CN107301400A - A kind of semantic semi-supervised video picture segmentation method being oriented to - Google Patents
A kind of semantic semi-supervised video picture segmentation method being oriented to Download PDFInfo
- Publication number
- CN107301400A CN107301400A CN201710487525.6A CN201710487525A CN107301400A CN 107301400 A CN107301400 A CN 107301400A CN 201710487525 A CN201710487525 A CN 201710487525A CN 107301400 A CN107301400 A CN 107301400A
- Authority
- CN
- China
- Prior art keywords
- semantic
- mrow
- mask
- pixel
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- 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
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- 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
- G06T2207/10016—Video; Image sequence
-
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
A kind of semantic semi-supervised video picture segmentation method being oriented to proposed in the present invention, its main contents include：Convolutional neural networks extract feature, semantic selection and it is semantic propagate, be combined display model with semanteme priori by condition stub device, training network, its process is, first feature is extracted with convolutional neural networks, semantic instance partitioning algorithm is recycled to be used as input, estimate the semanteme of object to be split, then display model is combined with semantic priori by condition stub device, finally train framework, to determine the foreground pixel of specific image, with weights initialisation convolutional neural networks and it is finely adjusted and iteration within the testing time.The influence that the present invention can overcome the change of light or block, effectively extracts the useful information in video, greatly reduces the plenty of time checked video and spent, man power and material；Segmentation is finer, and the degree of accuracy also increases.
Description
Technical field
The present invention relates to video object segmentation field, more particularly, to a kind of semantic semi-supervised object video being oriented to point
Segmentation method.
Background technology
In informationized society of today, video can provide the abundant and comprehensive information content to us, therefore it is more next
More paid attention to by industries such as Modern Traffic, the network media and computer visions.But the letter contained by general original video
Breath amount is all very big, and even most is all little for the meaning of industry research and practical application for which part.Therefore, Wo Menxu
Video is reduced, extract wherein useful information.Video Object Segmentation Technology is exactly the one kind grown up in recent years
The important basic technology of video effective information is extracted, it has been widely used in traffic flow video monitoring, industrial automation prison
In the actual production lives such as control, security protection, network multimedia interaction and video compression coding.However, original method is vulnerable to
The change of light or the influence blocked, and semi-supervised, therefore practical application effect and bad can not be realized.
The present invention proposes a kind of semantic semi-supervised video picture segmentation method being oriented to, and is first extracted with convolutional neural networks
Feature, recycles semantic instance partitioning algorithm as input, estimates the semanteme of object to be split, then will by condition stub device
Display model is combined with semantic priori, is finally trained framework, to determine the foreground pixel of specific image, is used within the testing time
Weights initialisation convolutional neural networks are simultaneously finely adjusted and iteration.The influence that the present invention can overcome the change of light or block, has
Effect extracts the useful information in video, greatly reduces the plenty of time checked video and spent, man power and material；Segmentation is more smart
Carefully, the degree of accuracy also increases.
The content of the invention
The problem of for being vulnerable to light change or blocking influence, it is an object of the invention to provide a kind of semantic guiding
Semi-supervised video picture segmentation method, first extracts feature with convolutional neural networks, recycles semantic instance partitioning algorithm as defeated
Enter, estimate the semanteme of object to be split, then display model is combined with semantic priori by condition stub device, finally trained
Framework, to determine the foreground pixel of specific image, with weights initialisation convolutional neural networks and is finely adjusted within the testing time
And iteration.
To solve the above problems, the present invention provides a kind of semantic semi-supervised video picture segmentation method being oriented to, its is main
Content includes：
(1) convolutional neural networks extract feature；
(2) semantic selection and semantic propagation；
(3) display model is combined with semantic priori by condition stub device；
(4) training network.
Wherein, described convolutional neural networks extract feature, and backbone network is used as using VGG16 convolutional neural networks；Remove
It is fully connected layer and last pond layer, increases space characteristics resolution ratio；Connection is skipped in addition, extracts the feature of hypercolumn, is gathered
Close the multi-scale information from different layers；Second, third, the 4th and the 5th convolutional layer block merge layer accordingly before, from it
Among extract output characteristic figure；Then characteristic pattern is adjusted, makes it identical with input picture size, and they are connected into formation
The characteristic of hypercolumn.
Wherein, described semantic selection and semantic propagation, by the use of semantic instance partitioning algorithm as input, estimate to be split
The semanteme of object；Multitask cascade (MNC) is selected as input example partitioning algorithm；MNC is a multi-stage network, by
Three major part compositions：Network (RPN) and area-of-interest (ROI)-Intelligence Classifier are proposed in shared convolutional layer, region.
Further, described semantic selection, semantic selection occurs in the frame of video first, according to the True Data of demarcation
The mask (being in semi-supervised framework, wherein the true mask of the first frame is input) of mask selection matching object；Selection sense is emerging
Interesting region, is classified, and it is overlapping that the True Data of demarcation and example are segmented into proposal.
Further, described semantic propagation, semantic propagation stage occurs after the first frame, by what is estimated in the first frame
Semanteme travel to after frame；Example segmentation mask is filtered using the estimation of first round prospect, and selects pond top
With object.
Wherein, it is described to be combined display model with semantic priori by condition stub device, use complete convolutional network
Intensive label, be typically expressed as the classification problem of each pixel；Thus, it can be understood that the overall situation slided on the entire image
Grader, and prospect or background label are distributed to by each pixel according to display model；If by the language before final classification
Justice merge, can as example (or one group of example) most possible in front frame mask.
Further, described pixel, for each pixel i, estimates the probability of the foreground pixel of given image：p(i|
I)；Probability can be decomposed into by the sum of k conditional probability of prior weight：
Wherein, K=2.
Further, described condition stub device, builds two condition stub devices, and one is focused on foreground pixel, another
Lay particular emphasis on background pixel；Case-based Reasoning segmentation output estimation priori p (k | I)；Specifically, if pixel is split positioned at example
In mask, then pixel depends on foreground classification device；And if background class mask departs from example segmentation mask, then background class
Device is more important；In an experiment, it regard the space smoothing of selected mask as semantic priori using Gaussian filter.
Further, the layer of described condition stub device, condition stub device can be integrated in end-to-end trainable mode
In a network；The layer is using two prognostic chart f1And f2, and use the semantic weight map ω obtained in advance as input；Wherein
Each input element is multiplied with weight mapping, is then added with the respective element in another mapping：
fout(x, y)=ω (x, y) f1(x,y)+(1-ω(x,y))f2(x,y) (2)
Similarly, in backpropagation step, according to weight map by top gtopGradient travel to two parts：
g1(x, y)=ω (x, y) gtop(x,y) (3)
g2(x, y)=(1- ω (x, y)) gtop(x,y) (4)
Respectively as shown in above formula.
Wherein, described training network, first, uses the VGG convolution of the weights initialisation of the training in advance architecture
The part of neutral net；The purpose of training framework is to determine the foreground pixel of specific image；
Then, it is absorbed in the special object to be split in video sequence study display model, weight is used within the testing time
Initialization convolutional neural networks are simultaneously finely adjusted, and are taken several iterations；In order to produce segmentation in each frame, to video sequence
Special object application trim network, obtains the mask corresponding with object, with the transmission of single forward direction.
Brief description of the drawings
Fig. 1 is a kind of system flow chart of the semantic semi-supervised video picture segmentation method being oriented to of the present invention.
Fig. 2 is a kind of schematic flow sheet of the semantic semi-supervised video picture segmentation method being oriented to of the present invention.
Fig. 3 is that the semantic selection and semanteme of a kind of semantic semi-supervised video picture segmentation method being oriented to of the present invention are propagated.
Fig. 4 is a kind of condition stub device of the semantic semi-supervised video picture segmentation method being oriented to of the present invention.
Embodiment
It should be noted that in the case where not conflicting, the feature in embodiment and embodiment in the application can phase
Mutually combine, the present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings.
Fig. 1 is a kind of system flow chart of the semantic semi-supervised video picture segmentation method being oriented to of the present invention.Mainly include
Convolutional neural networks extract feature, semantic selection and it is semantic propagate, by condition stub device by display model and semanteme priori phase
With reference to training network.
Convolutional neural networks extract feature, and backbone network is used as using VGG16 convolutional neural networks；Removal be fully connected layer and
Last pond layer, increases space characteristics resolution ratio；Connection is skipped in addition, extracts the feature of hypercolumn, and polymerization comes from different layers
Multi-scale information；Second, third, the 4th and the 5th convolutional layer block merge layer accordingly before, extract defeated among them
Go out characteristic pattern；Then characteristic pattern is adjusted, makes it identical with input picture size, and they are connected to the spy for forming hypercolumn
Property.
Display model is combined with semantic priori by condition stub device, using the intensive label of complete convolutional network,
It is typically expressed as the classification problem of each pixel；Thus, it can be understood that the global classification device slided on the entire image, and
Prospect or background label are distributed to by each pixel according to display model；, can be with if the semanteme before final classification merged
It is used as the mask of example (or one group of example) most possible in front frame.
For each pixel i, the probability of the foreground pixel of given image is estimated：p(i|I)；Probability can be decomposed into by elder generation
The sum of k conditional probability of preceding weighting：
Wherein, K=2.
Training network, first, uses the portion of the VGG convolutional neural networks of the weights initialisation of the training in advance architecture
Point；The purpose of training framework is to determine the foreground pixel of specific image；
Then, it is absorbed in the special object to be split in video sequence study display model, weight is used within the testing time
Initialization convolutional neural networks are simultaneously finely adjusted, and are taken several iterations；In order to produce segmentation in each frame, to video sequence
Special object application trim network, obtains the mask corresponding with object, with the transmission of single forward direction.
Fig. 2 is a kind of schematic flow sheet of the semantic semi-supervised video picture segmentation method being oriented to of the present invention.First use convolution
Neutral net extracts feature, recycles semantic instance partitioning algorithm as input, estimates the semanteme of object to be split, then pass through
Display model is combined by condition stub device with semantic priori, finally trains framework, to determine the foreground pixel of specific image,
With weights initialisation convolutional neural networks and it is finely adjusted and iteration in testing time.
Fig. 3 is that the semantic selection and semanteme of a kind of semantic semi-supervised video picture segmentation method being oriented to of the present invention are propagated.
By the use of semantic instance partitioning algorithm as input, the semanteme of object to be split is estimated；Select multitask cascade (MNC) conduct
Input example partitioning algorithm；MNC is a multi-stage network, is made up of three major parts：Net is proposed in shared convolutional layer, region
Network (RPN) and area-of-interest (ROI)-Intelligence Classifier.
Semantic selection occurs in the frame of video first, is selected to match the mask of object according to the True Data mask of demarcation
(being in semi-supervised framework, wherein the true mask of the first frame is input)；Area-of-interest is selected, is classified, will demarcated
The segmentation proposal of True Data and example it is overlapping.
Semantic propagation stage occurs after the first frame, the frame after the semanteme estimated in the first frame is traveled to；Use
The estimation of first round prospect is filtered to example segmentation mask, and selects matching object at the top of pond.
Fig. 4 is a kind of condition stub device of the semantic semi-supervised video picture segmentation method being oriented to of the present invention.Build two
Condition stub device, an emphasis foreground pixel, another lays particular emphasis on background pixel；Case-based Reasoning segmentation output estimation priori p
(k|I)；Specifically, if pixel is located in example segmentation mask, pixel depends on foreground classification device；And if background
Mask of classifying departs from example segmentation mask, then background class device is more important；In an experiment, using Gaussian filter by selected mask
Space smoothing be used as semantic priori.
Condition stub device can be integrated in a network in end-to-end trainable mode；The layer is using two prognostic chart f1With
f2, and use the semantic weight map ω obtained in advance as input；Wherein each input element is multiplied with weight mapping, then
It is added with the respective element in another mapping：
fout(x, y)=ω (x, y) f1(x,y)+(1-ω(x,y))f2(x,y) (2)
Similarly, in backpropagation step, according to weight map by top gtopGradient travel to two parts：
g1(x, y)=ω (x, y) gtop(x,y) (3)
g2(x, y)=(1- ω (x, y)) gtop(x,y) (4)
Respectively as shown in above formula.
For those skilled in the art, the present invention is not restricted to the details of above-described embodiment, in the essence without departing substantially from the present invention
In the case of refreshing and scope, the present invention can be realized with other concrete forms.In addition, those skilled in the art can be to this hair
Bright to carry out various changes and modification without departing from the spirit and scope of the present invention, these improvement and modification also should be regarded as the present invention's
Protection domain.Therefore, appended claims are intended to be construed to include preferred embodiment and fall into all changes of the scope of the invention
More and modification.
Claims (10)
1. a kind of semantic semi-supervised video picture segmentation method being oriented to, it is characterised in that mainly carried including convolutional neural networks
Take feature (one)；Semantic selection and semantic propagation (two)；Display model is combined with semantic priori by condition stub device
(3)；Training network (four).
2. extract feature (one) based on the convolutional neural networks described in claims 1, it is characterised in that use VGG16 convolution
Neutral net is used as backbone network；Remove and be fully connected layer and last pond layer, increase space characteristics resolution ratio；Add the company of skipping
Connect, extract the feature of hypercolumn, polymerize the multi-scale information from different layers；Second, third, the 4th and the 5th convolutional layer
Block merges before layer accordingly, and output characteristic figure is extracted among them；Then characteristic pattern is adjusted, makes itself and input picture size
It is identical, and they are connected to the characteristic for forming hypercolumn.
3. based on the semantic selection described in claims 1 and semantic propagation (two), it is characterised in that utilize semantic instance segmentation
Algorithm estimates the semanteme of object to be split as input；Multitask cascade (MNC) is selected to be calculated as input example segmentation
Method；MNC is a multi-stage network, is made up of three major parts：Shared convolutional layer, region are proposed network (RPN) and felt emerging
Interesting region (ROI)-Intelligence Classifier.
4. based on the semantic selection described in claims 3, it is characterised in that semantic selection occurs in the frame of video first, root
(it is according to the mask of the True Data mask selection matching object of demarcation in semi-supervised framework, wherein the true mask of the first frame
For input)；Area-of-interest is selected, is classified, it is overlapping that the True Data of demarcation and example are segmented into proposal.
5. propagated based on the semanteme described in claims 3, it is characterised in that semantic propagation stage occurs after the first frame,
Frame after the semanteme estimated in first frame is traveled to；Example segmentation mask is filtered using the estimation of first round prospect,
And select matching object at the top of pond.
6. based on display model is combined into (three) with semantic priori by condition stub device described in claims 1, it is special
Levy and be, using the intensive label of complete convolutional network, be typically expressed as the classification problem of each pixel；Accordingly, it is to be understood that
For the global classification device slided on the entire image, and prospect or background label are distributed to by each picture according to display model
Element；If the semanteme before final classification merged, example (or one group of example) most possible in front frame can be used as
Mask.
7. based on the pixel described in claims 6, it is characterised in that for each pixel i, estimate the prospect picture of given image
The probability of element：p(i|I)；Probability can be decomposed into by the sum of k conditional probability of prior weight：
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>|</mo>
<mi>I</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>K</mi>
</munderover>
<mi>p</mi>
<mo>(</mo>
<mrow>
<mi>i</mi>
<mo>|</mo>
<mi>I</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
<mo>)</mo>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>|</mo>
<mi>I</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, K=2.
8. based on the condition stub device described in claims 6, it is characterised in that build two condition stub devices, an emphasis
Foreground pixel, another lays particular emphasis on background pixel；Case-based Reasoning segmentation output estimation priori p (k | I)；Specifically, if
Pixel is located in example segmentation mask, then pixel depends on foreground classification device；And if background class mask departs from example point
Mask is cut, then background class device is more important；In an experiment, it regard the space smoothing of selected mask as semanteme using Gaussian filter
Priori.
9. the layer based on the condition stub device described in claims 8, it is characterised in that condition stub device can with it is end-to-end can
The mode of training is integrated in a network；The layer is using two prognostic chart f1And f2, and use semantic in advance as input acquisition
Weight map ω；Wherein each input element is multiplied with weight mapping, is then added with the respective element in another mapping：
fout(x, y)=ω (x, y) f1(x,y)+(1-ω(x,y))f2(x,y) (2)
Similarly, in backpropagation step, according to weight map by top gtopGradient travel to two parts：
g1(x, y)=ω (x, y) gtop(x,y) (3)
g2(x, y)=(1- ω (x, y)) gtop(x,y) (4)
Respectively as shown in above formula.
10. based on the training network (four) described in claims 1, it is characterised in that first, at the beginning of using the weight of training in advance
The part of the VGG convolutional neural networks of the beginningization architecture；The purpose of training framework is to determine the foreground pixel of specific image；
Then, it is absorbed in the special object to be split in video sequence study display model, it is initial with weight within the testing time
Change convolutional neural networks and be finely adjusted, take several iterations；In order to produce segmentation in each frame, to the specific of video sequence
Object application trim network, obtains the mask corresponding with object, with the transmission of single forward direction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710487525.6A CN107301400A (en) | 2017-06-23 | 2017-06-23 | A kind of semantic semi-supervised video picture segmentation method being oriented to |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710487525.6A CN107301400A (en) | 2017-06-23 | 2017-06-23 | A kind of semantic semi-supervised video picture segmentation method being oriented to |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107301400A true CN107301400A (en) | 2017-10-27 |
Family
ID=60135037
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710487525.6A Withdrawn CN107301400A (en) | 2017-06-23 | 2017-06-23 | A kind of semantic semi-supervised video picture segmentation method being oriented to |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107301400A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107704862A (en) * | 2017-11-06 | 2018-02-16 | 深圳市唯特视科技有限公司 | A kind of video picture segmentation method based on semantic instance partitioning algorithm |
| CN108053410A (en) * | 2017-12-11 | 2018-05-18 | 厦门美图之家科技有限公司 | Moving Object Segmentation method and device |
| CN108256562A (en) * | 2018-01-09 | 2018-07-06 | 深圳大学 | Well-marked target detection method and system based on Weakly supervised space-time cascade neural network |
| CN108875732A (en) * | 2018-01-11 | 2018-11-23 | 北京旷视科技有限公司 | Model training and example dividing method, device and system and storage medium |
| CN108898618A (en) * | 2018-06-06 | 2018-11-27 | 上海交通大学 | A kind of Weakly supervised video object dividing method and device |
| CN109635812A (en) * | 2018-11-29 | 2019-04-16 | 中国科学院空间应用工程与技术中心 | The example dividing method and device of image |
| CN109902800A (en) * | 2019-01-22 | 2019-06-18 | 北京大学 | The method of multistage backbone network detection generic object based on quasi- Feedback Neural Network |
| CN109949317A (en) * | 2019-03-06 | 2019-06-28 | 东南大学 | Based on the semi-supervised image instance dividing method for gradually fighting study |
| CN109948616A (en) * | 2019-03-26 | 2019-06-28 | 北京迈格威科技有限公司 | Image detecting method, device, electronic equipment and computer readable storage medium |
| CN110009556A (en) * | 2018-01-05 | 2019-07-12 | 广东欧珀移动通信有限公司 | Image background weakening method, device, storage medium and electronic equipment |
| CN110889851A (en) * | 2018-09-11 | 2020-03-17 | 苹果公司 | Robust use of semantic segmentation for depth and disparity estimation |
| CN111046917A (en) * | 2019-11-20 | 2020-04-21 | 南京理工大学 | Object-based enhanced target detection method based on deep neural network |
| CN111292334A (en) * | 2018-12-10 | 2020-06-16 | 北京地平线机器人技术研发有限公司 | Panoramic image segmentation method and device and electronic equipment |
| CN111541900A (en) * | 2020-04-28 | 2020-08-14 | 济南浪潮高新科技投资发展有限公司 | Security and protection video compression method, device, equipment and storage medium based on GAN |
| CN111652930A (en) * | 2020-06-04 | 2020-09-11 | 上海媒智科技有限公司 | Image target detection method, system and equipment |
| CN112166438A (en) * | 2018-03-13 | 2021-01-01 | 雷哥尼公司 | Deterministic token data generation and artificial intelligence training approaches |
| CN112351928A (en) * | 2018-07-10 | 2021-02-09 | 铁路视像有限公司 | Railway obstacle detection method and system based on track segmentation |
| CN112489060A (en) * | 2020-12-07 | 2021-03-12 | 北京医准智能科技有限公司 | System and method for pneumonia focus segmentation |
-
2017
- 2017-06-23 CN CN201710487525.6A patent/CN107301400A/en not_active Withdrawn
Non-Patent Citations (1)
| Title |
|---|
| SERGI CEALLES等: "Semantically-Guided Video Object Segmentation", 《HTTPS://ARXIV.ORG/ABS/1704.01926》 * |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107704862A (en) * | 2017-11-06 | 2018-02-16 | 深圳市唯特视科技有限公司 | A kind of video picture segmentation method based on semantic instance partitioning algorithm |
| CN108053410A (en) * | 2017-12-11 | 2018-05-18 | 厦门美图之家科技有限公司 | Moving Object Segmentation method and device |
| CN110009556A (en) * | 2018-01-05 | 2019-07-12 | 广东欧珀移动通信有限公司 | Image background weakening method, device, storage medium and electronic equipment |
| US11410277B2 (en) | 2018-01-05 | 2022-08-09 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and device for blurring image background, storage medium and electronic apparatus |
| CN108256562A (en) * | 2018-01-09 | 2018-07-06 | 深圳大学 | Well-marked target detection method and system based on Weakly supervised space-time cascade neural network |
| CN108256562B (en) * | 2018-01-09 | 2022-04-15 | 深圳大学 | Salient target detection method and system based on weak supervision time-space cascade neural network |
| CN108875732A (en) * | 2018-01-11 | 2018-11-23 | 北京旷视科技有限公司 | Model training and example dividing method, device and system and storage medium |
| CN108875732B (en) * | 2018-01-11 | 2022-07-12 | 北京旷视科技有限公司 | Model training and instance segmentation method, device and system and storage medium |
| CN112166438A (en) * | 2018-03-13 | 2021-01-01 | 雷哥尼公司 | Deterministic token data generation and artificial intelligence training approaches |
| CN108898618B (en) * | 2018-06-06 | 2021-09-24 | 上海交通大学 | Weak surveillance video object segmentation method and device |
| CN108898618A (en) * | 2018-06-06 | 2018-11-27 | 上海交通大学 | A kind of Weakly supervised video object dividing method and device |
| CN112351928B (en) * | 2018-07-10 | 2023-11-10 | 铁路视像有限公司 | Railway obstacle detection method and system based on track segmentation |
| CN112351928A (en) * | 2018-07-10 | 2021-02-09 | 铁路视像有限公司 | Railway obstacle detection method and system based on track segmentation |
| JP7343531B2 (en) | 2018-07-10 | 2023-09-12 | レール ビジョン リミテッド | Method and system for detecting railway obstacles based on rail segmentation |
| US12079721B2 (en) | 2018-07-10 | 2024-09-03 | Rail Vision Ltd. | Method and system for railway obstacle detection based on rail segmentation |
| JP2021530388A (en) * | 2018-07-10 | 2021-11-11 | レール ビジョン リミテッドRail Vision Ltd | Methods and systems for detecting railroad obstacles based on rail segmentation |
| CN110889851A (en) * | 2018-09-11 | 2020-03-17 | 苹果公司 | Robust use of semantic segmentation for depth and disparity estimation |
| CN109635812A (en) * | 2018-11-29 | 2019-04-16 | 中国科学院空间应用工程与技术中心 | The example dividing method and device of image |
| CN111292334B (en) * | 2018-12-10 | 2023-06-09 | 北京地平线机器人技术研发有限公司 | Panoramic image segmentation method and device and electronic equipment |
| CN111292334A (en) * | 2018-12-10 | 2020-06-16 | 北京地平线机器人技术研发有限公司 | Panoramic image segmentation method and device and electronic equipment |
| CN109902800B (en) * | 2019-01-22 | 2020-11-27 | 北京大学 | Method for detecting general object by using multi-stage backbone network based on quasi-feedback neural network |
| CN109902800A (en) * | 2019-01-22 | 2019-06-18 | 北京大学 | The method of multistage backbone network detection generic object based on quasi- Feedback Neural Network |
| CN109949317A (en) * | 2019-03-06 | 2019-06-28 | 东南大学 | Based on the semi-supervised image instance dividing method for gradually fighting study |
| CN109948616B (en) * | 2019-03-26 | 2021-05-25 | 北京迈格威科技有限公司 | Image detection method and device, electronic equipment and computer readable storage medium |
| CN109948616A (en) * | 2019-03-26 | 2019-06-28 | 北京迈格威科技有限公司 | Image detecting method, device, electronic equipment and computer readable storage medium |
| CN111046917A (en) * | 2019-11-20 | 2020-04-21 | 南京理工大学 | Object-based enhanced target detection method based on deep neural network |
| CN111541900B (en) * | 2020-04-28 | 2022-05-17 | 山东浪潮科学研究院有限公司 | Security and protection video compression method, device, equipment and storage medium based on GAN |
| CN111541900A (en) * | 2020-04-28 | 2020-08-14 | 济南浪潮高新科技投资发展有限公司 | Security and protection video compression method, device, equipment and storage medium based on GAN |
| CN111652930A (en) * | 2020-06-04 | 2020-09-11 | 上海媒智科技有限公司 | Image target detection method, system and equipment |
| CN111652930B (en) * | 2020-06-04 | 2024-02-27 | 上海媒智科技有限公司 | Image target detection method, system and equipment |
| CN112489060A (en) * | 2020-12-07 | 2021-03-12 | 北京医准智能科技有限公司 | System and method for pneumonia focus segmentation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107301400A (en) | A kind of semantic semi-supervised video picture segmentation method being oriented to | |
| CN113486726B (en) | Rail transit obstacle detection method based on improved convolutional neural network | |
| CN109509192B (en) | Semantic segmentation network integrating multi-scale feature space and semantic space | |
| Liu et al. | Multi-objective convolutional learning for face labeling | |
| CN109886286A (en) | Object detection method, target detection model and system based on cascade detectors | |
| CN108537824B (en) | Feature map enhanced network structure optimization method based on alternating deconvolution and convolution | |
| CN109543502A (en) | A kind of semantic segmentation method based on the multiple dimensioned neural network of depth | |
| CN108021923A (en) | A kind of image characteristic extracting method for deep neural network | |
| CN112489050A (en) | Semi-supervised instance segmentation algorithm based on feature migration | |
| CN110889360A (en) | Crowd counting method and system based on switching convolutional network | |
| CN111931572B (en) | Target detection method for remote sensing image | |
| Adiba et al. | Transfer learning and U-Net for buildings segmentation | |
| Sun et al. | A Metaverse text recognition model based on character-level contrastive learning | |
| Chacon-Murguia et al. | Moving object detection in video sequences based on a two-frame temporal information CNN | |
| CN112802048B (en) | Method and device for generating layer generation countermeasure network with asymmetric structure | |
| Chen et al. | Ship tracking for maritime traffic management via a data quality control supported framework | |
| Shahbaz et al. | Moving object detection based on deep atrous spatial features for moving camera | |
| Gong et al. | MSAug: Multi-Strategy Augmentation for rare classes in semantic segmentation of remote sensing images | |
| CN117011530A (en) | Underwater image instance segmentation method, system, storage medium and electronic equipment | |
| Colovic et al. | End-to-end training of hybrid CNN-CRF models for semantic segmentation using structured learning | |
| CN112200055B (en) | Pedestrian attribute identification method, system and device of combined countermeasure generation network | |
| Ran et al. | Adaptive fusion and mask refinement instance segmentation network for high resolution remote sensing images | |
| Dong et al. | SiameseDenseU‐Net‐based Semantic Segmentation of Urban Remote Sensing Images | |
| Norelyaqine et al. | Architecture of Deep Convolutional Encoder‐Decoder Networks for Building Footprint Semantic Segmentation | |
| Wang et al. | Remote sensing image semantic segmentation based on cascaded Transformer |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20171027 |
|
| WW01 | Invention patent application withdrawn after publication |