CN110084290A - Method, apparatus, electronic equipment and the computer readable storage medium of training classifier - Google Patents
Method, apparatus, electronic equipment and the computer readable storage medium of training classifier Download PDFInfo
- Publication number
- CN110084290A CN110084290A CN201910292542.3A CN201910292542A CN110084290A CN 110084290 A CN110084290 A CN 110084290A CN 201910292542 A CN201910292542 A CN 201910292542A CN 110084290 A CN110084290 A CN 110084290A
- Authority
- CN
- China
- Prior art keywords
- attribute value
- data items
- subclass
- training
- classifier
- 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
- G06F18/24—Classification techniques
-
- 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
-
- 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
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The present disclosure discloses a kind of methods of trained classifier, it is characterized in that, it include: to obtain training set, the training set includes that the first subclass and second subset are closed, data items in first subclass correspond to first property value, corresponding second attribute value of data items in the second subset conjunction, the first property value are less than second attribute value；The data items of the first quantity in second subset conjunction are corresponded into the first property value, the data items of the second quantity in first subclass are corresponded into second attribute value；Gather training classifier according to the training.Method, apparatus, electronic equipment and the computer readable storage medium for the training classifier that the embodiment of the present disclosure provides, for the data items in the subclass in the training set of classifier, change its corresponding attribute value, the classifier is enabled to export more smooth classification results.
Description
Technical field
This disclosure relates to field of information processing more particularly to a kind of method, apparatus, electronic equipment and the meter of trained classifier
Calculation machine readable storage medium storing program for executing.
Background technique
With the progress of internet the relevant technologies, application relevant to image is more abundant, such as can be by image
Analysis adjust automatically image attribute to obtain different display effects.
By taking more popular U.S. face application as an example, have the application of U.S. face function can be by classifier to the face of shooting
Image classify and corrected according to classification results.Specifically, industry is based on to amount of calculation and counting accuracy
Consider, target category can be divided into standard class to the left by typical classifier, standard class, standard class to the right totally three classifications, often
A classification is corresponding with specific attribute value, such as the classifier for classifying according to shape of face, and standard class to the left is corresponding
Shape of face attribute value is, for example, 0, represents sharp face, and the corresponding shape of face attribute value of standard class is, for example, 1, represents standard shape of face, standard deviation
The corresponding shape of face attribute value of right class is, for example, 2, represents round face, and the classification results that classifier exports the image of input include needle
To the probability distribution of each target category, wherein for probability distribution of all categories and be 1, such as aforementioned shape of face attribute
The probability distribution of three classifications is (0,0.01,0.99), then according to the probability distribution and corresponding shape of face attribute value of all categories
It can be desired for 0*0+1*0.01+2*0.99=1.99 with the shape of face attribute value of calculating input image, so that it is determined that input picture
In face belong to standard class to the right i.e. round face, and then function of the correction of " thin face " to realize U.S. face is carried out based on the expectation 1.99
Energy.
Since classifier is by being trained to a large amount of data items, in order to make the classification results of classifier
It is more significant, it is often well-chosen according to target category for training the data items of classifier.Such as aforementioned
The classifier classified according to shape of face can tend to select with sharper when selecting the data items for training classifier
The image tagged of shape of face be standard class to the left (such as by attribute value of described image be labeled as 0), select with more round face
The image tagged of type is standard class to the right (such as the attribute value of described image is labeled as 2), according to these well-chosen numbers
The classification results of the classifier trained according to project, output can be more significant, i.e., classification results are for some target class
Higher probability can not shown as.Therefore it will lead to following problem: since the classification results of classifier output are more significant, thus
Leading to the expectation of the attribute value calculated based on the classification results can be distorted, and then carries out correction according to the expectation of the distortion and will lead to
Excessively, justing think may belong to a kind of slightly round but be not the shape of face justified very much for the shape of face in the image of input for correction, but point
Class device is likely to provide the classification results of such as (0,0.01,0.99), and then based on the calculated expectation 1.99 of the classification results
The correction for carrying out " thin face " causes correction excessive.
Summary of the invention
The embodiment of the present disclosure provides the method for training classifier, device, electronic equipment and computer readable storage medium,
For the data items in the subclass in the training set of classifier, change its corresponding attribute value, so that the classifier
More smooth classification results can be exported.
In a first aspect, the embodiment of the present disclosure provides a kind of method of trained classifier characterized by comprising obtain instruction
Practice set, the training set includes that the first subclass and second subset are closed, and the data items in first subclass are corresponding
First property value, the second subset close in corresponding second attribute value of data items, the first property value is less than described the
Two attribute values；By the second subset close in the data items of the first quantity correspond to the first property value, by described the
The data items of the second quantity in one subclass correspond to second attribute value；According to the training set training classification
Device.
Further, first quantity and second quantity meet one in following relationship: first quantity
It is equal with second quantity and be greater than 0；First quantity and second quantity are unequal.
Further, the training set further includes third subclass, and the data items in the third subclass are corresponding
Third attribute value, the third attribute value are greater than second attribute value；Before gathering training classifier according to the training,
The method also includes: the data items of the third quantity in second subset conjunction are corresponded into the third attribute value, it will
The data items of the 4th quantity in the third subclass correspond to second attribute value.
Further, the third quantity and the 4th quantity meet one in following relationship: the third quantity
It is equal with the 4th quantity and be greater than 0；The third quantity and the 4th quantity are unequal.
Further, the data items in the training set include attribute value label, and the attribute value label is for marking
Remember the corresponding attribute value of the data items.
Further, the classification results of the classifier output include corresponding with each subclass of the training set
Each class probability, with it is described training set each subclass correspondingly each class probability and be 1.
Further, the data items for the subclass that the training set includes correspond respectively to different attribute values,
Corresponding attribute value is less than in the subclass of second attribute value, the corresponding first property value of first subclass and institute
The distance for stating the second attribute value is minimum.
Further, the data items for the subclass that the training set includes correspond respectively to different attribute values,
Corresponding attribute value is greater than in the subclass of second attribute value, the corresponding third attribute value of the third subclass and institute
The distance for stating the second attribute value is minimum.
Second aspect, the embodiment of the present disclosure provide a kind of device of trained classifier characterized by comprising obtain mould
Block, for obtaining trained set, the training set includes that the first subclass and second subset are closed, the number of first subclass
First property value is corresponded to according to project, the data items that the second subset is closed correspond to the second attribute value, and the first property value is small
In second attribute value；Attribute value respective modules, the data items pair for the first quantity in closing the second subset
First property value described in Ying Yu is also used to the data items of the second quantity in first subclass corresponding to described second
Attribute value；Training module, for gathering training classifier according to the training.
Further, first quantity and second quantity meet one in following relationship: first quantity
It is equal with second quantity and be greater than 0；First quantity and second quantity are unequal.
Further, the training set further includes third subclass, and the data items in the third subclass are corresponding
Third attribute value, the third attribute value are greater than second attribute value；The attribute value respective modules are also used to: by described
The data items of third quantity in two subclass correspond to the third attribute value, by the 4th number in the third subclass
The data items of amount correspond to second attribute value.
Further, the third quantity and the 4th quantity meet one in following relationship: the third quantity
It is equal with the 4th quantity and be greater than 0；The third quantity and the 4th quantity are unequal.
Further, the data items in the training set include attribute value label, and the attribute value label is for marking
Remember the corresponding attribute value of the data items.
Further, the classification results of the classifier output include corresponding with each subclass of the training set
Each class probability, with it is described training set each subclass correspondingly each class probability and be 1.
Further, the data items for the subclass that the training set includes correspond respectively to different attribute values,
Corresponding attribute value is less than in the subclass of second attribute value, the corresponding first property value of first subclass and institute
The distance for stating the second attribute value is minimum.
Further, the data items for the subclass that the training set includes correspond respectively to different attribute values,
Corresponding attribute value is greater than in the subclass of second attribute value, the corresponding third attribute value of the third subclass and institute
The distance for stating the second attribute value is minimum.
The third aspect, the embodiment of the present disclosure provide a kind of electronic equipment, comprising: memory, it is computer-readable for storing
Instruction；And the one or more processors coupled with the memory, for running the computer-readable instruction, so that institute
State the method that any trained classifier in aforementioned first aspect is realized when processor operation.
Fourth aspect, the embodiment of the present disclosure provide a kind of non-transient computer readable storage medium, which is characterized in that described
Non-transient computer readable storage medium stores computer instruction, when the computer instruction is computer-executed, so that institute
The method for stating any trained classifier that computer executes in aforementioned first aspect.
The present disclosure discloses method, apparatus, electronic equipment and the computer readable storage mediums of a kind of trained classifier.Its
Described in training classifier method characterized by comprising obtain training set, it is described training set include the first subset
It closes and second subset is closed, the data items in first subclass correspond to first property value, the number in the second subset conjunction
According to corresponding second attribute value of project, the first property value is less than second attribute value；By the second subset close in the
The data items of one quantity correspond to the first property value, by the data items pair of the second quantity in first subclass
Second attribute value described in Ying Yu；Gather training classifier according to the training.The training classifier that the embodiment of the present disclosure provides
Method, apparatus, electronic equipment and computer readable storage medium, for the data in the subclass in the training set of classifier
Project changes its corresponding attribute value, the classifier is enabled to export more smooth classification results.
Above description is only the general introduction of disclosed technique scheme, in order to better understand the technological means of the disclosure, and
It can be implemented in accordance with the contents of the specification, and to allow the above and other objects, features and advantages of the disclosure can be brighter
Show understandable, it is special below to lift preferred embodiment, and cooperate attached drawing, detailed description are as follows.
Detailed description of the invention
In order to illustrate more clearly of the embodiment of the present disclosure or technical solution in the prior art, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this public affairs
The some embodiments opened for those of ordinary skill in the art without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is the flow chart of the embodiment of the method one for the training classifier that the embodiment of the present disclosure provides；
Fig. 2 is the flow chart of the embodiment of the method two for the training classifier that the embodiment of the present disclosure provides；
Fig. 3 is the structural schematic diagram of the Installation practice for the training classifier that the embodiment of the present disclosure provides；
Fig. 4 is the structural schematic diagram of the electronic equipment provided according to the embodiment of the present disclosure.
Specific embodiment
Illustrate embodiment of the present disclosure below by way of specific specific example, those skilled in the art can be by this specification
Disclosed content understands other advantages and effect of the disclosure easily.Obviously, described embodiment is only the disclosure
A part of the embodiment, instead of all the embodiments.The disclosure can also be subject to reality by way of a different and different embodiment
It applies or applies, the various details in this specification can also be based on different viewpoints and application, in the spirit without departing from the disclosure
Lower carry out various modifications or alterations.It should be noted that in the absence of conflict, the feature in following embodiment and embodiment can
To be combined with each other.Based on the embodiment in the disclosure, those of ordinary skill in the art are without creative efforts
Every other embodiment obtained belongs to the range of disclosure protection.
It should be noted that the various aspects of embodiment within the scope of the appended claims are described below.Ying Xian
And be clear to, aspect described herein can be embodied in extensive diversified forms, and any specific structure described herein
And/or function is only illustrative.Based on the disclosure, it will be understood by one of ordinary skill in the art that one described herein
Aspect can be independently implemented with any other aspect, and can combine the two or both in these aspects or more in various ways.
For example, carry out facilities and equipments in terms of any number set forth herein can be used and/or practice method.In addition, can make
With other than one or more of aspect set forth herein other structures and/or it is functional implement this equipment and/or
Practice the method.
It should also be noted that, diagram provided in following embodiment only illustrates the basic structure of the disclosure in a schematic way
Think, component count, shape and the size when only display is with component related in the disclosure rather than according to actual implementation in diagram are drawn
System, when actual implementation kenel, quantity and the ratio of each component can arbitrarily change for one kind, and its assembly layout kenel can also
It can be increasingly complex.
In addition, in the following description, specific details are provided for a thorough understanding of the examples.However, fields
The skilled person will understand that the aspect can be practiced without these specific details.
The method of the training classifier provided in this embodiment can be executed by the device of a trained classifier, the dress
It sets and can be implemented as software, can be implemented as hardware, be also implemented as the combination of software and hardware, such as the training classification
The device of device includes computer equipment, to execute the training classifier provided in this embodiment by the computer equipment
Method, as understood by those skilled in the art, computer equipment can be desk-top or portable computer device, can also be shifting
Dynamic terminal device etc..
Fig. 1 is the flow chart of the embodiment of the method one for the training classifier that the embodiment of the present disclosure provides, as shown in Figure 1, this
The method of the training classifier of open embodiment includes the following steps:
Step S101 obtains training set, and the training set includes that the first subclass and second subset are closed, and described first
Data items in subclass correspond to first property value, corresponding second attribute value of data items in the second subset conjunction, institute
First property value is stated less than second attribute value；
The embodiment of the present disclosure wants to that classifier is made to export more smooth classification results, therefore first in step S101
Middle acquisition training set, to be handled the data items in training set to train classifier.It is worth noting that this
Disclosed embodiment can be based on data based on the data items training classifier in training set, the classifier trained
One or more attributes of project classify to the data items of input.For convenient for clearly describing embodiment of the disclosure,
Later will to close training classifier by training set of images, and classifier be based on image attributes to image data item into
It is described for row training and classification.Those skilled in the art can understand, and the classifier in the embodiment of the present disclosure can be with
Image is trained and is classified based on multiple images attribute, such as according to multiple face characters training classifier, to pass through
Classifier exports the classification to the age of who object, also for example according to multiple images attribute training classifier, thus by dividing
The classification (such as classifier to the image classification of input or is identified as personage and still builds) of object in class device output image；And
And those skilled in the art can understand, open embodiment is not only applicable only to classify to image, and the disclosure is implemented
Example is also applied to any required scene based on the attribute of data items to data classification, such as network security field
Scape, the embodiment of the present disclosure can also gather training classifier based on the training of network security data, and be based on network security attribute
Classify to network security data.
As previously mentioned, the data items in the embodiment of the present disclosure can be image.Correspondingly, the category in the embodiment of the present disclosure
Property can be image attributes.As will be clear to the skilled person, image attributes may include color attribute, brightness attribute,
Gray scale attribute, the essential attributes such as shape attribute can also include complicated image attribute according to the content of image, such as portrait
It may include face character (shape of face attribute, nose type attribute, ocular form attribute, ear type attribute, nozzle type attribute, eyebrow type attribute, hair style category
Property etc.), portrait attribute (height attribute, fat or thin attribute, leg type attribute, hand-type attribute etc.), background attribute etc., the face character
It can be characterized by color attribute, gray scale attribute, brightness attribute, and/or shape attribute etc. with portrait attribute etc., and the face
Color attribute, gray scale attribute, brightness attribute, and/or shape attribute etc. may include pixel color, pixel grey scale, pixel intensity,
And/or location of pixels etc..And those skilled in the art can define, and the attribute value in the embodiment of the present disclosure can be used for quantifying
Its corresponding image attributes, such as brightness attribute, bigger brightness attribute value may mean that brightness is bigger, for ocular form
Attribute, smaller ocular form attribute value may mean that the eyes in image are smaller, and those skilled in the art can according to need pair
The value range of the attribute of image and the corresponding attribute value of the attribute of image and characterization content etc. carry out any definition.
Optionally, first property value described in the embodiment of the present disclosure and second attribute value correspond to the same image category
Property or the first property value include the first property value of the image attributes, and second attribute value includes the image attributes
The second attribute value.Such as the training of classifier acquired in step s101 is gathered, which is for basis
Ocular form attribute in image is classified, as an example, ocular form attribute value is for example for characterizing the size of eyes in image, then
Image in the first subclass both corresponds to the first ocular form attribute value (for example, 1, it is smaller to represent eyes), closes in second subset
In image both correspond to the second ocular form attribute value (for example, 7, it is larger to represent eyes), the first ocular form attribute value be less than institute
State the second ocular form attribute value.
The data items of the first quantity in second subset conjunction are corresponded to the first property value by step S102,
The data items of the second quantity in first subclass are corresponded into second attribute value；
For training set acquired in step s101, by the data item of the first quantity in second subset conjunction
Mesh corresponds to the first property value, this is it can be appreciated that by the data items of the first quantity described in second subset conjunction
It is included into the first subclass, since the data items in the first subclass correspond to the first property value, then for from described
The data items that second subset closes first quantity for being included into first subclass also then correspond to first attribute
Value.Based on example above-mentioned, the first number of images originally belonged in second subset conjunction both corresponds to the second ocular form attribute value (example
For example 7, it is larger to represent eyes), i.e., this originally belongs to the eyes size in the image for the first quantity that second subset is closed generally all
Be it is biggish, be included into the first subclass or correspond to first property value after, although in the image of first quantity
Eyes or biggish, but first quantity have more ox-eyed image will correspond to first property value (for example, 1,
It is smaller to represent eyes).
Equally in step s 102, the data items of the second quantity in first subclass are also corresponded to described the
Two attribute values, this is it can be appreciated that be included into second subset for the data items of the second quantity described in first subclass
It closes, due to corresponding second attribute value of data items that second subset is closed, then originally belonging to the first subclass by aforementioned operation
In the second quantity data after being included into the second subset and closing, will no longer correspond to first property value, and correspond to the
Two attribute values.Based on example above-mentioned, originally belongs to the second number of images in the first subclass and both correspond to the first ocular form category
Property value (for example, 1, it is smaller to represent eyes), i.e. eyes size in the image of second quantity for originally belonging to the first subclass
It is typically lesser, after being included into second subset and closing or correspond to after the second attribute value, although second quantity
Eyes in image or lesser, but the image with lesser eyes of second quantity will correspond to the second attribute value
(for example, 7, it is larger to represent eyes).
In the embodiment of the present disclosure, first quantity and second quantity are the integer more than or equal to 0.At one
In optional embodiment, at least one of first quantity and second quantity are greater than 0；In another optional implementation
In example, first quantity is equal with second quantity and is greater than 0.In another optional embodiment, first number
Measure unequal with second quantity, such as first quantity and second quantity are different positive integer or described
One in first quantity and the second quantity is 0, and another is positive integer.
Optionally, in the embodiment of the present disclosure, the data items in the training set include attribute value label, the attribute
Value label is for marking the corresponding attribute value of the data items.Correspondingly, by the first quantity in second subset conjunction
Data items correspond to the first property value, comprising: by the category of the data items of the first quantity in second subset conjunction
Property value label be labeled as first property value；The data items of the second quantity in first subclass are corresponded to described second
Attribute value, comprising: the attribute value label of the data items of the second quantity in first subclass is labeled as the second attribute
Value.
Step S103 gathers training classifier according to the training.
In step s 103 according to the data items training classifier in training set, so that trained classifier
Can the data items of input be classified or be identified.In an alternative embodiment, the classification of the classifier output
It as a result include each subclass each class probability correspondingly with the training set, each subclass with the training set
One-to-one each class probability and be 1.
Such as the understanding of those skilled in the art, the classifier in the embodiment of the present disclosure be by a large amount of data items into
Row training obtains, and after the completion of training, for the data items of input, classifier can provide the result of classification or identification.
By taking classifier is trained and is classified to image data item based on an image attributes as an example, for training the instruction of the classifier
Practicing set includes great amount of images data items, for view data item purpose one image attributes, the great amount of images data
Project marks in advance to be had, can be by carrying out data to the great amount of images data items during training classifier
Processing, identifies its attribute value and the corresponding attribute value of the attribute value of identification is compared and is used to correct by comparison result
Classifier, so that it is various based on convolutional Neural net to complete training, such as LeNet, AlexNet, GoogLeNet to classifier etc.
The classifier of network is applicable in the embodiment of the present disclosure.
As described in disclosure background technique, in order to which the classification results for exporting classifier are more significant, for training classification
The data items of device are often well-chosen according to target category, such as the classification of some classification for image attributes,
The data items with more distinguishing feature can be selected.As ocular form attribute value is used to characterize the big of eyes in image in aforementioned exemplary
Small, training the corresponding first ocular form attribute value of image in the first subclass of set to represent for 1, eyes are smaller, and second subset is closed
In the corresponding second ocular form attribute value of image be that 7 to represent eyes larger, then will tend to select in construction training set
The lesser image of eyes in image corresponds to the first ocular form attribute value 1, and the biggish image of eyes in image is selected to correspond to
First ocular form attribute value 7, so that the classification results of classifier are more significant.After based on above-mentioned data training classifier, when
Eyes size in the image of input is bigger but reaches the degree of the second ocular form attribute value 7 not yet, such as according to ocular form
Attribute value is quantified, and the corresponding attribute value of eyes size of the image of input is 3, when what is provided by the non-embodiment of the present disclosure
When the classifier trained of method of training classifier classifies to the image of the input, classification knot that classifier may export
Fruit is (0.9,0.1), then the ocular form of the image of the input calculated according to the classification results of classifier is desired for 1*0.9+7*
0.1=1.6, the expectation based on 1.6 are likely to result in correction excessively to the U.S. face of ocular form progress of the image of input.
And in the step S102 of the embodiment of the present disclosure, by the data items pair of the first quantity in second subset conjunction
The data items of the second quantity in first subclass are corresponded to second attribute by first property value described in Ying Yu
Value passes through step S103 training classifier, it will the classification results of smoothsort device later.Such as aforementioned ocular form attribute value
Example, the classification results that may export of classifier by embodiment of the present disclosure training are (0.7,0.3), then according to classification
The ocular form of the image for the input that the classification results of device calculate is desired for 1*0.7+7*0.3=2.8, in this image with input
Corresponding 3 deviation of attribute value of eyes size is smaller, so that the expectation based on 2.8 will to the U.S. face of ocular form progress of the image of input
Obtain more preferably effect.
As shown in Fig. 2, in the embodiment two of the training method of the classifier of the disclosure, in step corresponding with step S101
In rapid S201, the training set of acquisition further includes third subclass, and the data items in the third subclass correspond to
Third attribute value, the third attribute value are greater than second attribute value.That is, step S201 includes: acquisition training set
It closes, the training set includes the first subclass, and second subset is closed and third subclass, the data in first subclass
Project corresponds to first property value, the second subset close in corresponding second attribute value of data items, in the third subclass
Data items correspond to third attribute value, the first property value is less than second attribute value, and the third attribute value is greater than
Second attribute value.Optionally, first property value described in the embodiment of the present disclosure, second attribute value and the third
Attribute value corresponds to the same image attributes or the first property value includes the first property value of the image attributes, described
Second attribute value includes the second attribute value of the image attributes, and the third attribute value includes the third attribute of the image attributes
Value.
It in the step S202 corresponding with step S102, while being also according to the training set training classifier
Before, further includes: the data items of the third quantity in second subset conjunction are corresponded into the third attribute value, it will be described
The data items of the 4th quantity in third subclass correspond to second attribute value.That is, step S202 include: by
The data items of the first quantity in the second subset conjunction correspond to the first property value, will be in first subclass
The data items of second quantity correspond to second attribute value, by the data items of the third quantity in second subset conjunction
Corresponding to the third attribute value, the data items of the 4th quantity in the third subclass are corresponded into second attribute
Value；
In the step S203 corresponding with step S103, comprising: gather training classifier according to the training.
In the embodiment two of the training method of the classifier of the above-mentioned disclosure, the training set includes the first subset
It closes, second subset is closed and third subclass, three subclass respectively correspond different attribute values, in step S202, not only
The data items of the first quantity in second subset conjunction are corresponded into the first property value, also close the second subset
In third quantity data items correspond to the third attribute value.Pass through step S203 training classifier later, it will flat
The classification results of sliding classifier.
It is for being carried out being classified as example according to the shape of face attribute in image with classifier, shape of face attribute value is for example for table
The shape of face in image is levied, in step s 201 the corresponding shape of face attribute of image in the first subclass of acquired training set
Value is, for example, 0, represents sharp face, and the corresponding shape of face attribute value of image in second subset conjunction is, for example, 1, represents standard shape of face, the
The corresponding shape of face attribute value of image in three subclass is, for example, 2, represents round face.In order to keep the classification results of classifier more aobvious
It writes, in construction training set, it is intended to which the image for selecting the shape of face in image sharper corresponds to attribute value 0, selects image
In the relatively round image of shape of face correspond to attribute value 2, therefore training set acquired in step S201 is often according to above-mentioned
What tendency was constructed, but in step S202, the data items of the first quantity in second subset conjunction are corresponded into institute
First property value is stated, the data items of the second quantity in first subclass are corresponded into second attribute value, by institute
State second subset close in the data items of the first quantity correspond to the first property value, by the second subset close in the
The data items of three quantity correspond to the third attribute value, and later by step S203 training classifier, this will smoothly divide
The classification results of class device.Such as quantified according to shape of face attribute value, the attribute value of the shape of face of the image of input is 1.5 i.e. partially round
But be not it is especially round, when the classifier that the method for the training classifier provided by the non-embodiment of the present disclosure is trained is to the input
Image when being classified, the classification results that classifier may export are (0.01,0.09,0.9), then according to point of classifier
The ocular form of the image for the input that class result calculates is desired for 0*0.01+1*0.09+2*0.9=1.81, based on 1.81 expectation
U.S. face is carried out to the shape of face of the image of input and is likely to result in correction excessively (such as the target corrected when to shape of face attribute value U.S. face
Shape of face attribute value is 1, then based on 1.81 it is desirable that correction 0.81, but the shape of face attribute value of the image actually inputted
It is 1.5, therefore shape of face will be corrected into 1.5-0.81=0.69 on the basis of 1.5, the target shape of face category relative to correction
Property value 1 that correction has occurred is excessive).And be by the classification results that may export of classifier of embodiment of the present disclosure training (0.1,
0.2,0.7), then the ocular form of the image of the input calculated according to the classification results of classifier is desired for 0*0.1+1*0.2+2*
0.7=1.6, corresponding 1.5 deviation of attribute value of shape of face in this and the image of input is smaller, thus based on 1.5 expectation to defeated
The shape of face of the image entered, which carries out U.S. face, will obtain more preferably effect.
Optionally, in the embodiment of the present disclosure, the data items in the training set include attribute value label, the attribute
Value label is for marking the corresponding attribute value of the data items.Correspondingly, by the third quantity in second subset conjunction
Data items correspond to the third attribute value, comprising: by the category of the data items of the third quantity in second subset conjunction
Property value label be labeled as third attribute value；The data items of the 4th quantity in the third subclass are corresponded to described second
Attribute value, comprising: the attribute value label of the data items of the 4th quantity in the third subclass is labeled as the second attribute
Value.
In the embodiment of the present disclosure, the third quantity and the 4th quantity are the integer more than or equal to 0.At one
In optional embodiment, at least one of the third quantity and the 4th quantity are greater than 0；In another optional implementation
In example, the third quantity is equal with the 4th quantity and is greater than 0.In another optional embodiment, the third number
Measure unequal with the 4th quantity, such as the third quantity and the 4th quantity are different positive integer or described
One in third quantity and the 4th quantity is 0, and another is positive integer.
In an alternative embodiment, first quantity, the second quantity, third quantity and the 4th quantity are positive
Integer and equal.
In another optional embodiment, first quantity, the second quantity, in third quantity and the 4th quantity
At least one is less than or equal to the 30% of the data items quantity of the subclass corresponding to it, such as first quantity, and second
Quantity, third quantity and the 4th quantity are respectively less than the 30% of the data items quantity of the subclass corresponding to it.
In an alternative embodiment, the subclass that the training set includes corresponds respectively to different attribute values,
In the subclass that corresponding attribute value is less than second attribute value, the corresponding first property value of first subclass with
The distance of second attribute value is minimum.In another optional embodiment, the subclass that the training set includes is distinguished
Corresponding to different attribute values, in the subclass that corresponding attribute value is greater than second attribute value, the third subset
It is minimum at a distance from second attribute value to close corresponding third attribute value.Training set in the embodiment of the present disclosure may include
More subclass, and the subclass (in data items) in the training set corresponds respectively to different attribute values, but
It is that it is right can be only changed to the adjacent subclass institute of attribute value when changing attribute value corresponding to the data items in subclass
The attribute value answered, such as the data items in the first subclass can be corresponded to the second attribute value, but cannot be by the first son
Data items in set correspond to third attribute value, can have biggish error in this way to avoid the classifier trained.
Fig. 3 show the structural schematic diagram of 300 embodiment of device of the training classifier of embodiment of the present disclosure offer, such as schemes
Shown in 3, described device includes obtaining module 301, attribute value respective modules 302 and training module 303.
Module 301 is obtained, for obtaining trained set, the training set includes that the first subclass and second subset are closed,
The data items of first subclass correspond to first property value, corresponding second attribute of the data items that the second subset is closed
Value, the first property value are less than second attribute value；
Attribute value respective modules 302, the data items for the first quantity in closing the second subset correspond to institute
First property value is stated, is also used to the data items of the second quantity in first subclass corresponding to second attribute
Value；
Training module 303, for gathering training classifier according to the training
The method that Fig. 3 shown device can execute Fig. 1 and/or embodiment illustrated in fig. 2, the portion that the present embodiment is not described in detail
Point, it can refer to the related description to Fig. 1 and/or embodiment illustrated in fig. 2.The implementation procedure and technical effect of the technical solution referring to
Description in Fig. 1 and/or embodiment illustrated in fig. 2, details are not described herein.
Below with reference to Fig. 4, it illustrates the structural representations for the electronic equipment 400 for being suitable for being used to realize the embodiment of the present disclosure
Figure.Electronic equipment in the embodiment of the present disclosure can include but is not limited to such as mobile phone, laptop, digital broadcasting and connect
Receive device, PDA (personal digital assistant), PAD (tablet computer), PMP (portable media player), car-mounted terminal (such as vehicle
Carry navigation terminal) etc. mobile terminal and such as number TV, desktop computer etc. fixed terminal.Electricity shown in Fig. 4
Sub- equipment is only an example, should not function to the embodiment of the present disclosure and use scope bring any restrictions.
As shown in figure 4, electronic equipment 400 may include processing unit (such as central processing unit, graphics processor etc.)
401, random access can be loaded into according to the program being stored in read-only memory (ROM) 402 or from storage device 408
Program in memory (RAM) 403 and execute various movements appropriate and processing.In RAM 403, it is also stored with electronic equipment
Various programs and data needed for 400 operations.Processing unit 401, ROM 402 and RAM 403 pass through bus or communication line
404 are connected with each other.Input/output (I/O) interface 405 is also connected to bus or communication line 404.
In general, following device can connect to I/O interface 405: including such as touch screen, touch tablet, keyboard, mouse, figure
As the input unit 406 of sensor, microphone, accelerometer, gyroscope etc.；Including such as liquid crystal display (LCD), loudspeaking
The output device 407 of device, vibrator etc.；Storage device 408 including such as tape, hard disk etc.；And communication device 409.It is logical
T unit 409 can permit electronic equipment 400 and wirelessly or non-wirelessly be communicated with other equipment to exchange data.Although Fig. 4 shows
The electronic equipment 400 with various devices is gone out, it should be understood that being not required for implementing or having all dresses shown
It sets.It can alternatively implement or have more or fewer devices.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be carried on computer-readable medium
On computer program, which includes the program code for method shown in execution flow chart.In such reality
It applies in example, which can be downloaded and installed from network by communication device 409, or from storage device 408
It is mounted, or is mounted from ROM 402.When the computer program is executed by processing unit 401, the embodiment of the present disclosure is executed
Method in the above-mentioned function that limits.
It should be noted that the above-mentioned computer-readable medium of the disclosure can be computer-readable signal media or meter
Calculation machine readable storage medium storing program for executing either the two any combination.Computer readable storage medium for example including but be not limited to
Electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor system, device or device, or any above combination.It is computer-readable
The more specific example of storage medium can include but is not limited to: have electrical connection, the portable computing of one or more conducting wires
Machine disk, hard disk, random access storage device (RAM), read-only memory (ROM), erasable programmable read only memory (EPROM
Or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic memory device or above-mentioned
Any appropriate combination.In the disclosure, computer readable storage medium can be it is any include or storage program it is tangible
Medium, the program can be commanded execution system, device or device use or in connection.And in the disclosure,
Computer-readable signal media may include in a base band or as the data-signal that carrier wave a part is propagated, wherein carrying
Computer-readable program code.The data-signal of this propagation can take various forms, including but not limited to electromagnetic signal,
Optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be other than computer readable storage medium
Any computer-readable medium, which can send, propagates or transmit for by instruction execution
System, device or device use or program in connection.The program code for including on computer-readable medium can
To transmit with any suitable medium, including but not limited to: electric wire, optical cable, RF (radio frequency) etc. are above-mentioned any appropriate
Combination.
Above-mentioned computer-readable medium can be included in above-mentioned electronic equipment；It is also possible to individualism, and not
It is fitted into the electronic equipment.
Above-mentioned computer-readable medium carries one or more program, when said one or multiple programs are by the electricity
When sub- equipment executes, so that the method that the electronic equipment executes the training classifier in above-described embodiment.
The calculating of the operation for executing the disclosure can be write with one or more programming languages or combinations thereof
Machine program code, above procedure design language include object oriented program language-such as Java, Smalltalk, C++,
It further include conventional procedural programming language-such as " C " language or similar programming language.Program code can be complete
It executes, partly executed on the user computer on the user computer entirely, being executed as an independent software package, part
Part executes on the remote computer or executes on a remote computer or server completely on the user computer.It is relating to
And in the situation of remote computer, remote computer can include local area network (LAN) or wide area network by the network-of any kind
(WAN) one be connected to subscriber computer, or, it may be connected to outer computer (such as using ISP come
It is connected by internet).
Flow chart and block diagram in attached drawing are illustrated according to the system of the various embodiments of the disclosure, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, program segment or code of table, a part of the module, program segment or code include one or more use
The executable instruction of the logic function as defined in realizing.It should also be noted that in some implementations as replacements, being marked in box
The function of note can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are actually
It can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it to infuse
Meaning, the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart can be with holding
The dedicated hardware based system of functions or operations as defined in row is realized, or can use specialized hardware and computer instruction
Combination realize.
Being described in unit involved in the embodiment of the present disclosure can be realized by way of software, can also be by hard
The mode of part is realized.Wherein, the title of unit does not constitute the restriction to the unit itself under certain conditions.
Above description is only the preferred embodiment of the disclosure and the explanation to institute's application technology principle.Those skilled in the art
Member is it should be appreciated that the open scope involved in the disclosure, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from design disclosed above, it is carried out by above-mentioned technical characteristic or its equivalent feature
Any combination and the other technical solutions formed.Such as features described above has similar function with (but being not limited to) disclosed in the disclosure
Can technical characteristic replaced mutually and the technical solution that is formed.
Claims (11)
1. a kind of method of trained classifier characterized by comprising
Training set is obtained, the training set includes that the first subclass and second subset are closed, the number in first subclass
First property value is corresponded to according to project, corresponding second attribute value of data items in the second subset conjunction, the first property value
Less than second attribute value；
The data items of the first quantity in second subset conjunction are corresponded into the first property value, by first subset
The data items of the second quantity in conjunction correspond to second attribute value；
Gather training classifier according to the training.
2. the method for trained classifier according to claim 1, which is characterized in that first quantity and second number
Amount meets one in following relationship:
First quantity is equal with second quantity and is greater than 0；
First quantity and second quantity are unequal.
3. the method for trained classifier according to claim 1, which is characterized in that the training set further includes third
Gather, the data items in the third subclass correspond to third attribute value, and the third attribute value is greater than second attribute
Value；
Before gathering training classifier according to the training, the method also includes:
The data items of third quantity in second subset conjunction are corresponded into the third attribute value, by the third subset
The data items of the 4th quantity in conjunction correspond to second attribute value.
4. the method for trained classifier according to claim 3, which is characterized in that the third quantity and the 4th number
Amount meets one in following relationship:
The third quantity is equal with the 4th quantity and is greater than 0；
The third quantity and the 4th quantity are unequal.
5. according to claim 1 to the method for any training classifier in 4, which is characterized in that in the training set
Data items include attribute value label, the attribute value label is for marking the corresponding attribute value of the data items.
6. the method for trained classifier according to claim 1, which is characterized in that the classification results of the classifier output
Including each subclass each class probability correspondingly with the training set, one by one with each subclass of the training set
Corresponding each class probability and be 1.
7. the method for trained classifier according to claim 1, which is characterized in that the subclass that the training set includes
Data items correspond respectively to different attribute values, corresponding attribute value be less than second attribute value subclass
In, the corresponding first property value of first subclass is minimum at a distance from second attribute value.
8. the method for trained classifier according to claim 3 or 4, which is characterized in that the son that the training set includes
The data items of set correspond respectively to different attribute values, are greater than the subset of second attribute value in corresponding attribute value
In conjunction, the corresponding third attribute value of the third subclass is minimum at a distance from second attribute value.
9. a kind of device of trained classifier characterized by comprising
Module is obtained, for obtaining trained set, the training set includes that the first subclass and second subset are closed, and described first
The data items of subclass correspond to first property value, corresponding second attribute value of the data items that the second subset is closed, and described the
One attribute value is less than second attribute value；
Attribute value respective modules, the data items for the first quantity in closing the second subset correspond to described first and belong to
Property value, be also used to by the data items of the second quantity in first subclass correspond to second attribute value；
Training module, for gathering training classifier according to the training.
10. a kind of electronic equipment, comprising:
Memory, for storing computer-readable instruction；And
Processor, for running the computer-readable instruction, so that realizing according to claim 1-8 when the processor is run
Any one of described in training classifier method.
11. a kind of non-transient computer readable storage medium, for storing computer-readable instruction, when the computer-readable finger
When order is executed by computer, so that the computer perform claim requires the side of training classifier described in any one of 1-8
Method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910292542.3A CN110084290B (en) | 2019-04-12 | 2019-04-12 | Method, apparatus, electronic device and computer-readable storage medium for training classifier |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910292542.3A CN110084290B (en) | 2019-04-12 | 2019-04-12 | Method, apparatus, electronic device and computer-readable storage medium for training classifier |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110084290A true CN110084290A (en) | 2019-08-02 |
| CN110084290B CN110084290B (en) | 2021-03-05 |
Family
ID=67414816
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910292542.3A Active CN110084290B (en) | 2019-04-12 | 2019-04-12 | Method, apparatus, electronic device and computer-readable storage medium for training classifier |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110084290B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103886330A (en) * | 2014-03-27 | 2014-06-25 | 西安电子科技大学 | Classification method based on semi-supervised SVM ensemble learning |
| CN105426826A (en) * | 2015-11-09 | 2016-03-23 | 张静 | Tag noise correction based crowd-sourced tagging data quality improvement method |
| US20160379091A1 (en) * | 2015-06-23 | 2016-12-29 | Adobe Systems Incorporated | Training a classifier algorithm used for automatically generating tags to be applied to images |
| CN107305565A (en) * | 2016-04-21 | 2017-10-31 | 富士通株式会社 | Information processor, information processing method and message processing device |
| CN107992887A (en) * | 2017-11-28 | 2018-05-04 | 东软集团股份有限公司 | Classifier generation method, sorting technique, device, electronic equipment and storage medium |
| CN109308318A (en) * | 2018-08-14 | 2019-02-05 | 深圳大学 | Training method, device, equipment and the medium of cross-domain texts sentiment classification model |
-
2019
- 2019-04-12 CN CN201910292542.3A patent/CN110084290B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103886330A (en) * | 2014-03-27 | 2014-06-25 | 西安电子科技大学 | Classification method based on semi-supervised SVM ensemble learning |
| US20160379091A1 (en) * | 2015-06-23 | 2016-12-29 | Adobe Systems Incorporated | Training a classifier algorithm used for automatically generating tags to be applied to images |
| CN105426826A (en) * | 2015-11-09 | 2016-03-23 | 张静 | Tag noise correction based crowd-sourced tagging data quality improvement method |
| CN107305565A (en) * | 2016-04-21 | 2017-10-31 | 富士通株式会社 | Information processor, information processing method and message processing device |
| CN107992887A (en) * | 2017-11-28 | 2018-05-04 | 东软集团股份有限公司 | Classifier generation method, sorting technique, device, electronic equipment and storage medium |
| CN109308318A (en) * | 2018-08-14 | 2019-02-05 | 深圳大学 | Training method, device, equipment and the medium of cross-domain texts sentiment classification model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110084290B (en) | 2021-03-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20210248355A1 (en) | Face key point detection method and apparatus, storage medium, and electronic device | |
| CN108510472B (en) | Method and apparatus for handling image | |
| CN108898185A (en) | Method and apparatus for generating image recognition model | |
| CN107909065A (en) | The method and device blocked for detecting face | |
| CN106599789A (en) | Video class identification method and device, data processing device and electronic device | |
| CN109087377B (en) | Method and apparatus for handling image | |
| CN109993150A (en) | The method and apparatus at age for identification | |
| CN108491809A (en) | The method and apparatus for generating model for generating near-infrared image | |
| CN108509892A (en) | Method and apparatus for generating near-infrared image | |
| CN108228463A (en) | For detecting the method and apparatus of initial screen time | |
| CN106489141A (en) | For method that content is classified and electronic equipment | |
| CN109947989A (en) | Method and apparatus for handling video | |
| CN110047122A (en) | Render method, apparatus, electronic equipment and the computer readable storage medium of image | |
| CN109344752A (en) | Method and apparatus for handling mouth image | |
| CN110458173A (en) | Method and apparatus for generating article color value | |
| CN110070551A (en) | Rendering method, device and the electronic equipment of video image | |
| CN109815365A (en) | Method and apparatus for handling video | |
| CN108491825A (en) | information generating method and device | |
| CN109598304A (en) | Disaggregated model calibration method, device, equipment and readable medium | |
| CN110069974A (en) | Bloom image processing method, device and electronic equipment | |
| CN110210501A (en) | Virtual objects generation method, electronic equipment and computer readable storage medium | |
| CN110263255A (en) | Acquisition methods, system, server and the storage medium of customer attribute information | |
| CN109948699A (en) | Method and apparatus for generating characteristic pattern | |
| CN110191356A (en) | Video reviewing method, device and electronic equipment | |
| CN110287816A (en) | Car door motion detection method, device and computer readable storage medium |
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 | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 100041 B-0035, 2 floor, 3 building, 30 Shixing street, Shijingshan District, Beijing. Patentee after: Douyin Vision Co.,Ltd. Address before: 100041 B-0035, 2 floor, 3 building, 30 Shixing street, Shijingshan District, Beijing. Patentee before: Tiktok vision (Beijing) Co.,Ltd. Address after: 100041 B-0035, 2 floor, 3 building, 30 Shixing street, Shijingshan District, Beijing. Patentee after: Tiktok vision (Beijing) Co.,Ltd. Address before: 100041 B-0035, 2 floor, 3 building, 30 Shixing street, Shijingshan District, Beijing. Patentee before: BEIJING BYTEDANCE NETWORK TECHNOLOGY Co.,Ltd. |
|
| CP01 | Change in the name or title of a patent holder |