Noise data modification method, device and electronic device for training set

Abstract

The present invention is applicable to the field of data processing technology, and provides a method, device and electronic device for modifying noise data in a training set, wherein the method comprises: clustering all pictures in a face picture training set to obtain multiple categories; grouping pictures in a target category according to annotation information to obtain multiple picture groups; replacing the original annotation information of all pictures in the target picture group with the i-th annotation information in the target annotation set, training a preset model, and obtaining the i-th recognition rate of the trained model; looping the process of obtaining the i-th recognition rate N times, with i taking different values ​​in each loop until i takes all values ​​from 1 to N to obtain N recognition rates; in the case where there is a recognition rate greater than the target recognition rate among the N recognition rates, modifying the annotation information of all pictures in the target picture group to the annotation information corresponding to the maximum recognition rate. The present invention can reduce the labor cost of correcting noise data.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to a noise data modification method, device and electronic equipment for a training set. Background Art

[0002] In recent years, deep learning has been widely used in image classification, target detection, scene recognition and other fields. In practical applications, a large number of training sets are needed to train neural networks. In order to ensure the training effect, the data in the training set is usually labeled manually. When the manual labeling is inaccurate, there will be some incorrectly labeled data in the training set, that is, noisy data. At present, the noisy data in the training set is usually corrected manually.

[0003] However, when there is a lot of noisy data, manual correction takes a long time and has high labor costs. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a method, device and electronic device for modifying noise data of a training set to solve the problem of high labor cost when correcting noise data in the prior art.

[0005] A first aspect of an embodiment of the present invention provides a method for modifying noise data for a training set, comprising:

[0006] Cluster all the images in the face image training set to obtain multiple categories;

[0007] The images in the target category are grouped according to the annotation information to obtain multiple image groups; wherein the target category is any one of the multiple categories, the images in each image group have the same annotation information, and the annotation information corresponding to different image groups is different;

[0008] The original annotation information of all pictures in the target picture group is replaced with the i-th annotation information in the target annotation set, and the preset model is trained based on the replaced target picture group and the picture groups other than the target picture group in the multiple picture groups, and the i-th recognition rate of the trained model is obtained; wherein the target picture group is any one of the multiple picture groups, the target annotation set is a set of all annotation information corresponding to the target category except the original annotation information, and the target annotation set includes N kinds of annotation information, 1≤i≤N, i and N are both positive integers;

[0009] The process of looping N times or more to obtain the i-th recognition rate, i takes a different value in each loop, until i takes all values ​​from 1 to N, and N recognition rates are obtained;

[0010] When there is a recognition rate greater than the target recognition rate among the N recognition rates, the annotation information of all pictures in the target picture group is modified to the annotation information corresponding to the maximum recognition rate among the N recognition rates; wherein the target recognition rate is the recognition rate of the model obtained after training the preset model based on multiple picture groups.

[0011] Optionally, cluster all images in the face image training set to obtain multiple categories, including:

[0012] All images in the face image training set are clustered according to the maximum and minimum K-means clustering algorithm to obtain multiple categories.

[0013] Optionally, group the images in the target category according to the annotation information, including:

[0014] Get the annotation information of all images in the target category;

[0015] Remove the repeated annotation information from the annotation information of all images to obtain the remaining annotation information;

[0016] According to each of the remaining annotation information, the images in the target category are grouped.

[0017] Optionally, after obtaining N recognition rates, the method further includes:

[0018] Get the maximum recognition rate among N recognition rates;

[0019] If the maximum recognition rate is greater than the target recognition rate, then there is a recognition rate greater than the target recognition rate among the N recognition rates; otherwise, there is no recognition rate greater than the target recognition rate among the N recognition rates.

[0020] Optionally, obtain the i-th recognition rate of the trained model, including:

[0021] The trained model is tested using a preset test set to obtain an i-th recognition rate; wherein the preset test set used each time the trained model is tested is the same.

[0022] A second aspect of an embodiment of the present invention provides a device for modifying noise data, including:

[0023] The clustering module is used to cluster all the images in the face image training set to obtain multiple categories;

[0024] A grouping module, used to group the images in the target category according to the annotation information to obtain multiple image groups; wherein the target category is any one of the multiple categories, the images in each image group have the same annotation information, and different image groups have different corresponding annotation information;

[0025] An acquisition module is used to replace the original annotation information of all pictures in the target picture group with the i-th annotation information in the target annotation set, and train the preset model based on the replaced target picture group and the picture group other than the target picture group in the multiple picture groups, and obtain the i-th recognition rate of the trained model; wherein the target picture group is any one of the multiple picture groups, the target annotation set is a set of all annotation information corresponding to the target category except the original annotation information, and the target annotation set includes N kinds of annotation information, 6≤i≤N, and i and N are both positive integers;

[0026] A loop module, used for looping N times or more to obtain the i-th recognition rate, i takes a different value in each loop, until i takes all values ​​from 6 to N, and obtains N recognition rates;

[0027] A modification module is used to modify the annotation information of all pictures in the target picture group to the annotation information corresponding to the maximum recognition rate among the N recognition rates when there is a recognition rate greater than the target recognition rate among the N recognition rates; wherein the target recognition rate is the recognition rate of the model obtained after training the preset model based on multiple picture groups.

[0028] Optionally, the clustering module is also used to:

[0029] All images in the face image training set are clustered according to the maximum and minimum K-means clustering algorithm to obtain multiple categories.

[0030] Optionally, the grouping module is also used to:

[0031] Get the annotation information of all images in the target category;

[0032] Remove the repeated annotation information from the annotation information of all images to obtain the remaining annotation information;

[0033] According to each of the remaining annotation information, the images in the target category are grouped.

[0034] Optionally, the noise data modification device further includes a judgment module, which is used to:

[0035] The maximum recognition rate among the N recognition rates is obtained; if the maximum recognition rate is greater than the target recognition rate, then there is a recognition rate greater than the target recognition rate among the N recognition rates, otherwise there is no recognition rate greater than the target recognition rate among the N recognition rates.

[0036] Optionally, the acquisition module is also used to:

[0037] The trained model is tested using a preset test set to obtain an i-th recognition rate; wherein the preset test set used each time the trained model is tested is the same.

[0038] A third aspect of an embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the first aspect when executing the computer program.

[0039] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0040] The embodiment of the present invention can first divide the pictures in the face picture training set into multiple categories, and then divide each category into multiple picture groups according to the annotation information of the pictures. After that, the annotation information of all pictures in the target picture group can be replaced with the annotation information in the target annotation set in sequence, and the recognition rate of the model obtained by training corresponding to each replacement can be obtained. Then, the target recognition rate corresponding to the target picture group and the recognition rate obtained by all replacements can be compared. If there is a recognition rate greater than the target recognition rate among all the recognition rates obtained by replacement, the annotation information of all pictures in the target picture group can be modified to the annotation information corresponding to the maximum recognition rate among all the recognition rates obtained by replacement. Since the recognition rate obtained by each replacement can reflect the probability of whether all pictures in the target picture group after the replacement are noise data, the recognition rate can be used to judge the noise data in the face picture training set. At the same time, it can be considered that the annotation information corresponding to the maximum recognition rate is the correct annotation information, and then the annotation information of all pictures in the target picture group can be modified to the annotation information corresponding to the maximum recognition rate, thereby correcting the noise data.

[0041] In this way, not only the accuracy of noise data correction can be guaranteed, but also the labor cost can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0043] Figure 1 A flowchart of a method for modifying noise data of a training set provided by an embodiment of the present invention;

[0044] Figure 2A method for modifying noise data of a training set provided by an embodiment of the present invention is a flowchart of the steps of the method for modifying noise data of a training set;

[0045] Figure 3 A schematic diagram of a noise data modification device provided by an embodiment of the present invention;

[0046] Figure 4 A schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0048] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0049] As described in the background technology, when manual annotation is inaccurate, some noise data will be present in the training set. At present, the noise data in the training set is usually corrected manually. However, when there is a lot of noise data, manual correction takes a long time and the labor cost is high.

[0050] In order to solve the problems in the prior art, an embodiment of the present invention provides a method, device and electronic device for modifying noise data of a training set. The following first introduces the method for modifying noise data of a training set provided by an embodiment of the present invention.

[0051] The applicant has found through research that when the number of face images in the face image training set is large, the annotation information of the newly added face images, such as the names of different parts of the face image such as lips, nose, eyebrows, forehead, etc., will most likely overlap with the annotation information of an existing face image in the face image training set. Based on this, when a face image is manually annotated incorrectly, the correct annotation information can be found from the set consisting of the annotation information of all images in the face image training set. In addition, the less noise data in the training sample, the higher the recognition rate of the face recognition model trained based on the training sample. Based on the above concept, a noise data modification method for a training set is proposed.

[0052] The execution subject of the noise data modification method for the training set provided in the embodiment of the present invention may be a noise data modification device, and the noise data modification device may be an electronic device with data processing capability, such as a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device may be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook or a personal digital assistant (PDA), etc., and the non-mobile electronic device may be a server, a network attached storage (NAS) or a personal computer (PC), etc., which is not specifically limited in the embodiment of the present invention.

[0053] like Figure 1 As shown, the noise data modification method for a training set provided by an embodiment of the present invention may include the following steps:

[0054] Step S110: cluster all the images in the face image training set to obtain multiple categories.

[0055] In some embodiments, considering that there are many pictures and categories in the face picture training set, all pictures in the face picture training set can be clustered to obtain multiple categories. In this way, pictures with a high degree of correlation can be classified into one category. It is worth mentioning that pictures belonging to the same category also have a high degree of correlation between their annotation information. For example, taking a face picture training set containing a large number of eye pictures as an example, when clustering, these eye pictures can be classified into one category. Correspondingly, the annotation information of these eye pictures, such as Ruifeng eyes, Danfeng eyes, round eyes, drooping eyes, squinting eyes, etc., also has a high degree of correlation with each other.

[0056] In some embodiments, all images in the face image training set may be clustered according to a maximum-minimum K-means clustering algorithm to obtain multiple categories.

[0057] Step S120: group the pictures in the target category according to the annotation information to obtain a plurality of picture groups.

[0058] In some embodiments, the target category may be any one of the multiple categories obtained above, the pictures in each picture group have the same annotation information, and different picture groups correspond to different annotation information.

[0059] For example, there are 100 pictures in the target category with 10 types of marking information in total. The pictures with the same marking information in these 100 pictures can be divided into one picture group. For example, if there are 20 pictures with round eye marking information in 100 pictures, these 20 pictures can be divided into one picture group.

[0060] Optional, such as Figure 2 As shown, the images in the target category are grouped according to the annotation information, including:

[0061] Step S210: Obtain the annotation information of all images in the target category.

[0062] In some embodiments, the tag information of all pictures in the target category may be read sequentially to obtain the tag information of all pictures in the target category.

[0063] Step S220: remove duplicate annotation information from the annotation information of all pictures to obtain remaining annotation information.

[0064] In some embodiments, after the annotation information of all images in the target category is obtained, repeated annotation information may be removed from the annotation information to obtain remaining annotation information.

[0065] It should be noted that the above process is only a way to implement the acquisition of the remaining annotation information, and is not a limitation on the acquisition of the remaining annotation information. For example, after reading the annotation information of each image in the target category in turn, it is also possible to determine whether the annotation information of the image is repeated with the acquired annotation information. If so, the annotation information of the image may not be marked as the remaining annotation information.

[0066] Step S230: group the images in the target category according to each type of annotation information in the remaining annotation information.

[0067] In some embodiments, after the remaining annotation information is obtained, pictures with the same annotation information can be searched among all pictures of the target category according to each type of annotation information in the remaining annotation information, and then the found pictures are divided into a group. In this way, the pictures in the target category can be grouped.

[0068] Through the processing of the above embodiment, pictures with the same tag information in the target category can be divided into one group.

[0069] Step S130, the original annotation information of all pictures in the target picture group is replaced with the i-th annotation information in the target annotation set, and based on the replaced target picture group and picture groups other than the target picture group in multiple picture groups, the preset model is trained, and the i-th recognition rate of the trained model is obtained.

[0070] In some embodiments, the target picture group can be any one of multiple picture groups, and the target annotation set can be a set of annotation information in the target category except the original annotation information of all pictures in the target picture group. The target annotation set can include N types of annotation information, 1≤i≤N, and i and N are both positive integers.

[0071] It should be noted that the above N can be used to refer to the types of annotation information in the target annotation set. If the types of annotation information in the target annotation set are 10, then N is 10.

[0072] In some embodiments, the preset model can be any model with face image recognition function, such as a residual network model. Considering that the target image grouping needs to be replaced with multiple annotation information in the future, the preset model will be trained multiple times accordingly, and the number of target image groupings may also be large. In order to reduce the cost of model training, a training method that takes into account both training cost and training effect can be adopted. In addition, the preset model can also adopt a model with lower complexity.

[0073] It should be noted that the recognition rate can reflect the probability of whether all pictures in the target picture group are noise data. If the recognition rate is higher, the probability of all pictures in the target picture group being noise data is lower.

[0074] In some embodiments, each time the trained model is tested, the same preset test set can be used. In this way, except for the labeling information, no other information is sent to change. In this way, the change in recognition rate will only be related to the labeling information, which can improve the recognition accuracy of the final noise data.

[0075] Step S140, the process of looping N times or more to obtain the i-th recognition rate, i takes a different value in each loop, until i takes all values ​​from 1 to N, and N recognition rates are obtained.

[0076] In some embodiments, the processing method of step 130 is followed, and the loop is repeated N times, and i takes different values ​​in each loop until i takes all values ​​from 1 to N. For example, i can be taken in order from 1 to N, so that the first recognition rate, the second recognition rate, ..., the Nth recognition rate can be obtained in sequence.

[0077] After the processing of the above embodiment, N recognition rates can be obtained. It should be noted that the recognition rate obtained by each replacement can reflect the probability of whether all pictures in the target picture group after the replacement are noise data.

[0078] Step S150: When there is a recognition rate greater than the target recognition rate among the N recognition rates, the labeling information of all pictures in the target picture group is modified to the labeling information corresponding to the maximum recognition rate among the N recognition rates.

[0079] In some embodiments, the target recognition rate may be the recognition rate of a model obtained after training a preset model based on multiple image groups. It should be noted that the multiple image groups here refer to image groups in which the annotation information of the images is the initial annotation information, that is, image groups in which the annotation information has not been replaced.

[0080] It should be noted that if the target recognition rate is greater than all N recognition rates, it can be considered that the initial annotation information of all pictures in the target picture group is more accurate than all the annotation information in the target annotation set, and it can be considered that all pictures in the target picture group are not noise data. Correspondingly, if there is a recognition rate greater than the target recognition rate among the N recognition rates, it indicates that the initial annotation information of all pictures in the target picture group may be inaccurate. At this time, it can be considered that all pictures in the target picture group are noise data. Afterwards, the annotation information of all pictures in the target picture group can be modified to the annotation information corresponding to the maximum recognition rate among the N recognition rates.

[0081] Specifically, the annotation information corresponding to the maximum recognition rate among the N recognition rates may be the annotation information of the target image group used when training the model with the maximum recognition rate.

[0082] In some embodiments, after obtaining N recognition rates, the maximum recognition rate among the N recognition rates may be obtained. Thereafter, the maximum recognition rate may be compared with the target recognition rate. If the maximum recognition rate is greater than the target recognition rate, it indicates that there is a recognition rate greater than the target recognition rate among the N recognition rates. If the maximum recognition rate is less than the target recognition rate, it indicates that there is no recognition rate greater than the target recognition rate among the N recognition rates.

[0083] It should be noted that after obtaining the result of whether the annotation information of all pictures in the target picture group is modified, the other picture groups in the target category can be modified in the same way until all picture groups in the target category have completed the modification process. After that, the other categories can be modified in the way of modifying the target category. In this way, the modification of the noise data in the entire face picture training set can be completed.

[0084] In an embodiment of the present invention, the pictures in the face picture training set are first divided into multiple categories, and then each category is divided into multiple picture groups according to the annotation information of the pictures. After that, the annotation information of all pictures in the target picture group can be replaced with the annotation information in the target annotation set in sequence, and the recognition rate of the model obtained by training corresponding to each replacement can be obtained. Then, the target recognition rate corresponding to the target picture group and the recognition rates obtained by all replacements can be compared. If there is a recognition rate greater than the target recognition rate among all the recognition rates obtained by replacement, the annotation information of all pictures in the target picture group can be modified to the annotation information corresponding to the maximum recognition rate among all the recognition rates obtained by replacement. Since the recognition rate obtained by each replacement can reflect the probability of whether all pictures in the target picture group after the replacement are noise data, the recognition rate can be used to judge the noise data in the face picture training set. At the same time, it can be considered that the annotation information corresponding to the maximum recognition rate is the correct annotation information, and then the annotation information of all pictures in the target picture group can be modified to the annotation information corresponding to the maximum recognition rate, thereby correcting the noise data.

[0085] In this way, not only the accuracy of noise data correction can be guaranteed, but also the labor cost can be reduced.

[0086] Based on the noise data modification method for a training set provided in the above embodiment, the present invention also provides a specific implementation of a noise data modification device applied to the noise data modification method for a training set. Please refer to the following embodiment.

[0087] Based on the noise data modification method for a training set provided in the above embodiment, the present invention also provides a specific implementation of a noise data modification device applied to the noise data modification method for a training set. Please refer to the following embodiment.

[0088] like Figure 3 As shown, a noise data modification device for a training set is provided, the device comprising:

[0089] A clustering module 310 is used to cluster all images in the face image training set to obtain multiple categories;

[0090] A grouping module 320 is used to group the pictures in the target category according to the annotation information to obtain multiple picture groups; wherein the target category is any one of the multiple categories, the pictures in each picture group have the same annotation information, and different picture groups have different corresponding annotation information;

[0091] An acquisition module 330 is used to replace the original annotation information of all pictures in the target picture group with the i-th annotation information in the target annotation set, and train the preset model based on the replaced target picture group and the picture group other than the target picture group in the multiple picture groups, and obtain the i-th recognition rate of the trained model; wherein the target picture group is any one of the multiple picture groups, the target annotation set is a set of all annotation information corresponding to the target category except the original annotation information, and the target annotation set includes N kinds of annotation information, 6≤i≤N, and i and N are both positive integers;

[0092] A loop module 340 is used to loop the process of obtaining the i-th recognition rate for N or more times, wherein i takes a different value in each loop until i takes all values ​​from 6 to N, and N recognition rates are obtained;

[0093] The modification module 350 is used to modify the annotation information of all pictures in the target picture group to the annotation information corresponding to the maximum recognition rate among the N recognition rates when there is a recognition rate greater than the target recognition rate among the N recognition rates; wherein the target recognition rate is the recognition rate of the model obtained after training the preset model based on multiple picture groups.

[0094] Optionally, the clustering module is also used to:

[0095] All images in the face image training set are clustered according to the maximum and minimum K-means clustering algorithm to obtain multiple categories.

[0096] Optionally, the grouping module is also used to:

[0097] Get the annotation information of all images in the target category;

[0098] Remove the repeated annotation information from the annotation information of all images to obtain the remaining annotation information;

[0099] According to each of the remaining annotation information, the images in the target category are grouped.

[0100] Optionally, the noise data modification device for the training set further includes a judgment module, which is used to:

[0101] The maximum recognition rate among the N recognition rates is obtained; if the maximum recognition rate is greater than the target recognition rate, then there is a recognition rate greater than the target recognition rate among the N recognition rates, otherwise there is no recognition rate greater than the target recognition rate among the N recognition rates.

[0102] Optionally, the acquisition module is also used to:

[0103] The trained model is tested using a preset test set to obtain an i-th recognition rate; wherein the preset test set used each time the trained model is tested is the same.

[0104] In an embodiment of the present invention, the pictures in the face picture training set are first divided into multiple categories, and then each category is divided into multiple picture groups according to the annotation information of the pictures. After that, the annotation information of all pictures in the target picture group can be replaced with the annotation information in the target annotation set in sequence, and the recognition rate of the model obtained by training corresponding to each replacement can be obtained. Then, the target recognition rate corresponding to the target picture group and the recognition rates obtained by all replacements can be compared. If there is a recognition rate greater than the target recognition rate among all the recognition rates obtained by replacement, the annotation information of all pictures in the target picture group can be modified to the annotation information corresponding to the maximum recognition rate among all the recognition rates obtained by replacement. Since the recognition rate obtained by each replacement can reflect the probability of whether all pictures in the target picture group after the replacement are noise data, the recognition rate can be used to judge the noise data in the face picture training set. At the same time, it can be considered that the annotation information corresponding to the maximum recognition rate is the correct annotation information, and then the annotation information of all pictures in the target picture group can be modified to the annotation information corresponding to the maximum recognition rate, thereby correcting the noise data.

[0105] In this way, not only the accuracy of noise data correction can be guaranteed, but also the labor cost can be reduced.

[0106] Figure 4 FIG. 1 is a schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 4 As shown, the electronic device 4 of this embodiment includes: a processor 40, a memory 41, and a computer program 42 stored in the memory 41 and executable on the processor 40. When the processor 40 executes the computer program 42, the steps in the above-mentioned noise data modification method for training sets are implemented. Alternatively, when the processor 40 executes the computer program 42, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0107] Exemplarily, the computer program 42 may be divided into one or more modules / units, which are stored in the memory 41 and executed by the processor 40 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 42 in the electronic device 4. For example, the computer program 42 may be divided into a clustering module, a grouping module, an acquisition module, a loop module, and a modification module, and the specific functions of each module are as follows:

[0108] The clustering module is used to cluster all the images in the face image training set to obtain multiple categories;

[0109] A grouping module, used to group the images in the target category according to the annotation information to obtain multiple image groups; wherein the target category is any one of the multiple categories, the images in each image group have the same annotation information, and different image groups have different corresponding annotation information;

[0110] An acquisition module is used to replace the original annotation information of all pictures in the target picture group with the i-th annotation information in the target annotation set, and train the preset model based on the replaced target picture group and the picture group other than the target picture group in the multiple picture groups, and obtain the i-th recognition rate of the trained model; wherein the target picture group is any one of the multiple picture groups, the target annotation set is a set of all annotation information corresponding to the target category except the original annotation information, and the target annotation set includes N kinds of annotation information, 6≤i≤N, and i and N are both positive integers;

[0111] A loop module, used for looping N times or more to obtain the i-th recognition rate, i takes a different value in each loop, until i takes all values ​​from 6 to N, and obtains N recognition rates;

[0112] A modification module is used to modify the annotation information of all pictures in the target picture group to the annotation information corresponding to the maximum recognition rate among the N recognition rates when there is a recognition rate greater than the target recognition rate among the N recognition rates; wherein the target recognition rate is the recognition rate of the model obtained after training the preset model based on multiple picture groups.

[0113] The electronic device 4 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The electronic device may include, but is not limited to, a processor 40 and a memory 41. Those skilled in the art will appreciate that Figure 4 It is only an example of the electronic device 4 and does not constitute a limitation of the electronic device 4. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0114] The processor 40 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0115] The memory 41 may be an internal storage unit of the electronic device 4, such as a hard disk or memory of the electronic device 4. The memory 41 may also be an external storage device of the electronic device 4, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 4. Further, the memory 41 may also include both an internal storage unit and an external storage device of the electronic device 4. The memory 41 is used to store the computer program and other programs and data required by the electronic device. The memory 41 may also be used to temporarily store data that has been output or is to be output.

[0116] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0117] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0118] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0119] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0120] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0121] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0122] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0123] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for modifying noise data for a training set, characterized in that: include: Cluster all the images in the face image training set to obtain multiple categories; Grouping the pictures in the target category according to the annotation information to obtain a plurality of picture groups; wherein the target category is any one of the plurality of categories, the pictures in each of the picture groups have the same annotation information, and different picture groups correspond to different annotation information; The original annotation information of all pictures in the target picture group is replaced with the i-th annotation information in the target annotation set, and based on the replaced target picture group and the picture groups other than the target picture group in the multiple picture groups, the preset model is trained, and the i-th recognition rate of the trained model is obtained; wherein the target picture group is any one of the multiple picture groups, the target annotation set is a set of all annotation information corresponding to the target category except the original annotation information, and the target annotation set includes N kinds of annotation information, 1≤i≤N, i and N are both positive integers; The process of looping N times or more to obtain the i-th recognition rate, i takes a different value in each loop, until i takes all values ​​from 1 to N, and N recognition rates are obtained; When there is a recognition rate greater than the target recognition rate among the N recognition rates, the annotation information of all pictures in the target picture group is modified to the annotation information corresponding to the maximum recognition rate among the N recognition rates; wherein the target recognition rate is the recognition rate of the model obtained after training the preset model based on the multiple picture groups.

2. The noise data modification method for a training set according to claim 1, characterized in that: The above method clusters all the images in the face image training set to obtain multiple categories, including: All images in the face image training set are clustered according to the maximum minimum K-means clustering algorithm to obtain multiple categories.

3. The noise data modification method for a training set according to claim 1, characterized in that: The grouping of images in the target category according to the annotation information includes: Obtaining the annotation information of all images in the target category; Eliminate duplicate annotation information from the annotation information of all the images to obtain remaining annotation information; The pictures in the target category are grouped according to each type of annotation information in the remaining annotation information.

4. The noise data modification method for a training set according to claim 1, characterized in that: After obtaining the N recognition rates, the method further includes: Obtaining the maximum recognition rate among the N recognition rates; If the maximum recognition rate is greater than the target recognition rate, then there is a recognition rate greater than the target recognition rate among the N recognition rates; otherwise, there is no recognition rate greater than the target recognition rate among the N recognition rates.

5. The noise data modification method for a training set according to any one of claims 1 to 4, characterized in that: The step of obtaining the i-th recognition rate of the trained model includes: The trained model is tested using a preset test set to obtain the i-th recognition rate; wherein the preset test set used each time the trained model is tested is the same.

6. A device for modifying noise data of a training set, characterized in that: include: The clustering module is used to cluster all the images in the face image training set to obtain multiple categories; A grouping module, used to group the pictures in the target category according to the annotation information to obtain a plurality of picture groups; wherein the target category is any one of the plurality of categories, the pictures in each picture group have the same annotation information, and the annotation information corresponding to different picture groups is different; An acquisition module, used to replace the original annotation information of all pictures in the target picture group with the i-th annotation information in the target annotation set, and train the preset model based on the replaced target picture group and the picture groups other than the target picture group in the multiple picture groups, and obtain the i-th recognition rate of the trained model; wherein the target picture group is any one of the multiple picture groups, the target annotation set is a set of all annotation information corresponding to the target category except the original annotation information, and the target annotation set includes N kinds of annotation information, 6≤i≤N, i and N are both positive integers; A loop module, used for looping N times or more to obtain the i-th recognition rate, i takes a different value in each loop, until i takes all values ​​from 6 to N, and obtains N recognition rates; A modification module is used to modify the annotation information of all pictures in the target picture group to the annotation information corresponding to the maximum recognition rate among the N recognition rates when there is a recognition rate greater than the target recognition rate among the N recognition rates; wherein the target recognition rate is the recognition rate of the model obtained after training the preset model based on the multiple picture groups.

7. The device for modifying noise data of a training set according to claim 6, characterized in that: The clustering module is also used to: All images in the face image training set are clustered according to the maximum minimum K-means clustering algorithm to obtain multiple categories.

8. The device for modifying noise data of a training set according to claim 6, characterized in that: The grouping module is also used for: Obtaining the annotation information of all images in the target category; Eliminate duplicate annotation information from the annotation information of all the images to obtain remaining annotation information; The pictures in the target category are grouped according to each type of annotation information in the remaining annotation information.

9. The device for modifying noise data of a training set according to claim 6, characterized in that: The device also includes a judging module, which is used to: The maximum recognition rate among the N recognition rates is obtained; if the maximum recognition rate is greater than the target recognition rate, then there is a recognition rate greater than the target recognition rate among the N recognition rates, otherwise there is no recognition rate greater than the target recognition rate among the N recognition rates.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

Images