CN109697457A - Object weighs the training method of identifying system, object recognition methods and relevant device again - Google Patents
Object weighs the training method of identifying system, object recognition methods and relevant device again Download PDFInfo
- Publication number
- CN109697457A CN109697457A CN201811416733.8A CN201811416733A CN109697457A CN 109697457 A CN109697457 A CN 109697457A CN 201811416733 A CN201811416733 A CN 201811416733A CN 109697457 A CN109697457 A CN 109697457A
- Authority
- CN
- China
- Prior art keywords
- feature vector
- image
- feature
- group
- similarity
- 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
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
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 discloses a kind of training method of object weight identifying system, object recognition methods and relevant device again and carries out the accuracy that object identifies again using this feature extractor to improve to improve the performance of feature extractor.Method includes: that current group training sample image is input in feature extractor, obtains the feature vector of each training sample image, the feature vector constitutive characteristic Vector Groups of each training sample image；Using each feature vector in described eigenvector group as a node, described eigenvector group is handled by random walk module to obtain the similarity matrix for characterizing the similarity between each feature vector, the product of described eigenvector group and the similarity matrix is input in classifier, the classification results classified to current group training sample image are obtained；The parameter of the feature extractor and classifier is adjusted according to the classification results, and next group of training sample image is handled based on feature extractor adjusted, classifier.
Description
Technical field
The present invention relates to computer deep learning field, in particular to the training method of a kind of object weight identifying system and its
Device, a kind of object recognition methods and its device, a kind of processing equipment again.
Background technique
Object weight identification technology plays essential work in fields such as intelligent video monitoring, robot, automatic Pilots
With.Give subject image to be retrieved, object weight identification technology be intended to retrieve from the image that different cameral takes with it is described
The identical associated picture of object.Camera perspective, gestures of object, the influence blocked etc. are so that object weight identification mission has phase
When challenge.
Have benefited from emerging in large numbers for depth learning technology, object weight identification technology is quickly grown in recent years.Current advanced side
Method is all based on deep learning mostly, mainly consists of two parts, i.e. feature extractor and loss function.Feature extractor generally by
Convolutional neural networks are constituted, and loss function is then used as the training of supervisory signals guide features extractor.Therefore, feature extractor
Performance quality directly affects the accuracy that object identifies again, at present the feature extractor technical problem poor there are performance, with
And there is the poor technical problem of the precision for using this feature extractor progress object to identify again.
Summary of the invention
The embodiment of the present invention is in a first aspect, provide the training method and its device of following object weight identifying system, to improve
The performance of feature extractor:
A kind of training method of object weight identifying system, sets gradually after the feature extractor of object weight identifying system
There are random walk module and classifier, following instruction is successively carried out to object weight identifying system according to multiple groups training sample image
Practice:
Current group training sample image is input in feature extractor, the feature vector of each training sample image is obtained,
The feature vector constitutive characteristic Vector Groups of each training sample image；
Using each feature vector in described eigenvector group as a node, by random walk module to the spy
Sign Vector Groups are handled to obtain the similarity matrix for characterizing the similarity between each feature vector；
The product of described eigenvector group and the similarity matrix is input in classifier, is obtained to current group training
The classification results that sample image is classified；
The parameter of the feature extractor and classifier is adjusted according to the classification results, is based on feature extraction adjusted
Device, classifier handle next group of training sample image.
A kind of training device of object weight identifying system, sets gradually after the feature extractor of object weight identifying system
There are random walk module and classifier, described device includes:
Input unit, for multiple groups training sample image to be sequentially inputted to training unit；
Training unit, for according to current group of training sample image receiving to object weight identifying system carry out with
Lower training:
Current group training sample image is input in feature extractor, the feature vector of each training sample image is obtained,
The feature vector constitutive characteristic Vector Groups of each training sample image；
Using each feature vector in described eigenvector group as a node, by random walk module to the spy
Sign Vector Groups are handled to obtain the similarity matrix for characterizing the similarity between each feature vector；
The product of described eigenvector group and the similarity matrix is input in classifier, is obtained to current group training
The classification results that sample image is classified；
The parameter of the feature extractor and classifier is adjusted according to the classification results, is based on feature extraction adjusted
Device, classifier handle next group of training sample image.
A kind of processing equipment, including memory, and the one or more processors with memory communication connection；Institute
The instruction for being stored with and being executed in memory by one or more of processors is stated, described instruction is by one or more of places
It manages device to execute, so that one or more of processors realize the training method of preceding weight identifying system.
The technical solution that first aspect of the embodiment of the present invention provides is provided with random between feature extractor and classifier
Migration module passes through the available similarity that can be indicated between the feature vector of training sample image of the random walk module
Similarity matrix, the product of feature vector group and similarity matrix that the feature vector of training sample image forms is input to
Classifier.Using technical solution of the present invention, on the one hand, using the feature vector of each training sample image as a node, pass through
Random walk module is handled, and actual optimization is that (communication distance refers to random walk for communication distance between two nodes
The primary expected time back and forth is completed between two nodes), unlike the distance between prior art immediate constraint training sample image,
The difficulty of optimization is slowed down, so that optimization has greater flexibility；On the other hand, random walk becomes similar difference
Small, the part loss of different class inheriteds contribution becomes larger, and can be conducive to acquire and has more identification in training sample image
Feature；In another aspect, the result of random walk is not only decided by individualized training sample image, same group of training sample is further depended on
Other training sample images in image, in the training process, the composition of one group of training sample image are possible with many kinds, more
Add abundant and diversity, over-fitting can be slowed down.Therefore, feature extractor can integrally be improved using technical solution of the present invention
Performance.
Second aspect of the embodiment of the present invention provides object recognition methods and its device again, to improve the standard that object identifies again
True property:
A kind of object recognition methods again, which comprises
Picture library image in image to be searched and picture library is input in feature extraction network, obtain each image to be searched and
The feature vector of picture library image；The feature extraction network is the object weight identifying system for first passing through aforementioned first aspect in advance and providing
Training method training obtain；
For every image to be searched, execute following steps: calculate the image to be searched respectively with each picture library image
The similarity of feature, and each picture library image is arranged to obtain image sequence according to the sequence of similarity from high to low；From the figure
As M images before choosing in sequence, image sequence corresponding with the image to be searched is obtained.
A kind of object weight identification device, described device include:
Feature extraction unit, for the picture library image in image to be searched and picture library to be input to the feature extraction network
In, obtain the feature vector of each image to be searched and picture library image；The feature extractor provides to first pass through first aspect in advance
Object weight identifying system training device training obtain；
Weight recognition unit executes following steps: calculating the image difference to be searched for being directed to every image to be searched
With the similarity of the feature of each picture library image, and each picture library image is arranged to obtain image according to the sequence of similarity from high to low
Sequence；M image before choosing from described image sequence obtains image sequence corresponding with the image to be searched.
The technical solution that second aspect of the embodiment of the present invention provides, is logical for carrying out the feature extractor that object identifies again
The training method training for crossing the offer of previous embodiment one obtains, and since the performance of feature extractor is higher, carries out to image
The result of feature extraction is more accurate, extracts result progress object according to this feature and identifies again, can be improved what object identified again
Accuracy.
The third aspect of the embodiment of the present invention provides a kind of object recognition methods and its device again, is identified again with improving object
Accuracy:
A kind of object recognition methods again, is provided with random walk model, method includes: after feature extraction network
Picture library image in image to be searched and picture library is input in the feature extraction network, each figure to be searched is obtained
The feature vector of picture and picture library image；
For every image to be searched, following steps are executed:
Similarity of the image to be searched respectively with the feature of each picture library image is calculated, and by each picture library image according to phase
It arranges to obtain image sequence like the sequence of degree from high to low；
M images before being chosen from described image sequence, the feature vector constitutive characteristic Vector Groups of the M images；
Using each feature vector in described eigenvector group as a node, by random walk module to the spy
Sign Vector Groups are handled to obtain the similarity matrix for characterizing the similarity between each feature vector；
Described eigenvector group is multiplied with the similarity matrix, obtains the new feature vector of M images；
The similarity of the feature vector of the image to be searched respectively with the new feature vector of M images is calculated, by the M
Image is resequenced according to the sequence of similarity from high to low, obtains image sequence corresponding with the image to be searched
Column.
A kind of object weight identification device, is provided with random walk model, described device includes: after feature extraction network
Feature extraction unit, for the picture library image in image to be searched and picture library to be input to the feature extraction network
In, obtain the feature vector of each image to be searched and picture library image；
Weight recognition unit executes following steps: calculating the image difference to be searched for being directed to every image to be searched
With the similarity of the feature of each picture library image, and each picture library image is arranged to obtain image according to the sequence of similarity from high to low
Sequence；M images before being chosen from described image sequence, the feature vector constitutive characteristic Vector Groups of the M images；It will be described
Each feature vector in feature vector group as a node, by random walk module to described eigenvector group at
Reason obtains characterizing the similarity matrix of the similarity between each feature vector；By described eigenvector group and the similarity matrix
It is multiplied, obtains the new feature vector of M images；The feature vector for calculating the image to be searched opens the new spy of image with M respectively
The similarity for levying vector resequences the M images according to the sequence of similarity from high to low, obtain with it is described to
Search for the corresponding image sequence of image.
The third aspect of the embodiment of the present invention, weight recognition unit are obtaining M corresponding with image to be searched figures from picture library
As after, the feature vector of the M images is also input to random walk module with obtain the feature vector of M images of characterization it
Between similarity similarity matrix, then obtain M after the feature vector of M images is multiplied with similarity matrix and open the new of images
Feature vector, then calculate based on new feature vector the similarity of image to be searched and M images, and be based on the similarity pair
M images are resequenced, due to considering the characteristic similarity of M images during rearrangement, so that
It sorts more accurate, more accurate image sequence corresponding with image to be searched can be obtained, identified again to improve object
Accuracy.
Fourth aspect of the embodiment of the present invention provides a kind of processing equipment, including memory, and communicates with the memory
The one or more processors of connection；The instruction that can be executed by one or more of processors is stored in the memory,
Described instruction is executed by one or more of processors so that one or more of processors realize aforementioned second aspect or
The object recognition methods again that the third aspect provides.
Detailed description of the invention
Attached drawing is used to provide further understanding of the present invention, and constitutes part of specification, with reality of the invention
It applies example to be used to explain the present invention together, not be construed as limiting the invention.
Fig. 1 is the training pattern of the training characteristics extractor constructed in the embodiment of the present invention；
Fig. 2 is one of the flow chart of training method of feature extractor in the embodiment of the present invention；
Fig. 3 is random walk module in the embodiment of the present invention to one of the schematic diagram for the treatment of process of feature vector group；
Fig. 4 is to obtain one of the flow chart of similarity matrix based on treatment process shown in Fig. 3 in the embodiment of the present invention；
Fig. 5 be in the embodiment of the present invention random walk module to the two of the schematic diagram of the treatment process of feature vector group；
Fig. 6 be the embodiment of the present invention in obtained based on treatment process shown in Fig. 3 similarity matrix flow chart two；
Fig. 7 is the schematic diagram that object identifies again in the embodiment of the present invention；
Fig. 8 is one of the method flow diagram that object identifies again in the embodiment of the present invention；
Fig. 9 is the specific example that the corresponding associated picture of image to be searched is determined in the embodiment of the present invention；
Figure 10 is the two of the method flow diagram that object identifies again in the embodiment of the present invention；
Figure 11 is the structural schematic diagram of the training device of object weight identifying system in the embodiment of the present invention；
Figure 12 is one of the structural schematic diagram of object weight identification device in the embodiment of the present invention；
Figure 13 is the second structural representation of object weight identification device in the embodiment of the present invention.
Specific embodiment
Technical solution in order to enable those skilled in the art to better understand the present invention, below in conjunction with of the invention real
The attached drawing in example is applied, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described implementation
Example is only a part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, this field is common
Technical staff's every other embodiment obtained without making creative work, all should belong to protection of the present invention
Range.
In technical solution of the present invention, the concrete application scene that object identifies again can be that pedestrian identifies again, vehicle identifies again
Deng the application does not make considered critical.
Embodiment one
The embodiment of the present invention one provides a kind of training method of object weight identifying system, in the feature of object weight identifying system
It is disposed with random walk module and classifier after extractor, as shown in Figure 1, the output of the feature extractor is random
The input of migration module, the product of the output of the output and random walk module of feature extractor are the input of classifier.
It referring to fig. 2, is the flow chart of the training method of object provided in an embodiment of the present invention weight identifying system, according to multiple groups
Training sample image successively carries out following training to object weight identifying system, for ease of description, the embodiment of the present invention will every time just
It is known as current group training sample image in one group of training sample image of processing:
Current group training sample image is input in feature extractor by step 101, obtains the spy of each training sample image
Levy vector, the feature vector constitutive characteristic Vector Groups of each training sample image；
Step 102, using each feature vector in described eigenvector group as a node, pass through random walk module
Described eigenvector group is handled to obtain the similarity matrix for characterizing the similarity between each feature vector；
The product of described eigenvector group and the similarity matrix is input in classifier by step 103, is obtained to working as
The classification results that preceding group of training sample image is classified；
Step 104, the parameter that the feature extractor and classifier are adjusted according to the classification results, based on adjusted
Feature extractor, classifier handle next group of training sample image.
In the embodiment of the present invention, multiple and different target objects, each mesh may include in each group of training sample image
Mark object corresponds to multiple images, for example, target object is pedestrian, one group of training sample figure in the case where pedestrian identifies application scenarios again
As may include 16 pedestrians, corresponding 8 images of each pedestrian.
In some alternative embodiments, feature extractor is trained according to multiple groups training sample image, can be adopted
Use stochastic gradient descent as optimizer, momentum is set as 0.9.It can certainly be trained using existing training method, this
Application does not do considered critical to training method.
In some alternative embodiments, the random walk module to the treatment process of feature vector group as shown in figure 3,
In the step 102, described eigenvector group is handled to obtain by random walk module is characterized between each feature vector
Similarity similarity matrix, can specifically be realized by process as shown in Figure 4, which includes:
Step A1, described eigenvector group is subjected to standardization processing, obtains first eigenvector group；
Step A2, the first eigenvector group is subjected to transposition, obtains second feature Vector Groups；
Step A3, the first eigenvector group is multiplied to obtain similarity matrix with second feature Vector Groups；
Step A4, standardization processing is carried out to the similarity matrix.
In some alternative embodiments, the standardization processing in the step A1 can be L2 standardization processing or L1
Standardization processing etc..
In some alternative embodiments, the random walk module to the treatment process of feature vector group as shown in figure 5,
In the step 102, described eigenvector group is handled to obtain by random walk module is characterized between each feature vector
Similarity similarity matrix, can specifically be realized by process as shown in FIG. 6, which includes:
Step B1, Nonlinear Processing is carried out to described eigenvector group using the first preset nonlinear processing algorithms, obtained
To third feature Vector Groups；
Step B2, standardization processing is carried out to the third feature Vector Groups, obtains fourth feature Vector Groups；
Step B3, transposition is carried out to the fourth feature Vector Groups, obtains fifth feature Vector Groups；
Step B4, Nonlinear Processing is carried out to described eigenvector group using the second preset nonlinear processing algorithms, obtained
To sixth feature Vector Groups；
Step B5, standardization processing is carried out to the sixth feature Vector Groups, obtains seventh feature Vector Groups；
Step B6, the fifth feature Vector Groups are multiplied with the seventh feature Vector Groups, obtain similarity matrix；
Step B7, standardization processing is carried out to the similarity matrix.
In the embodiment of the present invention, there is step B1~step B3 strictly successive to execute sequence, step B4~step B6 tool
Have it is stringent it is successive execute sequence, but there is no stringent successively execution suitable between B1~step B3 and step B4~step B6
Sequence can also first carry out step B1~step B3 and execute step B4~step B6 again, be also possible to first carry out rapid B4~step B6
Step B1~step B3 is executed again, is also possible to step B1~step B3 and step B4~step B6 and is executed parallel, the application is not
Do considered critical.
In the embodiment of the present invention, first nonlinear processing algorithms and the second nonlinear processing algorithms can it is identical can also
With nonlinear function, the neural network for having parameter etc. not identical, such as that printenv can be used.
In the embodiment of the present invention, the N in Fig. 3 and Fig. 5 is characterized the quantity of Vector Groups, and d includes by each feature vector
Characteristic dimension quantity.Feature extractor output feature vector group indicated with X, the similarity matrix that random walk module obtains
Indicated with W, the degree matrix of W is indicated with D, degree matrix D be a diagonal matrix, diagonal entry be defined as each row element of W it
With that is,P indicates the similarity matrix after standardization, the product X ' table of feature vector group and similarity matrix
Show, then can indicate are as follows:
P=D-1W formula (1)
X'=PX formula (2)
In some alternative embodiments, by the first eigenvector group and second feature vector in abovementioned steps A3
During group multiplication obtains similarity matrix, by the fifth feature Vector Groups and the seventh feature vector in step B6
Group is multiplied obtain similarity matrix during, calculate two features being respectively from two different feature vector groups to
It can be calculated using the cosine similarity with Gaussian kernel when similarity between amount, specific calculate can be such as following formula (3):
In formula (3), wijIndicate ith feature vector x i in two different characteristic Vector Groups and j-th of feature to
Measure yjBetween similarity, the quantity for the feature vector that n includes by a feature vector group, σ be Gaussian kernel wideband coefficients.
Certainly, in the embodiment of the present invention, two can also be respectively to calculate by existing similarity calculation algorithm
The similarity between two feature vectors in a different feature vector group, the application do not do considered critical.
In some alternative embodiments, in the method flow of earlier figures 4 and Fig. 6, in step A4 and step B7, specifically
Can using softmax standardization, max-min standardization or 0 mean value standardization etc. normalized fashions to similarity matrix into
Row standardization processing.
For carrying out standardization processing to similarity matrix using softmax standardization, (4) are counted according to the following formula
It calculates:
Embodiment two
The embodiment of the present invention two provides a kind of object recognition methods again, random trip is provided with after feature extraction network
Model is walked, as shown in Figure 7.The process of object recognition methods again is as shown in Figure 8, comprising:
Picture library image in image to be searched and picture library is input in the feature extraction network by step 201, is obtained each
The feature vector of image and picture library image to be searched；
Step 202 is directed to every image to be searched, following steps 202A~step 202E is executed, to obtain every wait search
The corresponding image sequence of rope image:
Step 202A, the image to be searched similarity with the feature of each picture library image respectively is calculated, and by each picture library
Image arranges to obtain image sequence according to the sequence of similarity from high to low；
Step 202B, M images before being chosen from described image sequence, the feature vector constitutive characteristic of the M images
Vector Groups；
Step 202C, using each feature vector in described eigenvector group as a node, pass through random walk mould
Block handles described eigenvector group to obtain the similarity matrix for characterizing the similarity between each feature vector；
Step 202D, described eigenvector group is multiplied with the similarity matrix, obtain the new feature of M image to
Amount；
Step 202E, the feature vector for calculating the image to be searched is similar to the new feature vector of M images respectively
The M images are resequenced according to the sequence of similarity from high to low, are obtained corresponding with the image to be searched by degree
Image sequence.
In the embodiment of the present invention, M can be a fixed constant previously according to empirical value setting, be also possible to preset
The product of picture library total number of images amount in proportion threshold value and picture library, such as picture library total number of images is N, preset proportion threshold value is
10%, then M is N multiplied by 10%.
As shown in Figure 9, it is assumed that need to be searched for picture library A and picture library B, the image that multiple video cameras are passed back is contained in picture library B,
Assuming that in picture library A to be searched comprising d image to be searched (with P1, P2 ..., Pd indicates), include N picture library images in picture library B
(with t1, t2 ..., tN indicate), by taking a certain image Pi to be searched as an example, in picture library B search and the image Pi phase to be searched
The image of pass is described in detail:
Firstly, d image and N to be searched picture library image input feature vectors are extracted in network, the feature of each image is obtained
Vector；
Secondly, the feature vector for calculating image Pi to be searched is similar between the feature vector of N picture library images respectively
Degree is ranked up to obtain image sequence according to the sequence of similarity from high to low to N picture library images；
Then, M images before being chosen from aforementioned image sequence, the feature vector constitutive characteristics of the M images to
Amount group；Feature vector be input in the migration module, obtain the new feature vector of M image；
Subsequently, using each feature vector in preceding feature Vector Groups as a node, pass through random walk module
Described eigenvector group is handled to obtain the similarity matrix for characterizing the similarity between each feature vector；
Finally, described eigenvector group is multiplied with the similarity matrix, the new feature vector of M images is obtained；Meter
The similarity of the feature vector of the image Pi to be searched respectively with the new feature vector of M images is calculated, the M images are pressed
It resequences according to the sequence of similarity from high to low, obtains image sequence corresponding with the image Pi to be searched.
In some alternative embodiments, in the step 202C, by random walk module to described eigenvector group
It is handled to obtain the similarity matrix for characterizing the similarity between each feature vector, be specifically included: by described eigenvector group
Standardization processing is carried out, first eigenvector group is obtained；By the first eigenvector group carry out transposition, obtain second feature to
Amount group；The first eigenvector group is multiplied to obtain similarity matrix with second feature Vector Groups；To the similarity matrix
Carry out standardization processing.For details, reference can be made to the related contents in embodiment one, and details are not described herein.
In some alternative embodiments, in the step 202C, by random walk module to described eigenvector group
It is handled to obtain the similarity matrix for characterizing the similarity between each feature vector, be specifically included: is non-using preset first
Linear process algorithm carries out Nonlinear Processing to described eigenvector group, obtains third feature Vector Groups；To the third feature
Vector Groups carry out standardization processing, obtain fourth feature Vector Groups；Transposition is carried out to the fourth feature Vector Groups, obtains the 5th
Feature vector group；Nonlinear Processing is carried out to described eigenvector group using the second preset nonlinear processing algorithms, obtains the
Six feature vector groups；Standardization processing is carried out to the sixth feature Vector Groups, obtains seventh feature Vector Groups；By the described 5th
Feature vector group is multiplied with the seventh feature Vector Groups, obtains similarity matrix；Standardize to the similarity matrix
Processing.For details, reference can be made to relevant contents in embodiment one, and details are not described herein.
In some alternative embodiments, the feature extractor in the embodiment of the present invention two, which can be, pre- first passes through other instructions
The mode training of white silk obtains, and is also possible to the training method of pre- any one object weight identifying system for first passing through the offer of embodiment one
Training obtains, and the application does not make considered critical.
Embodiment three
Based on the identical design of previous embodiment one, the embodiment of the present invention three provides a kind of object recognition methods again, the party
The process of method is as shown in Figure 10, comprising:
Picture library image in image to be searched and picture library is input in feature extraction network by step 301, is obtained respectively wait search
The feature vector of rope image and picture library image；The feature extraction network is to first pass through any one object that embodiment one provides in advance
The training method training of weight identifying system obtains (specifically can related content in detailed in Example one, details are not described herein)；
Step 302 is directed to every image to be searched, following steps 302A~step 302B is executed, to obtain every wait search
The corresponding image sequence of rope image:
Step 302A, the image to be searched similarity with the feature of each picture library image respectively is calculated, and by each picture library
Image arranges to obtain image sequence according to the sequence of similarity from high to low；
Step 302B, M image before choosing from described image sequence obtains image corresponding with the image to be searched
Sequence.
Example IV
The same idea of the training method of object weight identifying system based on the offer of previous embodiment one, the embodiment of the present invention
Four provide a kind of training device 1 of object weight identifying system, set gradually after the feature extractor of object weight identifying system
There are random walk module and classifier, the structure of the device 1 is as shown in figure 11, comprising:
Input unit 11, for multiple groups training sample image to be sequentially inputted to training unit 12；
Training unit 12, for being carried out according to the current group of training sample image received to object weight identifying system
It trains below:
Current group training sample image is input in feature extractor, the feature vector of each training sample image is obtained,
The feature vector constitutive characteristic Vector Groups of each training sample image；
Using each feature vector in described eigenvector group as a node, by random walk module to the spy
Sign Vector Groups are handled to obtain the similarity matrix for characterizing the similarity between each feature vector；
The product of described eigenvector group and the similarity matrix is input in classifier, is obtained to current group training
The classification results that sample image is classified；
The parameter of the feature extractor and classifier is adjusted according to the classification results, is based on feature extraction adjusted
Device, classifier handle next group of training sample image.
In some alternative embodiments, the training unit 12 by random walk module to described eigenvector group into
Row processing obtains characterizing the similarity matrix of the similarity between each feature vector, specifically includes: by described eigenvector group into
Row standardization processing obtains first eigenvector group；The first eigenvector group is subjected to transposition, obtains second feature vector
Group；The first eigenvector group is multiplied to obtain similarity matrix with second feature Vector Groups；To the similarity matrix into
Row standardization processing.Specifically can related content in detailed in Example one, details are not described herein.
In some alternative embodiments, the training unit 12 by random walk module to described eigenvector group into
Row processing obtains characterizing the similarity matrix of the similarity between each feature vector, specifically includes: non-thread using preset first
Property Processing Algorithm to described eigenvector group carry out Nonlinear Processing, obtain third feature Vector Groups；To the third feature to
Amount group carries out standardization processing, obtains fourth feature Vector Groups；Transposition is carried out to the fourth feature Vector Groups, obtains the 5th spy
Levy Vector Groups；Nonlinear Processing is carried out to described eigenvector group using the second preset nonlinear processing algorithms, obtains the 6th
Feature vector group；Standardization processing is carried out to the sixth feature Vector Groups, obtains seventh feature Vector Groups；It is special by the described 5th
Sign Vector Groups are multiplied with the seventh feature Vector Groups, obtain similarity matrix；The similarity matrix is carried out at standardization
Reason.Specifically can related content in detailed in Example one, details are not described herein.
Embodiment five
Same idea based on the object recognition methods again that previous embodiment two provides, the embodiment of the present invention five provide one kind
Object weight identification device 2, is provided with random walk module after feature extraction network, and described device 2 as shown in figure 12, is wrapped
It includes:
Feature extraction unit 21, for the picture library image in image to be searched and picture library to be input to the feature extraction net
In network, the feature vector of each image to be searched and picture library image is obtained；
Weight recognition unit 22 executes following steps: calculating the image to be searched point for being directed to every image to be searched
Not with the similarity of the feature of each picture library image, and each picture library image is arranged to obtain figure according to the sequence of similarity from high to low
As sequence；M images before being chosen from described image sequence, the feature vector constitutive characteristic Vector Groups of the M images；By institute
Each feature vector in feature vector group is stated as a node, described eigenvector group is carried out by random walk module
Processing obtains characterizing the similarity matrix of the similarity between each feature vector；By described eigenvector group and the similarity moment
Battle array is multiplied, and obtains the new feature vector of M images；The feature vector for calculating the image to be searched opens the new of image with M respectively
The similarity of feature vector resequences the M images according to the sequence of similarity from high to low, obtain with it is described
The corresponding image sequence of image to be searched.
In some alternative embodiments, weight recognition unit 22 carries out described eigenvector group by random walk module
Processing obtains characterizing the similarity matrix of the similarity between each feature vector, specifically includes: described eigenvector group is carried out
Standardization processing obtains first eigenvector group；The first eigenvector group is subjected to transposition, obtains second feature vector
Group；The first eigenvector group is multiplied to obtain similarity matrix with second feature Vector Groups；To the similarity matrix into
Row standardization processing.Specifically can related content in detailed in Example one, details are not described herein.
In some alternative embodiments, weight recognition unit 22 carries out described eigenvector group by random walk module
Processing obtains characterizing the similarity matrix of the similarity between each feature vector, specifically includes: non-linear using preset first
Processing Algorithm carries out Nonlinear Processing to described eigenvector group, obtains third feature Vector Groups；To the third feature vector
Group carries out standardization processing, obtains fourth feature Vector Groups；Transposition is carried out to the fourth feature Vector Groups, obtains fifth feature
Vector Groups；Nonlinear Processing is carried out to described eigenvector group using the second preset nonlinear processing algorithms, obtains the 6th spy
Levy Vector Groups；Standardization processing is carried out to the sixth feature Vector Groups, obtains seventh feature Vector Groups；By the fifth feature
Vector Groups are multiplied with the seventh feature Vector Groups, obtain similarity matrix；Standardization processing is carried out to the similarity matrix.
Specifically can related content in detailed in Example one, details are not described herein.
In the embodiment of the present invention five, the feature extractor can obtain to pass through the training of other training methods in advance,
The training device training for the object weight identifying system that can be provided by example IV obtains.
Embodiment six
Same idea based on the object recognition methods again that previous embodiment three provides, the embodiment of the present invention six provide one kind
Object weight identification device 3,3 structure of device are as shown in figure 13, comprising:
Feature extraction unit 31, for the picture library image in image to be searched and picture library to be input to the feature extraction net
In network, the feature vector of each image to be searched and picture library image is obtained；The feature extractor is to first pass through previous embodiment in advance
The training device training of the four object weight identifying systems provided obtains (content of specific detailed in Example, details are not described herein)；
Weight recognition unit 32 executes following steps: calculating the image to be searched point for being directed to every image to be searched
Not with the similarity of the feature of each picture library image, and each picture library image is arranged to obtain figure according to the sequence of similarity from high to low
As sequence；M image before choosing from described image sequence obtains image sequence corresponding with the image to be searched.
Embodiment seven
The same idea of the training method of object weight identifying system based on the offer of previous embodiment one, the embodiment of the present invention
Seven provide a kind of processing equipment, and the processing equipment includes memory, and one or more with memory communication connection
A processor；
The instruction that can be executed by one or more of processors is stored in the memory, described instruction is by described one
A or multiple processors execute, so that one or more of processors realize that any one feature provided in embodiment one mentions
Take the training method of device.
Embodiment eight
Same idea based on the object recognition methods again that previous embodiment two provides, the embodiment of the present invention eight provide one kind
Processing equipment, the processing equipment include: memory, and the one or more processors with memory communication connection；
The instruction that can be executed by one or more of processors is stored in the memory, described instruction is by described one
A or multiple processors execute, so that one or more of processors are realized as any one embodiment mentions in embodiment two
The recognition methods again of the object of confession.
Embodiment nine
Same idea based on the object recognition methods again that previous embodiment three provides, the embodiment of the present invention nine provide one kind
Processing equipment, the processing equipment include: memory, and the one or more processors with memory communication connection；
The instruction that can be executed by one or more of processors is stored in the memory, described instruction is by described one
A or multiple processors execute, so that one or more of processors realize the object provided such as embodiment three the side of identification again
Method.
Basic principle of the invention is described in conjunction with specific embodiments above, however, it is desirable to, it is noted that general to this field
For logical technical staff, it is to be understood that the whole or any steps or component of methods and apparatus of the present invention can be any
Computing device (including processor, storage medium etc.) perhaps in the network of computing device with hardware firmware, software or they
Combination is realized that this is that those of ordinary skill in the art use the basic of them in the case where having read explanation of the invention
What programming skill can be achieved with.
Those of ordinary skill in the art will appreciate that all or part of the steps that realization above-described embodiment method carries is can
To instruct relevant hardware to complete by program, the program be can store in a kind of computer readable storage medium,
The program when being executed, includes the steps that one or a combination set of embodiment of the method.
It, can also be in addition, each functional unit in each embodiment of the present invention can integrate in a processing module
It is that each unit physically exists alone, can also be integrated in two or more units in a module.Above-mentioned integrated mould
Block both can take the form of hardware realization, can also be realized in the form of software function module.The integrated module is such as
Fruit is realized and when sold or used as an independent product in the form of software function module, also can store in a computer
In read/write memory medium.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the present invention, which can be used in one or more,
The shape for the computer program product implemented in usable storage medium (including but not limited to magnetic disk storage and optical memory etc.)
Formula.
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Although the above embodiment of the present invention has been described, created once a person skilled in the art knows basic
Property concept, then additional changes and modifications can be made to these embodiments.So it includes upper that the following claims are intended to be interpreted as
It states embodiment and falls into all change and modification of the scope of the invention.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (18)
1. a kind of training method of object weight identifying system, which is characterized in that object weight identifying system feature extractor it
After be disposed with random walk module and classifier, according to multiple groups training sample image successively to the object weight identifying system
Carry out following training:
Current group training sample image is input in feature extractor, the feature vector of each training sample image is obtained, it is each to instruct
Practice the feature vector constitutive characteristic Vector Groups of sample image；
Using each feature vector in described eigenvector group as a node, by random walk module to the feature to
Amount group is handled to obtain the similarity matrix for characterizing the similarity between each feature vector；
The product of described eigenvector group and the similarity matrix is input in classifier, is obtained to current group training sample
The classification results that image is classified；
The parameter of the feature extractor and classifier is adjusted according to the classification results, based on feature extractor adjusted,
Classifier handles next group of training sample image.
2. the method according to claim 1, wherein being carried out by random walk module to described eigenvector group
Processing obtains characterizing the similarity matrix of the similarity between each feature vector, specifically includes:
Described eigenvector group is subjected to standardization processing, obtains first eigenvector group；
The first eigenvector group is subjected to transposition, obtains second feature Vector Groups；
The first eigenvector group is multiplied to obtain similarity matrix with second feature Vector Groups；
Standardization processing is carried out to the similarity matrix.
3. the method according to claim 1, wherein being carried out by random walk module to described eigenvector group
Processing obtains characterizing the similarity matrix of the similarity between each feature vector, specifically includes:
Using the first preset nonlinear processing algorithms to described eigenvector group carry out Nonlinear Processing, obtain third feature to
Amount group；
Standardization processing is carried out to the third feature Vector Groups, obtains fourth feature Vector Groups；
Transposition is carried out to the fourth feature Vector Groups, obtains fifth feature Vector Groups；
Using the second preset nonlinear processing algorithms to described eigenvector group carry out Nonlinear Processing, obtain sixth feature to
Amount group；
Standardization processing is carried out to the sixth feature Vector Groups, obtains seventh feature Vector Groups；
The fifth feature Vector Groups are multiplied with the seventh feature Vector Groups, obtain similarity matrix；
Standardization processing is carried out to the similarity matrix.
4. a kind of recognition methods again of object, which is characterized in that be provided with random walk model, method after feature extraction network
Include:
Picture library image in image to be searched and picture library is input in the feature extraction network, obtain each image to be searched and
The feature vector of picture library image；
For every image to be searched, following steps are executed:
Similarity of the image to be searched respectively with the feature of each picture library image is calculated, and by each picture library image according to similarity
Sequence from high to low arranges to obtain image sequence；
M images before being chosen from described image sequence, the feature vector constitutive characteristic Vector Groups of the M images；
Using each feature vector in described eigenvector group as a node, by random walk module to the feature to
Amount group is handled to obtain the similarity matrix for characterizing the similarity between each feature vector；
Described eigenvector group is multiplied with the similarity matrix, obtains the new feature vector of M images；
The similarity of the feature vector of the image to be searched respectively with the new feature vector of M images is calculated, by the M figures
As resequencing according to the sequence of similarity from high to low, image sequence corresponding with the image to be searched is obtained.
5. according to the method described in claim 4, it is characterized in that, being carried out by random walk module to described eigenvector group
Processing obtains characterizing the similarity matrix of the similarity between each feature vector, specifically includes:
Described eigenvector group is subjected to standardization processing, obtains first eigenvector group；
The first eigenvector group is subjected to transposition, obtains second feature Vector Groups；
The first eigenvector group is multiplied to obtain similarity matrix with second feature Vector Groups；
Standardization processing is carried out to the similarity matrix.
6. according to the method described in claim 4, it is characterized in that, being carried out by random walk module to described eigenvector group
Processing obtains characterizing the similarity matrix of the similarity between each feature vector, specifically includes:
Using the first preset nonlinear processing algorithms to described eigenvector group carry out Nonlinear Processing, obtain third feature to
Amount group；
Standardization processing is carried out to the third feature Vector Groups, obtains fourth feature Vector Groups；
Transposition is carried out to the fourth feature Vector Groups, obtains fifth feature Vector Groups；
Using the second preset nonlinear processing algorithms to described eigenvector group carry out Nonlinear Processing, obtain sixth feature to
Amount group；
Standardization processing is carried out to the sixth feature Vector Groups, obtains seventh feature Vector Groups；
The fifth feature Vector Groups are multiplied with the seventh feature Vector Groups, obtain similarity matrix；
Standardization processing is carried out to the similarity matrix.
7. method according to any one of claim 4 to 6, which is characterized in that the feature extractor is to first pass through power in advance
Benefit requires the training method training of 1~3 described in any item object weight identifying systems to obtain.
8. a kind of recognition methods again of object, which is characterized in that the described method includes:
Picture library image in image to be searched and picture library is input in feature extraction network, each image to be searched and picture library are obtained
The feature vector of image；The feature extraction network is to first pass through object weight described in 1~3 any one of preceding claims in advance
The training method training of identifying system obtains；
For every image to be searched, following steps are executed:
Similarity of the image to be searched respectively with the feature of each picture library image is calculated, and by each picture library image according to similarity
Sequence from high to low arranges to obtain image sequence；
M image before choosing from described image sequence obtains image sequence corresponding with the image to be searched.
9. a kind of training device of object weight identifying system, which is characterized in that object weight identifying system feature extractor it
After be disposed with random walk module and classifier, described device includes:
Input unit, for multiple groups training sample image to be sequentially inputted to training unit；
Training unit, for carrying out following instruction to object weight identifying system according to the current group of training sample image received
Practice:
Current group training sample image is input in feature extractor, the feature vector of each training sample image is obtained, it is each to instruct
Practice the feature vector constitutive characteristic Vector Groups of sample image；
Using each feature vector in described eigenvector group as a node, by random walk module to the feature to
Amount group is handled to obtain the similarity matrix for characterizing the similarity between each feature vector；
The product of described eigenvector group and the similarity matrix is input in classifier, is obtained to current group training sample
The classification results that image is classified；
The parameter of the feature extractor and classifier is adjusted according to the classification results, based on feature extractor adjusted,
Classifier handles next group of training sample image.
10. device according to claim 9, which is characterized in that the training unit is by random walk module to described
Feature vector group is handled to obtain the similarity matrix for characterizing the similarity between each feature vector, is specifically included:
Described eigenvector group is subjected to standardization processing, obtains first eigenvector group；
The first eigenvector group is subjected to transposition, obtains second feature Vector Groups；
The first eigenvector group is multiplied to obtain similarity matrix with second feature Vector Groups；
Standardization processing is carried out to the similarity matrix.
11. device according to claim 9, which is characterized in that the training unit is by random walk module to described
Feature vector group is handled to obtain the similarity matrix for characterizing the similarity between each feature vector, is specifically included:
Using the first preset nonlinear processing algorithms to described eigenvector group carry out Nonlinear Processing, obtain third feature to
Amount group；
Standardization processing is carried out to the third feature Vector Groups, obtains fourth feature Vector Groups；
Transposition is carried out to the fourth feature Vector Groups, obtains fifth feature Vector Groups；
Using the second preset nonlinear processing algorithms to described eigenvector group carry out Nonlinear Processing, obtain sixth feature to
Amount group；
Standardization processing is carried out to the sixth feature Vector Groups, obtains seventh feature Vector Groups；
The fifth feature Vector Groups are multiplied with the seventh feature Vector Groups, obtain similarity matrix；
Standardization processing is carried out to the similarity matrix.
12. a kind of object weight identification device, which is characterized in that be provided with random walk model, institute after feature extraction network
Stating device includes:
Feature extraction unit, for the picture library image in image to be searched and picture library to be input in the feature extraction network,
Obtain the feature vector of each image to be searched and picture library image；
Weight recognition unit executes following steps for being directed to every image to be searched: calculating the image to be searched respectively and respectively
The similarity of the feature of picture library image, and each picture library image is arranged to obtain image sequence according to the sequence of similarity from high to low
Column；M images before being chosen from described image sequence, the feature vector constitutive characteristic Vector Groups of the M images；By the spy
Each feature vector in Vector Groups is levied as a node, described eigenvector group is handled by random walk module
Obtain characterizing the similarity matrix of the similarity between each feature vector；By described eigenvector group and the similarity matrix phase
Multiply, obtains the new feature vector of M images；The feature vector for calculating the image to be searched opens the new feature of image with M respectively
The M images are resequenced according to the sequence of similarity from high to low, are obtained with described wait search by the similarity of vector
The corresponding image sequence of rope image.
13. device according to claim 12, which is characterized in that weight recognition unit is by random walk module to the spy
Sign Vector Groups are handled to obtain the similarity matrix for characterizing the similarity between each feature vector, are specifically included:
Described eigenvector group is subjected to standardization processing, obtains first eigenvector group；
The first eigenvector group is subjected to transposition, obtains second feature Vector Groups；
The first eigenvector group is multiplied to obtain similarity matrix with second feature Vector Groups；
Standardization processing is carried out to the similarity matrix.
14. device according to claim 12, which is characterized in that the heavy recognition unit is by random walk module to institute
It states feature vector group to be handled to obtain the similarity matrix for characterizing the similarity between each feature vector, specifically include:
Using the first preset nonlinear processing algorithms to described eigenvector group carry out Nonlinear Processing, obtain third feature to
Amount group；
Standardization processing is carried out to the third feature Vector Groups, obtains fourth feature Vector Groups；
Transposition is carried out to the fourth feature Vector Groups, obtains fifth feature Vector Groups；
Using the second preset nonlinear processing algorithms to described eigenvector group carry out Nonlinear Processing, obtain sixth feature to
Amount group；
Standardization processing is carried out to the sixth feature Vector Groups, obtains seventh feature Vector Groups；
The fifth feature Vector Groups are multiplied with the seventh feature Vector Groups, obtain similarity matrix；
Standardization processing is carried out to the similarity matrix.
15. 1~13 described in any item devices according to claim 1, which is characterized in that the feature extractor is to first pass through in advance
The training device training of the described in any item object weight identifying systems of claim 8~10 obtains.
16. a kind of object weight identification device, which is characterized in that described device includes:
Feature extraction unit, for the picture library image in image to be searched and picture library to be input in the feature extraction network,
Obtain the feature vector of each image to be searched and picture library image；The feature extractor is to first pass through claim 8~10 times in advance
The training device training of the weight identifying system of object described in one obtains；
Weight recognition unit executes following steps for being directed to every image to be searched: calculating the image to be searched respectively and respectively
The similarity of the feature of picture library image, and each picture library image is arranged to obtain image sequence according to the sequence of similarity from high to low
Column；M image before choosing from described image sequence obtains image sequence corresponding with the image to be searched.
17. a kind of processing equipment, which is characterized in that including memory, and one or more with memory communication connection
A processor；
Be stored with the instruction that can be executed by one or more of processors in the memory, described instruction by one or
Multiple processors execute, so that one or more of processors realize object weight according to any one of claims 1 to 3
The training method of identifying system.
18. a kind of processing equipment, which is characterized in that including memory, and one or more with memory communication connection
A processor；
Be stored with the instruction that can be executed by one or more of processors in the memory, described instruction by one or
Multiple processors execute, so that one or more of processors are realized such as claim 4~7, object described in any one of 9
Recognition methods again.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811416733.8A CN109697457B (en) | 2018-11-26 | 2018-11-26 | Training method of object re-recognition system, object re-recognition method and related equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811416733.8A CN109697457B (en) | 2018-11-26 | 2018-11-26 | Training method of object re-recognition system, object re-recognition method and related equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109697457A true CN109697457A (en) | 2019-04-30 |
| CN109697457B CN109697457B (en) | 2021-01-08 |
Family
ID=66230129
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811416733.8A Active CN109697457B (en) | 2018-11-26 | 2018-11-26 | Training method of object re-recognition system, object re-recognition method and related equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109697457B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112529038A (en) * | 2020-11-11 | 2021-03-19 | 联想(北京)有限公司 | Method and device for identifying mainboard material and storage medium |
| CN114360008A (en) * | 2021-12-23 | 2022-04-15 | 上海清鹤科技股份有限公司 | Generation method of face authentication model, authentication method, equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090034791A1 (en) * | 2006-12-04 | 2009-02-05 | Lockheed Martin Corporation | Image processing for person and object Re-identification |
| CN101949241A (en) * | 2010-09-10 | 2011-01-19 | 上海交通大学 | Intelligent switch device and method of car door |
| US20160092736A1 (en) * | 2014-09-30 | 2016-03-31 | C/O Canon Kabushiki Kaisha | System and method for object re-identification |
| CN108062562A (en) * | 2017-12-12 | 2018-05-22 | 北京图森未来科技有限公司 | A kind of object recognition methods and device again |
| CN108108754A (en) * | 2017-12-15 | 2018-06-01 | 北京迈格威科技有限公司 | The training of identification network, again recognition methods, device and system again |
| CN108491872A (en) * | 2018-03-16 | 2018-09-04 | 深圳市商汤科技有限公司 | Target recognition methods and device, electronic equipment, program and storage medium again |
-
2018
- 2018-11-26 CN CN201811416733.8A patent/CN109697457B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090034791A1 (en) * | 2006-12-04 | 2009-02-05 | Lockheed Martin Corporation | Image processing for person and object Re-identification |
| CN101949241A (en) * | 2010-09-10 | 2011-01-19 | 上海交通大学 | Intelligent switch device and method of car door |
| US20160092736A1 (en) * | 2014-09-30 | 2016-03-31 | C/O Canon Kabushiki Kaisha | System and method for object re-identification |
| CN108062562A (en) * | 2017-12-12 | 2018-05-22 | 北京图森未来科技有限公司 | A kind of object recognition methods and device again |
| CN108108754A (en) * | 2017-12-15 | 2018-06-01 | 北京迈格威科技有限公司 | The training of identification network, again recognition methods, device and system again |
| CN108491872A (en) * | 2018-03-16 | 2018-09-04 | 深圳市商汤科技有限公司 | Target recognition methods and device, electronic equipment, program and storage medium again |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112529038A (en) * | 2020-11-11 | 2021-03-19 | 联想(北京)有限公司 | Method and device for identifying mainboard material and storage medium |
| CN112529038B (en) * | 2020-11-11 | 2023-07-25 | 联想(北京)有限公司 | Method and device for identifying main board material and storage medium |
| CN114360008A (en) * | 2021-12-23 | 2022-04-15 | 上海清鹤科技股份有限公司 | Generation method of face authentication model, authentication method, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109697457B (en) | 2021-01-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Huang et al. | Point cloud labeling using 3d convolutional neural network | |
| CN108197532B (en) | The method, apparatus and computer installation of recognition of face | |
| CN107944020A (en) | Facial image lookup method and device, computer installation and storage medium | |
| CN109190635A (en) | Target tracking method, device and electronic equipment based on classification CNN | |
| CN107688823A (en) | A kind of characteristics of image acquisition methods and device, electronic equipment | |
| CN110443969A (en) | A kind of fire point detecting method, device, electronic equipment and storage medium | |
| CN104933428B (en) | A kind of face identification method and device based on tensor description | |
| CN109117897A (en) | Image processing method, device and readable storage medium storing program for executing based on convolutional neural networks | |
| CN110516677A (en) | A kind of neural network recognization model, target identification method and system | |
| CN111008631B (en) | Image association method and device, storage medium and electronic device | |
| CN110069959A (en) | A kind of method for detecting human face, device and user equipment | |
| CN108304859A (en) | Image-recognizing method and cloud system | |
| Sapijaszko et al. | An overview of recent convolutional neural network algorithms for image recognition | |
| CN107948586A (en) | Trans-regional moving target detecting method and device based on video-splicing | |
| CN107886110A (en) | Method for detecting human face, device and electronic equipment | |
| CN109697457A (en) | Object weighs the training method of identifying system, object recognition methods and relevant device again | |
| CN109685805A (en) | A kind of image partition method and device | |
| CN116503399A (en) | Insulator pollution flashover detection method based on YOLO-AFPS | |
| CN115393633A (en) | Data processing method, electronic device, storage medium, and program product | |
| CN108021693A (en) | A kind of image search method and device | |
| CN114707604A (en) | Twin network tracking system and method based on space-time attention mechanism | |
| CN114723784A (en) | Pedestrian motion trajectory prediction method based on domain adaptation technology | |
| CN113762331A (en) | Relational self-distillation method, apparatus and system, and storage medium | |
| Kim et al. | Automatic labeled LiDAR data generation based on precise human model | |
| CN111626212A (en) | Method and device for identifying object in picture, storage medium and electronic device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200331 Address after: 101300, No. two, 1 road, Shunyi Park, Zhongguancun science and Technology Park, Beijing, Shunyi District Applicant after: BEIJING TUSENZHITU TECHNOLOGY Co.,Ltd. Address before: 200120 Shanghai, Nanhui, new towns around the Lake Road West, No. two, building C, Applicant before: TUSIMPLE, Inc. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |