CN109255791A - A kind of shape collaboration dividing method based on figure convolutional neural networks - Google Patents
A kind of shape collaboration dividing method based on figure convolutional neural networks Download PDFInfo
- Publication number
- CN109255791A CN109255791A CN201810797764.6A CN201810797764A CN109255791A CN 109255791 A CN109255791 A CN 109255791A CN 201810797764 A CN201810797764 A CN 201810797764A CN 109255791 A CN109255791 A CN 109255791A
- Authority
- CN
- China
- Prior art keywords
- pieces
- sub
- shape
- distance
- node
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T19/00—Manipulating 3D models or images for computer graphics
- G06T19/20—Editing of 3D images, e.g. changing shapes or colours, aligning objects or positioning parts
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2219/00—Indexing scheme for manipulating 3D models or images for computer graphics
- G06T2219/20—Indexing scheme for editing of 3D models
- G06T2219/2021—Shape modification
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Architecture (AREA)
- Computer Graphics (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The shape based on figure convolutional network that the invention discloses a kind of cooperates with dividing method, and the method for the invention includes: one group of given shape to be excessively cut into sub-pieces, and construct the relationship graph model between sub-pieces;It is specified that label is carried out to part sub-pieces therein;The sub-pieces label information marked is propagated to and other is not marked in sub-pieces by structure figures convolutional network.Picture scroll product network application is cooperateed with segmentation field in shape by the present invention, and compared to current other methods, the present invention can obtain the higher result of accuracy rate.
Description
Technical field
The present invention relates to the modeling of the geometry of graphics and analytical technologies, can be widely used in three-dimensional with for technical foundation
The fields such as game, modeling, emulation.
Background technique
Shape segmentations, which refer to, is cut into one group of limited amount for shape, respectively has simple semantic sub- shape.The technology
It can be widely used in every field of graphics, such as 3d gaming, modeling, emulation, the pattern-recognition of model etc..The work of early stage
It is concentrated mainly in the segmentation to single shape, but efficiency is lower, there are many scholars to study recently while dividing one group of shape,
And establish the corresponding relationship between them, it may be assumed that shape collaboration segmentation.It can effectively assist solving the problems, such as many shape processing, such as
Modeling, model index and texture mapping etc..
Currently, for this problem, there are the methods of unsupervised approaches and supervision.Unsupervised method can be automatic, efficient
Collaboration segmentation carried out to shape, however its result is dependent on given data set, i.e., to different data sets, collaboration segmentation
As a result accuracy rate can have very big difference.On the other hand, the collaboration segmentation result based on measure of supervision is depended on and has been marked
Training set, give enough training sets, it can obtain the very high segmentation result of accuracy rate.However the shortcomings that such methods, exists
In first it also rely on given training dataset, secondly its to the time complexity of the training process of labeled data very
It is high.
Summary of the invention
The present invention designs a kind of novel collaboration parted pattern method based on figure convolutional neural networks, can dig to abundant
Graph model data are dug, semi-supervised model is analyzed.It mainly comprises the processes of
Step 1: shape over-segmentation
Segmentation is normalized in given threedimensional model set;Each threedimensional model is divided into 30 sub-pieces.
Step 2: graph model building
Feature extraction is carried out to each sub-pieces of Models Sets;Respectively extract shape diameter function, conformal factor, in shape under
Text, average geodesic distance and the geodesic distance to bottom;Draft piIndicate i-th of sub-pieces, hk,iIndicate k-th of spy of the sub-pieces
Sign description;All features of the sub-pieces are attached, the feature for forming the sub-pieces describes xi=[h1,i,h2,i,...,h5,i];
For any two sub-pieces piAnd pj, the similarity distance d (p between them is calculated using EMD distancei,pj)=EMD (xi,xj), it obtains
It is defined to initial distance;By applying Gaussian kernel, final two sub-pieces p are obtainediAnd pjDistance definition ai,j=exp (- d (pi,
pj)/2σ2);The root Ju distance defines graph model G=(V, E).The node of figure indicates each sub-pieces, and is described with its feature
The weight of the attribute of the sub-pieces, the side of figure is defined using the distance of two sub-pieces;The distance between corresponding sub-pieces constitutes phase
Like matrix W.
Step 3: figure convolutional neural networks
The given graph model of root Ju, defines two layers of picture scroll product network model, and the structure of each layer of figure is remained unchanged, saved
Point and the raw dynamic of making of the category on side adjust;Drafting the output of the last one node is Z.Each layer of neural network be described as one it is non-linear
Function H(l+1)=f (Hl, A), wherein H (0)=X, H (l)=Z;By f () functional expansion, the structure of neural network is defined are as follows:WhereinFor the adjacency matrix of figure, I is unit matrix.WlFor the weight of every layer matrix.φ () is defined as activation primitive, common activation primitive such as RELU definition
Are as follows: RELU ()=max (0);
Two layers of figure convolutional network is unfolded, the output of neural network is obtained are as follows:
Wherein W(0)For the weight for being input to hidden layer
Matrix, WlFor hidden layer to the matrix of output.WhereinError between the loss function predicted value and a reference value of definition:
Wherein YLFor the index of label.
Step 4: shape segmentations result
It is that root Ju step 3 is predicted as a result, the label information of each node is reflected into shape, obtain shape
Segmentation and semantic results.
The present invention is possessed compared with the existing technology the utility model has the advantages that the present invention can support semi-supervised shape cooperates with to divide
It cuts, i.e., only needs less markup information, accurate can predict the other semantic informations for not marking part of threedimensional model.
Compared to other methods, in the case where only needing a small number of markup informations, the precision for obtaining result is higher for it.
Detailed description of the invention
Fig. 1 inventive network block flow diagram;
Fig. 2 model is divided into different sub-pieces;
The product network structure signal of Fig. 3 picture scroll;
Fig. 4 model interoperability segmentation result example.
Specific embodiment
Clear, complete description is carried out to technical solution of the present invention by the way that example is embodied in conjunction with attachment.
1. network structure
As shown in Figure 1, system is broadly divided into three steps, first is excessively cut into shape 30 sub-pieces.Then pass through spy
Sign is extracted, and constructs a graph model to these sub-pieces.Finally by picture scroll product network model, study obtains other models on figure
Segmentation result.
2. model over-segmentation is illustrated
Fig. 2 shows that we carry out over-segmentation to model, and system carries out cutting, each model to model using normalization segmentation
We are cut into 30 sub-pieces.Their boundary and characteristic curve are almost the same.After converting over-segmentation for segmentation problem
The clustering problem of sub-pieces.
3. network exports explanation
Fig. 3 shows our picture scroll product network architecture.The model of system is broadly divided into two layers.Graph structure is as defeated
Emit each node after entering into network model and the characteristic information of itself is sent to neighbor node after transformation.This step
It is to carry out extraction transformation in the characteristic information to node.Then each node is received to assemble the characteristic information of neighbor node
Come.This step is merged in the partial structurtes information to node.The information aggregation of front is done later finally by transformation
Nonlinear transformation increases the ability to express of model.
The figure convolutional neural networks of design have following property as common convolutional neural networks: 1), local parameter it is shared,
Operator is to be suitable for each node, is shared everywhere.2), receptive field is proportional to the number of plies, most at the beginning of, each node contains
The information of immediate neighbor, then while calculating the second layer, can be included the information of the neighbours of neighbours, participate in the letter of operation in this way
Breath is just more more abundant.The number of plies is more, and receptive field is just wider, participates in the information of operation with regard to more.
GCN model is likewise supplied with three kinds of properties of deep learning, 1), (feature extracts hierarchical structure in layer, one layer of ratio
One layer more abstract, more advanced);2), nonlinear transformation (ability to express for increasing model);3), end-to-end training (does not need to go again
Define any rule, it is only necessary to node one label of figure, allow model oneself to learn, fusion feature information and structural information.)
By the study and transmission to node, it can carry out end-to-end to node diagnostic information and structural information simultaneously
It practises.
4. model result example
Fig. 4 illustrates the segmentation result using this method on some data sets of this model, identical semantic segmentation portion
Us are divided to indicate using identical color.
Claims (1)
1. a kind of shape based on figure convolutional network cooperates with dividing method, which is characterized in that this method specifically includes the following steps:
Step 1: shape over-segmentation
Segmentation is normalized in given threedimensional model set;Each threedimensional model is divided into 30 sub-pieces;
Step 2: graph model building
Feature extraction is carried out to each sub-pieces of Models Sets;Respectively extract shape diameter function, conformal factor, Shape context,
Average geodesic distance and the geodesic distance to bottom;Draft piIndicate i-th of sub-pieces, hk,iIndicate that k-th of feature of the sub-pieces is retouched
It states;All features of the sub-pieces are attached, the feature for forming the sub-pieces describes xi=[h1,i,h2,i,...,h5,i];For
Any two sub-pieces piAnd pj, the similarity distance d (p between them is calculated using EMD distancei,pj)=EMD (xi,xj), it obtains just
Beginning distance definition;By applying Gaussian kernel, final two sub-pieces p are obtainediAnd pjDistance definition ai,j=exp (- d (pi,pj)/2
σ2);The root Ju distance defines graph model G=(V, E);The node of figure indicates each sub-pieces, and describes the son with its feature
The weight of the attribute of piece, the side of figure is defined using the distance of two sub-pieces;The distance between corresponding sub-pieces constitutes similar square
Battle array W;
Step 3: figure convolutional neural networks
The given graph model of root Ju defines two layers of picture scroll product network model, and the structure of each layer of figure remains unchanged, node and
The raw dynamic of making of the category on side adjusts;Drafting the output of the last one node is Z;Each layer of neural network is described as a nonlinear function
H(l+1)=f (Hl, A), wherein H (0)=X, H (l)=Z;By f () functional expansion, the structure of neural network is defined are as follows:WhereinFor the adjacency matrix of figure, I is unit matrix;WlFor the weight of every layer matrix;φ () is defined as activation primitive, common activation primitive such as RELU definition
Are as follows: RELU ()=max (0);
Two layers of figure convolutional network is unfolded, the output of neural network is obtained are as follows:Wherein W(0)For the weight matrix for being input to hidden layer, WlFor
The matrix of hidden layer extremely output; WhereinError between the loss function predicted value and a reference value of definition:Its
Middle YLFor the index of label;
Step 4: shape segmentations result
It is that root Ju step 3 is predicted as a result, the label information of each node to be reflected into the segmentation for obtaining shape in shape
And semantic results.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810797764.6A CN109255791A (en) | 2018-07-19 | 2018-07-19 | A kind of shape collaboration dividing method based on figure convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810797764.6A CN109255791A (en) | 2018-07-19 | 2018-07-19 | A kind of shape collaboration dividing method based on figure convolutional neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109255791A true CN109255791A (en) | 2019-01-22 |
Family
ID=65048967
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810797764.6A Pending CN109255791A (en) | 2018-07-19 | 2018-07-19 | A kind of shape collaboration dividing method based on figure convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109255791A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109934826A (en) * | 2019-02-28 | 2019-06-25 | 东南大学 | A kind of characteristics of image dividing method based on figure convolutional network |
CN110210330A (en) * | 2019-05-13 | 2019-09-06 | 清华大学 | Electromagnetic signal recognition methods and device based on Tacit Knowledge structure figures convolutional network |
CN110210487A (en) * | 2019-05-30 | 2019-09-06 | 上海商汤智能科技有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
CN110211134A (en) * | 2019-05-30 | 2019-09-06 | 上海商汤智能科技有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
CN110276777A (en) * | 2019-06-26 | 2019-09-24 | 山东浪潮人工智能研究院有限公司 | A kind of image partition method and device based on depth map study |
CN110288026A (en) * | 2019-06-27 | 2019-09-27 | 山东浪潮人工智能研究院有限公司 | A kind of image partition method and device practised based on metric relation graphics |
CN110334704A (en) * | 2019-06-21 | 2019-10-15 | 浙江大学宁波理工学院 | Threedimensional model interest point extraction method and system based on Layered Learning |
CN110427835A (en) * | 2019-07-11 | 2019-11-08 | 清华大学 | The electromagnet signal recognition method and device of figure convolutional network and transfer learning |
CN110674829A (en) * | 2019-09-26 | 2020-01-10 | 哈尔滨工程大学 | Three-dimensional target detection method based on graph convolution attention network |
CN110705613A (en) * | 2019-09-19 | 2020-01-17 | 创新奇智(青岛)科技有限公司 | Object classification method |
CN110934587A (en) * | 2019-12-13 | 2020-03-31 | 深圳龙岗智能视听研究院 | Alzheimer disease auxiliary diagnosis method based on atlas neural network |
CN112634281A (en) * | 2020-12-10 | 2021-04-09 | 浙江大学 | Grid segmentation method based on graph convolution network |
WO2021079233A1 (en) * | 2019-10-23 | 2021-04-29 | International Business Machines Corporation | New framework for few-shot temporal action localization |
CN116778027A (en) * | 2023-08-22 | 2023-09-19 | 中国空气动力研究与发展中心计算空气动力研究所 | Curved surface parameterization method and device based on neural network |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102915451A (en) * | 2012-10-18 | 2013-02-06 | 上海交通大学 | Dynamic texture identification method based on chaos invariant |
CN103413310A (en) * | 2013-08-15 | 2013-11-27 | 中国科学院深圳先进技术研究院 | Collaborative segmentation method and device |
CN103699789A (en) * | 2013-12-17 | 2014-04-02 | 中国科学院深圳先进技术研究院 | Method and device for shape correspondence |
CN103700094A (en) * | 2013-12-09 | 2014-04-02 | 中国科学院深圳先进技术研究院 | Interactive collaborative shape segmentation method and device based on label propagation |
-
2018
- 2018-07-19 CN CN201810797764.6A patent/CN109255791A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102915451A (en) * | 2012-10-18 | 2013-02-06 | 上海交通大学 | Dynamic texture identification method based on chaos invariant |
CN103413310A (en) * | 2013-08-15 | 2013-11-27 | 中国科学院深圳先进技术研究院 | Collaborative segmentation method and device |
CN103700094A (en) * | 2013-12-09 | 2014-04-02 | 中国科学院深圳先进技术研究院 | Interactive collaborative shape segmentation method and device based on label propagation |
CN103699789A (en) * | 2013-12-17 | 2014-04-02 | 中国科学院深圳先进技术研究院 | Method and device for shape correspondence |
Non-Patent Citations (1)
Title |
---|
THOMAS N. KIPF E.T.: "Semi-supervised classification with graph convolutional networks", 《ICLR 2017》 * |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109934826B (en) * | 2019-02-28 | 2023-05-12 | 东南大学 | Image feature segmentation method based on graph convolution network |
CN109934826A (en) * | 2019-02-28 | 2019-06-25 | 东南大学 | A kind of characteristics of image dividing method based on figure convolutional network |
CN110210330A (en) * | 2019-05-13 | 2019-09-06 | 清华大学 | Electromagnetic signal recognition methods and device based on Tacit Knowledge structure figures convolutional network |
CN110210487A (en) * | 2019-05-30 | 2019-09-06 | 上海商汤智能科技有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
CN110211134A (en) * | 2019-05-30 | 2019-09-06 | 上海商汤智能科技有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
CN110334704B (en) * | 2019-06-21 | 2022-10-21 | 浙江大学宁波理工学院 | Three-dimensional model interest point extraction method and system based on layered learning |
CN110334704A (en) * | 2019-06-21 | 2019-10-15 | 浙江大学宁波理工学院 | Threedimensional model interest point extraction method and system based on Layered Learning |
CN110276777A (en) * | 2019-06-26 | 2019-09-24 | 山东浪潮人工智能研究院有限公司 | A kind of image partition method and device based on depth map study |
CN110276777B (en) * | 2019-06-26 | 2022-03-22 | 山东浪潮科学研究院有限公司 | Image segmentation method and device based on depth map learning |
CN110288026A (en) * | 2019-06-27 | 2019-09-27 | 山东浪潮人工智能研究院有限公司 | A kind of image partition method and device practised based on metric relation graphics |
CN110288026B (en) * | 2019-06-27 | 2021-08-10 | 山东浪潮科学研究院有限公司 | Image segmentation method and device based on metric relation graph learning |
CN110427835A (en) * | 2019-07-11 | 2019-11-08 | 清华大学 | The electromagnet signal recognition method and device of figure convolutional network and transfer learning |
CN110427835B (en) * | 2019-07-11 | 2022-03-11 | 清华大学 | Electromagnetic signal identification method and device for graph convolution network and transfer learning |
CN110705613A (en) * | 2019-09-19 | 2020-01-17 | 创新奇智(青岛)科技有限公司 | Object classification method |
CN110705613B (en) * | 2019-09-19 | 2021-06-11 | 创新奇智(青岛)科技有限公司 | Object classification method |
CN110674829A (en) * | 2019-09-26 | 2020-01-10 | 哈尔滨工程大学 | Three-dimensional target detection method based on graph convolution attention network |
WO2021079233A1 (en) * | 2019-10-23 | 2021-04-29 | International Business Machines Corporation | New framework for few-shot temporal action localization |
US11164039B2 (en) | 2019-10-23 | 2021-11-02 | International Business Machines Corporation | Framework for few-shot temporal action localization |
GB2604071A (en) * | 2019-10-23 | 2022-08-24 | Ibm | New framework for few-shot temporal action localization |
GB2604071B (en) * | 2019-10-23 | 2023-02-15 | Ibm | New framework for few-shot temporal action localization |
US11727686B2 (en) | 2019-10-23 | 2023-08-15 | International Business Machines Corporation | Framework for few-shot temporal action localization |
CN110934587A (en) * | 2019-12-13 | 2020-03-31 | 深圳龙岗智能视听研究院 | Alzheimer disease auxiliary diagnosis method based on atlas neural network |
CN112634281A (en) * | 2020-12-10 | 2021-04-09 | 浙江大学 | Grid segmentation method based on graph convolution network |
CN116778027A (en) * | 2023-08-22 | 2023-09-19 | 中国空气动力研究与发展中心计算空气动力研究所 | Curved surface parameterization method and device based on neural network |
CN116778027B (en) * | 2023-08-22 | 2023-11-07 | 中国空气动力研究与发展中心计算空气动力研究所 | Curved surface parameterization method and device based on neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109255791A (en) | A kind of shape collaboration dividing method based on figure convolutional neural networks | |
CN109145939B (en) | Semantic segmentation method for small-target sensitive dual-channel convolutional neural network | |
CN105631479B (en) | Depth convolutional network image labeling method and device based on non-equilibrium study | |
CN104850633B (en) | A kind of three-dimensional model searching system and method based on the segmentation of cartographical sketching component | |
CN107403183A (en) | The intelligent scissor method that conformity goal is detected and image segmentation is integrated | |
CN102004922B (en) | High-resolution remote sensing image plane extraction method based on skeleton characteristic | |
CN106203395A (en) | Face character recognition methods based on the study of the multitask degree of depth | |
CN107247938A (en) | A kind of method of high-resolution remote sensing image City Building function classification | |
CN111161278B (en) | Deep network aggregation-based fundus image focus segmentation method | |
CN108763376A (en) | Syncretic relation path, type, the representation of knowledge learning method of entity description information | |
CN101639935A (en) | Digital human serial section image segmentation method based on geometric active contour target tracking | |
CN108009286A (en) | A kind of Sketch Searching method based on deep learning | |
CN110135459A (en) | A kind of zero sample classification method based on double triple depth measure learning networks | |
CN112925908A (en) | Attention-based text classification method and system for graph Attention network | |
CN110807485B (en) | Method for fusing two-classification semantic segmentation maps into multi-classification semantic map based on high-resolution remote sensing image | |
CN105868706A (en) | Method for identifying 3D model based on sparse coding | |
CN102073713A (en) | Optimal bag-of-words (BoW) model based three-dimensional model searching method | |
CN103810303A (en) | Image search method and system based on focus object recognition and theme semantics | |
CN110210431A (en) | A kind of point cloud classifications method based on cloud semantic tagger and optimization | |
CN108829810A (en) | File classification method towards healthy public sentiment | |
CN116187200A (en) | Lightweight three-dimensional CAD model classification and retrieval method based on graph convolution network | |
CN102024029B (en) | Local visual attention-based color image retrieving method | |
CN102270343B (en) | Image segmentation method based on Ising graph model | |
Zheng et al. | YOLOv4-lite–based urban plantation tree detection and positioning with high-resolution remote sensing imagery | |
CN111985204A (en) | Customs import and export commodity tax number prediction method |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190122 |
|
RJ01 | Rejection of invention patent application after publication |