CN112017305A - Three-dimensional model automatic coloring method based on reinforcement learning - Google Patents
Three-dimensional model automatic coloring method based on reinforcement learning Download PDFInfo
- Publication number
- CN112017305A CN112017305A CN202010876342.5A CN202010876342A CN112017305A CN 112017305 A CN112017305 A CN 112017305A CN 202010876342 A CN202010876342 A CN 202010876342A CN 112017305 A CN112017305 A CN 112017305A
- Authority
- CN
- China
- Prior art keywords
- model
- semantic segmentation
- training
- function
- network
- 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
Classifications
-
- 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
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B33—ADDITIVE MANUFACTURING TECHNOLOGY
- B33Y—ADDITIVE MANUFACTURING, i.e. MANUFACTURING OF THREE-DIMENSIONAL [3-D] OBJECTS BY ADDITIVE DEPOSITION, ADDITIVE AGGLOMERATION OR ADDITIVE LAYERING, e.g. BY 3-D PRINTING, STEREOLITHOGRAPHY OR SELECTIVE LASER SINTERING
- B33Y50/00—Data acquisition or data processing for additive manufacturing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2200/00—Indexing scheme for image data processing or generation, in general
- G06T2200/04—Indexing scheme for image data processing or generation, in general involving 3D image data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
- G06T2207/10012—Stereo images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2219/2012—Colour editing, changing, or manipulating; Use of colour codes
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Architecture (AREA)
- Computer Graphics (AREA)
- Computer Hardware Design (AREA)
- Chemical & Material Sciences (AREA)
- Manufacturing & Machinery (AREA)
- Materials Engineering (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an automatic coloring method of a three-dimensional model based on reinforcement learning, which is characterized in that training is continuously carried out based on an initial data set, the training is carried out and the expansion and completion are carried out, then the reinforcement learning technology is adopted, an induction network carries out iterative self-learning by using a newly added model, finally, a progressive function enhancement network is adopted for an input model to obtain object-level semantic segmentation, and efficient geometric matching based on a discrete Hausdorff distance is carried out according to model target function description appointed by a user or in a 'geometric-material-function' data set to obtain an object-level printable multi-material intelligent distribution result of the input model. Compared with the prior art, the method has the advantages that the multi-material distribution and coloring are automatically carried out on the three-dimensional model, the printing model which meets the multi-material distribution information of the three-dimensional model specified by a user can be accurately manufactured, the actual manufacturing requirements in the fields of aviation, aerospace, medical treatment and the like are greatly met, and the method has good application prospect.
Description
Technical Field
The invention relates to the technical field of three-dimensional model printing, in particular to a three-dimensional model automatic coloring method based on reinforcement learning.
Background
The three-dimensional printing technology, also called additive manufacturing technology in general, is a new strategic direction in the mechanical field of "china manufacturing 2025" by virtue of its brand new manufacturing concept and unique technical advantages, has been widely applied in many fields, and promotes transformation and upgrade of the traditional manufacturing mode. China is in the key stage of deep integration of informatization and industrialization and upgrading of industries, safety, reliability and autonomous controllability become national strategies for informatization development, and further promotion of industrial manufacturing informatization has important strategic significance for improving industrial competitiveness in China. The domestic three-dimensional printing market is rapidly increased, and is expected to reach 350 billion yuan in 2020. However, compared with developed countries, the autonomous innovation capability of the three-dimensional printing technology in China is insufficient, and particularly in the aspect of processing software of three-dimensional printing, the problems of low data processing efficiency, poor stability and universality and the like exist.
At present, three-dimensional printing can only be manufactured by using one material in the same model, the requirements of important fields such as aviation, aerospace, medical treatment and the like cannot be met, and the application range and the technical development of the three-dimensional printing are severely limited. With the advent of multi-extrusion head technology, a variety of hardware devices supporting multi-material three-dimensional printing are in the row.
The automatic coloring method of the three-dimensional model in the prior art cannot be directly expanded to a multi-material three-dimensional printing technology, and cannot perform automatic multi-material distribution of the model according to the target function of the three-dimensional model.
Disclosure of Invention
The invention aims to provide a three-dimensional model automatic coloring method based on reinforcement learning, which adopts a progressive function enhancement network to obtain object-level semantic segmentation, and according to the model target function description appointed by the user or the high-efficiency geometric matching method based on the discrete Hausdorff distance in the 'geometry-material-function' data set, the object-level printable multi-material intelligent distribution result of the input model is obtained, the method is utilized to automatically perform multi-material distribution and coloring on the three-dimensional model, the method is simple and convenient, has high efficiency, can accurately manufacture the printing model which accords with the three-dimensional model multi-material distribution information specified by a user, greatly meets the actual manufacturing requirements in the fields of aviation, aerospace, medical treatment and the like, and has good application prospect.
The specific technical scheme for realizing the purpose of the invention is as follows: a three-dimensional model automatic coloring method based on reinforcement learning is characterized in that a progressive function enhancement network is adopted to obtain object-level semantic segmentation, a geometric matching method based on discrete Hausdorff distance is carried out according to model target function description appointed by a user or a 'geometry-material-function' data set, an object-level printable multi-material intelligent distribution result of an input model is obtained, and multi-material distribution and coloring of a three-dimensional model specifically comprise the following steps:
step 1: completion and expansion of data sets
1.1) by collecting an initial data set containing geometric information, material information, and a coarse functional type description;
1.2) clustering the models by using a K-Means algorithm, and stopping training when the loss value of each type is less than a specified threshold value;
1.3) manually marking each clustering core with high-precision function description so as to obtain a first batch of complete function label core models.
Step 2: self-iterative learning of models
2.1) inputting the obtained core model with the complete functional label into a deep learning neural network, adopting a reinforcement learning strategy, inducing the network to carry out self-iterative learning by using the newly added model, and determining an energy function value through the label and the network output;
2.2) training a deep neural network model on the training set according to the energy function values and the training parameters; the training parameters are as follows: the learning rate was 0.00025;
2.3) stopping training when the energy function value of the deep neural network on the training set reaches a set threshold value;
2.4) model labeling is continuously self-perfected along with the continuous expansion of the data set quantity, thereby obtaining a 'geometry-material-function' data set supporting online optimization.
And step 3: multi-material intelligent distribution
3.1) inputting the obtained data set into a semantic segmentation network to obtain object-level semantic segmentation of the data set, and determining a segmentation loss cost function value through the object-level semantic segmentation and label data;
3.2) training a semantic segmentation network on the training set according to the segmentation loss cost function value and the training parameters; the learning rate of the training parameters is 0.00025;
3.3) stopping training when the segmentation loss cost function value of the semantic segmentation network on the training set is smaller than a set threshold value;
and 3.4) carrying out efficient geometric matching on the data set based on the discrete Hausdorff distance to obtain an object-level printable multi-material intelligent distribution result of the input model.
The completion and expansion of step 1 specifically includes:
A1) collecting the structure and information of the three-dimensional model through network search and database search, and adding the three-dimensional model structure and information into the existing three-dimensional model data set for data set expansion;
A2) carrying out three-dimensional model rotation, cutting and scaling processing operations on the expanded data set;
A3) and clustering the models by using a K-Means algorithm, manually comparing and checking each clustering core, and marking with high-precision function description so as to obtain a first batch of complete function label core models.
The self-iterative learning of step 2 specifically includes:
B1) inputting the complete functional label core model into a deep learning neural network, and determining that the output is the function and material label information of the three-dimensional model;
B2) according to the function and material label information of the three-dimensional model obtained by the deep learning neural network, comparing the function and material label information with the label of the originally input high-precision function description, and calculating each output energy function value by a reinforcement learning method;
B3) and optimizing the neural network according to the energy function values and the training parameters, and improving the accuracy of network output labels, so that model labeling is continuously and self-perfected along with the continuous expansion of the data set quantity, and a 'geometry-material-function' data set supporting online optimization is realized.
The multi-material intelligent distribution of the step 3 specifically comprises:
C1) acquiring parameters of a trained deep convolutional neural network U-Net from the network, determining model initial parameters of a deep semantic segmentation network according to the parameters of the U-Net, taking a training set as the input of the deep semantic segmentation network to obtain output semantic segmentation maps with the same dimensionality, determining a Dice loss function of the semantic segmentation maps by comparing the ground-route of the training set with the output semantic segmentation maps, and training the deep semantic segmentation network by using the training set on the basis of the initial parameters;
C2) and updating parameters in the deep semantic segmentation network on the training set in a back propagation mode according to the Dice loss function value of the semantic segmentation graph.
The multi-material intelligent distribution of the step 3 specifically comprises:
D1) inputting the model to be colored into the deep semantic segmentation network, and outputting a corresponding semantic segmentation graph;
D2) and comparing the semantic segmentation graph with the models existing in the database, calculating the discrete Hausdorff distance to perform efficient geometric matching, obtaining an object-level printable multi-material intelligent distribution result of the input model, and outputting the object-level printable multi-material intelligent distribution result.
Compared with the prior art, the method has the advantages of correlating the actual functions of the model with the material distribution thereof, automatically distributing and coloring the three-dimensional model by multiple materials, high intelligent degree, low manual design cost, simple method and high efficiency, can accurately manufacture the printing model which meets the three-dimensional model multiple material distribution information specified by a user, greatly meets the actual manufacturing requirements in the fields of aviation, aerospace, medical treatment and the like, and has good application prospect.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a flow diagram of model self-iteration;
FIG. 3 is a flow chart of semantic segmentation level intelligent distribution matching.
Detailed Description
The invention is illustrated in further detail below by means of specific examples:
example 1
Referring to fig. 1, the present invention automatically colors the three-dimensional model according to the following steps:
step 1: completion and expansion of data sets
S111: and constructing an initial data set through network search and database search, continuously training based on the initial data set, and expanding and completing the initial data set.
S112: clustering the models by using a K-Means algorithm, manually comparing and checking each clustering core, and marking with high-precision function description to obtain a first batch of complete function label core models, wherein the K-Means algorithm is a cost function defined by the following formula a:
wherein:
represents and x(i)The nearest cluster center point.
S113: and (4) adopting a reinforcement learning technology to induce the network to carry out iterative self-learning by using the newly added model.
S114: and for the input model, obtaining object-level semantic segmentation by adopting a progressive function enhancement network, and performing efficient geometric matching based on a discrete Hausdorff distance according to the model target function description specified by a user or in a 'geometry-material-function' data set to obtain an object-level printable multi-material intelligent distribution result of the input model.
Step 2: self-iterative learning of models
Referring to fig. 2, the method for inducing the network to perform iterative self-learning by using a newly added model by using a reinforcement learning technology mainly comprises the following steps:
s211: inputting the obtained core model with the complete function label into a deep learning neural network, adopting a reinforcement learning strategy, inducing the network to carry out self-iterative learning by using a newly added model, determining an energy function value through the label and the network output, encouraging the network to carry out reinforcement learning according to the good and bad evaluation state of the energy value function, carrying out self-iterative learning by the network through a maximized energy value function, and calculating the energy function value by the following formula b:
wherein: s represents a state; pi represents a behavior strategy; gtRepresenting the reward at time t.
S212: and training a deep neural network model on a training set according to the energy function value and a training parameter, wherein the learning rate of the training parameter is 0.00025.
S213: and when the energy function value of the deep neural network on the training set reaches a set threshold value, stopping training.
S214: with the continuous expansion of the data set quantity, model labeling is continuously and self-perfected, so that a 'geometry-material-function' data set supporting online optimization is obtained.
And step 3: multi-material intelligent distribution
Referring to fig. 3, an object-level semantic segmentation is obtained by using a progressive function enhancement network, and an object-level printable multi-material intelligent distribution result of an input model is obtained by performing efficient geometric matching based on a discrete Hausdorff distance according to a model target function description specified by a user or in a "geometry-material-function" data set, which mainly comprises the following steps:
s311: obtaining parameters of a trained deep convolutional neural network U-Net from the network, determining model initial parameters of a deep semantic segmentation network according to the parameters of the U-Net, taking a training set as the input of the deep semantic segmentation network, training the deep semantic segmentation network by using the training set on the basis of the initial parameters to obtain output semantic segmentation maps with the same dimension, and determining a Dice loss function of the semantic segmentation maps by comparing a ground-route of the training set with the output semantic segmentation maps, wherein the Dice loss function is defined by the following formula c:
wherein: x is an output semantic segmentation graph; y is a corresponding group-route semantic segmentation graph.
S312: and updating training parameters, and updating parameters in the deep semantic segmentation network on a training set in a back propagation mode according to the Dice loss function value of the semantic segmentation graph.
S313: and inputting the model to be colored into the deep semantic segmentation network, and outputting a corresponding semantic segmentation graph.
S314: comparing the semantic segmentation graph with a model existing in a database, calculating and carrying out efficient geometric matching based on a discrete Hausdorff distance, obtaining an object-level printable multi-material intelligent distribution result of an input model and outputting the result, wherein the Hausdorff distance is a measure for describing the similarity degree between two sets of point sets, and the Hausdorff distance is calculated by the following formula d:
H(A,B)＝max(h(A,B),h(B,A)) (d)；
wherein: A. b represents two sets of point sets; | L | · | represents a distance norm (e.g., L2 or euclidean distance) between point set a and point set B; h (a, B) ═ max (a ∈ a) min (B ∈ B) | | | a-B |; h (B, a) ═ max (B ∈ B) min (a ∈ a) | | | B-a |.
The above examples are only for further illustration of the present invention and are not intended to limit the present invention, and all equivalent implementations of the present invention should be included within the scope of the claims of the present invention.
Claims (9)
1. A three-dimensional model automatic coloring method based on reinforcement learning is characterized in that a progressive function enhancement network is adopted to obtain object-level semantic segmentation, a geometric matching method based on a discrete Hausdorff distance is carried out according to model target function description appointed by a user or a 'geometry-material-function' data set, an object-level printable multi-material intelligent distribution result of an input model is obtained, and multi-material distribution and coloring of a three-dimensional model specifically comprise the following steps:
step 1: completion and expansion of data sets
1.1) by collecting an initial data set containing geometric information, material information, and a coarse functional type description;
1.2) clustering the models by using a K-Means algorithm, and stopping training when the loss value of each type is less than a specified threshold value;
1.3) manually labeling the function description of each clustering core, thereby obtaining a first batch of complete function label core models;
step 2: self-iterative learning of models
2.1) inputting a core model with a complete functional label into a deep learning neural network, adopting a reinforcement learning strategy, inducing the network to carry out self-iterative learning by using a newly added model, and determining an energy function value through the label and network output;
2.2) training a deep neural network model on a training set according to the energy function value and the training parameters;
2.3) stopping training when the energy function value of the deep neural network on the training set reaches a set threshold value;
2.4) continuously expanding the data set quantity and continuously self-perfecting the model label, thereby obtaining a 'geometry-material-function' data set supporting online optimization;
and step 3: multi-material intelligent distribution
3.1) inputting the data set into a semantic segmentation network to obtain object-level semantic segmentation of the data set, and determining a segmentation loss cost function value through the object-level semantic segmentation and label data;
3.2) training a semantic segmentation network on a training set according to the segmentation loss cost function value and the training parameters;
3.3) stopping training when the segmentation loss cost function value of the semantic segmentation network on the training set is smaller than a set threshold value;
and 3.4) carrying out geometrical matching on the data set based on the discrete Hausdorff distance to obtain an object-level printable multi-material intelligent distribution result of the input model.
2. The method for automatically coloring a three-dimensional model based on reinforcement learning of claim 1, wherein the completion and expansion of the data set of step 1 specifically comprises:
A1) collecting the structure and information of the three-dimensional model through network search and database search, and adding the three-dimensional model structure and information into the existing three-dimensional model data set for data set expansion;
A2) carrying out three-dimensional model rotation, cutting and scaling processing operations on the expanded data set;
A3) and clustering the three-dimensional models by using a K-Means algorithm, manually comparing and checking each clustering core, and marking labels with function description so as to obtain a first batch of complete function label core models.
3. The method for automatically coloring a three-dimensional model based on reinforcement learning of claim 1, wherein the self-iterative learning of the model of step 2 specifically comprises:
B1) inputting the complete functional label core model into a deep learning neural network, and determining that the output is the function and material label information of the three-dimensional model;
B2) calculating each output energy function value by adopting a reinforcement learning method according to the comparison between the function and material label information of the three-dimensional model and the label of the originally input function description;
B3) and optimizing the neural network according to the energy function values and the training parameters, improving the accuracy of network output labels, and continuously expanding the volume of the data set and continuously and automatically perfecting model labels, thereby obtaining a 'geometry-material-function' data set supporting online optimization.
4. The reinforcement learning-based three-dimensional model automatic coloring method according to claim 1, wherein the intelligent distribution of the multiple materials of the step 3 specifically comprises:
C1) acquiring parameters of a trained deep convolutional neural network U-Net from the network, determining model initial parameters of a deep semantic segmentation network according to the parameters of the U-Net, taking a training set as the input of the deep semantic segmentation network to obtain output semantic segmentation maps with the same dimensionality, determining a Dice loss function of the semantic segmentation maps by comparing the ground-route of the training set with the output semantic segmentation maps, and training the deep semantic segmentation network by using the training set on the basis of the initial parameters;
C2) updating parameters in a deep semantic segmentation network on a training set in a back propagation mode according to the Dice loss function value of the semantic segmentation map;
C3) inputting the model to be colored into the deep semantic segmentation network, and outputting a corresponding semantic segmentation graph;
C4) and comparing the semantic segmentation graph with the models in the database, calculating the discrete Hausdorff distance and performing geometric matching to obtain an object-level printable multi-material intelligent distribution result of the input model.
5. The method of claim 1, wherein the training parameter is a learning rate of 0.00025.
6. The method for automatically coloring a three-dimensional model based on reinforcement learning of claim 1, wherein the K-Means algorithm is a cost function defined by the following formula a:
wherein:
represents and x(i)The nearest cluster center point.
7. The method of claim 1, wherein the energy function value is calculated by the following equation b:
wherein: s represents a state; pi represents a behavior strategy; gtRepresenting the reward at time t.
8. The reinforcement learning-based three-dimensional model automatic coloring method according to claim 1, wherein the Dice loss function is defined by the following formula c:
wherein: x is an output semantic segmentation graph; y is a corresponding group-route semantic segmentation graph.
9. The method for automatically coloring a three-dimensional model based on reinforcement learning of claim 1, wherein the Hausdorff distance is calculated by the following formula d:
H(A,B)＝max(h(A,B),h(B,A)) (d)；
wherein: A. b is two groups of point sets; and | l | · | is a distance equation between the point set A and the point set B.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010876342.5A CN112017305A (en) | 2020-08-25 | 2020-08-25 | Three-dimensional model automatic coloring method based on reinforcement learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010876342.5A CN112017305A (en) | 2020-08-25 | 2020-08-25 | Three-dimensional model automatic coloring method based on reinforcement learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112017305A true CN112017305A (en) | 2020-12-01 |
Family
ID=73502727
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010876342.5A Pending CN112017305A (en) | 2020-08-25 | 2020-08-25 | Three-dimensional model automatic coloring method based on reinforcement learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112017305A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104875381A (en) * | 2014-02-27 | 2015-09-02 | 致伸科技股份有限公司 | Three-dimensional printing apparatus having coloring function |
| CN105625720A (en) * | 2016-01-05 | 2016-06-01 | 江苏敦超电子科技有限公司 | Multi-material building three-dimensional printing and molding method |
| CN108274742A (en) * | 2017-01-06 | 2018-07-13 | 三纬国际立体列印科技股份有限公司 | Three-dimensional printing coloring method and three-dimensional printing system |
| CN110084245A (en) * | 2019-04-04 | 2019-08-02 | 中国科学院自动化研究所 | The Weakly supervised image detecting method of view-based access control model attention mechanism intensified learning, system |
| US20190385021A1 (en) * | 2018-06-18 | 2019-12-19 | Drvision Technologies Llc | Optimal and efficient machine learning method for deep semantic segmentation |
-
2020
- 2020-08-25 CN CN202010876342.5A patent/CN112017305A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104875381A (en) * | 2014-02-27 | 2015-09-02 | 致伸科技股份有限公司 | Three-dimensional printing apparatus having coloring function |
| CN105625720A (en) * | 2016-01-05 | 2016-06-01 | 江苏敦超电子科技有限公司 | Multi-material building three-dimensional printing and molding method |
| CN108274742A (en) * | 2017-01-06 | 2018-07-13 | 三纬国际立体列印科技股份有限公司 | Three-dimensional printing coloring method and three-dimensional printing system |
| US20190385021A1 (en) * | 2018-06-18 | 2019-12-19 | Drvision Technologies Llc | Optimal and efficient machine learning method for deep semantic segmentation |
| CN110084245A (en) * | 2019-04-04 | 2019-08-02 | 中国科学院自动化研究所 | The Weakly supervised image detecting method of view-based access control model attention mechanism intensified learning, system |
Non-Patent Citations (3)
| Title |
|---|
| XU SHENG ETC.: "3D object recognition using multi-moment and neural network", 《IEEEXPLORE》 * |
| 刘艳申: "工艺亭子三维造型与3D打印", 《电子测试》 * |
| 李明爱等: "基于多个并行CMAC神经网络的强化学习方法", 《系统仿真学报》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020034632A1 (en) | Mbd-based three-dimensional process designing method and platform for typical automobile machined part | |
| CN111291678B (en) | Face image clustering method and device based on multi-feature fusion | |
| CN108595763A (en) | Die assembly design method based on model index technology | |
| CN102254192B (en) | Method and system for semi-automatic marking of three-dimensional (3D) model based on fuzzy K-nearest neighbor | |
| CN107679562B (en) | Analysis processing method and device for three-dimensional model | |
| CN107330458A (en) | A kind of fuzzy C-means clustering method of minimum variance clustering of optimizing initial centers | |
| CN101477529B (en) | Three-dimensional object retrieval method and apparatus | |
| CN108664574A (en) | Input method, terminal device and the medium of information | |
| CN102929894A (en) | Online clustering visualization method of text | |
| CN110119772B (en) | Three-dimensional model classification method based on geometric shape feature fusion | |
| CN104156433A (en) | Image retrieval method based on semantic mapping space construction | |
| CN109359135B (en) | Time sequence similarity searching method based on segment weight | |
| CN104392253A (en) | Interactive classification labeling method for sketch data set | |
| CN110348488A (en) | A kind of modal identification method based on local density's peak value cluster | |
| CN104809186A (en) | Constructing method for mold design and manufacturing knowledge base | |
| CN103279746A (en) | Method and system for identifying faces based on support vector machine | |
| CN109446320A (en) | A kind of biological medicine technology Topics Crawling method based on LDA | |
| CN102646198B (en) | Mode recognition method of mixed linear SVM (support vector machine) classifier with hierarchical structure | |
| CN102063719B (en) | Local matching method of three-dimensional model | |
| CN113593004A (en) | Method and system for customizing three-dimensional model space decoration combination design scheme | |
| CN110765781A (en) | Man-machine collaborative construction method for domain term semantic knowledge base | |
| CN112017305A (en) | Three-dimensional model automatic coloring method based on reinforcement learning | |
| CN116798028A (en) | Automatic dimension marking method for three-dimensional part | |
| Wu et al. | Mvpn: Multi-view prototype network for 3d shape recognition | |
| CN115034837A (en) | Product sales prediction method, equipment and medium based on knowledge graph |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20201201 |
|
| WD01 | Invention patent application deemed withdrawn after publication |