CN101458714A - Three-dimensional model search method based on precision geodesic - Google Patents
Three-dimensional model search method based on precision geodesic Download PDFInfo
- Publication number
- CN101458714A CN101458714A CNA2008102473381A CN200810247338A CN101458714A CN 101458714 A CN101458714 A CN 101458714A CN A2008102473381 A CNA2008102473381 A CN A2008102473381A CN 200810247338 A CN200810247338 A CN 200810247338A CN 101458714 A CN101458714 A CN 101458714A
- Authority
- CN
- China
- Prior art keywords
- dimensional model
- model
- dimensional
- sampled point
- geodesic
- 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
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a three-dimensional model searching method based on an accurate geodesic line, the three-dimensional model of the invention is stored in a three-dimensional model database of a server terminal, the method of the invention submits the three-dimensional model need to be searched to the server terminal by a client terminal, the server terminal performs a matching operation for the three-dimensional model submitted by the client terminal and the three-dimensional model stored in the three-dimensional model database and outputs the three-dimensional model conforming to the matching regulation in the three-dimensional model database. The method not only keeps good invariability of rotation, translation and scale, but also extracts characteristic vectors from the geodesic line information of a sampling point, can perform good model match, the searching efficiency can be improved greatly, meanwhile, the accurate geodesic line algorithm can enable the geodesic distance of any point to be more accurate, the searching accuracy can be further perfected, and the insensitivity of simplifying the model by the searching result can be ensured.
Description
Technical field
The invention belongs to the multimedia information retrieval field, particularly a kind of based on accurate geodesic method for searching three-dimension model.
Background technology
Along with visual, the digitized develop rapidly of three-dimensional model, increasing three-dimensional model and database thereof have occurred recent years on the network, so this problem of three-dimensional model search becomes the focus that current sciemtifec and technical sphere is paid close attention to.How rationally to find relevant three-dimensional model resource exactly, made many useful researchs in this respect both at home and abroad.A three-dimensional model search engine efficiently can not only greatly shorten time of web search, improves the efficient and the quality of our scientific research, also can be the actual production service for life to have bigger practical value more easily.
Some relatively reasonably method for searching three-dimension model have been arranged at present both at home and abroad; Wherein Masaki Hilaga has proposed a kind ofly based on approximate geodesic MRG figure matching algorithm in paper " Topology Matching for Fully AutomaticSimilarity Estimation of 3D Shapes ", and A.Ben Hamza and Hamid Krim have also introduced based on approximate geodesic method for searching three-dimension model at the paper " Geodesic Object Representation and Recognition " of successively delivering and " GeodesicMatching of Triangulated Surfaces ". These methods are significantly improved with respect to original D2 algorithm performance, and it becomes better to the rotation and the stability of translation.It is not enough that but this wherein also exists:at first the complexity of MRG figure matching algorithm is higher, makes up and mates in the MRG figure process and might cause whole Model Matching misalignment because of the error of certain details; And the statistical method that A.Ben Hamza and Hamid Krim propose also relates to the mathematical computations of relative complex.They do not have to extract a complete relatively proper vector from geodesic line information, are unfavorable for adopting the method for pattern match to retrieve.In addition, not accurate geodesic line in the middle of the algorithm that they adopt, thereby when three-dimensional model is simplified processing after, can not guarantee the consistance that geodesic line is measured, obtain by approximate treatment that also there is certain error in the geodesic distance of point-to-point transmission in the middle of the model.
Summary of the invention
(1) technical matters that will solve
The technical problem to be solved in the present invention is to overcome the deficiencies in the prior art, provide a kind of based on accurate geodesic method for searching three-dimension model, it has not only kept model rotation, translation and convergent-divergent good stable, and from accurate geodesic line information, extract classification and the proper vector of mating, so just greatly improved recall precision.Owing to adopted accurate geodesic algorithm, it can calculate the bee-line between each sampled point accurately, can improve the precision of retrieval well simultaneously, and has kept the insensitivity for the model simplification processing.
(2) technical scheme
At above problem, it is a kind of based on accurate geodesic method for searching three-dimension model that the present invention proposes.Three-dimensional model is stored in the three-dimensional modeling data storehouse of server end, described method is submitted the three-dimensional model that needs retrieval to by client to described server end, the three-dimensional model of storing in three-dimensional model that server end is submitted to according to client and the three-dimensional modeling data storehouse carries out matching operation, and export the three-dimensional model that meets matched rule in the three-dimensional modeling data storehouse, wherein:
Described matching operation comprises the steps:
(1) three-dimensional model of being stored in the three-dimensional modeling data storehouse of three-dimensional model that client is submitted to and server end carries out the geodesic line characteristic information and extracts operation, obtains the three-dimensional model characteristic information of each three-dimensional model of the three-dimensional model characteristic information of client and server end respectively;
Described geodesic line characteristic information extracts operation and further comprises the steps:
(11) obtain the accurate geodesic line of each sampled point other each sampled points in this three-dimensional model on the three-dimensional model apart from v;
(12) to each sampled point, the accurate geodesic line of calculating these other each points in the model apart from sum μ (v);
(13) utilize formula
Each sampled point is carried out normalization handle, wherein p is each sampled point, and S is the surface area of three-dimensional model, μ
n(v) represent the normalized value of n sampled point;
(14) histogram information and the corresponding proper vector of structure sampled point.At first 0-1 is divided into the m equal portions, wherein m is an integer and greater than 1, each sampled point is according to normalized value μ
n(v) put different equal portions under.Add up each equal portions sampled point number, make up histogram.Again thereby each equal portions sampled point number is got each equal portions proportion divided by the total number of sampled point, constitute m dimensional feature vector Wm, Wm is the three-dimensional model characteristic information, the above-mentioned m dimensional feature vector that obtains in the three-dimensional model of representing respectively to be stored the three-dimensional modeling data storehouse of the three-dimensional model submitted to from client and server end with Wx and Wi;
(2) by formula ‖ W
x-W
i‖ computed range length Δ Wi is the three-dimensional model that meets matched rule in the three-dimensional modeling data storehouse apart from the pairing three-dimensional model of Wi of length Δ Wi minimum.
Wherein, step (11) also comprises the steps:
(111) at first from three-dimensional model, select a summit as sampled point arbitrarily, this point is labeled as handles, and add the sampled point tabulation to;
(112) the accurate geodesic line of calculating this sampled point other each sampled points in this three-dimensional model is apart from v, find wherein geodesic distance less than
Each summit, wherein the surface area of S representative model is included into these summits in the middle of the similar point of this sampled point, and is labeled as and handles, and does not add the sampled point tabulation to;
(113) to sampled point tabulation is added it in a chosen distance summit farthest from untreated summit, repeats the operation of above-mentioned (112), and all summits all are labeled as and handle in model; Wherein in the step (112), calculate accurate geodesic line and be apart from the method for v:
(1121) setting this sampled point is source point, makees window with the form of ray divergence on each triangle surface and increases, and each limit in the model is decomposed into some wickets;
(1122) in the process that window increases, deviation may take place in the geodesic line only concave point place in model, and the point that deviation takes place is called pseudo-source point, iterates thus until the window propagation process of finishing whole model;
(1123) obtain the geodesic line distance of source point other each sampled points in the model again by the method for feedback, promptly accurately geodesic line apart from v.
Wherein, described step (1) also comprises the steps:
(15) the m dimensional feature vector that obtains in the three-dimensional model that adopts the SVM method that the three-dimensional modeling data storehouse of server end is stored is classified, as criteria for classification, the average of the m dimensional feature vector of each group of classification back gained is as the eigenwert Wy of this classification with the numerical value degree of approach of proper vector;
In the described step (2), earlier by formula ‖ W
x-Wy ‖ computed range length Δ Wy finds the pairing group of Wy apart from length Δ Wy minimum, and the m dimensional feature vector in this group is represented with Wi.
Wherein, described step (15) also comprises the steps:
(151) the arbitrary m dimensional feature vector that obtains in the three-dimensional model of at first selecting the three-dimensional modeling data storehouse to be stored is as training standard, adopt SVM algorithm three-dimensional model and the difference that it is close with it to distinguish than large-sized model, make up a thick classification thus, and calculate the characteristic vector W of the average of the proper vector of each model in this rough sort as this classification
0
(152) the m dimensional feature vector of selecting any three-dimensional model again in the bigger three-dimensional model of the difference of remainder is as training standard, and the operation of repeating step (151) obtains a new rough sort and calculates the characteristic vector W y of this classification.
(153) the SVM algorithm is carried out in the operation of repeating step (152) repeatedly, until whole three-dimensional modeling data storehouse is divided into several thick classification.
Wherein, also comprise when meeting the three-dimensional model of matched rule in the output three-dimensional modeling data storehouse, each three-dimensional model is sorted according to distance length Δ Wi order from small to large, select the forward three-dimensional model of some ranks to export as result for retrieval.
Wherein, server end also is provided with the three-dimensional model property data base, and the three-dimensional model characteristic information of each three-dimensional model is stored in this three-dimensional model property data base.
(3) beneficial effect
The present invention's advantage compared with prior art is: for the unchangeability that has not only kept the good rotation of three-dimensional model, translation and convergent-divergent, and from the geodesic line information of sampled point, extract proper vector, can carry out Model Matching well, recall precision is greatly improved.Adopt accurate geodesic algorithm simultaneously, make in the model geodesic distance of arbitrfary point more accurate, further improve the precision of retrieval, and guaranteed insensitivity for model simplification.And in the process that makes up database, adopt the SVM algorithm that the three-dimensional model of database is carried out rough sort, quicken the speed and the efficient of retrieval.
Description of drawings
Fig. 1 is main modular and the flow process based on accurate geodesic method for searching three-dimension model of the present invention;
Fig. 2 is the basic procedure of point sampling of the present invention;
Fig. 3 is the main process that vectorial characteristic information of the present invention extracts;
Fig. 4 is a match retrieval output result's of the present invention main process;
Fig. 5 is an an example of the present invention: the retrieval effectiveness figure of input elephant three-dimensional model.
Embodiment
Below in conjunction with accompanying drawing the specific embodiment of the present invention is elaborated.
One, the main modular and the flow process based on accurate geodesic method for searching three-dimension model of embodiment of the present invention
Be illustrated in figure 1 as main modular and the flow process based on accurate geodesic method for searching three-dimension model of the present invention, method of the present invention is based on the client and server end, client is used to submit to three-dimensional model information to be retrieved, Server Side Include has been stored the three-dimensional modeling data storehouse of three-dimensional model, and through extracting the geodesic line proper vector, and the property data base that the geodesic line proper vector of three-dimensional model is obtained after according to SVM algorithm rough sort.
The three-dimensional model that the client input is to be retrieved, at first extract the operation of geodesic line proper vector earlier according to the pattern of server end, and then submit to server end, and and the property data base of server end in grouped data compare, determine and the immediate concrete classification of three-dimensional model to be retrieved, and then three-dimensional model to be retrieved and each three-dimensional model in the concrete classification that obtains carried out the rate of exchange, find the three-dimensional model of coupling, and the output result.
Two, the basic procedure of the point sampling of embodiment of the present invention of the present invention
The basic procedure of point sampling of the present invention is at first handled each model in the three-dimensional modeling data storehouse as shown in Figure 2, obtains the three-dimensional model sampling point information, obtains sampled point to the accurate bee-line of other each points.Concrete steps are as follows:
(1) extracts sampled point, at first from model, select a summit arbitrarily, this point is labeled as handles, and add the sampled point tabulation to;
(2) calculate the geodesic line distance of these other each points in the model, find wherein geodesic distance less than
Each summit, the surface area of S representative model wherein is included into these summits in the middle of this fundamental point one class, and is labeled as and handles, and does not add the sampled point tabulation to, the scope point is labeled and finishes on every side;
(3) to sampled point tabulation is added it in a chosen distance summit farthest from untreated summit, repeats the operation of (2), and all summits all are labeled as and handle in model.
Wherein calculating accurate geodesic concrete grammar is:
A, to set this point be source point, makees window with the form of ray divergence on each triangle surface to increase, and each limit in the model is decomposed into wicket one by one.
B, in the process that window increases, deviation may take place in the geodesic line only concave point place in model, the point that deviation takes place is called pseudo-source point, iterates thus until the window propagation process of finishing whole model.
C, obtain the geodesic line distance and the Actual path of source point other each points in the model by the method for feedback again, promptly in the model source point to the accurate bee-line of other each point.
Three, the main process of the vectorial characteristic information extraction of embodiment of the present invention
Be illustrated in figure 3 as the main process that vectorial characteristic information of the present invention extracts, concrete steps are as follows:
(1) geodesic line that each sampled point is calculated these other each points in the model apart from sum μ (v);
(2) utilize formula
Each sampled point is carried out normalization handle, wherein p is each sampled point;
(3) histogram information and the corresponding proper vector of structure sampled point.At first 0-1 is divided into the m equal portions, choosing per 0.05 is equal portions, and each sampled point (v) puts different equal portions under according to normalized value μ n.Add up each equal portions sampled point number, make up histogram.Again thereby each equal portions sampled point number is got each equal portions proportion divided by the total number of sampled point, constitute the m dimensional feature vector;
The process in the three-dimensional modeling data storehouse that four, the structure of embodiment of the present invention is complete
Geodesic line information according to each model sampled point adopts the SVM method that each three-dimensional model in the database is carried out rough sort, makes up complete three-dimensional modeling data storehouse with this.Concrete steps are as follows:
(1) at first select in the database the geodesic proper vector of three-dimensional model as training standard, adopt SVM algorithm three-dimensional model and the difference that it is close with it to distinguish than large-sized model, make up a thick classification thus, and calculate the characteristic vector W of the average of the proper vector of each model in this rough sort as this classification
0
(2) proper vector of selecting a three-dimensional model again in the bigger model of remaining difference is as training standard, and the operation of repetition (1) obtains a new rough sort and calculates the characteristic vector W y of this classification.
(3) repeat operation in (2), carries out the SVM algorithm k time, until being the individual thick classification of k with the entire database Preliminary division, and set up the proper vector of classification separately, finish the structure of database thus.
Five, the model to be retrieved that client is provided of embodiment of the present invention carries out the process of sampling processing
The model to be retrieved that client is provided processing such as sample makes up corresponding geodesic line proper vector, concrete steps with Figure 3 shows that the main process that vectorial characteristic information of the present invention extracts is identical with the word segment explanation.
Six, the match retrieval of embodiment of the present invention and generate the process of result for retrieval
Be illustrated in figure 4 as match retrieval output result's of the present invention main flow process,, calculate similarity distance and ordering, finally generate result for retrieval the aspect of model of client submission and the characteristic matching in the database.Concrete steps are as follows:
(1) relatively preliminary, at first the characteristic vector W y with each rough sort in three-dimensional model characteristic vector W x to be retrieved and the database compares, by formula ‖ W
x-Wy ‖ calculates it apart from length, and a class of chosen distance minimum (or former class) is as the Primary Location of model to be retrieved.
(2) distinguish in detail, from the rough sort of determining, select the proper vector of three-dimensional model separately, calculate comparison, obtain similarity distance separately with the characteristic vector W i of model to be retrieved.
(3) output result, by sorting in proper order from small to large, s three-dimensional model shows output before selecting, and finishes retrieval with each similarity distance.
Be illustrated in figure 5 as the result for retrieval design sketch of embodiment of the present invention, what the left side showed is the elephant three-dimensional model of input, and the result that retrieval is come out is presented at the right side, and ranking is forward more, and then itself and three-dimensional model to be retrieved are approaching more.
Claims (6)
1, a kind of based on accurate geodesic method for searching three-dimension model, described three-dimensional model is stored in the three-dimensional modeling data storehouse of server end, described method is submitted the three-dimensional model that needs retrieval to by client to described server end, the three-dimensional model of storing in three-dimensional model that server end is submitted to according to client and the three-dimensional modeling data storehouse carries out matching operation, and export the three-dimensional model that meets matched rule in the three-dimensional modeling data storehouse, it is characterized in that:
Described matching operation comprises the steps:
(1) three-dimensional model of being stored in the three-dimensional modeling data storehouse of three-dimensional model that client is submitted to and server end carries out the geodesic line characteristic information and extracts operation, obtains the three-dimensional model characteristic information of each three-dimensional model of the three-dimensional model characteristic information of client and server end respectively;
Described geodesic line characteristic information extracts operation and further comprises the steps:
(11) obtain the accurate geodesic line of each sampled point other each sampled points in this three-dimensional model on the three-dimensional model apart from v;
(12) to each sampled point, the accurate geodesic line of calculating these other each points in the model apart from sum μ (v);
(13) utilize formula
Each sampled point is carried out normalization handle, wherein p is each sampled point, and S is the surface area of three-dimensional model, μ
n(v) represent the normalized value of n sampled point;
(14) make up the histogram information and the corresponding proper vector of sampled point, at first 0-1 is divided into the m equal portions, wherein m is an integer and greater than 1, each sampled point is according to normalized value μ
n(v) put different equal portions under, add up each equal portions sampled point number, make up histogram, again thereby each equal portions sampled point number is got each equal portions proportion divided by the total number of sampled point, constitute m dimensional feature vector Wm, Wm is the three-dimensional model characteristic information, the above-mentioned m dimensional feature vector that obtains in the three-dimensional model of representing respectively to be stored the three-dimensional modeling data storehouse of the three-dimensional model submitted to from client and server end with Wx and Wi;
(2) pass through formula || W
x-W
i|| computed range length Δ Wi is the three-dimensional model that meets matched rule in the three-dimensional modeling data storehouse apart from the pairing three-dimensional model of Wi of length Δ Wi minimum.
2, as claimed in claim 1ly it is characterized in that based on accurate geodesic method for searching three-dimension model step (11) comprises the steps:
(111) at first from three-dimensional model, select a summit as sampled point arbitrarily, this point is labeled as handles, and add the sampled point tabulation to;
(112) the accurate geodesic line of calculating this sampled point other each sampled points in this three-dimensional model is apart from v, find wherein geodesic distance less than
Each summit, wherein the surface area of S representative model is included into these summits in the middle of the similar point of this sampled point, and is labeled as and handles, and does not add the sampled point tabulation to;
(113) to sampled point tabulation is added it in a chosen distance summit farthest from untreated summit, repeats the operation of above-mentioned (112), and all summits all are labeled as and handle in model; Wherein in the step (112), calculate accurate geodesic line and be apart from the method for v:
(1121) setting this sampled point is source point, makees window with the form of ray divergence on each triangle surface and increases, and each limit in the model is decomposed into some wickets;
(1122) in the process that window increases, deviation may take place in the geodesic line only concave point place in model, and the point that deviation takes place is called pseudo-source point, iterates thus until the window propagation process of finishing whole model;
(1123) obtain the geodesic line distance of source point other each sampled points in the model again by the method for feedback, promptly accurately geodesic line apart from v.
3, as claimed in claim 1 or 2ly it is characterized in that based on accurate geodesic method for searching three-dimension model described step (1) also comprises the steps:
(15) the m dimensional feature vector that obtains in the three-dimensional model that adopts the SVM method that the three-dimensional modeling data storehouse of server end is stored is classified, as criteria for classification, the average of the m dimensional feature vector of each group of classification back gained is as the eigenwert Wy of this classification with the numerical value degree of approach of proper vector;
In the described step (2), pass through formula earlier || W
x-Wy|| computed range length Δ Wy finds the pairing group of Wy apart from length Δ Wy minimum, and the m dimensional feature vector in this group is represented with Wi.
4, as claimed in claim 3ly it is characterized in that based on accurate geodesic method for searching three-dimension model described step (15) also comprises the steps:
(151) the arbitrary m dimensional feature vector that obtains in the three-dimensional model of at first selecting the three-dimensional modeling data storehouse to be stored is as training standard, adopt SVM algorithm three-dimensional model and the difference that it is close with it to distinguish than large-sized model, make up a thick classification thus, and calculate the characteristic vector W of the average of the proper vector of each model in this rough sort as this classification
0
(152) the m dimensional feature vector of selecting any three-dimensional model again in the bigger three-dimensional model of the difference of remainder is as training standard, and the operation of repeating step (151) obtains a new rough sort and calculates the characteristic vector W y of this classification;
(153) the SVM algorithm is carried out in the operation of repeating step (152) repeatedly, until whole three-dimensional modeling data storehouse is divided into several thick classification.
5, as claimed in claim 4 based on accurate geodesic method for searching three-dimension model, it is characterized in that, when meeting the three-dimensional model of matched rule in the output three-dimensional modeling data storehouse, each three-dimensional model is sorted according to distance length Δ Wi order from small to large, select the forward three-dimensional model of some ranks to export as result for retrieval.
6, as claimed in claim 5ly it is characterized in that based on accurate geodesic method for searching three-dimension model server end also is provided with the three-dimensional model property data base, the three-dimensional model characteristic information of each three-dimensional model is stored in this three-dimensional model property data base.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2008102473381A CN101458714A (en) | 2008-12-30 | 2008-12-30 | Three-dimensional model search method based on precision geodesic |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2008102473381A CN101458714A (en) | 2008-12-30 | 2008-12-30 | Three-dimensional model search method based on precision geodesic |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101458714A true CN101458714A (en) | 2009-06-17 |
Family
ID=40769570
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNA2008102473381A Pending CN101458714A (en) | 2008-12-30 | 2008-12-30 | Three-dimensional model search method based on precision geodesic |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101458714A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101599077B (en) * | 2009-06-29 | 2011-01-05 | 清华大学 | Method for retrieving three-dimensional object |
| CN101714253B (en) * | 2009-12-04 | 2012-05-23 | 西安电子科技大学 | Interactive image segmentation correcting method based on geodesic active region models |
| CN104361347A (en) * | 2014-10-21 | 2015-02-18 | 浙江大学 | Numerically-controlled machine tool design module three-dimension model retrieval method based on single image |
| CN107491481A (en) * | 2017-07-10 | 2017-12-19 | 深圳三维盘酷网络科技有限公司 | LOD pattern searches method and system, the method for establishing LOD model databases and computer-readable storage medium |
| CN110019901A (en) * | 2017-09-13 | 2019-07-16 | 深圳三维盘酷网络科技有限公司 | Three-dimensional model search device, searching system, search method and computer readable storage medium |
| CN111089592A (en) * | 2019-12-13 | 2020-05-01 | 天津大学 | Method for calculating geodesic line in discrete curved surface |
-
2008
- 2008-12-30 CN CNA2008102473381A patent/CN101458714A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101599077B (en) * | 2009-06-29 | 2011-01-05 | 清华大学 | Method for retrieving three-dimensional object |
| CN101714253B (en) * | 2009-12-04 | 2012-05-23 | 西安电子科技大学 | Interactive image segmentation correcting method based on geodesic active region models |
| CN104361347A (en) * | 2014-10-21 | 2015-02-18 | 浙江大学 | Numerically-controlled machine tool design module three-dimension model retrieval method based on single image |
| CN104361347B (en) * | 2014-10-21 | 2017-10-03 | 浙江大学 | A kind of Digit Control Machine Tool design module method for searching three-dimension model based on single image |
| CN107491481A (en) * | 2017-07-10 | 2017-12-19 | 深圳三维盘酷网络科技有限公司 | LOD pattern searches method and system, the method for establishing LOD model databases and computer-readable storage medium |
| CN107491481B (en) * | 2017-07-10 | 2020-08-18 | 深圳三维盘酷网络科技有限公司 | LOD model searching method and system, method for establishing LOD model database and computer-readable storage medium |
| CN110019901A (en) * | 2017-09-13 | 2019-07-16 | 深圳三维盘酷网络科技有限公司 | Three-dimensional model search device, searching system, search method and computer readable storage medium |
| CN111089592A (en) * | 2019-12-13 | 2020-05-01 | 天津大学 | Method for calculating geodesic line in discrete curved surface |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kazakovtsev et al. | Genetic algorithm with fast greedy heuristic for clustering and location problems | |
| Zhang et al. | Prboost: Prompt-based rule discovery and boosting for interactive weakly-supervised learning | |
| CN109117560A (en) | Automobile typical machined part three-dimensional process design method and platform based on MBD | |
| CN108154430A (en) | A kind of credit scoring construction method based on machine learning and big data technology | |
| CN108920720A (en) | The large-scale image search method accelerated based on depth Hash and GPU | |
| CN101458714A (en) | Three-dimensional model search method based on precision geodesic | |
| CN111027140B (en) | Airplane standard part model rapid reconstruction method based on multi-view point cloud data | |
| CN110674407A (en) | Hybrid recommendation method based on graph convolution neural network | |
| CN110147421B (en) | Target entity linking method, device, equipment and storage medium | |
| CN112464611B (en) | Automatic PCB wiring system based on cloud-end collaborative intelligent processing | |
| CN104462196A (en) | Multi-feature-combined Hash information retrieval method | |
| CN113077501B (en) | End-to-end point cloud registration method based on feature learning | |
| CN106649440A (en) | Approximate repeated video retrieval method incorporating global R features | |
| CN103294832A (en) | Motion capture data retrieval method based on feedback study | |
| CN116991869A (en) | Method for automatically generating database query statement based on NLP language model | |
| Dai et al. | Cross-modal guidance network for sketch-based 3D shape retrieval | |
| CN111753067A (en) | Innovative assessment method, device and equipment for technical background text | |
| CN110825852B (en) | Long text-oriented semantic matching method and system | |
| CN117313849A (en) | Knowledge graph construction method and device for energy industry based on multi-source heterogeneous data fusion technology | |
| CN117010373A (en) | Recommendation method for category and group to which asset management data of power equipment belong | |
| CN102110159B (en) | CAD three-dimensional model retrieval method and system | |
| CN110245248A (en) | A kind of remote sensing image keyword query method | |
| CN115238705A (en) | Semantic analysis result reordering method and system | |
| CN112926340B (en) | Semantic matching model for knowledge point positioning | |
| CN105354264B (en) | A kind of quick adding method of theme label based on local sensitivity Hash |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Open date: 20090617 |