CN107330355A - A kind of depth pedestrian based on positive sample Constraints of Equilibrium identification method again - Google Patents
A kind of depth pedestrian based on positive sample Constraints of Equilibrium identification method again Download PDFInfo
- Publication number
- CN107330355A CN107330355A CN201710330206.4A CN201710330206A CN107330355A CN 107330355 A CN107330355 A CN 107330355A CN 201710330206 A CN201710330206 A CN 201710330206A CN 107330355 A CN107330355 A CN 107330355A
- Authority
- CN
- China
- Prior art keywords
- sample
- network
- positive sample
- equilibrium
- depth
- 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 42
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 238000012360 testing method Methods 0.000 claims description 24
- 230000006870 function Effects 0.000 claims description 22
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 241000208340 Araliaceae Species 0.000 claims description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 2
- 235000008434 ginseng Nutrition 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 4
- 238000013135 deep learning Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000001902 propagating effect Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- 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
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- 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
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of depth pedestrian based on positive sample Constraints of Equilibrium identification method again, the residual error net network structure that this method is used is succinct and is used widely, deep network structure enhances feature representation ability enough, and network structure need not be specifically designed;It was found that carrying out image characteristics extraction with residual error net grader, the accuracy rate that pedestrian identifies again just can be higher than most well-designed method;The method with triple loss is lost compared to two tuples, lift structure loss, which need not specially generate effective sample, just can reach similar effect, and using overall distributed intelligence, the gradient direction learnt is more steadily and surely effective;On the basis of lift structure loss, positive sample Constraints of Equilibrium is added, the distance of positive sample pair can be not only controlled, and positive sample can be balanced and adjusted the distance the gradient adjusted the distance with negative sample so that algorithm is easier training and boosting algorithm performance.
Description
Technical field
The present invention relates to deep learning and pedestrian marking field again, more particularly, to one kind based on positive sample balance about
The depth pedestrian of beam identification method again.
Background technology
In these years, pattern-recognition, machine learning, computer vision research field achieve it is impressive enter
Step.These progress have attracted video monitoring, the attention of law Safety Industry, while the sector is also to these intelligent algorithms and intelligence
The demand of energy system is also being constantly increasing.In the presence of the continuing to develop of Safety Industry, for face monitoring, fingerprint prison
Survey, other biological feature is monitored and the intellectual monitoring instrument of people and urban environment is widely used.These instruments are received
Collect substantial amounts of data, existed generally in the form of image or video, be that machine learning field brings new research topic,
And cause the problem of the larger interest of academia to be then that pedestrian identifies again in recent years.
Generally speaking, processing pedestrian identifies problem again two class methods, is conventional method and deep learning method respectively.Pass
In general system method needs to design or learns feature sane and with identification, and most of is shallow Model,
Feature representation is limited in one's ability;And deep learning network can by learn weight and reach automatic study to need observation which
Effective feature, it is not necessary to which such as traditional method needs engineer's feature, over nearest 1 year, starts increasing research
Person goes solution pedestrian to identify problem again with deep learning method, and has taken good progress.But, present deep learning side
Method utilizes the information of data local distribution mostly, and still uses less hidden layer, and network is not relatively deep, therefore algorithm
It can go up and also be obviously improved space.
The content of the invention
The present invention provides a kind of depth pedestrian based on positive sample Constraints of Equilibrium for improving feature representation ability and identified again
Method.
In order to reach above-mentioned technique effect, technical scheme is as follows:
A kind of depth pedestrian based on positive sample Constraints of Equilibrium identification method again, comprises the following steps:
S1:Input data training datasetWherein, N is
Sample size, d is image pixel, and c is the quantity of different pedestrians in training set, xiIt is the column vector of d dimensions, yi=[yi1,yi2,
yi3,…,yic]TBe c dimension column vector, element therein be equal to 1 or 0, andX=[x1,x2,x3,…,xN],
X is the matrix of d rows N row;
S2:Pre-training is carried out to network using softmax disaggregated models;
S3:Network is trained using the lift structure loss based on positive sample Constraints of Equilibrium;
S4:Feature extraction is carried out to test sample image;
S5:Arest neighbors KNN classification is carried out to test sample using obtained feature and then result is identified again.
Further, the detailed process of the step S2 is:
Training learning rate η and epoch the maximum times T of network is set, and network joined using softmax disaggregated models
Number W carries out pre-training, and the method for training is back-propagation algorithm, and specific step is as follows:
Initialization network parameter W first;If current epoch number is less than T, mini batch data is generatedThen
It is input to network to carry out propagating calculating forward, calculates the loss function value for obtaining this iterationThen basisCarry out back kick
Calculating is broadcast, gradient is calculatedFinally carry out network parameter renewal Network parameter is according to the rule
Constantly updated, until epoch number is equal to T.
Further, the detailed process of the step S3 is:
The loss function of depth network is changed to lift structure from cross entropy to lose, is for training datasetDefinitionThe set of positive sample pair in training sample is represented, andThen represent negative sample pair
Set, positive sample adjusts the distance as Di,j, and negative sample is adjusted the distance as Di,kAnd Di,lIt is the constant ginseng for controlling negative sample to adjust the distance
Number;
The loss function of depth network is changed to lift structure from cross entropy to lose, its specific formula is as follows:
Wherein:
Di,j=| | Ψ (xi)-Ψ(xj)||2
Because this loss function is unsmooth, a very bad Local Extremum of performance is easily absorbed in the training process,
The other function asks gradient also inconvenient, optimizes original function indirectly by optimizing its a smooth upper bound, and structure is damaged
Lose function and refer to the upper bound, wherein increase by two constant parameters β and λ, the former controls positive sample to adjust the distance, the latter's equilibrium gradient:
Training learning rate η and epoch the maximum times T of network, and three constant parameters α, β, λ are set again, using anti-
Depth network is trained to propagation algorithm, the network parameter W optimized is finally given.
Further, the detailed process of the step S4 is:
Feature extraction is carried out to test sample image, test sample image is by query set And test setThe purpose that pedestrian identifies again is to give a query set
Sample xiq, be in test setIn retrieve the image of identical pedestrian, if Ψ () represents depth convolutional network, then to query set
Depth characteristic Ψ (x are extracted with test setiq) and Ψ (xit)。
Further, the detailed process of the step S5 is:
Utilize obtained Ψ (xiq) and Ψ (xit), to query setIn each sample in test setIt is middle to be retrieved, return
The retrieval list including some images returned is in recognition result.
Compared with prior art, the beneficial effect of technical solution of the present invention is:
The residual error net network structure used is succinct and is used widely, and deep network structure enhances mark sheet enough
Danone power, and network structure need not be specifically designed;It was found that carrying out image characteristics extraction with residual error net grader, OK
The accuracy rate that people identifies again just can be higher than most well-designed method;Compared to the loss of two tuples and triple loss
Method, lift structure loss, which need not specially generate effective sample, just can reach similar effect, and using overall
Distributed intelligence, the gradient direction learnt more it is sane effectively.
Brief description of the drawings
Fig. 1 is flow chart of the present invention;
Fig. 2 is 50 layers of residual error schematic network structure;
Fig. 3 (a) is the situation schematic diagram that binary resists loss;
Fig. 3 (b) is the situation schematic diagram that ternary resists loss;
Fig. 3 (c) is that lift structure loses schematic diagram.
Embodiment
Accompanying drawing being given for example only property explanation, it is impossible to be interpreted as the limitation to this patent;
In order to more preferably illustrate the present embodiment, some parts of accompanying drawing have omission, zoomed in or out, and do not represent actual product
Size;
To those skilled in the art, it is to be appreciated that some known features and its explanation, which may be omitted, in accompanying drawing
's.
Technical scheme is described further with reference to the accompanying drawings and examples.
Embodiment 1
As shown in figure 1, a kind of depth pedestrian based on positive sample Constraints of Equilibrium identification method again, comprises the following steps:
S1:Input data training datasetWherein, N is
Sample size, d is image pixel, and c is the quantity of different pedestrians in training set, xiIt is the column vector of d dimensions, yi=[yi1,yi2,
yi3,…,yic]TBe c dimension column vector, element therein be equal to 1 or 0, andX=[x1,x2,x3,…,xN],
X is the matrix of d rows N row;
S2:Pre-training is carried out to network using softmax disaggregated models;
S3:Network is trained using lift structure loss;
S4:Feature extraction is carried out to test sample image;
S5:Arest neighbors KNN classification is carried out to test sample using obtained feature and then result is identified again.
Step S2 detailed process is:
Training learning rate η and epoch the maximum times T of network is set, and network joined using softmax disaggregated models
Number W carries out pre-training, and the method for training is back-propagation algorithm, and specific step is (as shown in Figure 2) as follows:
Initialization network parameter W first;If current epoch number is less than T, mini batch data is generatedThen
It is input to network to carry out propagating calculating forward, calculates the loss function value for obtaining this iterationThen basisCarry out back kick
Calculating is broadcast, gradient is calculatedFinally carry out network parameter renewal Network parameter is according to the rule
Constantly updated, until epoch number is equal to T.
Initialization network parameter W first;If current epoch number is less than T, mini batch data is generatedThen
It is input to network to carry out propagating calculating forward, calculates the loss function value for obtaining this iterationThen basisCarry out back kick
Calculating is broadcast, gradient is calculatedFinally carry out network parameter renewal Network parameter is according to the rule
Constantly updated, until epoch number is equal to T.
Step S3 detailed process is:
As shown in Fig. 3 (a)-(c), the loss function of depth network is changed to lift structure from cross entropy and lost, for instruction
Practicing data set isDefinitionThe set of positive sample pair in training sample is represented, andThen generation
The set of table negative sample pair, positive sample is adjusted the distance as Di,j, and negative sample is adjusted the distance as Di,kAnd Di,lIt is that control negative sample is adjusted the distance
Constant parameter;
The loss function of depth network is changed to lift structure from cross entropy to lose, its specific formula is as follows:
Wherein:
Di,j=| | Ψ (xi)-Ψ(xj)||2
Because this loss function is unsmooth, a very bad Local Extremum of performance is easily absorbed in the training process,
The other function asks gradient also inconvenient, optimizes original function indirectly by optimizing its a smooth upper bound, and structure is damaged
Lose function and refer to the upper bound, wherein increase by two constant parameters β and λ, the former controls positive sample to adjust the distance, the latter's equilibrium gradient:
Training learning rate η and epoch the maximum times T of network, and three constant parameters α, β, λ are set again, using anti-
Depth network is trained to propagation algorithm, the network parameter W optimized is finally given.Fig. 3 (a) represents binary confrontation loss
Situation, rectangle and triangle sample are negative samples pair, rectangle sample can be released when study outside dotted line circle, and just
To being likely to be towards circular sample, this is not inconsistent with intended effect.And Fig. 3 (b) represents ternary loss, similarly also having very much can
Rectangle sample can be pushed to circular sample.Therefore, the learning algorithm based on both losses needs carefully to generate training sample
This set, to reduce the appearance of case above.And for lifting loss function, because it considers each negative sample of sample simultaneously
This and positive sample, make use of integrally-built information, shown in such as Fig. 3 (c), can be well rectangle sample to similar sample
Push, so that preferably reduce variation in group, between-group variation increase.
Step S4 detailed process is:
Feature extraction is carried out to test sample image, test sample image is by query set And test setThe purpose that pedestrian identifies again is to give a query set
Sample xiq, be in test setIn retrieve the image of identical pedestrian, if Ψ () represents depth convolutional network, then to query set
Depth characteristic Ψ (x are extracted with test setiq) and Ψ (xit)。
Step S5 detailed process is:
Utilize obtained Ψ (xiq) and Ψ (xit), to query setIn each sample in test setIt is middle to be retrieved, return
The retrieval list including some images returned is in recognition result.
The same or analogous part of same or analogous label correspondence;
Position relationship is used for being given for example only property explanation described in accompanying drawing, it is impossible to be interpreted as the limitation to this patent;
Obviously, the above embodiment of the present invention is only intended to clearly illustrate example of the present invention, and is not pair
The restriction of embodiments of the present invention.For those of ordinary skill in the field, may be used also on the basis of the above description
To make other changes in different forms.There is no necessity and possibility to exhaust all the enbodiments.It is all this
Any modifications, equivalent substitutions and improvements made within the spirit and principle of invention etc., should be included in the claims in the present invention
Protection domain within.
Claims (5)
1. a kind of depth pedestrian based on positive sample Constraints of Equilibrium identification method again, it is characterised in that comprise the following steps:
S1:Input data training datasetWherein, N is sample
Quantity, d is image pixel, and c is the quantity of different pedestrians in training set, xiIt is the column vector of d dimensions, yi=[yi1,yi2,yi3,…,
yic]TBe c dimension column vector, element therein be equal to 1 or 0, andX=[x1,x2,x3,…,xN], X is d rows N
The matrix of row;
S2:Pre-training is carried out to network using softmax disaggregated models;
S3:Network is trained using the lift structure loss based on positive sample Constraints of Equilibrium;
S4:Feature extraction is carried out to test sample image;
S5:Arest neighbors KNN classification is carried out to test sample using obtained feature and then result is identified again.
2. the depth pedestrian according to claim 1 based on positive sample Constraints of Equilibrium identification method again, it is characterised in that institute
Stating step S2 detailed process is:
Training learning rate η and epoch the maximum times T of network is set, and using softmax disaggregated models to network parameter W
Pre-training is carried out, the method for training is back-propagation algorithm, and specific step is as follows:
Initialization network parameter W first;If current epoch number is less than T, mini batch data is generatedThenInput
To network propagate forward and calculate, calculate the loss function value for obtaining this iterationThen basisCarry out back-propagation meter
Calculate, calculate gradientFinally carry out network parameter renewal Network parameter is carried out according to the rule
Constantly update, until epoch number is equal to T.
3. the depth pedestrian according to claim 2 based on positive sample Constraints of Equilibrium identification method again, it is characterised in that institute
Stating step S3 detailed process is:
The loss function of depth network is changed to lift structure from cross entropy to lose, is for training datasetDefinitionThe set of positive sample pair in training sample is represented, andThen represent negative sample pair
Set, positive sample adjusts the distance as Di,j, and negative sample is adjusted the distance as Di,kAnd Di,lIt is the constant ginseng for controlling negative sample to adjust the distance
Number;
The loss function of depth network is changed to lift structure from cross entropy to lose, its specific formula is as follows:
Wherein:
Because this loss function is unsmooth, a very bad Local Extremum of performance is easily absorbed in the training process, in addition
The function asks gradient also inconvenient, and original function, structural penalties letter are optimized indirectly by optimizing its a smooth upper bound
Number refers to the upper bound, wherein increase by two constant parameters β and λ, the former controls positive sample to adjust the distance, the latter's equilibrium gradient:
Training learning rate η and epoch the maximum times T of network, and three constant parameters α, β, λ are set again, reverse biography is used
Broadcast algorithm to be trained depth network, finally give the network parameter W optimized.
4. the depth pedestrian according to claim 3 based on positive sample Constraints of Equilibrium identification method again, it is characterised in that institute
Stating step S4 detailed process is:
Feature extraction is carried out to test sample image, test sample image is by query set
And test setThe purpose that pedestrian identifies again is to give a query set sample xiq, be
Test setIn retrieve the image of identical pedestrian, if Ψ () represents depth convolutional network, then query set and test set are extracted
Depth characteristic Ψ (xiq) and Ψ (xit)。
5. the depth pedestrian according to claim 4 based on positive sample Constraints of Equilibrium identification method again, it is characterised in that institute
Stating step S5 detailed process is:
Utilize obtained Ψ (xiq) and Ψ (xit), to query setIn each sample in test setIt is middle to be retrieved, return
Retrieval list including some images is in recognition result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710330206.4A CN107330355B (en) | 2017-05-11 | 2017-05-11 | Deep pedestrian re-identification method based on positive sample balance constraint |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710330206.4A CN107330355B (en) | 2017-05-11 | 2017-05-11 | Deep pedestrian re-identification method based on positive sample balance constraint |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107330355A true CN107330355A (en) | 2017-11-07 |
CN107330355B CN107330355B (en) | 2021-01-26 |
Family
ID=60193737
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710330206.4A Expired - Fee Related CN107330355B (en) | 2017-05-11 | 2017-05-11 | Deep pedestrian re-identification method based on positive sample balance constraint |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107330355B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108229435A (en) * | 2018-02-01 | 2018-06-29 | 北方工业大学 | Method for pedestrian recognition |
CN108960342A (en) * | 2018-08-01 | 2018-12-07 | 中国计量大学 | Based on the image similarity calculation method for improving SoftMax loss function |
CN109117891A (en) * | 2018-08-28 | 2019-01-01 | 电子科技大学 | It merges social networks and names across the social media account matching process of feature |
CN109271852A (en) * | 2018-08-07 | 2019-01-25 | 重庆大学 | A kind of processing method that the pedestrian detection based on deep neural network identifies again |
CN109598191A (en) * | 2018-10-23 | 2019-04-09 | 北京市商汤科技开发有限公司 | Pedestrian identifies residual error network training method and device again |
WO2019196130A1 (en) * | 2018-04-12 | 2019-10-17 | 广州飒特红外股份有限公司 | Classifier training method and device for vehicle-mounted thermal imaging pedestrian detection |
CN111382793A (en) * | 2020-03-09 | 2020-07-07 | 腾讯音乐娱乐科技(深圳)有限公司 | Feature extraction method and device and storage medium |
CN113569111A (en) * | 2021-09-24 | 2021-10-29 | 腾讯科技(深圳)有限公司 | Object attribute identification method and device, storage medium and computer equipment |
CN113887561A (en) * | 2021-09-03 | 2022-01-04 | 广东履安实业有限公司 | Face recognition method, device, medium and product based on data analysis |
CN114764942A (en) * | 2022-05-20 | 2022-07-19 | 清华大学深圳国际研究生院 | Difficult positive and negative sample online mining method and face recognition method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130343642A1 (en) * | 2012-06-21 | 2013-12-26 | Siemens Corporation | Machine-learnt person re-identification |
CN104537356A (en) * | 2015-01-12 | 2015-04-22 | 北京大学 | Pedestrian re-identification method and device for carrying out gait recognition through integral scheduling |
CN104915643A (en) * | 2015-05-26 | 2015-09-16 | 中山大学 | Deep-learning-based pedestrian re-identification method |
CN105956606A (en) * | 2016-04-22 | 2016-09-21 | 中山大学 | Method for re-identifying pedestrians on the basis of asymmetric transformation |
-
2017
- 2017-05-11 CN CN201710330206.4A patent/CN107330355B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130343642A1 (en) * | 2012-06-21 | 2013-12-26 | Siemens Corporation | Machine-learnt person re-identification |
CN104537356A (en) * | 2015-01-12 | 2015-04-22 | 北京大学 | Pedestrian re-identification method and device for carrying out gait recognition through integral scheduling |
CN104915643A (en) * | 2015-05-26 | 2015-09-16 | 中山大学 | Deep-learning-based pedestrian re-identification method |
CN105956606A (en) * | 2016-04-22 | 2016-09-21 | 中山大学 | Method for re-identifying pedestrians on the basis of asymmetric transformation |
Non-Patent Citations (2)
Title |
---|
GUANWEN ZHANG ET AL.: "People Re-identifification using Deep Convolutional Neural Network", 《2014 INTERNATIONAL CONFERENCE ON COMPUTER VISION THEORY AND APPLICATION》 * |
张景祥等: "融合异构特征的子空间迁移学习算法", 《自动化学报》 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108229435A (en) * | 2018-02-01 | 2018-06-29 | 北方工业大学 | Method for pedestrian recognition |
CN108229435B (en) * | 2018-02-01 | 2021-03-30 | 北方工业大学 | Method for pedestrian recognition |
WO2019196130A1 (en) * | 2018-04-12 | 2019-10-17 | 广州飒特红外股份有限公司 | Classifier training method and device for vehicle-mounted thermal imaging pedestrian detection |
CN108960342A (en) * | 2018-08-01 | 2018-12-07 | 中国计量大学 | Based on the image similarity calculation method for improving SoftMax loss function |
CN108960342B (en) * | 2018-08-01 | 2021-09-14 | 中国计量大学 | Image similarity calculation method based on improved Soft-Max loss function |
CN109271852A (en) * | 2018-08-07 | 2019-01-25 | 重庆大学 | A kind of processing method that the pedestrian detection based on deep neural network identifies again |
CN109117891B (en) * | 2018-08-28 | 2022-04-08 | 电子科技大学 | Cross-social media account matching method fusing social relations and naming features |
CN109117891A (en) * | 2018-08-28 | 2019-01-01 | 电子科技大学 | It merges social networks and names across the social media account matching process of feature |
CN109598191A (en) * | 2018-10-23 | 2019-04-09 | 北京市商汤科技开发有限公司 | Pedestrian identifies residual error network training method and device again |
CN111382793A (en) * | 2020-03-09 | 2020-07-07 | 腾讯音乐娱乐科技(深圳)有限公司 | Feature extraction method and device and storage medium |
CN113887561A (en) * | 2021-09-03 | 2022-01-04 | 广东履安实业有限公司 | Face recognition method, device, medium and product based on data analysis |
CN113887561B (en) * | 2021-09-03 | 2022-08-09 | 广东履安实业有限公司 | Face recognition method, device, medium and product based on data analysis |
CN113569111B (en) * | 2021-09-24 | 2021-12-21 | 腾讯科技(深圳)有限公司 | Object attribute identification method and device, storage medium and computer equipment |
CN113569111A (en) * | 2021-09-24 | 2021-10-29 | 腾讯科技(深圳)有限公司 | Object attribute identification method and device, storage medium and computer equipment |
CN114764942A (en) * | 2022-05-20 | 2022-07-19 | 清华大学深圳国际研究生院 | Difficult positive and negative sample online mining method and face recognition method |
Also Published As
Publication number | Publication date |
---|---|
CN107330355B (en) | 2021-01-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107330355A (en) | A kind of depth pedestrian based on positive sample Constraints of Equilibrium identification method again | |
CN111985581B (en) | Sample-level attention network-based few-sample learning method | |
CN110750665A (en) | Open set domain adaptation method and system based on entropy minimization | |
CN108427921A (en) | A kind of face identification method based on convolutional neural networks | |
CN113378632A (en) | Unsupervised domain pedestrian re-identification algorithm based on pseudo label optimization | |
CN104462494B (en) | A kind of remote sensing image retrieval method and system based on unsupervised feature learning | |
CN115187610B (en) | Neuron morphological analysis method and device based on graph neural network and storage medium | |
CN110321862B (en) | Pedestrian re-identification method based on compact ternary loss | |
CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
CN112115993B (en) | Zero sample and small sample evidence photo anomaly detection method based on meta-learning | |
CN112633403A (en) | Graph neural network classification method and device based on small sample learning | |
CN111598167B (en) | Small sample image identification method and system based on graph learning | |
CN108960342A (en) | Based on the image similarity calculation method for improving SoftMax loss function | |
CN113032613B (en) | Three-dimensional model retrieval method based on interactive attention convolution neural network | |
Hui et al. | Research on face recognition algorithm based on improved convolution neural network | |
CN114359631A (en) | Target classification and positioning method based on coding-decoding weak supervision network model | |
US20230222768A1 (en) | Multiscale point cloud classification method and system | |
CN115797694A (en) | Display panel microdefect classification method based on multi-scale twin neural network | |
CN117152503A (en) | Remote sensing image cross-domain small sample classification method based on false tag uncertainty perception | |
CN116977844A (en) | Lightweight underwater target real-time detection method | |
CN110111365B (en) | Training method and device based on deep learning and target tracking method and device | |
CN115329120A (en) | Weak label Hash image retrieval framework with knowledge graph embedded attention mechanism | |
CN114419372A (en) | Multi-scale point cloud classification method and system | |
CN104573728B (en) | A kind of texture classifying method based on ExtremeLearningMachine | |
Zhang et al. | Application of gesture recognition based on simulated annealing bp neural network |
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: 20210126 Termination date: 20210511 |