Method and device for correcting training data set, electronic equipment and program product

Abstract

The invention is suitable for the technical field of artificial intelligence, and provides a training data set correction method and device, electronic equipment and a program product. The correction method of the training data set comprises the steps of obtaining a first detection result obtained by performing target detection on each image in an initial training set through a visual language deep learning algorithm; comparing the first detection result with a second detection result, and determining unknown corpus data in the initial training set according to a comparison result, the second detection result being a target detection result obtained by performing continuous frame detection on the initial training set; and correcting the annotation information of the unknown corpus data to obtain a target training set. According to the embodiment of the invention, the annotation information of the unknown corpus data in the training set can be more accurate, and the precision and robustness of the training object can be improved.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and in particular, relates to a method, device, electronic device and program product for correcting a training data set. Background Art

[0002] Model training usually relies on the labeling of training data sets. Traditional labeling methods need to be completed manually and the labeling results need to be confirmed by human eyes, which is inefficient. Some manual labeling solutions are optimized from the tool usage level, such as adding shortcut keys or labeling video streams, but they are still manual labeling in essence. The more advanced method is automatic labeling, but automatic labeling is mostly based on simulation environments and it is difficult to solve unknown situations in specific production environments. Summary of the invention

[0003] The embodiments of the present application provide a method, device, electronic device and program product for correcting a training data set, which can solve the problem of poor reliability of automatic labeling for unknown environments in related technologies.

[0004] A first aspect of an embodiment of the present application provides a method for correcting a training data set, including: obtaining a first detection result obtained by a visual language deep learning algorithm for performing target detection on each image in an initial training set; comparing the first detection result with a second detection result, and determining unknown corpus data in the initial training set according to the comparison result, wherein the second detection result is a target detection result obtained by performing continuous frame detection on the initial training set; and correcting the annotation information of the unknown corpus data to obtain a target training set.

[0005] In some embodiments of the first aspect, the labeling information of the unknown corpus data is corrected to obtain a target training set, including: performing trajectory correction according to the first detection result and the second detection result to obtain a corrected detection result; and correcting the labeling information of the unknown corpus data according to the corrected detection result to obtain a target training set.

[0006] In some embodiments of the first aspect, the trajectory correction is performed based on the first detection result and the second detection result to obtain a corrected detection result, including: assigning a life cycle to the detection frame of the unknown corpus data, and determining the trajectory of the detection target based on the detection results of each sample image within the life cycle; and fusing the first detection result and the second detection result based on the trajectory of the detection target to obtain the corrected detection result.

[0007] In some embodiments of the first aspect, the comparing the first detection result with the second detection result and determining unknown corpus data in the initial training set based on the comparison result includes: determining a position difference between the detection target in the first detection result and the second detection result; and in the initial training set, determining the training samples whose position difference is greater than a difference threshold as the unknown corpus data.

[0008] In some embodiments of the first aspect, the modifying the annotation information of the unknown corpus data to obtain the target training set includes: modifying the detection box position and / or text information of the image in the annotation information of the unknown corpus data to obtain the target training set, wherein the text information is used to describe the image content of the corresponding sample image.

[0009] In some embodiments of the first aspect, after obtaining the target training set, it also includes: retraining the visual language deep learning algorithm according to the target training set to obtain a retrained visual language deep learning algorithm, and the retrained visual language deep learning algorithm is used to generate annotation information for the image.

[0010] In some embodiments of the first aspect, the initial training set includes vocabulary data of multiple vocabulary libraries, each vocabulary library corresponding to an operating environment of the robot.

[0011] A second aspect of an embodiment of the present application provides a device for correcting a training data set, including: a result acquisition unit, used to obtain a first detection result obtained by a visual language deep learning algorithm for performing target detection on each image in an initial training set; an unknown corpus data determination unit, used to compare the first detection result with the second detection result, and determine unknown corpus data in the initial training set according to the comparison result, wherein the second detection result is a target detection result obtained by performing continuous frame detection on the initial training set; and a correction unit, used to correct the annotation information of the unknown corpus data to obtain a target training set.

[0012] A third aspect of an embodiment of the present application 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 above-mentioned method for correcting a training data set when executing the computer program.

[0013] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for correcting the above-mentioned training data set are implemented.

[0014] A fifth aspect of an embodiment of the present application provides a computer program product. When the computer program product is run on an electronic device, the electronic device executes the steps of the above-mentioned method for correcting a training data set.

[0015] In an implementation of the present application, a first detection result is obtained by obtaining a visual language deep learning algorithm to perform target detection on each image in an initial training set, and the first detection result is compared with a second detection result obtained by performing continuous frame detection on the initial training set. According to the comparison result, unknown corpus data is determined in the initial training set, and the annotation information of the unknown corpus data is corrected to obtain a target training set. This can make the annotation information of the unknown corpus data in the training set more accurate, and make the annotation information of the unknown environment more reliable, which helps to improve the accuracy and robustness of the training objects. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, 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 application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0017] Figure 1 It is a schematic diagram of an implementation flow of a method for correcting a training data set provided in an embodiment of the present application;

[0018] Figure 2 It is a structural schematic diagram of a training data set correction device provided in an embodiment of the present application;

[0019] Figure 3 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are protected by the present application.

[0021] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0022] In the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0023] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0024] Traditional labeling methods require manual completion and visual confirmation of the labeling results, which is inefficient. Some manual labeling solutions are optimized from the tool usage level, such as adding shortcut keys or labeling video streams, but they are still essentially manual labeling. The more advanced method is automatic labeling, but automatic labeling is mostly based on simulation environments and is difficult to solve unknown situations in specific production environments.

[0025] In view of this, the present application proposes a method for correcting a training data set, which can make the annotation information of unknown corpus data in the training set more accurate, make the annotation information of unknown environments more reliable, and help improve the accuracy and robustness of the training object.

[0026] In order to illustrate the technical solution of the present application, a specific embodiment is provided below for illustration.

[0027] Figure 1 A schematic diagram of the implementation flow of a method for correcting a training data set provided in an embodiment of the present application is shown, and the method can be applied to an electronic device. The electronic device can be a computer, a mobile phone, a robot or other intelligent device.

[0028] Specifically, the method for correcting the training data set may include the following steps S101 to S103.

[0029] Step S101, obtaining a first detection result obtained by performing target detection on each image in an initial training set using a visual language deep learning algorithm.

[0030] Specifically, the initial training set can be a public training set or a training set obtained based on an actual environment, which is used to train the visual language deep learning algorithm.

[0031] The visual language deep learning algorithm may specifically refer to a visual language model, which can be used to perform target detection on input multimodal information. The multimodal information may include visual information (i.e., images) and text information. The visual language deep learning algorithm performs target detection on each image in the initial training set, and a first detection result may be obtained. The first detection result may include a detection box / initial position of the target in the image.

[0032] Step S102, comparing the first detection result with the second detection result, and determining unknown corpus data in the initial training set according to the comparison result.

[0033] The second detection result is the target detection result obtained by performing continuous frame detection on the initial training set. Continuous frame detection refers to joint detection of a group of images that are continuous in time sequence, that is, for each image, target detection can be performed in combination with its upper and lower frame images to obtain the second detection result.

[0034] In an embodiment of the present application, the first detection result is compared with the second detection result to obtain a comparison result. The comparison result can reflect the deviation between the first detection result and the second detection result. In the initial training set, if the deviation between the first detection result and the second detection result of the same image is obvious, it can be used as unknown corpus data. It can be understood that the unknown corpus data is usually some extreme scenarios, corresponding to unknown situations in a specific production environment.

[0035] Step S103, correcting the annotation information of the unknown corpus data to obtain a target training set.

[0036] In the implementation of the present application, for unknown corpus data, its annotation information can be modified so that the annotation information is more consistent with the actual situation presented in the image, and a target training set is obtained. The target training set is a training set obtained after the initial training set is modified.

[0037] In an implementation of the present application, a first detection result is obtained by obtaining a visual language deep learning algorithm to perform target detection on each image in an initial training set, and the first detection result is compared with a second detection result obtained by performing continuous frame detection on the initial training set. According to the comparison result, unknown corpus data is determined in the initial training set, and the annotation information of the unknown corpus data is corrected to obtain a target training set. This can make the annotation information of the unknown corpus data in the training set more accurate, and make the annotation information of the unknown environment more reliable, which helps to improve the accuracy and robustness of the training objects.

[0038] In some embodiments of the present application, before obtaining a first detection result obtained by the visual language deep learning algorithm by performing target detection on each image in an initial training set, the visual language deep learning algorithm may be initialized.

[0039] Specifically, the initial training set can be used to train the visual language deep learning algorithm to implement the initialization operation.

[0040] In some embodiments of the present application, the initial training set may include vocabulary data of multiple vocabulary libraries, each vocabulary library corresponding to an operating environment of the robot.

[0041] For example, the robot can operate in environments such as parks, shopping malls, and schools. For parks, shopping malls, schools, and other environments, a vocabulary can be established separately. The vocabulary can record text information corresponding to the sample image. The text information can be used to describe the image content of the sample image, and the image content includes but is not limited to the image background and image foreground (i.e., the detection target). In this way, the text information and the corresponding sample image can be used as a training sample to train the visual language deep learning algorithm. Combining the vocabulary corresponding to different operating environments into an initial training set can make the visual language deep learning algorithm universal for different operating environments and cover different business scenarios of the robot.

[0042] In some embodiments of the present application, comparing the first detection result with the second detection result, and determining unknown corpus data in the initial training set according to the comparison result may include: determining the position difference of the detection target in the first detection result and the second detection result. In the initial training set, the training samples whose position difference is greater than the difference threshold are determined as unknown corpus data.

[0043] Among them, the position difference can reflect the deviation between the detection boxes of the same detection target in the first detection result and the second detection result, and can be calculated by feature correlation, histogram distribution, text word feature distance or other existing algorithms, and this application does not impose any restrictions on this.

[0044] If the position difference is greater than the difference threshold, it means that there is a large deviation between the detection result of the visual language deep learning algorithm and the detection result of the continuous frame. At this time, the training sample can be determined as unknown corpus data.

[0045] In some embodiments of the present application, correcting the labeling information of unknown corpus data to obtain a target training set may include: performing trajectory correction according to the first detection result and the second detection result to obtain a corrected detection result, and correcting the labeling information of the unknown corpus data according to the corrected detection result to obtain a target training set.

[0046] Specifically, the trajectory of the detection target can be corrected by combining the first detection result and the second detection result to determine the position of the detection target in the current frame corresponding to the unknown corpus data, and obtain a corrected detection result. The corrected detection result is used as the corrected annotation information to form a target training set.

[0047] Specifically, in some embodiments of the present application, trajectory correction is performed based on the first detection result and the second detection result to obtain a corrected detection result, which may include: assigning a life cycle to the detection frame of the unknown corpus data, and determining the trajectory of the detection target based on the detection results of each sample image within the life cycle; based on the trajectory of the detection target, the first detection result and the second detection result are merged to obtain a corrected detection result.

[0048] Specifically, the life cycle may include unknown corpus data and sample images that form continuous frames with the unknown corpus data. The detection results of each sample image in the life cycle may include a first detection result and / or a second detection result of the sample image. Using the detection results of each sample image in the life cycle, the trajectory of the detection target can be determined to fuse the first detection result with the second detection result to obtain a corrected detection result. Among them, the fusion method can be weighted fusion, and the fusion weight can be set according to the detection position deviation (i.e., trajectory stability) between the first detection result and the second detection result and the upper and lower frame images.

[0049] In other words, pseudo-annotation (first detection result and second detection result) and trajectory correction of unknown corpus data can be performed. According to the relationship between the upper and lower frame tracking and the feature-related relationship within the life cycle, the image detection target life cycle is given, and according to the trajectory of the detection target in each frame, the target detection frame after trajectory correction is used as the corrected annotation information. In this way, by giving continuous frames a life cycle and integrating the visual language deep learning algorithm, the training data set can be effectively corrected.

[0050] In some embodiments of the present application, correcting the annotation information of the unknown corpus data to obtain the target training set may include: correcting the detection box position and / or text information of the image in the annotation information of the unknown corpus data to obtain the target training set. The text information can be used to describe the image content of the corresponding sample image. By correcting, the detection box position and text information of the training data set can be made more consistent with the content actually presented in the sample image.

[0051] After obtaining the target training set, in some embodiments of the present application, it may also include: retraining the visual language deep learning algorithm according to the target training set to obtain a retrained visual language deep learning algorithm. In this way, the retrained visual language deep learning algorithm can be applied in extreme scenarios in unknown production deployment environments with higher accuracy and robustness. The retrained visual language deep learning algorithm can be used to generate annotation information for images, making the automatic annotation results more accurate.

[0052] It should be noted that, for the sake of simplicity of description, the aforementioned method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited to the described order of actions, because according to the present application, certain steps can be performed in other orders.

[0053] like Figure 2 FIG. 2 is a schematic diagram of the structure of a training data set correction device 200 provided in an embodiment of the present application. The training data set correction device 200 is configured on an electronic device.

[0054] Specifically, the training data set correction device 200 may include:

[0055] A result acquisition unit 201 is used to acquire a first detection result obtained by performing target detection on each image in the initial training set by a visual language deep learning algorithm;

[0056] An unknown corpus data determination unit 202 is used to compare the first detection result with the second detection result, and determine the unknown corpus data in the initial training set according to the comparison result, wherein the second detection result is a target detection result obtained by performing continuous frame detection on the initial training set;

[0057] The correction unit 203 is used to correct the labeling information of the unknown corpus data to obtain a target training set.

[0058] In some embodiments of the present application, the correction unit 203 can be specifically used to: perform trajectory correction according to the first detection result and the second detection result to obtain a corrected detection result; and correct the annotation information of the unknown corpus data according to the corrected detection result to obtain a target training set.

[0059] In some embodiments of the present application, the correction unit 203 can be specifically used to: assign a life cycle to the detection frame of the unknown corpus data, and determine the trajectory of the detection target based on the detection results of each sample image within the life cycle; and fuse the first detection result with the second detection result based on the trajectory of the detection target to obtain the corrected detection result.

[0060] In some embodiments of the present application, the unknown corpus data determination unit 202 can be specifically used to: determine the position difference of the detection target in the first detection result and the second detection result; in the initial training set, determine the training samples whose position difference is greater than the difference threshold as the unknown corpus data.

[0061] In some embodiments of the present application, the correction unit 203 can be specifically used to: correct the detection box position and / or text information of the image in the annotation information of the unknown corpus data to obtain the target training set, and the text information is used to describe the image content of the corresponding sample image.

[0062] In some embodiments of the present application, the training data set correction device 200 may include a training unit for retraining the visual language deep learning algorithm according to the target training set to obtain a retrained visual language deep learning algorithm, and the retrained visual language deep learning algorithm is used to generate annotation information for the image.

[0063] In some embodiments of the present application, the initial training set includes vocabulary data of multiple vocabulary libraries, each vocabulary library corresponds to an operating environment of the robot.

[0064] It should be noted that, for the convenience and simplicity of description, the specific working process of the training data set correction device 200 can refer to Figure 1 The corresponding process of the method will not be repeated here.

[0065] like Figure 3 , which is a schematic diagram of an electronic device provided in an embodiment of the present application. Specifically, the electronic device 3 may include: a processor 30, a memory 31, and a computer program 32 stored in the memory 31 and executable on the processor 30, such as a correction program for a training data set. When the processor 30 executes the computer program 32, the steps in the above-mentioned correction method for each training data set are implemented, such as Figure 1 Alternatively, when the processor 30 executes the computer program 32, the functions of each module / unit in the above-mentioned device embodiments are realized, for example Figure 2 The functions of the result acquisition unit 201, the unknown corpus data determination unit 202 and the correction unit 203 are shown.

[0066] The computer program may be divided into one or more modules / units, which are stored in the memory 31 and executed by the processor 30 to complete the present application. 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 in the electronic device.

[0067] For example, the computer program can be divided into: a result acquisition unit, an unknown corpus data determination unit, and a correction unit. The specific functions of each unit are as follows: a result acquisition unit, used to obtain a first detection result obtained by the visual language deep learning algorithm for target detection of each image in the initial training set; an unknown corpus data determination unit, used to compare the first detection result with the second detection result, and determine the unknown corpus data in the initial training set according to the comparison result, wherein the second detection result is a target detection result obtained by continuous frame detection of the initial training set; a correction unit, used to correct the annotation information of the unknown corpus data to obtain a target training set.

[0068] The electronic device may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art will appreciate that Figure 3 These are merely examples of electronic devices and do not constitute a limitation of the electronic device. The electronic device may include more or fewer components than those 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.

[0069] The processor 30 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), off-the-shelf programmable gate arrays 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.

[0070] The memory 31 may be an internal storage unit of the electronic device, such as a hard disk or memory of the electronic device. The memory 31 may also be an external storage device of the electronic device, 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. Further, the memory 31 may also include both an internal storage unit of the electronic device and an external storage device. The memory 31 is used to store the computer program and other programs and data required by the electronic device. The memory 31 may also be used to temporarily store data that has been output or is to be output.

[0071] It should be noted that, for the convenience and brevity of description, the structure of the above electronic device can also refer to the specific description of the structure in the method embodiment, which will not be repeated here.

[0072] 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.

[0073] 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.

[0074] 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 this application.

[0075] In the embodiments provided in the present application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely 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.

[0076] 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.

[0077] In addition, each functional unit in each embodiment of the present application 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.

[0078] 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 application 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), random access memory (RAM), 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.

[0079] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person 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 application, and should all be included in the protection scope of the present application.

Claims

1. A method for correcting a training data set, characterized in that: include: Obtaining a first detection result obtained by performing target detection on each image in the initial training set using a visual language deep learning algorithm; Comparing the first detection result with the second detection result, and determining unknown corpus data in the initial training set according to the comparison result, wherein the second detection result is a target detection result obtained by performing continuous frame detection on the initial training set; The labeling information of the unknown corpus data is modified to obtain a target training set.

2. The method for correcting a training data set according to claim 1, characterized in that: The step of correcting the labeling information of the unknown corpus data to obtain a target training set includes: Performing trajectory correction according to the first detection result and the second detection result to obtain a corrected detection result; According to the corrected detection result, the labeling information of the unknown corpus data is corrected to obtain a target training set.

3. The method for correcting a training data set according to claim 2, characterized in that: The performing trajectory correction according to the first detection result and the second detection result to obtain a corrected detection result includes: Assigning a life cycle to the detection frame of the unknown corpus data, and determining a trajectory of the detection target according to the detection results of each sample image within the life cycle; The first detection result and the second detection result are fused according to the trajectory of the detection target to obtain the corrected detection result.

4. The method for correcting a training data set according to claim 1, wherein: The comparing the first detection result with the second detection result, and determining unknown corpus data in the initial training set according to the comparison result, comprises: Determine a position difference between the first detection result and the second detection result of the detection target; In the initial training set, the training samples whose position difference is greater than a difference threshold are determined as the unknown corpus data.

5. The method for correcting a training data set according to any one of claims 1 to 4, characterized in that: The step of correcting the labeling information of the unknown corpus data to obtain a target training set includes: In the annotation information of the unknown corpus data, the detection frame position and / or text information of the image is corrected to obtain the target training set, and the text information is used to describe the image content of the corresponding sample image.

6. The method for correcting a training data set according to any one of claims 1 to 4, characterized in that: After obtaining the target training set, the method further includes: The visual language deep learning algorithm is retrained according to the target training set to obtain a retrained visual language deep learning algorithm, and the retrained visual language deep learning algorithm is used to generate annotation information for the image.

7. The method for correcting a training data set according to any one of claims 1 to 4, characterized in that: The initial training set includes vocabulary data of multiple vocabulary libraries, and each vocabulary library corresponds to an operating environment of the robot.

8. A training data set correction device, characterized in that: include: A result acquisition unit, used to acquire a first detection result obtained by performing target detection on each image in the initial training set by the visual language deep learning algorithm; an unknown corpus data determination unit, configured to compare the first detection result with the second detection result, and determine unknown corpus data in the initial training set according to the comparison result, wherein the second detection result is a target detection result obtained by performing continuous frame detection on the initial training set; The correction unit is used to correct the labeling information of the unknown corpus data to obtain a target training set.

9. 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 for correcting the training data set according to any one of claims 1 to 7 are implemented.

10. A computer program product, characterized in that It comprises a computer program, which, when executed by a processor, implements the steps of the method for correcting a training data set as claimed in any one of claims 1 to 7.

Images