shown in Figure 10, the device may include at least one memory 1001 and at least one processor 1002. In addition, some other components can be included, such as communication ports, input/output controllers, network communication interfaces, etc. These components communicate via bus 1003 and so on.

At least one memory 1001 is used to store computer programs. In one example, the computer program can be understood as including various modules of the device shown in FIG. 8 or FIG. 9 . In addition, at least one memory 1001 may store an operating system and the like. Operating systems include but are not limited to: Android operating system, Symbian operating system, windows operating system, Linux operating system, etc.

At least one processor 1002 is used to call a computer program stored in at least one memory 1001 to execute the CAD model search method described in the example of this application. The processor 1002 can be a CPU, a processing unit/module, an ASIC, a logic module or a programmable gate array, etc., and it can receive and send data through a communication port.

The input/output controller has a display and an input device for inputting, outputting and displaying relevant data as a human-computer interaction module.

It will be understood that as used herein, "and/or" is intended to include any and all possible combinations of one or more of the associated listed items.

In addition, by using hierarchical indexed CAD tree model clustering samples, the most similar CAD model clusters can be quickly located, further improving the search efficiency of CAD models. And, since hierarchical clustering can reduce the size of each cluster almost exponentially, while the increase in hierarchies used for indexing is only linear, the present invention can support million/billion level CAD model searches, applicable Large-scale CAD model search engine.

Finally, the technical solution of the present invention can be implemented offline on the local server, so it has better security and privacy.

The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the present application. within the scope of protection.

Claims

A method for generating CAD model clustering and classification models, which is characterized by including:

For each CAD model in the database, the geometric information of the CAD model is extracted, the geometric information is structured, and the geometric information of the structured representation is converted into a vector of predefined length to obtain the CAD model's geometric information. vector;

The vectors of all CAD models in the database are clustered to obtain multiple CAD model clusters that do not exceed a set number, so that when searching for CAD models, the search range is positioned to a CAD model that does not exceed a set number. CAD model clustering;

For each CAD model cluster, all CAD models in the CAD model cluster are used as training samples to train to obtain a classification model corresponding to the CAD model cluster, so that when performing a CAD model search, the corresponding classification model is used to perform all operations. Describes the search for CAD models within CAD model clusters that do not exceed a set number of sizes.
The CAD model clustering and classification model generation method according to claim 1, wherein the vectors of all CAD models in the database are clustered to obtain a plurality of CAD model clusters not exceeding a set number. include:

Use the vectors of all CAD models in the database as the current CAD models to be clustered;

Cluster the CAD models currently to be clustered according to the principle of not exceeding the maximum number of clusters, and obtain the current number of clusters;

For each cluster in the current number of clusters, determine whether the size of the cluster exceeds the set number. If so, use the cluster exceeding the set number as the current CAD model to be clustered, And return to perform the operation of clustering the current CAD model to be clustered according to the principle of not exceeding the maximum number of clusters; otherwise, obtain a cluster whose size does not exceed the set number.
The CAD model clustering and classification model generation method according to claim 2, further comprising: constructing a hierarchical index CAD model tree using inclusion relationships between clusters generated in each iteration of the clustering process. shape clustering sample; the hierarchically indexed CAD model tree clustering sample includes CAD model clustering as a leaf node and CAD model clustering of parent nodes at all levels located above the leaf node, each parent node CAD Model clustering includes the CAD model clustering of each of its sub-nodes. The CAD model clustering as a leaf node is a CAD model clustering that does not exceed the set number. The samples of each CAD model clustering are located within the cluster. Vector representation of the center point of the CAD model.
The CAD model clustering and classification model generation method according to any one of claims 1 to 3, characterized in that, for each CAD model cluster, all CAD models within the CAD model cluster are used as The training samples are trained to obtain a classification model corresponding to the CAD model clustering, including:

For each CAD model cluster, screenshots of different perspectives of each CAD model in the CAD model cluster are used as input training samples, and the likelihood between the screenshots and each CAD model in the CAD model cluster is degree is used as an output training sample to train to obtain a classification model corresponding to the CAD model clustering; or include:

For each CAD model cluster, videos including different perspectives of each CAD model in the CAD model cluster are used as input training samples, and the similarity between the CAD model and each CAD model in the CAD model cluster is The probability is used as the output training sample to train to obtain a classification model corresponding to the CAD model clustering; or include:

For each CAD model cluster, the vector of each CAD model in the CAD model cluster is used as the input training sample, and the likelihood between the CAD model and each CAD model in the CAD model cluster is used as the output Training samples are used to train and obtain a classification model corresponding to the CAD model clustering.
A CAD model search method is characterized by including:

Receive the current CAD model of the input;

Extract the geometric information of the current CAD model, perform a structured representation of the geometric information, convert the structured geometric information into a vector of a predefined length, and obtain the vector of the current CAD model;

Based on the vector of the current CAD model, search for a CAD model cluster sample to obtain a target CAD model cluster that is most similar to the current CAD model and does not exceed a set number; the CAD model cluster sample includes: After clustering the vectors of all CAD models in a database, multiple CAD model clusters that do not exceed the set number are obtained;

Use the current CAD model as an input sample to input a classification model corresponding to the target CAD model cluster, and obtain the likelihood of each CAD model in the current CAD model and the target CAD model cluster; the classification The model is trained by using all CAD models in the target CAD model cluster as training samples;

According to the likelihood of each CAD model in the current CAD model and the leaf node CAD model cluster, a target CAD model is obtained.
The CAD model search method according to claim 5, wherein the CAD model clustering sample is a hierarchically indexed CAD model tree clustering sample; the hierarchical indexed CAD model tree clustering sample Including the CAD model clusters that do not exceed the set number as leaf nodes and the CAD model clusters of parent nodes at each level located above the leaf nodes. Each parent node CAD model cluster includes the CAD model clusters of its child nodes. CAD model clustering, the samples of each CAD model cluster are represented by the vector of the CAD model located at the center point within the cluster.
The CAD model search method according to claim 5 or 6, wherein the current CAD model is used as an input sample to input a classification model corresponding to the target CAD model cluster, and the current CAD model and the target CAD model are obtained. The likelihood of each CAD model in the target CAD model clustering includes:

Obtain screenshots of the current CAD model from different perspectives, use the screenshots from different perspectives as an input sample to input the classification model corresponding to the target CAD model clustering, and obtain the difference between the screenshot and the target CAD model clustering. The likelihood of each CAD model; or include:

Obtain videos of the current CAD model including different perspectives, use the video as an input sample to input a classification model corresponding to the target CAD model cluster, and obtain each of the current CAD model and the target CAD model cluster. Likelihood of the CAD model; alternatively including:

The vector of the current CAD model is used as an input sample to input the classification model corresponding to the target CAD model cluster, and the likelihood of each CAD model in the current CAD model and the target CAD model cluster is obtained.
A CAD model clustering and classification model generation device, which is characterized by including:

The first module is used to extract the geometric information of the CAD model for each CAD model in the database, perform a structured representation of the geometric information, and convert the structured representation of the geometric information into a vector of predefined length, Obtain the vector of the CAD model;

The second module is used to cluster the vectors of all CAD models in the database to obtain multiple CAD model clusters that do not exceed a set number, so that when searching for CAD models, the search range can be positioned to a specific location. Clustering of CAD models exceeding the set number size;

The third module is used for each CAD model cluster, using all CAD models in the CAD model cluster as training samples to train to obtain a classification model corresponding to the CAD model cluster, so that when searching for CAD models, use The corresponding classification model performs a search for CAD models within the CAD model clusters that do not exceed the set number.
A CAD model search device, characterized by including:

The fourth module is used to receive the input current CAD model, extract the geometric information of the current CAD model, perform a structured representation of the geometric information, and convert the structured representation of the geometric information into a vector of predefined length to obtain The vector of the current CAD model;

The fifth module is used to search for a CAD model clustering sample based on the vector of the current CAD model, and obtain a target CAD model cluster that is most similar to the current CAD model and does not exceed a set number; the CAD Model clustering samples include: multiple CAD model clusters that do not exceed the set number and size obtained after clustering the vectors of all CAD models in a database;

The sixth module is used to input the current CAD model as an input sample into a classification model corresponding to the target CAD model cluster, and obtain the similarity between the current CAD model and each CAD model in the target CAD model cluster. Probability; the classification model is obtained by training all CAD models in the target CAD model cluster as training samples;

The seventh module is used to obtain a target CAD model based on the likelihood of each CAD model in the current CAD model and the leaf node CAD model clustering.
A computer device, characterized by comprising at least one memory and at least one processor, wherein:

The at least one memory is used to store a computer program;

The at least one processor is configured to call a computer program stored in the at least one memory to execute the CAD model clustering and classification model generation method as claimed in any one of claims 1 to 4, or as claimed in claims 5 to 7 The CAD model search method described in any one of the above.
A computer-readable storage medium with a computer program stored thereon; characterized in that the computer program can be executed by a processor and implement the CAD model clustering and classification as claimed in any one of claims 1 to 4 A model generation method, or a CAD model search method according to any one of claims 5 to 7.