CN112784903A - Method, device and equipment for training target recognition model - Google Patents
Method, device and equipment for training target recognition model Download PDFInfo
- Publication number
- CN112784903A CN112784903A CN202110102511.4A CN202110102511A CN112784903A CN 112784903 A CN112784903 A CN 112784903A CN 202110102511 A CN202110102511 A CN 202110102511A CN 112784903 A CN112784903 A CN 112784903A
- Authority
- CN
- China
- Prior art keywords
- recognition model
- sample
- label
- training
- loss value
- 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
- 238000012549 training Methods 0.000 title claims abstract description 71
- 238000000034 method Methods 0.000 title claims abstract description 40
- 230000006870 function Effects 0.000 claims description 14
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 235000013361 beverage Nutrition 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Images
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The application relates to the technical field of artificial intelligence and discloses a method for training a target recognition model. The method comprises the following steps: obtaining a training sample; the training samples comprise a first label sample, a second label sample and a non-label sample; the label sample comprises a first label sample and a second label sample; obtaining a first alternative identification model according to the first label sample; obtaining a second alternative identification model according to the second label sample and the first alternative identification model; obtaining a third alternative recognition model according to the label-free sample and the second alternative recognition model; and obtaining the target recognition model according to the third alternative recognition model. The preset recognition model is trained by obtaining the training sample, so that the recognition accuracy of the third alternative recognition model on the same kind of target of the unlabeled sample can be improved, and the recognition accuracy of the target recognition model on the same kind of target of the long tail sample is improved. The application also discloses a device and equipment for training the target recognition model.
Description
Technical Field
The present application relates to the field of artificial intelligence technologies, and in particular, to a method, an apparatus, and a device for training a target recognition model.
Background
The identification of the object type is an indispensable link in many artificial intelligence processing processes, and with the development of a machine learning algorithm, the automatic identification of the object type by using a neural network model becomes possible. The long tail samples are samples with a low sample number, and compared with other training samples, the training samples of the long tail object are fewer, so that the trained neural network model is difficult to effectively identify the same type of target of the long tail sample.
In the process of implementing the embodiments of the present disclosure, it is found that at least the following problems exist in the related art: the accuracy rate of the identification result of the similar target of the long tail sample in the prior art is low.
Disclosure of Invention
The following presents a simplified summary in order to provide a basic understanding of some aspects of the disclosed embodiments. This summary is not an extensive overview nor is intended to identify key/critical elements or to delineate the scope of such embodiments but rather as a prelude to the more detailed description that is presented later.
The embodiment of the disclosure provides a method, a device and equipment for training a target recognition model, so as to improve the recognition accuracy of the target recognition model on similar targets of a long tail sample.
In some embodiments, the method for target recognition model training comprises:
obtaining a training sample; the training samples comprise a first label sample, a second label sample and an unlabeled sample; the first label samples are label samples with the number ratio reaching a preset value in the training samples; the second label sample is a label sample with the quantity ratio lower than the preset value in the training samples;
obtaining a first alternative identification model according to the first label sample;
obtaining the second alternative identification model according to the second label sample and the first alternative identification model;
obtaining a third alternative recognition model according to the label-free sample and the second alternative recognition model;
and obtaining a target recognition model according to the third candidate recognition model.
In some embodiments, the apparatus comprises: a processor and a memory storing program instructions, the processor being configured to, when executing the program instructions, perform the method for object recognition model training described above.
The method, the device and the equipment for training the target recognition model provided by the embodiment of the disclosure can realize the following technical effects: according to the scheme, the first label sample is obtained, the preset recognition model is trained to obtain the first alternative recognition model, and the recognition accuracy of the first alternative recognition model on the same kind of target of the first label sample can be improved; the first alternative recognition model is trained through the second label sample to obtain a second alternative recognition model, so that the recognition accuracy of the second alternative recognition model on the same kind of target of the second label sample can be improved; the second alternative recognition model is trained through the unlabeled sample to obtain a third alternative recognition model, so that the recognition accuracy of the third alternative recognition model on the similar target of the unlabeled sample can be improved, and the recognition accuracy of the target recognition model on the similar target of the long-tail sample can be improved.
The foregoing general description and the following description are exemplary and explanatory only and are not restrictive of the application.
Drawings
One or more embodiments are illustrated by way of example in the accompanying drawings, which correspond to the accompanying drawings and not in limitation thereof, in which elements having the same reference numeral designations are shown as like elements and not in limitation thereof, and wherein:
FIG. 1 is a schematic diagram of a method for target recognition model training provided by an embodiment of the present disclosure;
FIG. 2 is a schematic diagram of an apparatus for training a target recognition model according to an embodiment of the present disclosure.
Detailed Description
So that the manner in which the features and elements of the disclosed embodiments can be understood in detail, a more particular description of the disclosed embodiments, briefly summarized above, may be had by reference to the embodiments, some of which are illustrated in the appended drawings. In the following description of the technology, for purposes of explanation, numerous details are set forth in order to provide a thorough understanding of the disclosed embodiments. However, one or more embodiments may be practiced without these details. In other instances, well-known structures and devices may be shown in simplified form in order to simplify the drawing.
The terms "first," "second," and the like in the description and in the claims, and the above-described drawings of embodiments of the present disclosure, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It should be understood that the data so used may be interchanged under appropriate circumstances such that embodiments of the present disclosure described herein may be made. Furthermore, the terms "comprising" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions.
The term "plurality" means two or more unless otherwise specified.
In the embodiment of the present disclosure, the character "/" indicates that the preceding and following objects are in an or relationship. For example, A/B represents: a or B.
The term "and/or" is an associative relationship that describes objects, meaning that three relationships may exist. For example, a and/or B, represents: a or B, or A and B.
Referring to fig. 1, an embodiment of the present disclosure provides a method for training a target recognition model, including:
step S101, obtaining a training sample; the training samples comprise a first label sample, a second label sample and a non-label sample;
step S102, a first candidate identification model is obtained according to the first label sample;
step S103, obtaining a second alternative recognition model according to the second label sample and the first alternative recognition model;
step S104, obtaining a third alternative recognition model according to the unlabeled sample and the second alternative recognition model;
and step S105, obtaining a target recognition model according to the third candidate recognition model.
By adopting the method for training the target recognition model provided by the embodiment of the disclosure, the first alternative recognition model is obtained by obtaining the first label sample and training the preset recognition model, so that the recognition accuracy of the first alternative recognition model on the same kind of target of the first label sample can be improved; the first alternative recognition model is trained through the second label sample to obtain a second alternative recognition model, so that the recognition accuracy of the second alternative recognition model on the same kind of target of the second label sample can be improved; the second alternative recognition model is trained through the unlabeled sample to obtain a third alternative recognition model, so that the recognition accuracy of the third alternative recognition model on the similar target of the unlabeled sample can be improved, and the recognition accuracy of the target recognition model on the similar target of the long-tail sample can be improved.
Obtaining a training sample; the training samples comprise a first label sample, a second label sample and an unlabeled sample; the first label samples are label samples with the number ratio reaching a preset value in the training samples; the second label sample is a label sample with the quantity ratio lower than the preset value in the training samples;
optionally, the first label sample is a label sample whose number ratio in the training sample reaches a preset value, and the second label sample is a label sample whose number ratio in the training sample is lower than the preset value. Optionally, the preset value is 0.1%.
In some embodiments, the first label swatch is a non-long tail object swatch and the second label swatch is a long tail object swatch.
Optionally, obtaining a first candidate recognition model from the first label exemplar comprises: training a preset recognition model according to the first label sample to obtain a first recognition result; acquiring a first loss value of the first recognition result according to a preset first loss function; and determining a first candidate recognition model according to the first loss value.
Optionally, the preset recognition model is Resnet (residual network model), VGG (Visual Geometry Group), or *** lenet.
Optionally, the first loss function is
Optionally by calculation
Obtaining a first loss value of the first recognition result, wherein theta is a parameter of a preset recognition model, log L (theta) is a first loss value of the first recognition result under the parameter theta, and Pθ(x1Y) is the first recognition result under the parameter θ, x1Is the first label sample, and y is the object identification type. Optionally by calculating Pθ(x1,y)＝Pθ(x1Y) P (y) obtaining a first recognition result under the parameter θ, wherein Pθ(x1Y) is the first label sample x1Probability of being identified as y class, and p (y) probability of target identification type being y class.
Optionally, the target identification type is a type of a target object to be identified. In some embodiments, the target object to be identified is a fast-selling item such as a beverage, a food, or the like. In some embodiments, the target identification type is a type of beverage, such as: cola, sprite, etc.
Optionally, determining a first candidate recognition model according to the first loss value includes: and determining a first candidate recognition model according to the first loss value and the first determination link. Optionally, the first determining step includes: acquiring a first gradient value of a first loss value; adjusting parameters of a preset identification model according to the first gradient value to determine a first adjusted identification model; training the first adjustment recognition model according to the first label sample to obtain a first adjustment recognition result; acquiring a first adjustment loss value of a first adjustment identification result; and determining a first candidate recognition model according to the first adjustment loss value.
Optionally, the first loss value is differentiated to obtain a first gradient value.
Optionally, adjusting parameters of the preset recognition model according to the first gradient value includes: by calculating theta1Obtaining a parameter θ of the first modified recognition model as θ + g1Wherein, theta1For the first adjustment of the parameters of the recognition model, θ is a parameter in the predetermined recognition model, and g is a first gradient value.
In some embodiments, the identification model is preset as Resnet, and the parameters θ in Resnet include LR (Learning Rate), BS (Batch Size), N (Number), and the like. For example, if the first gradient value is-3, the parameters of the first adjusted recognition model are LR-3, BS-3, N-3, etc.
Optionally, determining a first candidate recognition model according to the first adjustment loss value includes: under the condition that the first adjustment loss value is smaller than or equal to a first set threshold value, determining a first adjustment recognition model corresponding to the first adjustment loss value as a first candidate recognition model; and under the condition that the first adjustment loss value is larger than the first set threshold, repeatedly executing the first determination step until the obtained first adjustment loss value is smaller than or equal to the first set threshold.
Therefore, the preset recognition model is trained through the first label sample to obtain the first alternative recognition model, and the recognition accuracy of the first alternative recognition model on the same kind of target of the first label sample can be improved.
Optionally, obtaining a second candidate recognition model according to the second label sample and the first candidate recognition model includes: training the first alternative recognition model according to the second label sample to obtain a second recognition result; acquiring a second loss value of a second recognition result according to a preset second loss function; and determining a second candidate recognition model according to the second loss value.
Optionally, the second loss function is
Optionally by calculation
Obtaining a second loss value of a second recognition result, wherein theta1Identifying a parameter, log L (θ), of the model for the first candidate1) To be at a parameter theta1A second loss value of the next second recognition result,
to be at a parameter theta1Second recognition result of X2Is the second label sample, y is the object identification type. Optionally by calculation
Obtained at the parameter theta1The second recognition result of, wherein,
is the second label sample x2Probability of being identified as y class, and p (y) probability of target identification type being y class.
Optionally, determining a second candidate recognition model according to the second loss value includes: and determining a second alternative recognition model according to the second loss value and a second determination link. Optionally, the second determining step includes: acquiring a second gradient value of a second loss value; adjusting parameters of the first candidate identification model according to the second gradient value to determine a second adjusted identification model; training the second adjustment recognition model according to the second label sample to obtain a second adjustment recognition result; acquiring a second adjustment loss value of a second adjustment identification result; and determining a second candidate recognition model according to the second adjustment loss value.
Optionally, the second loss value is differentiated to obtain a second gradient value.
Optionally, adjusting parameters of the first candidate recognition model according to the second gradient value includes: by calculating theta2＝θ1+g1Obtaining a parameter theta of a second adjusted recognition model2Wherein, theta2For the first adjustment, the parameters of the recognition model, theta1Identifying parameters in the model for the first candidate, g1Is a second gradient value.
Optionally, determining a second candidate recognition model according to the second adjustment loss value includes: determining a second adjustment recognition model corresponding to the second adjustment loss value as a second candidate recognition model under the condition that the second adjustment loss value is smaller than or equal to a second set threshold value; and under the condition that the second adjustment loss value is larger than a second set threshold value, repeatedly executing a second determination step until the obtained second adjustment loss value is smaller than or equal to the second set threshold value.
Therefore, the second alternative recognition model is obtained by training the first alternative recognition model through the second label sample, and the recognition accuracy of the second alternative recognition model on the same kind of target of the second label sample can be improved.
Optionally, obtaining the target recognition model according to a third candidate recognition model includes: and taking the third candidate recognition model as the target recognition model under the condition that the third candidate recognition model meets the preset condition.
Optionally, when the third candidate recognition model satisfies the preset condition, taking the third candidate recognition model as the target recognition model includes: obtaining a test sample; testing the third alternative model according to the test sample to obtain a test result; and taking the third candidate recognition model as the target recognition model under the condition that the test result meets the preset condition.
Optionally, the test result satisfies a preset condition, including that the recognition accuracy of the test result is 90%.
Optionally, obtaining a third candidate recognition model according to the unlabeled exemplar and the second candidate recognition model includes: training the second alternative recognition model according to the label-free sample to obtain a third recognition result; acquiring a third loss value of a third recognition result according to a preset third loss function; and determining a third candidate recognition model according to the third loss value.
Optionally, the third loss function is
Optionally by calculation
Obtaining a third loss value of a third recognition result, wherein theta2Identifying a parameter, log L (θ), of the model for the second candidate2) To be at a parameter theta2A third loss value of the next third recognition result,
to be at a parameter theta2As a result of the following third recognition result,
is the total probability of an unlabeled exemplar being identified as the target identification type. Optionally by calculation
A total probability is obtained that the unlabeled exemplars are identified as the target identification type, wherein,
as unlabeled sample x3Is identified as C1Probability of class, P (C)1) Identifying type as C for target1The probability of a class is determined by the probability of the class,
as unlabeled sample x3Is identified as C2Probability of class, P (C)2) Identifying type as C for target2Probability of class, x3Is an unlabeled sample.
Optionally, determining a third candidate recognition model according to the third loss value includes: and determining a third alternative recognition model according to a third determination link according to the third loss value. Optionally, the third determining step includes: acquiring a third gradient value of a third loss value; adjusting parameters of the second alternative recognition model according to the third gradient value to determine a third adjusted recognition model; training the third adjustment recognition model according to the label-free sample to obtain a third adjustment recognition result; acquiring a third adjustment loss value of a third adjustment identification result; and determining a third alternative recognition model according to the third adjustment loss value.
Optionally, a derivative of the third loss value is taken to obtain a third gradient value.
Optionally, adjusting parameters of the second candidate recognition model according to the third gradient value includes: by calculating theta3＝θ2+g2Obtaining a parameter theta of a third adjustment recognition model3Wherein, theta3For the third adjustment, θ2For the second candidate, identifying parameters in the model, g2Is the third gradient value.
Optionally, determining a third candidate recognition model according to the third adjustment loss value includes: determining a third adjustment recognition model corresponding to the third adjustment loss value as a third candidate recognition model under the condition that the third adjustment loss value is smaller than or equal to a third set threshold value; and under the condition that the third adjustment loss value is larger than a third set threshold value, repeatedly executing a third determination step until the obtained third adjustment loss value is smaller than or equal to the third set threshold value.
Therefore, the second alternative recognition model is trained through the unlabeled sample to obtain a third alternative recognition model, and the recognition accuracy of the third alternative recognition model on the same kind of target of the unlabeled sample can be improved.
Training the preset recognition model through the weighing scale training sample in the prior art, because the number of the first label samples is greater than that of the second label samples, the mode of the balance training sample has two types, including: equalizing the training samples in a manner of reducing the number of the first label samples; or, the training samples are equalized by increasing the number of second label samples. If the training samples are balanced in a mode of reducing the number of the first label samples, then training a preset recognition model easily leads the trained recognition model to be incapable of meeting the user requirements; it is costly to equalize the training samples by increasing the number of second label samples. If the preset recognition model is trained in an unbalanced training sample mode, because the sample is unbalanced, iteration times are too few when the model is easily trained, and therefore the recognition accuracy of the trained recognition model is low, and the requirements of users cannot be met.
According to the scheme, the first alternative recognition model is obtained by obtaining the first label sample and training the preset recognition model, so that the recognition accuracy of the first alternative recognition model on the first label sample can be improved; the first alternative recognition model is trained through the second label sample to obtain a second alternative recognition model, so that the recognition accuracy of the second alternative recognition model on the second label sample can be improved; the second alternative recognition model is trained through the unlabeled sample to obtain a third alternative recognition model, so that the recognition accuracy of the third alternative recognition model on the unlabeled sample can be improved; the target recognition model is obtained according to the third alternative recognition model, so that the recognition accuracy of the target recognition model on the similar target of the long tail sample can be improved.
As shown in fig. 2, an apparatus for training a target recognition model according to an embodiment of the present disclosure includes a processor (processor)100 and a memory (memory)101 storing program instructions. Optionally, the apparatus may also include a Communication Interface (Communication Interface)102 and a bus 103. The processor 100, the communication interface 102, and the memory 101 may communicate with each other via a bus 103. The communication interface 102 may be used for information transfer. The processor 100 may call program instructions in the memory 101 to perform the method for object recognition model training of the above embodiments.
Further, the program instructions in the memory 101 may be implemented in the form of software functional units and stored in a computer readable storage medium when sold or used as a stand-alone product.
The memory 101, which is a computer-readable storage medium, may be used for storing software programs, computer-executable programs, such as program instructions/modules corresponding to the methods in the embodiments of the present disclosure. The processor 100 executes functional applications and data processing, i.e., implements the methods for object recognition model training in the above embodiments, by executing program instructions/modules stored in the memory 101.
The memory 101 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the terminal device, and the like. In addition, the memory 101 may include a high-speed random access memory, and may also include a nonvolatile memory.
By adopting the device for training the target recognition model, which is provided by the embodiment of the disclosure, the first alternative recognition model is obtained by obtaining the first label sample and training the preset recognition model, so that the recognition accuracy of the first alternative recognition model on the same kind of target of the first label sample can be improved; the first alternative recognition model is trained through the second label sample to obtain a second alternative recognition model, so that the recognition accuracy of the second alternative recognition model on the same kind of target of the second label sample can be improved; the second alternative recognition model is trained through the unlabeled sample to obtain a third alternative recognition model, so that the recognition accuracy of the third alternative recognition model on the similar target of the unlabeled sample can be improved, and the recognition accuracy of the target recognition model on the similar target of the long-tail sample can be improved.
The embodiment of the disclosure provides a device, which comprises the above device for training the target recognition model. The equipment obtains a first alternative recognition model by obtaining a first label sample and training a preset recognition model, and can improve the recognition accuracy of the first alternative recognition model on the same kind of target of the first label sample; the first alternative recognition model is trained through the second label sample to obtain a second alternative recognition model, so that the recognition accuracy of the second alternative recognition model on the same kind of target of the second label sample can be improved; the second alternative recognition model is trained through the unlabeled sample to obtain a third alternative recognition model, so that the recognition accuracy of the third alternative recognition model on the similar target of the unlabeled sample can be improved, and the recognition accuracy of the target recognition model on the similar target of the long-tail sample can be improved.
Optionally, the device is a computer or the like.
Embodiments of the present disclosure provide a computer-readable storage medium storing computer-executable instructions configured to perform the above-described method for target recognition model training.
Embodiments of the present disclosure provide a computer program product comprising a computer program stored on a computer readable storage medium, the computer program comprising program instructions which, when executed by a computer, cause the computer to perform the above-described method for object recognition model training.
The computer-readable storage medium described above may be a transitory computer-readable storage medium or a non-transitory computer-readable storage medium.
The technical solution of the embodiments of the present disclosure may be embodied in the form of a software product, where the computer software product is stored in a storage medium and includes one or more instructions to enable a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method of the embodiments of the present disclosure. And the aforementioned storage medium may be a non-transitory storage medium comprising: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes, and may also be a transient storage medium.
The above description and drawings sufficiently illustrate embodiments of the disclosure to enable those skilled in the art to practice them. Other embodiments may incorporate structural, logical, electrical, process, and other changes. The examples merely typify possible variations. Individual components and functions are optional unless explicitly required, and the sequence of operations may vary. Portions and features of some embodiments may be included in or substituted for those of others. Furthermore, the words used in the specification are words of description only and are not intended to limit the claims. As used in the description of the embodiments and the claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. Similarly, the term "and/or" as used in this application is meant to encompass any and all possible combinations of one or more of the associated listed. Furthermore, the terms "comprises" and/or "comprising," when used in this application, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. Without further limitation, an element defined by the phrase "comprising an …" does not exclude the presence of other like elements in a process, method or apparatus that comprises the element. In this document, each embodiment may be described with emphasis on differences from other embodiments, and the same and similar parts between the respective embodiments may be referred to each other. For methods, products, etc. of the embodiment disclosures, reference may be made to the description of the method section for relevance if it corresponds to the method section of the embodiment disclosure.
Those of skill in the art would appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software may depend upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the disclosed embodiments. It can be clearly understood by the skilled person that, for convenience and brevity of description, the specific working processes of the system, the apparatus and the unit described above may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the embodiments disclosed herein, the disclosed methods, products (including but not limited to devices, apparatuses, etc.) may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units may be merely a logical division, and in actual implementation, there may be another division, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form. The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to implement the present embodiment. In addition, functional units in the embodiments of the present disclosure may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. In the description corresponding to the flowcharts and block diagrams in the figures, operations or steps corresponding to different blocks may also occur in different orders than disclosed in the description, and sometimes there is no specific order between the different operations or steps. For example, two sequential operations or steps may in fact be executed substantially concurrently, or they may sometimes be executed in the reverse order, depending upon the functionality involved. Each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
Claims (10)
1. A method for target recognition model training, comprising:
obtaining a training sample; the training samples comprise a first label sample, a second label sample and an unlabeled sample; the first label samples are label samples with the number ratio reaching a preset value in the training samples; the second label sample is a label sample with the quantity ratio lower than the preset value in the training samples;
obtaining a first alternative identification model according to the first label sample;
obtaining a second alternative identification model according to the second label sample and the first alternative identification model;
obtaining a third alternative recognition model according to the label-free sample and the second alternative recognition model;
and obtaining a target recognition model according to the third candidate recognition model.
2. The method of claim 1, wherein obtaining a first candidate recognition model from the first label exemplar comprises:
training a preset recognition model according to the first label sample to obtain a first recognition result;
acquiring a first loss value of the first identification result according to a preset first loss function;
and determining a first candidate recognition model according to the first loss value.
3. The method of claim 2, wherein determining a first candidate recognition model based on the first loss value comprises:
acquiring a first gradient value of the first loss value;
adjusting parameters of the preset identification model according to the first gradient value to determine a first adjusted identification model;
training the first adjustment recognition model according to the first label sample to obtain a first adjustment recognition result;
acquiring a first adjustment loss value of the first adjustment identification result;
and determining a first candidate recognition model according to the first adjustment loss value.
4. The method of claim 1, wherein obtaining the second candidate recognition model from the second label exemplar and the first candidate recognition model comprises:
training the first candidate recognition model according to the second label sample to obtain a second recognition result;
acquiring a second loss value of the second identification result according to the preset second loss function;
and determining a second candidate recognition model according to the second loss value.
5. The method of claim 4, wherein determining a second candidate recognition model based on the second loss value comprises:
acquiring a second gradient value of the second loss value;
adjusting parameters of the first candidate identification model according to the second gradient value to determine a second adjusted identification model;
training the second adjustment recognition model according to the second label sample to obtain a second adjustment recognition result;
acquiring a second adjustment loss value of the second adjustment identification result;
and determining a second candidate recognition model according to the second adjustment loss value.
6. The method of claim 1, wherein obtaining a target recognition model from the third candidate recognition model comprises:
and taking the third candidate recognition model as a target recognition model under the condition that the third candidate recognition model meets the preset condition.
7. The method of any of claims 1 to 6, wherein obtaining a third candidate recognition model from the unlabeled exemplars and the second candidate recognition model comprises:
training the second alternative recognition model according to the label-free sample to obtain a third recognition result;
acquiring a third loss value of the third recognition result according to a preset third loss function;
and determining a third candidate recognition model according to the third loss value.
8. The method of claim 7, wherein determining a third candidate recognition model based on the third loss value comprises:
acquiring a third gradient value of the third loss value;
adjusting parameters of the second candidate identification model according to the third gradient value to determine a third adjusted identification model;
training the third adjustment recognition model according to the label-free sample to obtain a third adjustment recognition result;
acquiring a third adjustment loss value of the third adjustment identification result;
and determining a third candidate recognition model according to the third adjustment loss value.
9. An apparatus for object recognition model training, comprising a processor and a memory having stored thereon program instructions, characterized in that the processor is configured to perform the method for object recognition model training according to any one of claims 1 to 8 when executing the program instructions.
10. An apparatus, characterized in that it comprises the device for object recognition model training according to claim 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110102511.4A CN112784903B (en) | 2021-01-26 | 2021-01-26 | Method, device and equipment for training target recognition model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110102511.4A CN112784903B (en) | 2021-01-26 | 2021-01-26 | Method, device and equipment for training target recognition model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112784903A true CN112784903A (en) | 2021-05-11 |
| CN112784903B CN112784903B (en) | 2023-12-12 |
Family
ID=75757776
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110102511.4A Active CN112784903B (en) | 2021-01-26 | 2021-01-26 | Method, device and equipment for training target recognition model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112784903B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111428757A (en) * | 2020-03-05 | 2020-07-17 | 支付宝(杭州)信息技术有限公司 | Model training method, abnormal data detection method and device and electronic equipment |
| CN111461164A (en) * | 2020-02-25 | 2020-07-28 | 清华大学 | Sample data set capacity expansion method and model training method |
| CN111582185A (en) * | 2020-05-11 | 2020-08-25 | 北京百度网讯科技有限公司 | Method and apparatus for recognizing image |
| CN111667066A (en) * | 2020-04-23 | 2020-09-15 | 北京旷视科技有限公司 | Network model training and character recognition method and device and electronic equipment |
| CN111931931A (en) * | 2020-09-29 | 2020-11-13 | 杭州迪英加科技有限公司 | Deep neural network training method and device for pathology full-field image |
-
2021
- 2021-01-26 CN CN202110102511.4A patent/CN112784903B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111461164A (en) * | 2020-02-25 | 2020-07-28 | 清华大学 | Sample data set capacity expansion method and model training method |
| CN111428757A (en) * | 2020-03-05 | 2020-07-17 | 支付宝(杭州)信息技术有限公司 | Model training method, abnormal data detection method and device and electronic equipment |
| CN111667066A (en) * | 2020-04-23 | 2020-09-15 | 北京旷视科技有限公司 | Network model training and character recognition method and device and electronic equipment |
| CN111582185A (en) * | 2020-05-11 | 2020-08-25 | 北京百度网讯科技有限公司 | Method and apparatus for recognizing image |
| CN111931931A (en) * | 2020-09-29 | 2020-11-13 | 杭州迪英加科技有限公司 | Deep neural network training method and device for pathology full-field image |
Non-Patent Citations (2)
| Title |
|---|
| YU LI ET AL: "Overcoming Classifier Imbalance for Long-tail Object Detection with Balanced Group Softmax", 《REAARXIV: 2006.10408V1》, pages 1 - 12 * |
| 熊智翔等: "使用少量有标签样本学习的方法", 《计算机应用》, pages 11 - 15 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112784903B (en) | 2023-12-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110147456B (en) | Image classification method and device, readable storage medium and terminal equipment | |
| EP3779774B1 (en) | Training method for image semantic segmentation model and server | |
| CN110502976B (en) | Training method of text recognition model and related product | |
| CN109858476B (en) | Tag expansion method and electronic equipment | |
| CN111382248B (en) | Question replying method and device, storage medium and terminal equipment | |
| CN115082920B (en) | Deep learning model training method, image processing method and device | |
| EP4036796A1 (en) | Automatic modeling method and apparatus for object detection model | |
| CN114419029B (en) | Training method of surface defect detection model, surface defect detection method and device | |
| CN111931809A (en) | Data processing method and device, storage medium and electronic equipment | |
| CN109685805B (en) | Image segmentation method and device | |
| CN112183326A (en) | Face age recognition model training method and related device | |
| CN112800226A (en) | Method for obtaining text classification model, method, device and equipment for text classification | |
| CN112765325A (en) | Vertical field corpus data screening method and system | |
| CN103279944A (en) | Image division method based on biogeography optimization | |
| CN112100378A (en) | Text classification model training method and device, computer equipment and storage medium | |
| CN116127060A (en) | Text classification method and system based on prompt words | |
| CN114049174A (en) | Method and device for commodity recommendation, electronic equipment and storage medium | |
| CN108984777B (en) | Customer service method, apparatus and computer-readable storage medium | |
| CN113379113A (en) | Method and device for predicting air conditioner operation mode and air conditioner | |
| CN112990753A (en) | Method, device and equipment for fault diagnosis of road side equipment | |
| CN116089616A (en) | Theme text acquisition method, device, equipment and storage medium | |
| CN104572820B (en) | The generation method and device of model, importance acquisition methods and device | |
| CN112784903A (en) | Method, device and equipment for training target recognition model | |
| CN116127188A (en) | Target feedback value determining method and device, electronic equipment and storage medium | |
| CN110851349B (en) | Page abnormity display detection method, terminal equipment and 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 |