CN115205496B - Digital twin model light weight method and system - Google Patents
Digital twin model light weight method and system Download PDFInfo
- Publication number
- CN115205496B CN115205496B CN202210685167.0A CN202210685167A CN115205496B CN 115205496 B CN115205496 B CN 115205496B CN 202210685167 A CN202210685167 A CN 202210685167A CN 115205496 B CN115205496 B CN 115205496B
- Authority
- CN
- China
- Prior art keywords
- panel
- patch
- model
- combination
- merging
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000005516 engineering process Methods 0.000 claims abstract description 17
- 238000002372 labelling Methods 0.000 claims description 40
- 238000012545 processing Methods 0.000 claims description 32
- 238000012549 training Methods 0.000 claims description 31
- 239000013598 vector Substances 0.000 claims description 29
- 230000000694 effects Effects 0.000 claims description 26
- 238000009877 rendering Methods 0.000 claims description 26
- 230000007246 mechanism Effects 0.000 claims description 15
- 230000009191 jumping Effects 0.000 claims description 9
- 238000003062 neural network model Methods 0.000 claims description 7
- 238000013145 classification model Methods 0.000 claims description 5
- 230000003190 augmentative effect Effects 0.000 claims description 2
- 239000000463 material Substances 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 230000000295 complement effect Effects 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 208000016261 weight loss Diseases 0.000 description 1
- 239000013585 weight reducing agent Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/30—Polynomial surface description
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/353—Clustering; Classification into predefined classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2210/00—Indexing scheme for image generation or computer graphics
- G06T2210/36—Level of detail
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computer Graphics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Geometry (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Image Generation (AREA)
- Processing Or Creating Images (AREA)
Abstract
The application provides a digital twin model light weight method and a digital twin model light weight system. The method is based on the fact that the digital twin model subjected to preliminary light weight treatment by adopting a conventional light weight technology comprises the steps of constructing model input, judging whether two patches in the patch combination need to be combined or not based on the model input according to a trained patch combining model, and combining the patches needing to be combined.
Description
Technical Field
The application relates to a computer technology, in particular to a digital twin model light weight method and a digital twin model light weight system.
Background
Digital twin model technology has become a new approach and method for design, construction, management and maintenance, and has been fully applied in modeling field. In addition to the design department, subsequent related construction and operation and maintenance management around the design effort has also begun to greatly expand the related applications of digital twin model technology. However, digital twin models are enormous in volume and have high performance requirements for hardware devices. This requires a lightweight process for the digital twin model.
Currently, there are many methods for lightening the digital twin model, such as instrucing, compression, LOD, and parameterisation. The methods can perform light weight processing on the digital twin model so as to improve the rendering speed. However, the requirements of users on rendering speed and rendering effect are higher and higher, and the conventional light-weight mode cannot meet the rendering requirements.
Disclosure of Invention
In view of the above, the application discloses a digital twin model light weight method. The method may include: obtaining a digital twin model after preliminary light weight treatment, wherein the model comprises a plurality of patches formed by model vertexes, and a first patch set is formed by the plurality of patches; repeating the following A, B, C, D steps until the patches in the first patch set are traversed; a, selecting a first panel from the first panel set, and determining a second panel adjacent to the first panel in the first panel set, wherein the adjacent is that the first panel and the second panel have the same side; b, generating model input based on the vertex coordinates included in the first panel, the vertex coordinates included in the second panel and the included angle between the first plane where the first panel is positioned and the second plane where the second panel is positioned; c, based on the model input, obtaining a classification result according to a pre-trained patch combination model, wherein the classification result indicates whether to carry out combination operation on the first patch and the second patch, and the patch combination model comprises a neural network model obtained by training based on a plurality of patch combination samples marked with marking information whether to carry out combination; wherein, the method is marked as merging when the degree of influencing the rendering effect of the digital twin model after merging is lower than a threshold value, and marked as not merging when the degree of influencing the rendering effect of the digital twin model after merging reaches the threshold value; d, merging the first panel and the second panel to obtain a merged panel, storing the merged panel into a second panel set, deleting the first panel and the second panel in the first panel set, jumping to the step A, and storing the first panel into the second panel set, deleting the first panel in the first panel set, jumping to the step A, when the classification result indicates that the first panel and the second panel do not perform the merging operation; and generating a digital twin model after further light weight processing based on the patches in the second patch set.
In some embodiments, the determining a second panel of the first panel set that is adjacent to the first panel includes: and determining the other surface pieces including any two vertexes of three vertexes included in the first surface piece set as the second surface piece.
In some embodiments, the generating the model input based on the vertex coordinates included in the first panel, the vertex coordinates included in the second panel, and an angle between a first plane in which the first panel is located and a second plane in which the second panel is located includes: obtaining vertex coordinates of three vertices included in the first panel, and determining a first normal vector of a first plane where the first panel is located according to the obtained vertex coordinates; obtaining vertex coordinates of three vertexes included in the second panel, and determining a second normal vector of a second plane where the second panel is located according to the obtained vertex coordinates; determining an included angle between the first plane and the second plane according to the first normal vector and the second normal vector; and generating model input based on the vertex coordinates of the three vertices included in the first panel and the vertex coordinates of the three vertices included in the second panel according to a preset model input generation strategy.
In some embodiments, the patch merge model is a classification model constructed based on a text processing model; the generating a strategy according to a preset model input, based on the vertex coordinates of three vertices included in the first panel, the vertex coordinates of three vertices included in the second panel, the included angle, generating the model input, includes: the vertex coordinates of three vertexes included in the first panel are used as words, and a text sequence is generated according to a preset sequence; based on the model input, obtaining a classification result according to a pre-trained patch merging model, wherein the classification result comprises the following steps: processing the text sequence by using a self-attention mechanism to obtain text characteristics; the text feature is a feature useful for determining whether to merge the first and second panels; and classifying based on the text features to obtain the classification result.
In some embodiments, the processing the text sequence to obtain text features using a self-attention mechanism includes: each word in the text sequence is used as a target word, and the attention weight between the target word and other words is determined based on a self-attention mechanism; the attention weight is used for indicating the association relation and the association degree between the other words and the target word; based on the attention weight and the dot multiplication operation of the other words, obtaining word characteristics corresponding to the target word; and obtaining the text characteristics according to the word characteristics corresponding to each word in the text sequence.
In some embodiments, the text processing model comprises a bert model.
In some embodiments, the patch is composed of three model vertices; the method further comprises the steps of: obtaining a plurality of dough sheet combinations, wherein each dough sheet combination comprises two adjacent dough sheets, and two identical vertexes in the two dough sheets form adjacent edges of the two dough sheets; labeling each panel combination to obtain labeling information about whether two panels in each panel combination are combined or not so as to obtain a plurality of panel combination samples; for each patch combination, for at least one patch in the patch combination, equidistant points are taken on a target central line of the patch, each obtained point and the other two vertexes of the patch form a new patch, and the labeling information of the patch combination is determined as the labeling information of a new patch combination formed by the new patch and the other patch in the patch combination so as to expand a patch combination sample; the target midline is a line segment formed by vertexes outside adjacent edges and midpoints of the adjacent edges in the surface patch; and performing supervised training on the patch merging model based on the expanded patch combination samples.
In some embodiments, the supervised training of the patch merge model based on the augmented patch combination samples includes: constructing an input sample based on the patch combinations in the patch combination sample; inputting the input sample into the dough sheet combination model to obtain prediction information aiming at whether two dough sheets in the dough sheet combination are combined or not; obtaining loss information based on the prediction result, the labeling information included in the dough sheet combined sample and a preset loss function; and adjusting model parameters of the patch merging model based on the loss information.
The application also provides a digital twin model light-weight system, which comprises: the acquisition unit is used for acquiring a digital twin model after the detail level technology LOD preliminary light weight treatment, wherein the model comprises a plurality of patches formed by model vertexes, and a first patch set is formed by the plurality of patches; a merging unit, configured to repeatedly perform the following four steps A, B, C, D until the patches in the first patch set are traversed; a, selecting a first panel from the first panel set, and determining a second panel adjacent to the first panel in the first panel set, wherein the adjacent is that the first panel and the second panel have the same side; b, generating model input based on the vertex coordinates included in the first panel, the vertex coordinates included in the second panel and the included angle between the first plane where the first panel is positioned and the second plane where the second panel is positioned; c, based on the model input, obtaining a classification result according to a pre-trained patch combination model, wherein the classification result indicates whether to carry out combination operation on the first patch and the second patch, and the patch combination model comprises a neural network model obtained by training based on a plurality of patch combination samples marked with marking information whether to carry out combination; d, merging the first panel and the second panel to obtain a merged panel, storing the merged panel into a second panel set, deleting the first panel and the second panel in the first panel set, storing the first panel into the second panel set, deleting the first panel in the first panel set, and jumping to the step A when the classification result indicates that the first panel and the second panel do not perform the merging operation; and the generating unit is used for generating a digital twin model after further light weight processing based on the patches in the second patch set.
In some embodiments, the patch is composed of three model vertices; the system further comprises: the training unit is used for obtaining a plurality of dough sheet combinations, wherein each dough sheet combination comprises two adjacent dough sheets, and two identical vertexes in the two dough sheets form adjacent edges of the two dough sheets; labeling each panel combination to obtain labeling information about whether two panels in each panel combination are combined or not so as to obtain a plurality of panel combination samples; for each patch combination, for at least one patch in the patch combination, equidistant points are taken on a target central line of the patch, each obtained point and the other two vertexes of the patch form a new patch, and the labeling information of the patch combination is determined as the labeling information of a new patch combination formed by the new patch and the other patch in the patch combination so as to expand a patch combination sample; the target midline is a line segment formed by vertexes outside adjacent edges and midpoints of the adjacent edges in the surface patch; and performing supervised training on the patch merging model based on the expanded patch combination samples.
In the foregoing scheme, the model input can be constructed based on the combination of the patches included in the digital twin model after the preliminary lightweight processing by adopting the conventional lightweight technology, then, whether two patches in the patch combination need to be combined or not is judged based on the model input according to the trained patch combination model, and then, the patches needing to be combined are combined.
Drawings
The drawings that are required for use in the description of the embodiments or the related art will be briefly described below.
FIG. 1 is a schematic flow chart of a digital twin model light weight method according to an embodiment of the present application;
FIG. 2 is a schematic diagram of a dough sheet merging process according to an embodiment of the present application;
FIG. 3 is a flow chart of a method for generating model inputs according to an embodiment of the present application;
FIG. 4 is a flow chart of a method for generating model inputs according to an embodiment of the present application;
FIG. 5 is a schematic flow chart of a training method for a dough combining model according to an embodiment of the present application;
FIG. 6 is a schematic view of a dough sheet assembly according to an embodiment of the present application;
fig. 7 is a schematic structural diagram of a digital twin model lightweight system according to an embodiment of the present application.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the application. Rather, they are merely examples of apparatus and methods consistent with aspects of the application as detailed in the accompanying claims.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or all possible combinations of one or more of the associated listed items. It will also be appreciated that the term "if," as used herein, may be interpreted as "at … …" or "at … …" or "responsive to a determination," depending on the context.
The application provides a digital twin model light weight method. According to the technical scheme disclosed by the method, the patches included in the digital twin model subjected to preliminary light weight treatment by adopting a conventional light weight technology can be combined by judging the patches needing to be combined through the neural network, so that the number of the patches is further reduced, the rendering effect is not influenced, the effect of the digital twin model is further light, the rendering speed of the model is improved, the rendering effect of the digital twin model is also ensured, and the customer requirements are met.
The following description of the embodiments is made with reference to the accompanying drawings. Referring to fig. 1, fig. 1 is a flow chart of a digital twin model light-weight method according to an embodiment of the application.
The digital twin model light weight method shown in fig. 1 can be applied to electronic equipment. The electronic equipment can execute the method by carrying software logic corresponding to the digital twin model light weight method. The type of electronic device may be a notebook computer, a server, a mobile phone, a palm top computer (Personal Digital Assistant, PDA), etc. The type of the electronic device is not particularly limited in the present application. The electronic device may also be a client device or a server device.
As shown in FIG. 1, the method may include S102-S106. The order of execution of the steps is not particularly limited, except as specifically described.
S102, acquiring a digital twin model after preliminary light weight processing.
The digital twin model is a three-dimensional model obtained by modeling a specific object scene through a three-dimensional laser point cloud modeling technology and a visual inertia technology. For example, in an active distribution network scene, coarse mode information of a distribution network equipment live-action space can be circularly acquired through a three-dimensional laser point cloud modeling technology and a visual inertia technology (VIO), BIM series software is applied to perform hierarchical division and data writing on a point cloud model, and part of core equipment is subjected to fine modeling in a manual modeling mode to obtain the digital twin model.
After the digital twin model is obtained, the conventional light weight technology can be adopted to carry out preliminary light weight treatment. For example, the preliminary weight reduction treatment may be performed using at least one of the technologies of instrucing, compression, LOD, and parameterisation. Regarding the steps of the preliminary light weight processing, reference may be made to the related art, and the present application will not be described in detail.
And deleting some vertexes of the digital twin model in the process of preliminary light weight treatment, folding some patches formed by the vertexes, and the like, wherein after the treatment is finished, the digital twin model still comprises a plurality of patches formed by the vertexes of the model, and a plurality of patches form a first patch set. The present application distinguishes between different sets of patches by name, namely a first set of patches and a second set of patches. The dough sheet may include a preset number of vertices. For example, 3, 4 or 5. The present application is described by taking the example in which the dough sheet includes 3 vertices.
S104, repeatedly executing the following steps A, B, C, D until the patches in the first patch set are traversed; a, selecting a first panel from the first panel set, and determining a second panel adjacent to the first panel in the first panel set, wherein the adjacent is that the first panel and the second panel have the same side; b, generating model input based on the vertex coordinates included in the first panel, the vertex coordinates included in the second panel and the included angle between the first plane where the first panel is positioned and the second plane where the second panel is positioned; c, based on the model input, obtaining a classification result according to a pre-trained patch combination model, wherein the classification result indicates whether to carry out combination operation on the first patch and the second patch, and the patch combination model comprises a neural network model obtained by training based on a plurality of patch combination samples marked with marking information whether to carry out combination; wherein, the method is marked as merging when the degree of influencing the rendering effect of the digital twin model after merging is lower than a threshold value, and marked as not merging when the degree of influencing the rendering effect of the digital twin model after merging reaches the threshold value; and D, merging the first panel and the second panel to obtain a merged panel, storing the merged panel into a second panel set, deleting the first panel and the second panel in the first panel set, jumping to the step A, and storing the first panel into the second panel set, deleting the first panel in the first panel set, jumping to the step A, under the condition that the classification result indicates that the first panel and the second panel do not perform the merging operation.
S104 is described below with reference to fig. 2.
Referring to fig. 2, fig. 2 is a schematic diagram of a dough sheet merging process according to an embodiment of the present application.
S202-S204 correspond to step A.
S202, selecting a first panel from the first panel set.
The first panel may be selected in any manner in some embodiments. For example, randomly extracting, selecting the patch with the largest or smallest number, and so on.
S204, determining a second panel adjacent to the first panel.
The adjacent means that the first panel and the second panel have the same side. For example, if panel 1 and panel 2 include the same edge, then it is determined that panel 1 is adjacent to panel 2.
In some embodiments, the second panel may be determined from among the other panels included in the first panel set, the other panels including any two vertices among the three vertices included in the first panel. The second panel can thus be determined quickly.
For example, the first set of patches includes 1, 2, 3, 4, 5, and 6 patches. Wherein 1 is a first panel comprising three vertices abc. In S204, from 2 to 6 panels, a panel containing ab, bc or ac is selected as the second panel corresponding to panel No. 1.
In some embodiments, there may be multiple second panels in the first panel set adjacent to the first panel, i.e., a determination is made as to whether to merge with the first panel for each second panel, and the merge is completed when it is required.
S206 corresponds to step B.
S206, generating model input.
Referring to fig. 3, fig. 3 is a flowchart of a method for generating model input according to an embodiment of the application. The steps illustrated in fig. 3 are complementary descriptions to S206. As shown in FIG. 3, the method may include S302-S308. The order of execution of the steps is not particularly limited, except as specifically described.
S302, vertex coordinates of three vertices included in the first panel are obtained, and a first normal vector of a first plane where the first panel is located is determined according to the obtained vertex coordinates.
In the step, three vertexes can be selected arbitrarily from the first panel to form two groups of vectors, and then the first normal vector is obtained by using a vector cross multiplication method.
S304, obtaining the vertex coordinates of three vertexes included in the second panel, and determining a second normal vector of a second plane where the second panel is located according to the obtained vertex coordinates.
In this step, three vertexes can be selected arbitrarily from the first panel to form two sets of vectors, and then the second normal vector of the vertexes is utilized by a vector cross multiplication method.
S306, determining an included angle between the first plane and the second plane according to the first normal vector and the second normal vector.
The step can construct a related formula based on a cosine theorem, and the included angle is obtained based on two normal vectors which are intersected.
S308, generating a strategy according to a preset model input, and generating the model input based on the vertex coordinates of the three vertices included in the first panel and the vertex coordinates of the three vertices included in the second panel, and the included angles.
The model input generation strategy can be preset according to requirements. In some embodiments, the policy is related to the type of neural network employed by the patch merge model. For example, the neural network is a convolutional neural network, and the strategy is to convert the vertex coordinates of three vertices included in the first panel and the vertex coordinates of three vertices included in the second panel into vectors with fixed lengths (for example, 128) respectively, and then splice the vectors to obtain an image matrix to obtain the model input. For another example, the neural network is a text processing network, the measurement is that the vertex coordinates of three vertices included in the first panel, the vertex coordinates of three vertices included in the second panel, and the included angles are combined into a text sequence.
The model input covers some information related to judging whether to merge the first panel and the second panel, information extraction can be carried out by using the panel merging model to obtain related information, and judging whether to merge based on the extracted information.
Through S302-S308, model input covering some information related to the judgment of whether to merge the first panel and the second panel can be generated, so that the subsequent judgment of whether to merge is facilitated.
S208 corresponds to step C.
S208, obtaining a classification result by using the patch merging model.
And the classification result indicates whether to carry out merging operation on the first panel and the second panel, and the panel merging model comprises a neural network model which is obtained by training based on a plurality of panel combination samples marked with marking information of whether to carry out merging. Wherein the method is marked as merging when the degree of influencing the rendering effect of the digital twin model after merging is lower than a threshold value, and marked as not merging when the degree of influencing the rendering effect of the digital twin model after merging reaches the threshold value. Subsequent embodiments of the present application will describe the process of training the patch merge model, and will not be described in detail herein.
In some embodiments, the patch merge model is a classification model constructed based on a text processing model. The text processing model may include various types of models. For example, word2Vec series models, bert (Bidirectional Encoder Representations from Transformers, bi-directional encoder) series models. In some embodiments, the text processing model is a bert model, so that the self-attention mechanism can be utilized to pay attention to the association relation and the association degree between the vertexes of the first panel and the second panel and between the vertexes and the included angles in the model input, so that the panel merging model pays attention to the interested information more, the accuracy of panel merging judgment is improved, and the light-weight effect of the digital twin model is further improved.
In this example, in S206, the vertex coordinates of the three vertices included in the first panel and the vertex coordinates of the three vertices included in the second panel may be used as words, and the text sequence may be generated according to a preset sequence. The text sequence is the model input.
The sequence can be set according to the requirements. For example, the order of the included angle panels may be in accordance with three vertex coordinates included in the first panel, three vertex coordinates included in the second panel, or may be in accordance with an order in which a first panel vertex coordinate is followed by a second panel vertex coordinate so as to intersect. The present application is not particularly limited to the order.
Referring to fig. 4, fig. 4 is a flowchart of a method for generating model input according to an embodiment of the application. The steps illustrated in fig. 4 are complementary descriptions to S208. As shown in fig. 4, the method may include S402-S404. The order of execution of the steps is not particularly limited, except as specifically described.
S402, processing the text sequence by using a self-attention mechanism to obtain text characteristics.
The text feature is a feature useful for determining whether to merge the first and second panels. In addition to characterizing the sizes of the first and second panels, the text features may be labeled with many other beneficial implicit features, which may be interpreted by a panel merge model, provided that the size of the panels obtained after merging the first and second panels is larger than the conventional information that is useful in determining whether to merge the panels, so that the panel merge model may further merge the panels that have little effect on the rendering effect of the digital twin model, thereby achieving a further light-weight effect.
In some embodiments, each word in the text sequence may be considered a target word, and the attention weight between the target word and other words is determined based on a self-attention mechanism; the attention weight is used for indicating the association relation and the association degree between the other words and the target word. Then, based on the attention weight and the dot multiplication operation of the other words, word characteristics corresponding to the target word can be obtained; and then, obtaining the text features according to the word features corresponding to each word in the text sequence. Therefore, the method can enable the face merging model to pay more attention to interested information based on the association relation and association degree of each word and other words in the text sequence, namely according to a self-attention mechanism, accuracy of face merging judgment is improved, and further light-weight effect of the digital twin model is improved.
S404, classifying based on the text features to obtain the classification result.
In this step, the text feature may be input as an input into a two-classifier. The classifier can obtain the classification result through a softmax layer. And obtaining a judging result of whether to combine the first panel and the second panel.
S210-S218 correspond to step D described above.
S210, judging whether the classification result indicates that the first panel and the second panel need to be combined.
S212, when the first panel and the second panel need to be combined, the first panel and the second panel are combined to obtain a combined panel.
In some embodiments, the vertices included in the first and second panels may be sequentially connected directly to obtain a new polygon, i.e., a merged panel. For example, the first patch includes abc three vertices, the second patch includes bcd three vertices, and a quadrilateral (merged patch) of abcd is generated after merging.
And S214, storing the combined patches to the second patch set.
The second panel set is used for storing the panels after the light weight treatment is carried out on the panels in the first panel set, namely, the panels comprise combined panels, and the panels in the first panel set are not required to be combined.
S216, deleting the first panel and the second panel in the first panel set.
S218, deleting the first panel in the first panel set under the condition that the first panel and the second panel do not need to be combined.
After S216 or S218 is completed, S220 determines whether the first patch set further includes a patch, if yes, jumps to S202, otherwise ends the merging flow.
And S104 is realized through the steps of S202-S220, and the following steps A, B, C, D are repeatedly executed until the patches in the first patch set are traversed, so that further light weight of the digital twin model is realized.
S106, generating a digital twin model after further light weight processing based on the patches in the second patch set.
In some embodiments, the further lightweight digital twin model may be represented by patches in the second set of patches. And when the rendering is performed subsequently, rendering is performed based on the patches stored in the second patch set.
Through the scheme recorded in S102-S106, model input can be built based on the dough sheet combination included in the digital twin model after preliminary light weight processing by adopting a conventional light weight technology, then whether two dough sheets in the dough sheet combination need to be combined or not is judged based on the model input according to the trained dough sheet combining model, and then the dough sheets needing to be combined are combined.
An embodiment for training the patch merge model is described below.
In the supervised training technology, a large number of training samples with labeling information are required to be constructed, so that model parameters are adjusted according to errors between prediction information obtained by predicting a sample input model and real labeling information, and the model training effect is achieved. Often, a large number of training samples are constructed, so that manpower and material resources are consumed, and the model training cost is high.
In order to solve the problem, the application provides a data enhancement method for special scenes of the patch combination, so that after a small number of labels are carried out, the sample size is automatically expanded in a data enhancement mode, the consumption of manpower and material resources is reduced, and the model training cost is reduced.
Referring to fig. 5, fig. 5 is a flowchart of a training method for a dough combining model according to an embodiment of the present application. As shown in fig. 5, the method includes S502-S508. The order of execution of the steps is not particularly limited, except as specifically described. The patch illustrated in this example is made up of three model vertices.
S502, obtaining a plurality of dough sheet combinations.
The dough sheet combination comprises two adjacent dough sheets, and two identical vertexes in the two dough sheets form adjacent edges of the two dough sheets.
In some embodiments, a number of combinations of faces may be selected from existing three-dimensional models with adjacent faces vertices.
S504, labeling is carried out on each panel combination, and labeling information on whether two panels in each panel combination are combined is obtained, so that a plurality of panel combination samples are obtained.
In some embodiments, a manual labeling manner may be adopted to label the panel combinations, so as to obtain labeling information about whether two panels in each panel combination are combined, thereby obtaining a plurality of panel combination samples. The step only needs to consume a small amount of manpower and material resources.
It should be noted that when labeling, the manual experience may be used, that is, when the degree of influencing the rendering effect of the original three-dimensional model after the two patches are combined is lower than the threshold, that is, labeling the two patches may be combined, or when the degree of influencing the rendering effect of the original three-dimensional model after the two patches are combined reaches the threshold, that is, labeling the two patches may not be combined.
The threshold is set according to the requirements. The rendering effect of the model after combining the patches can be compared with the previous rendering effect according to the manual work, the influence degree of the combined patches on the rendering effect is obtained, and then the comparison is carried out with the threshold value, so that the labeling information is determined.
In some embodiments, the dough sheet combinations may be marked in an automatic marking manner, so as to obtain marking information about whether two dough sheets in each dough sheet combination are combined, thereby obtaining a plurality of dough sheet combination samples. The step does not need to consume a small amount of manpower and material resources.
The threshold is set according to the requirements. The method can utilize a pre-trained three-dimensional model recognition model to recognize a new three-dimensional model after merging the patches and a previous original three-dimensional model, determine the influence degree of merging the patches on the rendering effect through the change condition of the confidence coefficient of the two recognition results, and then compare the influence degree with the threshold value to determine the labeling information.
For example, the confidence that the three-dimensional model is the active distribution network equipment is 0.9 by using the three-dimensional model to identify the three-dimensional model of the active distribution network equipment. And (3) obtaining a new three-dimensional model after the surface patches are combined, and obtaining the confidence of the new three-dimensional model as active distribution network equipment by utilizing a three-dimensional model identification model to be 0.85. I.e. the extent of the influence of the merging patches on the rendering effect is 0.05. Assuming that the threshold is 0.06, it is determined that annotation is allowed. Assuming that the threshold is 0.03, it is determined that labeling is not allowed.
S506, for each patch combination, for at least one patch in the patch combination, equidistant points are taken on a target central line of the patch, each obtained point and the other two vertexes of the patch form a new patch, and the labeling information of the patch combination is determined as the labeling information of a new patch combination formed by the new patch and the other patch in the patch combination, so as to expand a patch combination sample.
The target midline refers to a line segment formed by a vertex which is positioned outside an adjacent side and a midpoint of the adjacent side in the surface piece.
Because the range of the new triangular patch is smaller than that of the original triangular patch, namely the merging relation of the new triangular patch and the other triangular patch is the same as that of the original triangular patch and the other triangular patch, the marking information of the patch combination can be determined as the marking information of the new patch combination formed by the new patch and the other patch in the patch combination, thereby achieving the purpose of automatically expanding the patch combination sample and reducing the loss of manpower and material resources.
Referring to fig. 6, fig. 6 is a schematic diagram of a dough sheet assembly according to an embodiment of the present application.
As shown in fig. 6, the dough sheet combination comprises two triangular dough sheets. One of the patches includes a vertex abc and the other patch includes a vertex bcd. Where bc is the adjacent edge of the two patches. e is the midpoint of bc and ae is the midline of one of the patches. Points are equidistantly taken on ae and then combined with bc to form a plurality of new triangular patches. The labeling information of the new triangular face piece and the other triangular face piece is used for the labeling information of the original triangular face piece and the other triangular face piece, so that the purpose of expanding the face piece combined sample is achieved, and the loss of manpower and material resources is reduced.
S508, performing supervised training on the patch merging model based on the expanded patch combination samples.
In some embodiments, S508 may repeat the following steps according to a preset number of iterations, completing the supervised training:
constructing an input sample based on the patch combinations in the patch combination sample; inputting the input sample into the dough sheet combination model to obtain prediction information aiming at whether two dough sheets in the dough sheet combination are combined or not; obtaining loss information based on the prediction result, the labeling information included in the dough sheet combined sample and a preset loss function; and adjusting model parameters of the patch merging model based on the loss information. The loss function may be MSE, cross entropy, etc., and is not particularly limited herein.
Through S502-S508, training of the patch merging model can be completed under the condition of consuming a small amount of manpower and material resources, so that the patch merging model has the capability of judging whether two patches need to be merged or not, namely, the merging is judged to be allowed under the condition that the rendering effect of the original model is not influenced after the merging.
The application further provides a digital twin model light-weight system. Referring to fig. 7, fig. 7 is a schematic structural diagram of a digital twin model light-weight system according to an embodiment of the present application.
As shown in fig. 7, a digital twin model lightweight system 700 (hereinafter referred to as a system) may include an acquisition unit 710, and a combining unit 720 connected to the acquisition unit 710, and a generating unit 730 connected to the combining unit 720.
It should be noted that, the functional units (including the acquisition unit, the merging unit and the generating unit) referred to in the present application include corresponding software functional logic and a hardware device for executing the software functional logic.
A digital twin model lightweight system 700, the system 700 comprising:
an obtaining unit 710, configured to obtain a digital twin model after performing a detail level technique LOD preliminary light weight processing, where the model includes a plurality of patches formed by model vertices, and a plurality of the patches form a first patch set;
a merging unit 720, configured to repeatedly perform the following four steps A, B, C, D until the patches in the first patch set are traversed;
a, selecting a first panel from the first panel set, and determining a second panel adjacent to the first panel in the first panel set, wherein the adjacent is that the first panel and the second panel have the same side;
B, generating model input based on the vertex coordinates included in the first panel, the vertex coordinates included in the second panel and the included angle between the first plane where the first panel is positioned and the second plane where the second panel is positioned;
c, based on the model input, obtaining a classification result according to a pre-trained patch combination model, wherein the classification result indicates whether to carry out combination operation on the first patch and the second patch, and the patch combination model comprises a neural network model obtained by training based on a plurality of patch combination samples marked with marking information whether to carry out combination;
d, merging the first panel and the second panel to obtain a merged panel, storing the merged panel into a second panel set, deleting the first panel and the second panel in the first panel set, storing the first panel into the second panel set, deleting the first panel in the first panel set, and jumping to the step A when the classification result indicates that the first panel and the second panel do not perform the merging operation;
And a generating unit 730, configured to generate a digital twin model after further light weight processing based on the patches in the second patch set.
In some embodiments, the merging unit 720 is further configured to:
and determining the other surface pieces including any two vertexes of three vertexes included in the first surface piece set as the second surface piece.
In some embodiments, the merging unit 720 is further configured to:
obtaining vertex coordinates of three vertices included in the first panel, and determining a first normal vector of a first plane where the first panel is located according to the obtained vertex coordinates;
obtaining vertex coordinates of three vertexes included in the second panel, and determining a second normal vector of a second plane where the second panel is located according to the obtained vertex coordinates;
determining an included angle between the first plane and the second plane according to the first normal vector and the second normal vector;
and generating model input based on the vertex coordinates of the three vertices included in the first panel and the vertex coordinates of the three vertices included in the second panel according to a preset model input generation strategy.
In some embodiments, the patch merge model is a classification model constructed based on a text processing model;
the merging unit 720 is further configured to:
the vertex coordinates of three vertexes included in the first panel are used as words, and a text sequence is generated according to a preset sequence;
based on the model input, obtaining a classification result according to a pre-trained patch merging model, wherein the classification result comprises the following steps:
processing the text sequence by using a self-attention mechanism to obtain text characteristics; the text feature is a feature useful for determining whether to merge the first and second panels;
and classifying based on the text features to obtain the classification result.
In some embodiments, the merging unit 720 is further configured to:
each word in the text sequence is used as a target word, and the attention weight between the target word and other words is determined based on a self-attention mechanism; the attention weight is used for indicating the association relation and the association degree between the other words and the target word;
Based on the attention weight and the dot multiplication operation of the other words, obtaining word characteristics corresponding to the target word;
and obtaining the text characteristics according to the word characteristics corresponding to each word in the text sequence.
In some embodiments, the text processing model comprises a bert model.
In some embodiments, the patch is composed of three model vertices; the system 700 further comprises:
the training unit is used for obtaining a plurality of dough sheet combinations, wherein each dough sheet combination comprises two adjacent dough sheets, and two identical vertexes in the two dough sheets form adjacent edges of the two dough sheets;
labeling each panel combination to obtain labeling information about whether two panels in each panel combination are combined or not so as to obtain a plurality of panel combination samples;
for each patch combination, for at least one patch in the patch combination, equidistant points are taken on a target central line of the patch, each obtained point and the other two vertexes of the patch form a new patch, and the labeling information of the patch combination is determined as the labeling information of a new patch combination formed by the new patch and the other patch in the patch combination so as to expand a patch combination sample; the target midline is a line segment formed by vertexes outside adjacent edges and midpoints of the adjacent edges in the surface patch;
And performing supervised training on the patch merging model based on the expanded patch combination samples.
In some embodiments, the training unit is further to:
constructing an input sample based on the patch combinations in the patch combination sample;
inputting the input sample into the dough sheet combination model to obtain prediction information aiming at whether two dough sheets in the dough sheet combination are combined or not;
obtaining loss information based on the prediction result, the labeling information included in the dough sheet combined sample and a preset loss function;
and adjusting model parameters of the patch merging model based on the loss information.
In the foregoing scheme, model input can be constructed based on a patch combination included in a digital twin model subjected to preliminary lightweight processing by adopting a conventional lightweight technology, then whether two patches in the patch combination need to be combined or not is judged based on the model input according to a trained patch combining model, and then the patches needing to be combined are combined.
One skilled in the relevant art will recognize that one or more embodiments of the application may be provided as a method, system, or computer program product. Accordingly, one or more embodiments of the application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, one or more embodiments of the application may take the form of a computer program product on one or more computer-usable storage media (which may include, but are not limited to, magnetic disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
"and/or" in the present application means having at least one of them. The embodiments of the present application are described in a progressive manner, and the same and similar parts of the embodiments are all referred to each other, and each embodiment is mainly described in the differences from the other embodiments. In particular, for data processing apparatus embodiments, the description is relatively simple, as it is substantially similar to method embodiments, with reference to the description of method embodiments in part.
While the application contains many specific implementation details, these should not be construed as limiting the scope of any disclosure or the scope of the claims, but rather as primarily describing features of particular embodiments of the particular disclosure. Certain features that are described in this application in the context of separate embodiments can also be implemented in combination in a single embodiment. On the other hand, the various features described in the individual embodiments may also be implemented separately in the various embodiments or in any suitable subcombination. Furthermore, although features may be acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.
Similarly, although operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In some cases, multitasking and parallel processing may be advantageous. Moreover, the separation of various system modules and components in the embodiments described should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.
The foregoing description of the preferred embodiment(s) of the application is merely illustrative of the presently preferred embodiment(s) of the application, and is not intended to limit the embodiment(s) of the application to the particular form disclosed, since various modifications, equivalent arrangements, improvements, etc., may be made within the spirit and scope of the embodiment(s) of the application.
Claims (6)
1. A method for lightening a digital twin model, the method comprising:
obtaining a digital twin model after preliminary light weight treatment, wherein the model comprises a plurality of patches formed by model vertexes, and a first patch set is formed by the plurality of patches;
Repeating the following A, B, C, D steps until the patches in the first patch set are traversed;
a, selecting a first panel from the first panel set, and determining a second panel adjacent to the first panel in the first panel set, wherein the adjacent is that the first panel and the second panel have the same side; the determining the second panel adjacent to the first panel in the first panel set includes determining, as the second panel, other panels including any two vertices among three vertices included in the first panel, among other panels included in the first panel set and excluding the first panel;
b, generating model input based on the vertex coordinates included in the first panel, the vertex coordinates included in the second panel and the included angle between the first plane where the first panel is positioned and the second plane where the second panel is positioned; the step B comprises the steps of obtaining vertex coordinates of three vertexes included in the first panel, and determining a first normal vector of a first plane where the first panel is located according to the obtained vertex coordinates; obtaining vertex coordinates of three vertexes included in the second panel, and determining a second normal vector of a second plane where the second panel is located according to the obtained vertex coordinates; determining an included angle between the first plane and the second plane according to the first normal vector and the second normal vector; generating a strategy according to a preset model input, and generating the model input based on the vertex coordinates of three vertexes included in the first panel, and the vertex coordinates of three vertexes included in the second panel, wherein the included angles are included;
C, based on the model input, obtaining a classification result according to a pre-trained patch combination model, wherein the classification result indicates whether to carry out combination operation on the first patch and the second patch, and the patch combination model comprises a neural network model obtained by training based on a plurality of patch combination samples marked with marking information whether to carry out combination; wherein, the method is marked as merging when the degree of influencing the rendering effect of the digital twin model after merging is lower than a threshold value, and marked as not merging when the degree of influencing the rendering effect of the digital twin model after merging reaches the threshold value; the patch merging model is a classification model constructed based on a text processing model; generating a strategy according to a preset model input, wherein the generating a model input comprises the steps of generating the model input by using the vertex coordinates of three vertexes included in the first panel and the included angles based on the vertex coordinates of three vertexes included in the second panel, wherein the included angles are respectively used as words and are generated into a text sequence according to a preset sequence; the classifying result is obtained according to a pre-trained patch merging model based on the model input, and the classifying result comprises the steps of processing the text sequence by using a self-attention mechanism to obtain text characteristics; the text feature is a feature useful for determining whether to merge the first and second panels; classifying based on the text features to obtain the classification result; the text sequence is processed by using a self-attention mechanism to obtain text characteristics, wherein each word in the text sequence is used as a target word, and attention weights between the target word and other words are determined based on the self-attention mechanism; the attention weight is used for indicating the association relation and the association degree between the other words and the target word; based on the attention weight and the dot multiplication operation of the other words, obtaining word characteristics corresponding to the target word; according to word characteristics corresponding to each word in the text sequence, obtaining the text characteristics;
D, merging the first panel and the second panel to obtain a merged panel, storing the merged panel into a second panel set, deleting the first panel and the second panel in the first panel set, jumping to the step A, and storing the first panel into the second panel set, deleting the first panel in the first panel set, jumping to the step A, when the classification result indicates that the first panel and the second panel do not perform the merging operation;
and generating a digital twin model after further light weight processing based on the patches in the second patch set.
2. The method of claim 1, wherein the text processing model comprises a bert model.
3. The method of claim 1, wherein the patch is comprised of three model vertices; the method further comprises the steps of:
obtaining a plurality of dough sheet combinations, wherein each dough sheet combination comprises two adjacent dough sheets, and two identical vertexes in the two dough sheets form adjacent edges of the two dough sheets;
Labeling each panel combination to obtain labeling information about whether two panels in each panel combination are combined or not so as to obtain a plurality of panel combination samples;
for each patch combination, for at least one patch in the patch combination, equidistant points are taken on a target central line of the patch, each obtained point and the other two vertexes of the patch form a new patch, and the labeling information of the patch combination is determined as the labeling information of a new patch combination formed by the new patch and the other patch in the patch combination so as to expand a patch combination sample; the target midline is a line segment formed by vertexes outside adjacent edges and midpoints of the adjacent edges in the surface patch;
and performing supervised training on the patch merging model based on the expanded patch combination samples.
4. The method of claim 3, wherein the supervised training of the patch merge model based on the augmented patch combination samples comprises:
constructing an input sample based on the patch combinations in the patch combination sample;
inputting the input sample into the dough sheet combination model to obtain prediction information aiming at whether two dough sheets in the dough sheet combination are combined or not;
Obtaining loss information based on the prediction information, the labeling information included in the dough sheet combined sample and a preset loss function;
and adjusting model parameters of the patch merging model based on the loss information.
5. A digital twin model lightweight system, the system comprising:
the acquisition unit is used for acquiring a digital twin model after the detail level technology LOD preliminary light weight treatment, wherein the model comprises a plurality of patches formed by model vertexes, and a first patch set is formed by the plurality of patches;
a merging unit, configured to repeatedly perform the following four steps A, B, C, D until the patches in the first patch set are traversed;
a, selecting a first panel from the first panel set, and determining a second panel adjacent to the first panel in the first panel set, wherein the adjacent is that the first panel and the second panel have the same side; the determining the second panel adjacent to the first panel in the first panel set includes determining, as the second panel, other panels including any two vertices among three vertices included in the first panel, among other panels included in the first panel set and excluding the first panel;
B, generating model input based on the vertex coordinates included in the first panel, the vertex coordinates included in the second panel and the included angle between the first plane where the first panel is positioned and the second plane where the second panel is positioned; the step B comprises the steps of obtaining vertex coordinates of three vertexes included in the first panel, and determining a first normal vector of a first plane where the first panel is located according to the obtained vertex coordinates; obtaining vertex coordinates of three vertexes included in the second panel, and determining a second normal vector of a second plane where the second panel is located according to the obtained vertex coordinates; determining an included angle between the first plane and the second plane according to the first normal vector and the second normal vector; generating a strategy according to a preset model input, and generating the model input based on the vertex coordinates of three vertexes included in the first panel, and the vertex coordinates of three vertexes included in the second panel, wherein the included angles are included;
c, based on the model input, obtaining a classification result according to a pre-trained patch combination model, wherein the classification result indicates whether to carry out combination operation on the first patch and the second patch, and the patch combination model comprises a neural network model obtained by training based on a plurality of patch combination samples marked with marking information whether to carry out combination; the patch merging model is a classification model constructed based on a text processing model; generating a strategy according to a preset model input, wherein the generating a model input comprises the steps of generating the model input by using the vertex coordinates of three vertexes included in the first panel and the included angles based on the vertex coordinates of three vertexes included in the second panel, wherein the included angles are respectively used as words and are generated into a text sequence according to a preset sequence; the classifying result is obtained according to a pre-trained patch merging model based on the model input, and the classifying result comprises the steps of processing the text sequence by using a self-attention mechanism to obtain text characteristics; the text feature is a feature useful for determining whether to merge the first and second panels; classifying based on the text features to obtain the classification result; the text sequence is processed by using a self-attention mechanism to obtain text characteristics, wherein each word in the text sequence is used as a target word, and attention weights between the target word and other words are determined based on the self-attention mechanism; the attention weight is used for indicating the association relation and the association degree between the other words and the target word; based on the attention weight and the dot multiplication operation of the other words, obtaining word characteristics corresponding to the target word; according to word characteristics corresponding to each word in the text sequence, obtaining the text characteristics;
D, merging the first panel and the second panel to obtain a merged panel, storing the merged panel into a second panel set, deleting the first panel and the second panel in the first panel set, storing the first panel into the second panel set, deleting the first panel in the first panel set, and jumping to the step A when the classification result indicates that the first panel and the second panel do not perform the merging operation;
and the generating unit is used for generating a digital twin model after further light weight processing based on the patches in the second patch set.
6. The system of claim 5, wherein the patch is comprised of three model vertices; the system further comprises:
the training unit is used for obtaining a plurality of dough sheet combinations, wherein each dough sheet combination comprises two adjacent dough sheets, and two identical vertexes in the two dough sheets form adjacent edges of the two dough sheets;
labeling each panel combination to obtain labeling information about whether two panels in each panel combination are combined or not so as to obtain a plurality of panel combination samples;
For each patch combination, for at least one patch in the patch combination, equidistant points are taken on a target central line of the patch, each obtained point and the other two vertexes of the patch form a new patch, and the labeling information of the patch combination is determined as the labeling information of a new patch combination formed by the new patch and the other patch in the patch combination so as to expand a patch combination sample; the target midline is a line segment formed by vertexes outside adjacent edges and midpoints of the adjacent edges in the surface patch;
and performing supervised training on the patch merging model based on the expanded patch combination samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210685167.0A CN115205496B (en) | 2022-06-14 | 2022-06-14 | Digital twin model light weight method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210685167.0A CN115205496B (en) | 2022-06-14 | 2022-06-14 | Digital twin model light weight method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115205496A CN115205496A (en) | 2022-10-18 |
| CN115205496B true CN115205496B (en) | 2023-10-24 |
Family
ID=83575479
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210685167.0A Active CN115205496B (en) | 2022-06-14 | 2022-06-14 | Digital twin model light weight method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115205496B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117557740B (en) * | 2024-01-10 | 2024-04-09 | 四川见山科技有限责任公司 | Three-dimensional model segmentation level switching method and device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112784130A (en) * | 2021-01-27 | 2021-05-11 | 杭州网易云音乐科技有限公司 | Twin network model training and measuring method, device, medium and equipment |
| CN113723070A (en) * | 2021-08-20 | 2021-11-30 | 上海浦东发展银行股份有限公司 | Text similarity model training method, text similarity detection method and text similarity detection device |
| CN113888716A (en) * | 2021-10-08 | 2022-01-04 | 广东工业大学 | Scene lightweight method for digital twinning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10430548B2 (en) * | 2015-11-30 | 2019-10-01 | Airbus Operations Gmbh | Computer-implemented method for space frame design, space frame construction kit and space frame |
-
2022
- 2022-06-14 CN CN202210685167.0A patent/CN115205496B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112784130A (en) * | 2021-01-27 | 2021-05-11 | 杭州网易云音乐科技有限公司 | Twin network model training and measuring method, device, medium and equipment |
| CN113723070A (en) * | 2021-08-20 | 2021-11-30 | 上海浦东发展银行股份有限公司 | Text similarity model training method, text similarity detection method and text similarity detection device |
| CN113888716A (en) * | 2021-10-08 | 2022-01-04 | 广东工业大学 | Scene lightweight method for digital twinning |
Non-Patent Citations (1)
| Title |
|---|
| 复杂产品三维模型轻量化服务构建技术;薛俊杰等;系统仿真学报(第04期);第29-37页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115205496A (en) | 2022-10-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Lin et al. | Convolution in the cloud: Learning deformable kernels in 3d graph convolution networks for point cloud analysis | |
| Zhi et al. | Toward real-time 3D object recognition: A lightweight volumetric CNN framework using multitask learning | |
| US10176404B2 (en) | Recognition of a 3D modeled object from a 2D image | |
| US11823391B2 (en) | Utilizing soft classifications to select input parameters for segmentation algorithms and identify segments of three-dimensional digital models | |
| CN116258719B (en) | Flotation foam image segmentation method and device based on multi-mode data fusion | |
| CN108595558B (en) | Image annotation method based on data equalization strategy and multi-feature fusion | |
| CN111368636A (en) | Object classification method and device, computer equipment and storage medium | |
| CN107545025A (en) | Database is inquired about using morphological criteria | |
| CN114926469A (en) | Semantic segmentation model training method, semantic segmentation method, storage medium and terminal | |
| He et al. | Pointinst3d: Segmenting 3d instances by points | |
| CN115205496B (en) | Digital twin model light weight method and system | |
| CN109241442B (en) | Project recommendation method based on predictive value filling, readable storage medium and terminal | |
| CN117237808A (en) | Remote sensing image target detection method and system based on ODC-YOLO network | |
| CN108537887A (en) | Sketch based on 3D printing and model library 3-D view matching process | |
| CN110400370B (en) | Method for constructing semantic-level component template of three-dimensional CAD model | |
| Yang et al. | Xception-based general forensic method on small-size images | |
| CN115482387A (en) | Weak supervision image semantic segmentation method and system based on multi-scale class prototype | |
| CN110717405A (en) | Face feature point positioning method, device, medium and electronic equipment | |
| EP2852918B1 (en) | Method and apparatus for generating shape descriptor of a model | |
| CN116912486A (en) | Target segmentation method based on edge convolution and multidimensional feature fusion and electronic device | |
| KR102129060B1 (en) | Content-based 3d model retrieval method using a single depth image, 3d model retrieval server for performing the methods and computer readable recording medium thereof | |
| CN112597329B (en) | Real-time image retrieval method based on improved semantic segmentation network | |
| Flagg et al. | Searching document repositories using 3D model reconstruction | |
| CN103530656B (en) | Hidden structure learning-based image digest generation method | |
| JP6355372B2 (en) | 3D model feature extraction method and 3D model annotation system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |