Training set generation method and device, electronic equipment and computer readable storage medium

Abstract

The invention relates to the field of driving assistance, in particular to a training set generation method and device, electronic equipment and a computer readable storage medium. The training set generation method comprises the steps of generating simulation driving data based on a preset simulator; inputting the simulation driving data into a target generator to obtain generated data; wherein the target generator is a generator obtained based on generative adversarial network training; based on the generated data, a model training set of an advanced driver assistance system (ADAS) model is determined. According to the method, the training set acquisition efficiency can be improved, and the comprehensive road condition data can be acquired for training the ADAS model, so that the driving auxiliary performance of the ADAS model is improved.

Description

Technical Field

[0001] The present application relates to the field of driving assistance, and in particular to a training set generation method, device, electronic device and computer-readable storage medium. Background Art

[0002] The Advanced Driving Assistance System (ADAS) can use various sensors installed on the vehicle to sense the surrounding environment and collect data in real time while the vehicle is driving. It can identify, detect and track static and dynamic objects, and combine navigation map data to perform systematic calculations and analysis, so that the driver can be aware of possible dangers in advance, effectively increasing the comfort and safety of car driving.

[0003] The training equipment can collect driving data of vehicles in various real-life scenarios, such as video footage of road conditions, and use this driving data as training data for the ADAS model to train and verify the ADAS.

[0004] However, due to the climate, geographical and other limitations of the vehicle in the real driving environment, the above-mentioned collection method is inefficient in obtaining training sets and it is difficult to collect comprehensive road condition data. The ADAS model trained based on this method cannot flexibly provide driving assistance in various road conditions, affecting driving safety. Summary of the Invention

[0005] In view of the above, the embodiments of the present application provide a training set generation method, device, electronic device and computer-readable storage medium, which can not only improve the efficiency of training set acquisition, but also obtain comprehensive road condition data for training ADAS models, thereby improving the driving assistance performance of the ADAS model.

[0006] The present invention provides a method for generating a training set, including:

[0007] Generate simulated driving data based on a preset simulator;

[0008] Inputting the simulated driving data into a target generator to obtain generated data;

[0009] Wherein, the target generator is a generator obtained by training based on a generative adversarial network;

[0010] Based on the generated data, a model training set of an advanced driver assistance system (ADAS) model is determined.

[0011] The embodiment of the present application uses a target generator in combination with a simulator. The simulator can simulate the real environment and obtain simulated driving data. The generator obtained by training the generative adversarial network can learn the differences between the simulated driving data and the real driving data, so that the generated data obtained based on the simulated driving data can be close to the real driving data, thereby greatly reducing the optimization steps of the simulation parameters in the simulator.

[0012] Moreover, the embodiment of the present application uses generated data close to real driving data as the model training set of the ADAS model, which can reduce the dependence of the ADAS model training data acquisition process on the real road conditions. This can not only improve the efficiency of obtaining the training set, but also facilitate the acquisition of training data reflecting various road conditions while reducing the amount of real driving data collected, so as to improve the driving assistance performance of the ADAS model.

[0013] In some embodiments, obtaining simulated driving data based on a preset simulator includes:

[0014] Obtaining driving scenario parameters to be trained for the ADAS model;

[0015] The driving scene parameters to be trained are input into the simulator to obtain the simulated driving data.

[0016] In some embodiments, the generative adversarial network includes an initial generator and a discriminator, and the training steps of the target generator include:

[0017] generating first driving data based on the initial generator, wherein a label of the first driving data is forged data;

[0018] Inputting the first driving data into the discriminator of the generative adversarial network to obtain a result of determining the authenticity of the first driving data;

[0019] determining whether the discriminator has converged based on a label of the first driving data and a result of determining the authenticity of the first driving data;

[0020] If the discriminator converges, it is confirmed that the training of the initial generator is completed, and the target generator is obtained.

[0021] In some embodiments, generating the first driving data based on the initial generator includes:

[0022] generating second driving data based on the simulator;

[0023] The second driving data is input into the initial generator to obtain the first driving data.

[0024] The present application also provides a training set generation method, which is used to train the target generator described above. The target generator is a generator trained based on a generative adversarial network, and the generative adversarial network includes an initial generator and a discriminator. The training set generation method includes:

[0025] generating first driving data based on the initial generator, wherein a label of the first driving data is forged data;

[0026] inputting the first driving data into the discriminator to obtain a result of determining the authenticity of the first driving data;

[0027] determining whether the discriminator has converged based on a label of the first driving data and a result of determining the authenticity of the first driving data;

[0028] If the discriminator converges, it is confirmed that the training of the initial generator is completed, and the target generator is obtained.

[0029] The embodiment of the present application trains an initial generator based on a generative adversarial network, so that the initial generator can learn the differences between simulated driving data and real driving data to obtain a target generator for generating a training set, so that the generated data can be close to the real driving data, thereby greatly reducing the optimization steps of the simulation parameters in the simulator.

[0030] Moreover, the embodiment of the present application uses generated data close to real driving data as the model training set of the ADAS model, which can reduce the dependence of the ADAS model training data acquisition process on the real road conditions. This can not only improve the efficiency of obtaining the training set, but also facilitate the acquisition of training data reflecting various road conditions while reducing the amount of real driving data collected, so as to improve the driving assistance performance of the ADAS model.

[0031] In some embodiments, generating the first driving data based on the initial generator includes:

[0032] generating second driving data based on the simulator;

[0033] The second driving data is input into the initial generator to obtain the first driving data.

[0034] In some embodiments, determining whether the discriminator has converged based on the label of the first driving data and the authenticity determination result of the first driving data includes:

[0035] Acquire third driving data, where a label of the third driving data is real data;

[0036] inputting the third driving data into the discriminator to obtain a result of determining the authenticity of the third driving data;

[0037] determining an accuracy rate of authenticity determination based on the label of the first driving data, the authenticity determination result of the first driving data, the label of the third driving data, and the authenticity determination result of the third driving data;

[0038] Based on the accuracy of the authenticity judgment, it is determined whether the discriminator has converged.

[0039] The present application also provides a training set generation device, including:

[0040] A simulation module, used for generating simulated driving data based on a preset simulator;

[0041] a generating module, configured to input the simulated driving data into a target generator to obtain generated data;

[0042] Wherein, the target generator is a generator obtained by training based on a generative adversarial network;

[0043] A production module is used to determine a model training set of an advanced driver assistance system (ADAS) model based on the generated data.

[0044] An embodiment of the present application also provides an electronic device, which includes a processor and a memory, wherein the memory is used to store instructions, and the processor is used to call the instructions in the memory so that the electronic device executes the above-mentioned training set generation method.

[0045] An embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the above-mentioned training set generation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The steps of the training set generation method provided in accordance with an embodiment of the present application are as follows Figure 1 .

[0047] Figure 2 The steps of the training set generation method provided in another embodiment of the present application are as follows Figure 2 .

[0048] Figure 3 A schematic diagram of a scenario of a training target generator provided according to an embodiment of the present application.

[0049] Figure 4 The steps of the training set generation method provided in another embodiment of the present application are as follows Figure 3 .

[0050] Figure 5A schematic diagram of a scenario for obtaining a model training set according to an embodiment of the present application.

[0051] Figure 6 The figure is a schematic diagram of the structure of a training set generation device provided according to an embodiment of the present application.

[0052] Figure 7 This is a structural diagram of a training set generation device provided according to another embodiment of the present application.

[0053] Figure 8 This is a schematic structural diagram of an electronic device provided according to another embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to more clearly understand the above-mentioned objectives, features and advantages of the present application, the present application is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features therein can be combined with each other in the absence of conflict.

[0055] In the following description, many specific details are set forth to facilitate a full understanding of the present application. The described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application pertains. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0057] It should be further noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0058] In this application, "at least one" means one or more, and "more than one" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and drawings of this application are used to distinguish similar objects, rather than to describe a specific order or sequence.

[0059] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0060] The ADAS system can use various sensors installed on the vehicle to sense the surrounding environment and collect data in real time while the vehicle is driving, identify, detect and track static and dynamic objects, and combine navigation map data to perform systematic calculations and analysis, so that the driver can be aware of possible dangers in advance, effectively increasing the comfort and safety of car driving.

[0061] The training equipment can collect driving data of vehicles in various real-life scenarios, such as video footage of road conditions, and use this driving data as training data for the ADAS model to train, test, and verify the ADAS.

[0062] However, due to the limitations of climate, region, and other factors in the real driving environment of vehicles, the training set acquisition efficiency of the above collection method is low, and it is difficult to collect comprehensive road condition data. The ADAS model trained based on this method cannot flexibly provide driving assistance in various road conditions, affecting driving safety.

[0063] In some embodiments, a simulator can be used to simulate driving scenarios. For example, the simulator can internally create multiple driving scenario parameters, such as road condition parameters and climate parameters. After the user determines that the ADAS model's assistance effect is not good in a driving scenario, they can select certain driving scenario parameters in the simulator. The simulator can then simulate the driving scenario for the selected driving scenario parameters to obtain training data, so that the collected data is no longer restricted by the climate, region, etc. of the real scene.

[0064] However, in order to make the simulated road condition data (such as image information) closer to the real scene, the user needs to spend a lot of time adjusting the simulation parameters in the simulator. Therefore, the efficiency of using this solution to obtain training data is also low.

[0065] Moreover, the driving scenario parameters pre-stored in the simulator are limited, while the driving scenario parameters encountered during actual driving are flexible and changeable. The above method cannot obtain the driving data of the vehicle in various driving scenarios.

[0066] In view of the above, embodiments of the present application also provide a training set generation method, device, electronic device and computer-readable storage medium.

[0067] The training set generation method of the embodiment of the present application can be applied to one or more electronic devices. The electronic device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to a processor, a microprogrammed control unit (MCU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), an embedded device, etc. The electronic device can be a vehicle-mounted computer, a personal computer, a server, etc., but is not limited thereto.

[0068] Figure 1 The present invention provides a flowchart of the steps of a training set generation method provided in one embodiment of the present invention. The training set generation method is used to train a target generator based on a Generative Adversarial Network (GAN).

[0069] The Generative Adversarial Network (GAN) consists of an initial generator and a discriminator. The target generator is the trained initial generator.

[0070] The target generator is used to obtain generated data based on simulated driving data, and the generated data can be used to determine a model training set of the ADAS model.

[0071] According to different requirements, the order of the steps in the flowchart can be changed, and some steps can be omitted.

[0072] See Figure 1 As shown, the training set generation method may include the following steps.

[0073] Step 101: Generate first driving data based on an initial generator, where the label of the first driving data is forged data.

[0074] The first driving data is used to describe the road environment information forged by the initial generator, such as driving road images, surrounding obstacles, etc.

[0075] In some embodiments, the vehicle may be equipped with various sensors, such as optical writing radars, cameras, etc., to obtain road environment information. The initial generator may forge road environment information monitored by various sensors and use it as the first driving data.

[0076] The first driving data may also include forged vehicle motion parameters such as speed, acceleration, direction, etc. of the vehicle, but is not limited thereto.

[0077] In some embodiments, step 101 may include: obtaining random noise, and inputting the random noise into the initial generator to generate first driving data.

[0078] In other embodiments, reference Figure 2 As shown, step 101 may include:

[0079] Step 1011: Generate second driving data based on the simulator.

[0080] The simulator (ADAS Simulator) can be configured based on existing ADAS datasets.

[0081] refer to Figure 3 As shown, the existing ADAS dataset stores real data, which may include real driving data collected by sensors in the vehicle and / or data in a public ADAS dataset.

[0082] The second driving data is used to describe the road environment information simulated by the simulator, such as driving road images, surrounding obstacles, etc.

[0083] The second driving data may also include vehicle motion parameters such as the simulated vehicle's speed, acceleration, direction, etc., but is not limited thereto.

[0084] Step 1012: Input the second driving data into the initial generator to obtain the first driving data.

[0085] The embodiment of the present application uses the second driving data obtained based on the simulator as the input of the discriminator. On the one hand, it facilitates the generator to learn the difference between the second driving data (i.e., simulated driving data) and the first driving data. On the other hand, it can improve the training efficiency of the generator.

[0086] Step 102: input the first driving data into a discriminator to obtain a result of determining the authenticity of the first driving data.

[0087] Specifically, after the first driving data is input into the discriminator of the generative adversarial network, the discriminator can determine whether the first driving data is real data or forged data to obtain a authenticity judgment result.

[0088] Step 103 : determining whether the discriminator has converged based on the label of the first driving data and the authenticity judgment result of the first driving data.

[0089] In some embodiments, step 103 may include: determining the accuracy of the authenticity judgment performed by the discriminator based on the label of the first driving data and the authenticity judgment result of the first driving data. For example, if the authenticity judgment result of the first driving data is forged data, it means that the discriminator's judgment is correct; if the authenticity judgment result of the first driving data is real data, it means that the discriminator's judgment is wrong. Then, based on the accuracy, determining whether the discriminator has converged.

[0090] For example, when the accuracy rate exceeds a preset threshold, it indicates that the discriminator has reached a steady state, and the discriminator can be determined to have converged; if the accuracy rate is less than the preset threshold, it is determined that the discriminator has not converged.

[0091] Among them, the preset threshold can be set according to actual application requirements. For example, the preset threshold can be set to 0.5, or fluctuate around 0.5. This embodiment of the present application is not limited to this.

[0092] In some embodiments, step 103 may include:

[0093] Step 1031: Acquire third driving data, where the label of the third driving data is real data.

[0094] The third driving data is used to describe the road condition and environment information actually collected, for example, the road condition and environment information actually collected based on the sensors configured in the vehicle.

[0095] Reference again Figure 3 As shown, the third driving data may come from an existing ADAS dataset.

[0096] In some embodiments, the third driving data may also include actual collected vehicle motion parameters such as vehicle speed, acceleration, direction, etc., but is not limited thereto.

[0097] Step 1032: Input the third driving data into the discriminator to obtain a result of authenticity determination of the third driving data.

[0098] Specifically, after the third driving data is input into the discriminator, the discriminator can determine whether the third driving data is real data or forged data to obtain an authenticity determination result.

[0099] Step 1033 : Determine the accuracy of the authenticity judgment based on the label of the first driving data and the authenticity judgment result thereof and the third driving data and the authenticity judgment result thereof.

[0100] For example, if the authenticity judgment result of the first driving data is forged data, the authenticity judgment result is consistent with the label of the first driving data, indicating that the discriminator's judgment is correct; if the authenticity judgment result of the first driving data is real data, the authenticity judgment result is inconsistent with the label of the first driving data, indicating that the discriminator's judgment is wrong.

[0101] If the authenticity judgment result of the third driving data is forged data, the authenticity judgment result is inconsistent with the label of the third driving data, indicating that the discriminator has made an incorrect judgment; if the authenticity judgment result of the third driving data is real data, the authenticity judgment result is consistent with the label of the third driving data, indicating that the discriminator has made a correct judgment.

[0102] Based on the authenticity judgment result, it is determined whether the discriminator has made a correct judgment, thereby obtaining the accuracy rate of the discriminator's authenticity judgment.

[0103] Step 1034: Determine whether the discriminator has converged based on the accuracy of the authenticity judgment.

[0104] If the discriminator converges, it means that the training of the generative adversarial network is completed, and step 104 is executed; if the discriminator does not converge, the initial generator and discriminator can continue to be trained based on the generative adversarial network. For example, after updating the model parameters in the initial generator, step 101 is continued.

[0105] Step 104: confirm that the initial generator training is completed and obtain the target generator.

[0106] That is, the current initial generator is used as the target generator.

[0107] The target generator is used to obtain generated data and determine a model training set of the ADAS model based on the generated data. The generated data may include generated driving road images, etc.

[0108] For example, the target generator may obtain generated data based on simulated driving data, where the simulated driving data is generated by a simulator, and then use the generated data as data in a model training set of an ADAS model.

[0109] An embodiment of the present application can train an initial generator in a generative adversarial network based on an existing ADAS dataset, so that the initial generator can learn the differences between the second driving data (i.e., simulated driving data) and the third driving data (i.e., real driving data) to obtain a target generator.

[0110] Then, the target generator can be used in conjunction with the simulator to generate a large amount of generated data that is close to the actual road environment information, and the model training set of the ADAS model can be obtained based on the generated data, which can greatly reduce the difficulty of optimizing the simulation parameters of the simulator and reduce the amount of sensor data collected in the vehicle, which is conducive to including driving data of various road conditions in the model training set.

[0111] The above embodiment is a method for training a target generator. The embodiment of the present application also provides a training set generation method. The training set generation method of the embodiment of the present application can use the target generator trained in the above embodiment to obtain generated data, and determine the model training set of the ADAS model based on the generated data.

[0112] It can be understood that the training set generation method in the embodiment of the present application (i.e., the step of using the target generator to obtain the model training set of the ADAS model) and the training set generation method in the above embodiment (i.e., the step of training to obtain the target generator) can be executed on the same electronic device or on different electronic devices, and the embodiment of the present application is not limited to this.

[0113] refer to Figure 4 As shown, the training set generation method of the embodiment of the present application includes:

[0114] Step 401: Generate simulated driving data based on a preset simulator.

[0115] The preset simulator and the simulator in step 1011 may be the same simulator.

[0116] In some embodiments, reference Figure 5 As shown, step 401 may include: obtaining driving scene parameters to be trained for the ADAS model; inputting the driving scene parameters to be trained into the simulator to obtain the simulated driving data.

[0117] The driving scenario parameters may include, but are not limited to, climate information, such as rainy, sunny, or snowy weather, light information, road curvature, surrounding obstacles, lane information, and other driving environment information. The driving scenario parameters to be trained can be configured based on training requirements and are not limited in this embodiment of the present application.

[0118] For further reference, Figure 5 As shown, the simulator can also obtain sensor configuration information in the vehicle, where the sensor configuration information is used to describe the sensors configured in the vehicle, and then obtain the simulated driving data based on the sensor configuration information and the driving scene parameters.

[0119] Specifically, the simulator can simulate the driving data collected by the target sensor in the driving scene parameters and use the driving data as the simulated driving data; wherein the target sensor is a sensor configured in the vehicle determined based on the sensor configuration information.

[0120] Step 402: Input the simulated driving data into the target generator to obtain generated data.

[0121] Among them, the target generator is a generator obtained based on generative adversarial network training.

[0122] In some embodiments, the generative adversarial network includes an initial generator and a discriminator, and the training steps of the target generator may include: generating first driving data based on the initial generator, and the label of the first driving data is forged data; inputting the first driving data into the discriminator of the generative adversarial network to obtain the authenticity judgment result of the first driving data; based on the label of the first driving data and the authenticity judgment result of the first driving data, determining whether the discriminator converges; if the discriminator converges, confirming that the training of the initial generator is completed, and obtaining the target generator.

[0123] Furthermore, in some embodiments, generating the first driving data based on the initial generator may include: generating second driving data based on the simulator; and inputting the second driving data into the initial generator to obtain the first driving data.

[0124] In some embodiments, determining whether the discriminator has converged based on the label of the first driving data and the authenticity judgment result of the first driving data may include: obtaining third driving data, the label of the third driving data being real data; inputting the third driving data into the discriminator to obtain the authenticity judgment result of the third driving data; determining the accuracy of the authenticity judgment based on the label of the first driving data, the authenticity judgment result of the first driving data, the label of the third driving data, and the authenticity judgment result of the third driving data; and determining whether the discriminator has converged based on the accuracy of the authenticity judgment.

[0125] The training method of the above target generator can refer to Figure 1 Implementation details of the embodiment corresponding to steps 101 to 104 shown in FIG. Figure 2 The implementation details of the embodiment corresponding to steps 1011 to 104 are not repeated here.

[0126] Step 403: Determine a model training set for the ADAS model based on the generated data.

[0127] In some embodiments, the electronic device may add the generated data to the model training set of the ADAS model as training data of the model training set.

[0128] The embodiment of the present application uses a target generator in combination with a simulator. The simulator can simulate the real environment and obtain simulated driving data. The generator obtained by training the generative adversarial network can learn the differences between the simulated driving data and the real driving data, so that the generated data obtained based on the simulated driving data can be close to the real driving data, thereby greatly reducing the optimization steps of the simulation parameters in the simulator.

[0129] Moreover, the model training set of the ADAS model is obtained based on the generated data, which makes it easy to obtain training data reflecting various road conditions while reducing the amount of real driving data collected, so as to improve the driving assistance performance of the ADAS model and improve the efficiency of obtaining the training set.

[0130] Based on the same idea as the training set generation method in the above embodiment, the present application also provides a training set device, which can be used to execute the above training set method.

[0131] For ease of explanation, the structural diagram of the training set generation device embodiment only shows the parts related to the embodiment of the present application. Those skilled in the art will understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than shown in the diagram, or a combination of certain components, or a different arrangement of components.

[0132] In some embodiments, the training set generation device may include multiple modules, each of which may be programmable software instructions stored in a memory and executable by a processor. It is understood that in other embodiments, the modules may also be program instructions or firmware embedded in the processor.

[0133] refer to Figure 6 As shown, Figure 6 This is a structural diagram of a training set generation device 600 provided in an embodiment of the present application. The training set generation device includes a simulation module 601, a generation module 602 and a production module 603.

[0134] The simulation module 601 is used to generate simulated driving data based on a preset simulator.

[0135] The generation module 602 is used to input the simulated driving data into a target generator to obtain generated data, wherein the target generator is a generator obtained by training a generative adversarial network.

[0136] The preparation module 603 is configured to determine a model training set of an advanced driver assistance system (ADAS) model based on the generated data.

[0137] In some embodiments, the simulation module 601 is further used to obtain driving scene parameters to be trained for the ADAS model; and input the driving scene parameters to be trained into the simulator to obtain the simulated driving data.

[0138] In some embodiments, the generative adversarial network includes an initial generator and a discriminator, and the training set generating device may further include a training device, the training device being used to generate first driving data based on the initial generator, wherein the label of the first driving data is forged data; the first driving data is input into the discriminator of the generative adversarial network to obtain a result of judging the authenticity of the first driving data; based on the label of the first driving data and the result of judging the authenticity of the first driving data, it is determined whether the discriminator has converged; if the discriminator has converged, it is confirmed that the training of the initial generator is completed, and the target generator is obtained.

[0139] In some embodiments, the training device is further used to generate second driving data based on the simulator; and input the second driving data into the initial generator to obtain the first driving data.

[0140] refer to Figure 7 As shown, Figure 7 This is a schematic diagram of the structure of a training set generation device 700 provided in another embodiment of the present application. The training set generation device 700 is used to train the target generator described above. The target generator is a generator trained based on a generative adversarial network, which includes an initial generator and a discriminator.

[0141] The training set generating device 700 includes a generating module 701 , a distinguishing module 702 and a training module 703 .

[0142] A generating module 701 is configured to generate first driving data based on the initial generator, wherein a label of the first driving data is forged data;

[0143] a determination module 702, configured to input the first driving data into a discriminator and obtain a result of determining the authenticity of the first driving data;

[0144] The training module 703 is used to determine whether the discriminator has converged based on the label of the first driving data and the authenticity judgment result of the first driving data; if the discriminator has converged, it is confirmed that the training of the initial generator is completed, and the target generator is obtained.

[0145] In some embodiments, the generating module 701 may also be configured to generate second driving data based on the simulator; and input the second driving data into the initial generator to obtain the first driving data.

[0146] In some embodiments, the discrimination module 702 is further used to obtain third driving data, the label of the third driving data is real data; the third driving data is input into the discriminator to obtain the authenticity judgment result of the third driving data; the training module 703 is further used to determine the accuracy of the authenticity judgment based on the label of the first driving data, the authenticity judgment result of the first driving data, the label of the third driving data, and the authenticity judgment result of the third driving data; based on the accuracy of the authenticity judgment, determine whether the discriminator converges.

[0147] Figure 8 This is a schematic diagram of an embodiment of an electronic device of the present application.

[0148] The electronic device 100 includes a memory 20, a processor 30, and a computer program 40 stored in the memory 20 and executable on the processor 30. When the processor 30 executes the computer program 40, the steps in the above-mentioned training set generation method embodiment are implemented, such as Figure 1 Steps 101 to 104 shown, Figure 2 1011 to step 104 shown, or Figure 4 Steps 401 to 403 are shown.

[0149] For example, the computer program 40 can also be divided into one or more modules / units, which are stored in the memory 20 and executed by the processor 30. The one or more modules / units can be a series of computer program instruction segments that can perform specific functions, and the instruction segments are used to describe the execution process of the computer program 40 in the electronic device 100. For example, it can be divided into Figure 6 The simulation module 601, the generation module 602 and the production module 603 shown, or Figure 7 The generation module 701, the discrimination module 702 and the training module 703 are shown.

[0150] Those skilled in the art will understand that the schematic diagram is merely an example of the electronic device 100 and does not constitute a limitation on the electronic device 100. The electronic device 100 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 100 may also include input and output devices, network access devices, buses, etc.

[0151] The processor 30 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, a single-chip microcomputer, or the processor 30 may be any conventional processor, etc.

[0152] The memory 20 can be used to store computer programs 40 and / or modules / units. The processor 30 implements various functions of the electronic device 100 by running or executing the computer programs and / or modules / units stored in the memory 20 and accessing data stored in the memory 20. The memory 20 may mainly include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area may store data (such as audio data) generated based on the use of the electronic device 100. In addition, the memory 20 may include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device.

[0153] If the modules / units integrated in the electronic device 100 are implemented in the form of software functional units and sold or used as independent products, they 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 when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. 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, USB flash drive, mobile hard disk, magnetic 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.

[0154] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the electronic device embodiments described above are merely illustrative. For example, the division of the units described is merely a logical function division, and other division methods may be used in actual implementation.

[0155] In addition, the functional units in the various embodiments of the present application may be integrated into the same processing unit, or each unit may exist physically separately, or two or more units may be integrated into the same unit. The above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0156] It is obvious to those skilled in the art that the present application is not limited to the details of the above-mentioned exemplary embodiments, and that the present application can be implemented in other specific forms without departing from the spirit or basic characteristics of the present application. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or electronic devices stated in the electronic device claim can also be implemented by the same unit or electronic device through software or hardware. Words such as first and second are used to indicate names and do not indicate any particular order.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not limiting. Although the present application has been described in detail with reference to the above embodiments, ordinary technicians in this field should understand that the technical solution of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present application.

Claims

1. A training set generation method, characterized in that: include: Generate simulated driving data based on a preset simulator; Inputting the simulated driving data into a target generator to obtain generated data; Wherein, the target generator is a generator obtained by training based on a generative adversarial network; Based on the generated data, a model training set of an advanced driver assistance system (ADAS) model is determined.

2. The training set generation method according to claim 1, wherein: The obtaining of simulated driving data based on a preset simulator includes: Obtaining driving scenario parameters to be trained for the ADAS model; The driving scene parameters to be trained are input into the simulator to obtain the simulated driving data.

3. The training set generation method according to claim 1 or 2, wherein: The generative adversarial network includes an initial generator and a discriminator, and the training steps of the target generator include: generating first driving data based on the initial generator, wherein a label of the first driving data is forged data; Inputting the first driving data into the discriminator of the generative adversarial network to obtain a result of determining the authenticity of the first driving data; determining whether the discriminator has converged based on a label of the first driving data and a result of determining the authenticity of the first driving data; If the discriminator converges, it is confirmed that the training of the initial generator is completed, and the target generator is obtained.

4. The training set generation method according to claim 3, wherein: The generating first driving data based on the initial generator includes: generating second driving data based on the simulator; The second driving data is input into the initial generator to obtain the first driving data.

5. A training set generation method, characterized in that: The training set generation method is used to train the target generator according to claim 1, wherein the target generator is a generator trained based on a generative adversarial network, and the generative adversarial network includes an initial generator and a discriminator; the training set generation method includes: generating first driving data based on the initial generator, wherein a label of the first driving data is forged data; inputting the first driving data into the discriminator to obtain a result of determining the authenticity of the first driving data; determining whether the discriminator has converged based on a label of the first driving data and a result of determining the authenticity of the first driving data; If the discriminator converges, it is confirmed that the training of the initial generator is completed, and the target generator is obtained.

6. The training set generation method according to claim 5, wherein: The generating first driving data based on the initial generator includes: generating second driving data based on the simulator; The second driving data is input into the initial generator to obtain the first driving data.

7. The training set generation method according to claim 5, wherein: The determining whether the discriminator has converged based on the label of the first driving data and the authenticity judgment result of the first driving data includes: Acquire third driving data, where a label of the third driving data is real data; inputting the third driving data into the discriminator to obtain a result of determining the authenticity of the third driving data; determining an accuracy rate of authenticity determination based on the label of the first driving data, the authenticity determination result of the first driving data, the label of the third driving data, and the authenticity determination result of the third driving data; Based on the accuracy of the authenticity judgment, it is determined whether the discriminator has converged.

8. A training set generation device, characterized in that: include: A simulation module, used for generating simulated driving data based on a preset simulator; a generating module, configured to input the simulated driving data into a target generator to obtain generated data; Wherein, the target generator is a generator obtained by training based on a generative adversarial network; A production module is used to determine a model training set of an advanced driver assistance system (ADAS) model based on the generated data.

9. An electronic device comprising a processor and a memory, characterized in that: The memory is used to store instructions, and the processor is used to call the instructions in the memory, so that the electronic device executes the training set generation method according to any one of claims 1 to 4, or executes the training set generation method according to any one of claims 5 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, which, when executed on an electronic device, cause the electronic device to execute the training set generation method according to any one of claims 1 to 4, or to execute the training set generation method according to any one of claims 5 to 7.

Images