CN108564126A - A kind of special scenes generation method of the semantic control of fusion - Google Patents
A kind of special scenes generation method of the semantic control of fusion Download PDFInfo
- Publication number
- CN108564126A CN108564126A CN201810353922.9A CN201810353922A CN108564126A CN 108564126 A CN108564126 A CN 108564126A CN 201810353922 A CN201810353922 A CN 201810353922A CN 108564126 A CN108564126 A CN 108564126A
- Authority
- CN
- China
- Prior art keywords
- special scenes
- article
- label
- generator
- scene
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of special scenes generation method of the semantic control of fusion, including chooses several article figures and multiple different special scenes pictures for including the article；According to different attribute tags are made in special scenes picture the characteristics of special scenes, after special scenes picture cutting is handled, training sample is obtained；Structure fights network by the condition production that arbiter and generator form；It by article figure and label together as input, is input in generator, generates special scenes figure described in label；Including special scenes figure, target scene figure, article figure and label described in the label generated by generator are input in arbiter by the special scenes figure of article together as target scene figure, arbiter fights network by condition and carries out model training；Pending ware figure and the scene wanted are inputted trained model and can be obtained corresponding scene image in tag form.
Description
Technical field
The invention belongs to machine learning algorithm fields, specifically, relate to a kind of special scenes of the semantic control of fusion
Generation method.
Background technology
The special scenes generation of the semantic control of fusion refers to allowing computer to generate described in language by semanteme control
Scene.The pursuit that the world is always the mankind can really be described, the needs in the world are described in the birth of drawing derived from the mankind, to pole
Art has been achieved in the pursuit of cause.The invention of camera makes the mankind record the world to become easy, and after computer occurs, the mankind start to allow meter
Calculation machine oneself describes real world, and be thus born many generating algorithms.Traditional generating algorithm has gradient orientation histogram,
Scale invariant features transform etc., these algorithms realize the life of target using manual extraction feature with the combined method of shallow Model
At.Its solution follows four steps substantially：Image preprocessing → manual feature extracts → establishes model (grader/recurrence
Device) → output.And the thinking that deep learning algorithm solves computer vision is end-to-end (End to End), i.e., it is direct from input
To output, centre uses the automatic learning characteristic of neural network, avoids the troublesome operation of manual feature extraction.
Deep learning is an important branch of machine learning, because its recent years is due to many fields obtain important breakthrough
It receives significant attention.Production confrontation network (Generative Adversarial Networks, GAN) be 2014 by
A kind of production deep learning model of the propositions such as Goodfellow, the model just become computer vision research once proposition
One of field hot research direction.Due to the outstanding generative capacities of GAN so that GAN generates field in sample and obtains prominent achievement,
Secondly GAN is in image restoring and the mutual generation of reparation, image Style Transfer, text and image, the high quality generation of image etc.
Field also has become a project for having huge applications to be worth.GAN has also been added in many leading enterprises in industrial quarters simultaneously
In the tide of development.Such as the companies such as Facebook, Google, Apple.Based on the above research, GAN is to realize the semantic control of fusion
System generates the possibility that special scenes provide realization.But it can be directly realized by there is presently no a model and be controlled by semanteme
Generate different special scenes.
In order to solve the above problems, people are seeking always a kind of ideal technical solution.
Invention content
The purpose of the present invention is in view of the deficiencies of the prior art, to provide a kind of special scenes of the semantic control of fusion
Generation method.
To achieve the goals above, the technical solution adopted in the present invention is：A kind of special scenes of the semantic control of fusion
Generation method includes the following steps：
Step 1 chooses several article figures and multiple different special scenes pictures for including the article；
Step 2, according to different attribute tags are made in special scenes picture the characteristics of special scenes, by special scenes figure
After piece cutting processing, training sample is obtained, training sample includes article figure, the specific field for including the article corresponding with article figure
Scape figure and the label for describing the scene；
Step 3, structure fight network by the condition production that arbiter and generator form；
Step 4, by article figure and label together as input, be input in generator, generate label described in it is specific
Scene graph；
Step 5, the special scenes figure comprising article are as target scene figure, described in the label generated by generator
Special scenes figure, target scene figure, article figure and label are input in arbiter together, arbiter by condition fight network into
Row model training；
Pending ware figure and the scene wanted are inputted trained model by step 6 in tag form
It can be obtained corresponding scene image.
Based on above-mentioned, the label is the semantic label of binary form.
Based on above-mentioned, in step 1, the article figure is the article close up view crawled from shopping website.
Based on above-mentioned, in step 3, the production confrontation network is GAN models, the generation of the production confrontation network
Device is expressed asWherein, y is target scene image area, and x is original input picture, and l is target scene image
Domain label,For special scenes figure described in label；
The cost function of use condition GAN loses as the antagonism of model, wherein the cost function is
Wherein, D is arbiter, and G is generator.
The present invention has substantive distinguishing features outstanding and significant progress compared with the prior art, specifically：
The present invention fights network progress model training by building condition production, passes through artificial intelligence technology and replaces repeating
Labour, can greatly improve the working efficiency of the mankind, some simple scenes can be generated directly by system, do not have to waste people
Power is gone shooting, is made.Specified scene is generated by semantic control, need to only be provided needed for some scenes for different situations
Training sample, and for training sample make domain label, by training, it will be able to generate the image of given scenario.Side of the present invention
Method has broad application prospects, and shows that the image of commodity details can be generated by this method especially on shopping website, to save
About labour and resource.
Description of the drawings
Fig. 1 is the algorithm flow schematic diagram of the present invention.
Fig. 2 is a kind of design diagram of the special scenes generation method of the semantic control of fusion of the present invention.
Specific implementation mode
Below by specific implementation mode, technical scheme of the present invention will be described in further detail.
As depicted in figs. 1 and 2, the special scenes generation method of the semantic control of a kind of fusion, includes the following steps：
Step 1 crawls several article figures and multiple different special scenes pictures for including the article from shopping website；
Step 2, according to different attribute tags are made in special scenes picture the characteristics of special scenes, the label is two
The semantic label of binary form；By special scenes picture cutting handle after, obtain training sample, training sample include article figure,
The special scenes figure comprising the article corresponding with article figure and the label for describing the scene；
Step 3, structure fight network by the condition production that arbiter and generator form；
Step 4, by article figure and label together as input, be input in generator, generate label described in it is specific
Scene graph；
Step 5, the special scenes figure comprising article are as target scene figure, described in the label generated by generator
Special scenes figure, target scene figure, article figure and label are input in arbiter together, arbiter by condition fight network into
Row model training；
Pending ware figure and the scene wanted are inputted trained model by step 6 in tag form
It can be obtained corresponding scene image.
Specifically, in step 3, the production confrontation network is GAN models, the generator of the production confrontation network
It is expressed asWherein, y is target scene image area, and x is original input picture, and l is target scene image area
Label,For special scenes figure described in label；
In the method for the present invention, each input images of items corresponds to pairs of a target scene image area y and label l,
G is allow accurately to learn to generate special scenes.Arbiter study needs true picture with image classification, generator is generated
Association's deception arbiter, and arbiter generates probability distribution in input images of items and label, can specify label, realizes
The generation of semanteme control generator.The target of generator is that original goods image is converted to the real scene figure described by label
Picture, therefore the data set of training sample is provided as one group of respective image (x, y, l), wherein x is input images of items, y
It is corresponding target scene image, l is target scene image area label.
The cost function of use condition GAN loses as the antagonism of algorithm model, which is a minimum pole
Big double zero-sum game：
Wherein, D is arbiter, and G is generator.
The first item of function shows that arbiter keeps object function as big as possible, and judges when inputting real scene image
It is true picture.The Section 2 of function indicates that G (x, y, l) is as small as possible in the image that input generates, and therefore, loses letter
Several values is relatively large, and generator, which cheats arbiter and is mistakenly considered arbiter while input is true picture, to be attempted it
It is identified as fault image, two models of function are played until reaching Nash Equilibrium, make generator study to the semanteme of label
Feature, and be mapped with images of items.
Network is fought using the production of GAN models, generator inputs the original image of aiming field scene, target area image
With label as conditional-variable, while generating false special scenes, target area image and aiming field label and being replicated in input
And splice with input picture.Generator then attempts from input picture and provides and rebuild new scene in original domain label, and attempts
The special scenes that cannot be distinguished with real scene are generated, make to be not easy to be distinguished by arbiter.The two during fighting game,
The scene that generator generates is more and more true to nature, and arbiter distinguishes real scene image and pseudo- scene image is further difficult, to real
Now trained purpose.
Overall structure of the present invention is simple, reasonable design, using condition GAN as model framework.In order to realize semantic control
Function, algorithm model can receive the training data of multiple fields, and only use a generator and learn all available areas
Between mapping, this algorithm model does not learn fixed generation (for example, only from clothes to positive model) instead of, by article
Image and target information learn the object in input picture being flexibly generated corresponding scene as input.By using
Label indicates domain information, in the training process, generates aiming field label at random, training pattern is converted to input picture
Aiming field converts the input into any desired scene output in the training stage to realize through semantic control domain label, than
Such as input the model for generating that front is stood, hand takes packet, hand to hang down, one model met the requirements for including input clothes of output.
An article figure is inputted, the reasonable scene for including the article is generated.Which overcome two big difficulties, are multiple domain first
Generation, followed by generation are not present in inputting and rational scene.For the first situation, the present invention is by the label of training sample
It indicates in vector form, and corresponding with input picture, target scene, forms mapping, by the training process, generating at random
Input picture is neatly converted to aiming field by one aiming field label, training pattern.By doing so, in the rank using model
Duan Shixian inputs different labels by semantic control domain label for same input picture, you can obtains different fields
Scape realizes the generation of multiple domain.For the second situation, the present invention provides target scene image in the training stage and describes this
The label of scape fights the mapping of e-learning between the two by production, and image is mapped with the text of label,
In training process, generator acquires the graphical representation of text, and arbiter identifies true picture and generates image, rich by fighting
It plays chess, generator generates human eye and can not distinguish true and false special scenes image.
The invention algorithm model structure is simplified, is trained conveniently, is operated steadily, reliably, portable preferable, can be a variety of
It is used in special scenes.
Finally it should be noted that:The above embodiments are merely illustrative of the technical scheme of the present invention and are not intended to be limiting thereof；To the greatest extent
The present invention is described in detail with reference to preferred embodiments for pipe, those of ordinary skills in the art should understand that：Still
It can modify to the specific implementation mode of the present invention or equivalent replacement is carried out to some technical characteristics；Without departing from this hair
The spirit of bright technical solution should all cover within the scope of the technical scheme claimed by the invention.
Claims (4)
1. a kind of special scenes generation method of the semantic control of fusion, which is characterized in that include the following steps：
Step 1 chooses several article figures and multiple different special scenes pictures for including the article；
Step 2, according to different attribute tags are made in special scenes picture the characteristics of special scenes, special scenes picture is cut out
After cutting processing, training sample is obtained, training sample includes article figure, the special scenes figure comprising the article corresponding with article figure
And the label of the scene is described；
Step 3, structure fight network by the condition production that arbiter and generator form；
Step 4, by article figure and label together as input, be input in generator, generate label described in special scenes
Figure；
Step 5, the special scenes figure comprising article, will be specific described in the label generated by generator as target scene figure
Scene graph, target scene figure, article figure and label are input in arbiter together, and arbiter fights network by condition and carries out mould
Type training；
Pending ware figure and the scene wanted are inputted trained model by step 6 in tag form
Obtain corresponding scene image.
2. the special scenes generation method of the semantic control of fusion according to claim 1, it is characterised in that：The label is
The semantic label of binary form.
3. the special scenes generation method of the semantic control of fusion according to claim 2, it is characterised in that：In step 1, institute
It is the article close up view crawled from shopping website to state article figure.
4. the special scenes generation method of the semantic control of fusion according to claim 2, it is characterised in that：In step 3, institute
It is GAN models to state production confrontation network, and the generator of the production confrontation network is expressed asWherein,
Y is target scene image area, and x is original input picture, and l is target scene image area label,It is specific described in label
Scene graph；
The cost function of use condition GAN loses as the antagonism of model, wherein the cost function is
Wherein, D is arbiter, and G is generator.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810353922.9A CN108564126B (en) | 2018-04-19 | 2018-04-19 | Specific scene generation method fusing semantic control |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810353922.9A CN108564126B (en) | 2018-04-19 | 2018-04-19 | Specific scene generation method fusing semantic control |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108564126A true CN108564126A (en) | 2018-09-21 |
| CN108564126B CN108564126B (en) | 2022-04-19 |
Family
ID=63535888
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810353922.9A Active CN108564126B (en) | 2018-04-19 | 2018-04-19 | Specific scene generation method fusing semantic control |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108564126B (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109447137A (en) * | 2018-10-15 | 2019-03-08 | 聚时科技(上海)有限公司 | A kind of image local Style Transfer method based on factoring |
| CN109493417A (en) * | 2018-10-31 | 2019-03-19 | 深圳大学 | Three-dimension object method for reconstructing, device, equipment and storage medium |
| CN109584257A (en) * | 2018-11-28 | 2019-04-05 | 中国科学院深圳先进技术研究院 | A kind of image processing method and relevant device |
| CN109726718A (en) * | 2019-01-03 | 2019-05-07 | 电子科技大学 | A kind of visual scene figure generation system and method based on relationship regularization |
| CN109831352A (en) * | 2019-01-17 | 2019-05-31 | 柳州康云互联科技有限公司 | A kind of detection sample generation system and method for based on confrontation generation network in internet detection |
| CN109871898A (en) * | 2019-02-27 | 2019-06-11 | 南京中设航空科技发展有限公司 | A method of deposit training sample is generated using confrontation network is generated |
| CN110414593A (en) * | 2019-07-24 | 2019-11-05 | 北京市商汤科技开发有限公司 | Image processing method and device, processor, electronic equipment and storage medium |
| CN110516577A (en) * | 2019-08-20 | 2019-11-29 | Oppo广东移动通信有限公司 | Image processing method, device, electronic equipment and storage medium |
| CN110738276A (en) * | 2019-12-19 | 2020-01-31 | 北京影谱科技股份有限公司 | Image material generation method and device, electronic device and computer-readable storage medium |
| CN110766638A (en) * | 2019-10-31 | 2020-02-07 | 北京影谱科技股份有限公司 | Method and device for converting object background style in image |
| CN111563482A (en) * | 2020-06-18 | 2020-08-21 | 深圳天海宸光科技有限公司 | Gas station dangerous scene picture generation method based on GAN |
| CN111754389A (en) * | 2019-03-27 | 2020-10-09 | 通用汽车环球科技运作有限责任公司 | Semantic preserving style transfer |
| WO2021097845A1 (en) * | 2019-11-22 | 2021-05-27 | 驭势（上海）汽车科技有限公司 | Simulation scene image generation method, electronic device and storage medium |
| CN112966742A (en) * | 2021-03-05 | 2021-06-15 | 北京百度网讯科技有限公司 | Model training method, target detection method and device and electronic equipment |
| CN113487629A (en) * | 2021-07-07 | 2021-10-08 | 电子科技大学 | Image attribute editing method based on structured scene and text description |
| CN115086059A (en) * | 2022-06-30 | 2022-09-20 | 北京永信至诚科技股份有限公司 | Deception scene description file generation method and device based on specific language of deception domain |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106096531A (en) * | 2016-05-31 | 2016-11-09 | 安徽省云力信息技术有限公司 | A kind of traffic image polymorphic type vehicle checking method based on degree of depth study |
| CN106878632A (en) * | 2017-02-28 | 2017-06-20 | 北京知慧教育科技有限公司 | A kind for the treatment of method and apparatus of video data |
| US20170278402A1 (en) * | 2016-03-25 | 2017-09-28 | Toyota Jidosha Kabushiki Kaisha | Understanding Road Scene Situation and Semantic Representation of Road Scene Situation for Reliable Sharing |
| US20170308756A1 (en) * | 2016-04-26 | 2017-10-26 | Disney Enterprises, Inc. | Systems and Methods for Identifying Activities in Media Contents Based on Prediction Confidences |
| CN107743072A (en) * | 2017-07-04 | 2018-02-27 | 中国电力科学研究院 | Efficiently expansible network simulation scene generating method |
| CN107832558A (en) * | 2017-11-29 | 2018-03-23 | 闽江学院 | Digital stage intention scene intelligent generation method |
| CN107862293A (en) * | 2017-09-14 | 2018-03-30 | 北京航空航天大学 | Radar based on confrontation generation network generates colored semantic image system and method |
-
2018
- 2018-04-19 CN CN201810353922.9A patent/CN108564126B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170278402A1 (en) * | 2016-03-25 | 2017-09-28 | Toyota Jidosha Kabushiki Kaisha | Understanding Road Scene Situation and Semantic Representation of Road Scene Situation for Reliable Sharing |
| US20170308756A1 (en) * | 2016-04-26 | 2017-10-26 | Disney Enterprises, Inc. | Systems and Methods for Identifying Activities in Media Contents Based on Prediction Confidences |
| CN106096531A (en) * | 2016-05-31 | 2016-11-09 | 安徽省云力信息技术有限公司 | A kind of traffic image polymorphic type vehicle checking method based on degree of depth study |
| CN106878632A (en) * | 2017-02-28 | 2017-06-20 | 北京知慧教育科技有限公司 | A kind for the treatment of method and apparatus of video data |
| CN107743072A (en) * | 2017-07-04 | 2018-02-27 | 中国电力科学研究院 | Efficiently expansible network simulation scene generating method |
| CN107862293A (en) * | 2017-09-14 | 2018-03-30 | 北京航空航天大学 | Radar based on confrontation generation network generates colored semantic image system and method |
| CN107832558A (en) * | 2017-11-29 | 2018-03-23 | 闽江学院 | Digital stage intention scene intelligent generation method |
Non-Patent Citations (4)
| Title |
|---|
| AUGUSTUS ODENA 等: "Conditional Image Synthesis with Auxiliary Classifier GANs", 《PMLR 70》 * |
| LEVENT KARACAN 等: "Learning to Generate Images of Outdoor Scenes from Attributes and Semantic Layouts", 《ARXIV:1612.00215V1》 * |
| 徐一峰: "生成对抗网络理论模型和应用综述", 《金华职业技术学院学报》 * |
| 曹仰杰 等: "生成式对抗网络及其计算机视觉应用研究综述", 《中国图象图形学报》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109447137A (en) * | 2018-10-15 | 2019-03-08 | 聚时科技(上海)有限公司 | A kind of image local Style Transfer method based on factoring |
| CN109493417A (en) * | 2018-10-31 | 2019-03-19 | 深圳大学 | Three-dimension object method for reconstructing, device, equipment and storage medium |
| CN109493417B (en) * | 2018-10-31 | 2023-04-07 | 深圳大学 | Three-dimensional object reconstruction method, device, equipment and storage medium |
| CN109584257A (en) * | 2018-11-28 | 2019-04-05 | 中国科学院深圳先进技术研究院 | A kind of image processing method and relevant device |
| CN109584257B (en) * | 2018-11-28 | 2022-12-09 | 中国科学院深圳先进技术研究院 | Image processing method and related equipment |
| CN109726718A (en) * | 2019-01-03 | 2019-05-07 | 电子科技大学 | A kind of visual scene figure generation system and method based on relationship regularization |
| CN109726718B (en) * | 2019-01-03 | 2022-09-16 | 电子科技大学 | Visual scene graph generation system and method based on relation regularization |
| CN109831352B (en) * | 2019-01-17 | 2022-05-17 | 柳州康云互联科技有限公司 | Detection sample generation system and method based on countermeasure generation network for Internet detection |
| CN109831352A (en) * | 2019-01-17 | 2019-05-31 | 柳州康云互联科技有限公司 | A kind of detection sample generation system and method for based on confrontation generation network in internet detection |
| CN109871898A (en) * | 2019-02-27 | 2019-06-11 | 南京中设航空科技发展有限公司 | A method of deposit training sample is generated using confrontation network is generated |
| CN111754389A (en) * | 2019-03-27 | 2020-10-09 | 通用汽车环球科技运作有限责任公司 | Semantic preserving style transfer |
| CN111754389B (en) * | 2019-03-27 | 2024-04-19 | 通用汽车环球科技运作有限责任公司 | Preserving semantic style transfer |
| CN110414593A (en) * | 2019-07-24 | 2019-11-05 | 北京市商汤科技开发有限公司 | Image processing method and device, processor, electronic equipment and storage medium |
| CN110516577A (en) * | 2019-08-20 | 2019-11-29 | Oppo广东移动通信有限公司 | Image processing method, device, electronic equipment and storage medium |
| CN110516577B (en) * | 2019-08-20 | 2022-07-12 | Oppo广东移动通信有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN110766638A (en) * | 2019-10-31 | 2020-02-07 | 北京影谱科技股份有限公司 | Method and device for converting object background style in image |
| WO2021097845A1 (en) * | 2019-11-22 | 2021-05-27 | 驭势（上海）汽车科技有限公司 | Simulation scene image generation method, electronic device and storage medium |
| CN110738276A (en) * | 2019-12-19 | 2020-01-31 | 北京影谱科技股份有限公司 | Image material generation method and device, electronic device and computer-readable storage medium |
| CN111563482A (en) * | 2020-06-18 | 2020-08-21 | 深圳天海宸光科技有限公司 | Gas station dangerous scene picture generation method based on GAN |
| CN112966742A (en) * | 2021-03-05 | 2021-06-15 | 北京百度网讯科技有限公司 | Model training method, target detection method and device and electronic equipment |
| CN113487629A (en) * | 2021-07-07 | 2021-10-08 | 电子科技大学 | Image attribute editing method based on structured scene and text description |
| CN113487629B (en) * | 2021-07-07 | 2023-04-07 | 电子科技大学 | Image attribute editing method based on structured scene and text description |
| CN115086059A (en) * | 2022-06-30 | 2022-09-20 | 北京永信至诚科技股份有限公司 | Deception scene description file generation method and device based on specific language of deception domain |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108564126B (en) | 2022-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108564126A (en) | A kind of special scenes generation method of the semantic control of fusion | |
| You et al. | Robust visual-textual sentiment analysis: When attention meets tree-structured recursive neural networks | |
| CN106355170A (en) | Photo classifying method and device | |
| Zhang et al. | Prior aided streaming network for multi-task affective recognitionat the 2nd abaw2 competition | |
| Shome et al. | FedAffect: Few-shot federated learning for facial expression recognition | |
| Zhao et al. | Jdnet: A joint-learning distilled network for mobile visual food recognition | |
| Ren et al. | A deep and structured metric learning method for robust person re-identification | |
| CN107066979A (en) | A kind of human motion recognition method based on depth information and various dimensions convolutional neural networks | |
| Zhang et al. | An improved discriminative model prediction approach to real-time tracking of objects with camera as sensors | |
| CN115359394A (en) | Identification method based on multi-mode fusion and application thereof | |
| Shehada et al. | A lightweight facial emotion recognition system using partial transfer learning for visually impaired people | |
| Zhu et al. | NAGNet: A novel framework for real‐time students' sentiment analysis in the wisdom classroom | |
| Radpour et al. | Conditional generative adversarial networks for emoji synthesis with word embedding manipulation | |
| Zhu et al. | A brief review on cycle generative adversarial networks | |
| CN110491372A (en) | A kind of feedback information generating method, device, storage medium and smart machine | |
| Liu et al. | 3D action recognition using multi-temporal skeleton visualization | |
| Eitz et al. | Learning to classify human object sketches | |
| Gao et al. | Face attribute prediction with convolutional neural networks | |
| Li et al. | Human Pose Estimation Based on Lightweight Convolutional Neural Networks | |
| CN115809432B (en) | Crowd social relation extraction method, equipment and storage medium | |
| Matsumoto et al. | Kobe University, NICT and University of Siegen at TRECVID 2016 AVS Task. | |
| Hu et al. | Hand pose estimation with CNN-RNN | |
| Zhao et al. | Self-attention Hypergraph Pooling Network. | |
| Xu et al. | Video-based emotion recognition using aggregated features and spatio-temporal information | |
| Suhail | Graph neural network for situation recognition |
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 |