CN107066578A - A kind of 3D based on deep learning and transfer learning draws intelligent recommendation method - Google Patents
A kind of 3D based on deep learning and transfer learning draws intelligent recommendation method Download PDFInfo
- Publication number
- CN107066578A CN107066578A CN201710239554.0A CN201710239554A CN107066578A CN 107066578 A CN107066578 A CN 107066578A CN 201710239554 A CN201710239554 A CN 201710239554A CN 107066578 A CN107066578 A CN 107066578A
- Authority
- CN
- China
- Prior art keywords
- scene
- draws
- picture
- user
- design
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 23
- 238000013526 transfer learning Methods 0.000 title claims abstract description 20
- 238000013135 deep learning Methods 0.000 title claims abstract description 17
- 238000013461 design Methods 0.000 claims abstract description 26
- 238000013480 data collection Methods 0.000 claims abstract description 17
- 230000008447 perception Effects 0.000 claims abstract description 4
- 238000010276 construction Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000013473 artificial intelligence Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 abstract 1
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
Abstract
Intelligent recommendation method is drawn the invention discloses a kind of 3D based on deep learning and transfer learning, first with disclosed large sample scene image data collection, general scene image grader is obtained based on deep learning；Then the 3D collected by user draws scene image data collection and carries out transfer learning, and general Image Classifier is switched into 3D draws scene space grader；Then 3D is built by Hash perception algorithm and draws the information fingerprint storehouse that design recommends picture library；User's photographed scene picture is finally drawn into design storehouse with 3D and matches screening, matching candidate subset is obtained, calculated per the information fingerprint Hamming distance between pictures and user's picture in the subset, then by the minimum 3D figure intelligent recommendations of distance to user.The present invention is based on deep learning and transfer learning, realizes the design drawn in specific environment and space structure to 3D, also shortens the design cycle of 3D pictures.
Description
Technical field
It is more particularly to a kind of to be based on deep learning and migration the present invention relates to a kind of machine learning and image processing field
The 3D of study draws intelligent recommendation method.
Background technology
In recent years, bore hole 3D is drawn with the interaction of its special artistic expression, superpower vision impact and great entertaining
Property is received more and more attention and pursued, and the multiple fields such as finishing, advertisement, exhibition and household is covered, with wide hair
Exhibition prospect.It is a kind of using anti-perspective and the special art form of optical illusion principle that 3D, which is drawn, and need ingenious utilization environment and
Space structure fusion is created.Therefore according to specific environment and space structure carry out 3D draw design be it is more time-consuming, it is right
The experience and level of artist also has certain requirement.Under traditional mode, artist needs skilled with various in drawing
Depending on relation, and possess very strong spatial impression and can just design a secondary good 3D and draw.Some new person painters are also in default of warp
Testing can not independent design 3D picture, a vacancy as 3D picture drafting industries.
The content of the invention
Intelligent recommendation method is drawn the invention provides a kind of 3D based on deep learning and transfer learning, is which overcomed existing
The 3D based on deep learning and transfer learning of technology draws the weak point present in intelligent recommendation method.
The technical solution adopted for the present invention to solve the technical problems is：A kind of 3D based on deep learning and transfer learning
Intelligent recommendation method is drawn, is comprised the following steps：
S1：Build the Image Classifier RCLF based on common image data set；Wherein, common image data set BS is MIT
Computational science Places205 public data collection, convolution model Inception-ResNet is in Places205 public data collection for selection
After upper training, the Image Classifier RCLF that each scene is concentrated in an identification Places205 public data is obtained；
S2：3D based on the Image Classifier RCLF obtained in step S1 draws scene space transfer learning；
S21, keeps the parameter constant of Image Classifier RCLF other layers in addition to softmax layers, by softmax layers of ginseng
Number learning rate tunes up 2 times, turns the learning rate of full articulamentum other specification down half；
S22, the 3D that user oneself is collected draws contextual data collection SS and trains full articulamentum, and scene space letter is drawn to capture 3D
Breath, the 3D for finally obtaining one 4 classification draws scene space grader DCLF；
S3：Build 3D and draw the information fingerprint that design recommends picture library；
S31, collects substantial amounts of 3D and draws design image, be used as result data collection to be recommended；
S32, picture construction information fingerprint FPS is drawn to every 3D of collection；
S33, the Hash for then seeking each image in step S32 using Hash perception algorithm perceives fingerprint；
S4：3D based on reality scene draws design output；
S41, user shoots a reality scene picture Ps, is used as matching service request；
S42, is transmitted to grader DCLF identifications by the picture in step S41, obtains the Intelligent Recognition result to user's scene
scene；
S43, its corresponding information fingerprint fp is calculated to user's scene picture Ps using step S3 methods；
S44, draws searching during design recommends picture library in 3D and belongs to the sample set CSS that scene type is scene；
S45, by the information fingerprint cp of each pictures in sample set CSS and user's scene picture Ps information fingerprint
Fp asks the minimum 3D of Hamming distance in Hamming distance, sample set CSS to draw design picture for recommendation results Ds.
Further, the Places205 data sets be MIT computer science and Artificial Intelligence Laboratory collect more than two
1500000 scene pictures, a total of 205 scene types.
Further, the 3D, which draws contextual data collection SS, includes different color, metope, ground, wall ground and recessed corner face
3D is drawn.
Further, described information fingerprint FPS building process is to each passage pixel value zero setting of center picture region a.
Compared to prior art, the invention has the advantages that：The present invention is based on deep learning and transfer learning, real
The intelligent design that existing 3D is drawn, this method is in any environment space that user needs, it is to avoid because of artist's individual's inspiration and draw a picture
The problem of the problems such as experience, causes design cycle length and difficult design；And on the one hand this method saves the time of training, enters
And shorten the cycle that 3D draws design, the application field of deep learning has on the other hand been expanded significantly.
The present invention is described in further detail below in conjunction with drawings and Examples；But one kind of the present invention is based on depth
Practise and the 3D picture intelligent recommendation methods of transfer learning are not limited to embodiment.
Brief description of the drawings
Fig. 1 is the transfer learning procedural block diagram of the present invention；
Fig. 2 is the recommendation process block diagram of the present invention.
Embodiment
Embodiment, is referred to shown in Fig. 1, Fig. 2, and a kind of 3D based on deep learning and transfer learning of the invention draws intelligence
Recommendation method, comprises the following steps：
S1：The Image Classifier based on common image data set is built, RCLF is named as；Wherein, common image data set,
It is MIT computational science Places205 public data collection to be named as BS, and selection convolution model Inception-ResNet exists
After being trained on Places205 public data collection, the image point that each scene is concentrated in an identification Places205 public data is obtained
Class device RCLF；205 scenes that identifier RCLF can be concentrated to Places205 public datas are according to color, construction, environment etc.
Factor carries out classification processing；
S2：3D based on the Image Classifier RCLF obtained in step S1 draws scene space transfer learning；
S21, keeps the parameter constant of Image Classifier RCLF other layers in addition to softmax layers, by softmax layers of ginseng
Number learning rate tunes up 2 times, turns the learning rate of full articulamentum other specification down half, and the full articulamentum is to connect last layer
Connection intermediary between softmax layers and layer second from the bottom；
S22, the 3D that user oneself is collected draws contextual data collection SS and trains full articulamentum, and scene space letter is drawn to capture 3D
Breath, the 3D for finally obtaining one 4 classification draws scene space grader, is named as DCLF；The spatial classification device DCLF is by step S1
205 scenes of middle Image Classifier RCLF classification are further separated into 4 classes；
S3：Build 3D and draw the information fingerprint that design recommends picture library；
S31, collects substantial amounts of 3D and draws design image, be used as result data collection to be recommended；
S32, draws picture construction information fingerprint storehouse to every 3D of collection, is named as FPS；
S33, the Hash for then seeking each image in step S32 using Hash perception algorithm perceives fingerprint；
S4：3D based on reality scene draws design output；
S41, user shoots a reality scene picture, is designated as Ps, is used as matching service request；
S42, spatial classification device DCLF identifications are transmitted to by the picture Ps in step S41, obtain knowing the intelligence of user's scene
Other result, the result represents that the scene includes the features such as color, the composition of user's scene with scene；
S43, calculates its corresponding information fingerprint using step S3 methods to user's scene picture Ps, is designated as fp；
S44, searching belongs to scene type in 3D draws design recommendation picture library and feature classification in scene is immediate
Picture is designated as CSS as sample set, the sample set；
S45, by the information fingerprint of each pictures in sample set CSS, is designated as cp and user's scene picture Ps letter
Breath fingerprint fp asks the minimum 3D of Hamming distance in Hamming distance, sample set CSS to draw design picture for recommendation results, is designated as
Ds。
In the present embodiment, the Places205 data sets be MIT computer science and Artificial Intelligence Laboratory collect it is super
Cross 2,500,000 scene pictures, a total of 205 scene types；The 3D draw contextual data collection SS include different color,
Metope, ground, wall ground and the 3D in recessed corner face are drawn；Described information fingerprint base FPS building process is each to center picture region a
Passage pixel value zero setting, and center picture region a is the 50% of artwork length and width.
Above-described embodiment is only used for further illustrating that a kind of 3D based on deep learning and transfer learning of the present invention draws intelligence
Method can be recommended, but the invention is not limited in embodiment, every technical spirit according to the present invention is made to above example
Any simple modification, equivalent variations and modification, each fall within the protection domain of technical solution of the present invention.
Claims (4)
1. a kind of 3D based on deep learning and transfer learning draws intelligent recommendation method, it is characterised in that comprise the following steps：
S1：Build the Image Classifier RCLF based on common image data set；Wherein, common image data set BS is that MIT is calculated
Science Places205 public data collection, selection convolution model Inception-ResNet is instructed on Places205 public data collection
After white silk, the Image Classifier RCLF that each scene is concentrated in an identification Places205 public data is obtained；
S2：3D based on the Image Classifier RCLF obtained in step S1 draws scene space transfer learning；
S21, keeps the parameter constant of Image Classifier RCLF other layers in addition to softmax layers, by softmax layers of parametrics
Habit rate tunes up 2 times, turns the learning rate of full articulamentum other specification down half；
S22, the 3D that user oneself is collected draws contextual data collection SS and trains full articulamentum, and scene space information is drawn to capture 3D,
The 3D for finally obtaining one 4 classification draws scene space grader DCLF；
S3：Build 3D and draw the information fingerprint that design recommends picture library；
S31, collects substantial amounts of 3D and draws design image, be used as result data collection to be recommended；
S32, picture construction information fingerprint FPS is drawn to every 3D of collection；
S33, the Hash for then seeking each image in step S32 using Hash perception algorithm perceives fingerprint；
S4：3D based on reality scene draws design output；
S41, user shoots a reality scene picture Ps, is used as matching service request；
S42, is transmitted to grader DCLF identifications by the picture in step S41, obtains the Intelligent Recognition result to user's scene
scene；
S43, its corresponding information fingerprint fp is calculated to user's scene picture Ps using step S3 methods；
S44, draws searching during design recommends picture library in 3D and belongs to the sample set CSS that scene type is scene；
S45, the information fingerprint cp of each pictures in sample set CSS and user's scene picture Ps information fingerprint fp are asked
It is recommendation results Ds that the minimum 3D of Hamming distance, which draws design picture, in Hamming distance, sample set CSS.
2. a kind of 3D based on deep learning and transfer learning according to claim 1 draws intelligent recommendation method, its feature
It is：The Places205 data sets be MIT computer science and Artificial Intelligence Laboratory collect more than 2,500,000
Scene picture, a total of 205 scene types.
3. a kind of 3D based on deep learning and transfer learning according to claim 1 draws intelligent recommendation method, its feature
It is：The 3D, which draws contextual data collection SS, includes different color, metope, ground, wall ground and the 3D in recessed corner face pictures.
4. a kind of 3D based on deep learning and transfer learning according to claim 1 draws intelligent recommendation method, its feature
It is：Described information fingerprint FPS building process is to each passage pixel value zero setting of center picture region a.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710239554.0A CN107066578B (en) | 2017-04-13 | 2017-04-13 | 3D picture intelligent recommendation method based on deep learning and transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710239554.0A CN107066578B (en) | 2017-04-13 | 2017-04-13 | 3D picture intelligent recommendation method based on deep learning and transfer learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107066578A true CN107066578A (en) | 2017-08-18 |
| CN107066578B CN107066578B (en) | 2020-06-26 |
Family
ID=59599586
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710239554.0A Expired - Fee Related CN107066578B (en) | 2017-04-13 | 2017-04-13 | 3D picture intelligent recommendation method based on deep learning and transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107066578B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107526843A (en) * | 2017-09-22 | 2017-12-29 | 四川长虹电器股份有限公司 | A kind of KBS |
| CN111368120A (en) * | 2020-05-28 | 2020-07-03 | 广东博智林机器人有限公司 | Target fingerprint database construction method and device, electronic equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150324690A1 (en) * | 2014-05-08 | 2015-11-12 | Microsoft Corporation | Deep Learning Training System |
| CN105512289A (en) * | 2015-12-07 | 2016-04-20 | 郑州金惠计算机***工程有限公司 | Image retrieval method based on deep learning and Hash |
-
2017
- 2017-04-13 CN CN201710239554.0A patent/CN107066578B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150324690A1 (en) * | 2014-05-08 | 2015-11-12 | Microsoft Corporation | Deep Learning Training System |
| CN105512289A (en) * | 2015-12-07 | 2016-04-20 | 郑州金惠计算机***工程有限公司 | Image retrieval method based on deep learning and Hash |
Non-Patent Citations (2)
| Title |
|---|
| DEREK HOIEM 等: "Representations and techniques for 3D object recognition and scene interpretation", 《SYNTHESIS LECTURES ON ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING》 * |
| 蒋树强 等: "面向智能交互的图像识别技术综述与展望", 《计算机研究与发展》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107526843A (en) * | 2017-09-22 | 2017-12-29 | 四川长虹电器股份有限公司 | A kind of KBS |
| CN111368120A (en) * | 2020-05-28 | 2020-07-03 | 广东博智林机器人有限公司 | Target fingerprint database construction method and device, electronic equipment and storage medium |
| CN111368120B (en) * | 2020-05-28 | 2020-09-15 | 广东博智林机器人有限公司 | Target fingerprint database construction method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107066578B (en) | 2020-06-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107204010B (en) | A kind of monocular image depth estimation method and system | |
| CN105139445B (en) | Scene reconstruction method and device | |
| CN109218619A (en) | Image acquiring method, device and system | |
| CN106203277B (en) | Fixed lens based on SIFT feature cluster monitor video feature extraction method in real time | |
| CN108830252A (en) | A kind of convolutional neural networks human motion recognition method of amalgamation of global space-time characteristic | |
| CN108764065A (en) | A kind of method of pedestrian's weight identification feature fusion assisted learning | |
| CN106778604A (en) | Pedestrian's recognition methods again based on matching convolutional neural networks | |
| CN104732506A (en) | Character picture color style converting method based on face semantic analysis | |
| CN106529499A (en) | Fourier descriptor and gait energy image fusion feature-based gait identification method | |
| CN108121957A (en) | The method for pushing and device of U.S. face material | |
| Ward et al. | Depth director: A system for adding depth to movies | |
| CN108388882A (en) | Based on the gesture identification method that the overall situation-part is multi-modal RGB-D | |
| CN107392131A (en) | A kind of action identification method based on skeleton nodal distance | |
| CN108961220A (en) | A kind of image collaboration conspicuousness detection method based on multilayer convolution Fusion Features | |
| CN110163954A (en) | Three-dimensional house type model generating method, device, equipment and storage medium | |
| CN109740537A (en) | The accurate mask method and system of pedestrian image attribute in crowd's video image | |
| CN102567716A (en) | Face synthetic system and implementation method | |
| CN102034247A (en) | Motion capture method for binocular vision image based on background modeling | |
| CN109657634A (en) | A kind of 3D gesture identification method and system based on depth convolutional neural networks | |
| Liu et al. | Stereo video object segmentation using stereoscopic foreground trajectories | |
| CN108446605B (en) | Double interbehavior recognition methods under complex background | |
| CN107066578A (en) | A kind of 3D based on deep learning and transfer learning draws intelligent recommendation method | |
| Du et al. | Double-channel guided generative adversarial network for image colorization | |
| CN103646397B (en) | Real-time synthetic aperture perspective imaging method based on multisource data fusion | |
| CN114360018A (en) | Rendering method and device of three-dimensional facial expression, storage medium and electronic device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200626 |