Model training and map data processing method, device, equipment and storage medium

Abstract

The present disclosure provides a model training and map data processing method, device, equipment and storage medium, which relates to the computer field, and in particular to artificial intelligence fields such as autonomous driving, cloud computing, smart cockpits, and computer vision. The specific implementation scheme is: based on the original map data of multiple data sources and the data source information corresponding to the original map data, multiple first images containing map elements are generated; the original map data includes multiple map elements; based on the multiple first images, multiple second images for training the target model to be trained are generated; based on the multiple second images, the target model to be trained is trained to obtain a trained target model, and the target model is used to obtain the confidence of the input image based on the input image, and the confidence is used to indicate the correctness of the image. The embodiments of the present disclosure are helpful to improve the quality of map data.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to artificial intelligence fields such as autonomous driving, cloud computing, smart cockpits, and computer vision. Background Art

[0002] With the development of computer technology, the quality of maps and users' requirements for maps are getting higher and higher. At the same time, urban construction is developing rapidly. In order to ensure the accuracy of map data, map data needs to be updated in real time.

[0003] The original data used to update the map data comes from different sources, and the quality of the provided original map data is unstable. In order to ensure the quality of the updated map data, it is necessary to judge the quality of the original map data so as to control the quality of the map data. Summary of the invention

[0004] The present invention provides a model training and map data processing method, device, equipment and storage medium.

[0005] According to one aspect of the present disclosure, there is provided a method for generating a plurality of first images including map elements based on map original data from multiple data sources and data source information corresponding to the map original data; the map original data includes a plurality of map elements;

[0006] Generating a plurality of second images for training a target model to be trained according to the plurality of first images;

[0007] According to the plurality of second images, a target model to be trained is trained to obtain a trained target model, wherein the target model is used to obtain a confidence of the input image according to the input image, and the confidence is used to indicate the correctness of the image.

[0008] According to another aspect of the present disclosure, a map data processing method is provided, comprising:

[0009] generating a plurality of fifth images including map elements according to the original map data from multiple data sources;

[0010] The fifth image is processed using the target model to obtain the confidence of the fifth image; the target model is a trained target model provided by any embodiment of the present disclosure.

[0011] According to another aspect of the present disclosure, a model training device is provided, comprising:

[0012] A first image module, used to generate a plurality of first images containing map elements according to map original data from multiple data sources and data source information corresponding to the map original data; the map original data includes a plurality of map elements;

[0013] A second image module, used to generate a plurality of second images for training a target model to be trained according to the plurality of first images;

[0014] The training module is used to train a target model to be trained based on multiple second images to obtain a trained target model. The target model is used to obtain the confidence of the input image based on the input image, and the confidence is used to indicate the correctness of the image.

[0015] According to another aspect of the present disclosure, there is provided a map data processing device, comprising:

[0016] generating a plurality of fifth images including map elements according to the original map data from multiple data sources;

[0017] The fifth image is processed using the target model to obtain the confidence of the fifth image; the target model is a trained target model provided by any embodiment of the present disclosure.

[0018] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0019] at least one processor; and

[0020] a memory communicatively connected to the at least one processor; wherein,

[0021] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method in any embodiment of the present disclosure.

[0022] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided. The computer instructions are used to enable a computer to execute the method in any embodiment of the present disclosure.

[0023] According to another aspect of the present disclosure, a computer program product is provided, including a computer program / instruction, which implements the method in any embodiment of the present disclosure when the computer program / instruction is executed by a processor.

[0024] According to the technology disclosed in the present invention, a first image containing map elements is generated based on the original map data and the corresponding data source information, so that a target model can be trained based on the first image, so that the target model can judge the confidence of the image containing the map elements, which is conducive to screening out images that meet the confidence requirements to update the map data.

[0025] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0027] Figure 1 is a schematic diagram of a model training method according to an embodiment of the present disclosure;

[0028] Figure 2 is a schematic diagram of a model training method according to another embodiment of the present disclosure;

[0029] Figure 3 is a schematic diagram of a model training method according to another embodiment of the present disclosure;

[0030] Figure 4 is a schematic diagram of a model training method according to another embodiment of the present disclosure;

[0031] Figure 5 is a schematic diagram of a GAN framework according to an example of the present disclosure;

[0032] Figure 6 is a schematic diagram of a target model training process according to an example of the present disclosure;

[0033] Figure 7 is a schematic diagram of a model training device according to an embodiment of the present disclosure;

[0034] Figure 8 is a schematic diagram of a model training device according to another embodiment of the present disclosure;

[0035] Fig. 9 is a schematic diagram of a model training device according to another embodiment of the present disclosure;

[0036] Fig.10 is a schematic diagram of a model training device according to another embodiment of the present disclosure;

[0037] Fig.11 is a schematic diagram of a model training device according to another embodiment of the present disclosure;

[0038] Fig.12 It is a block diagram of an electronic device used to implement the model training method of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0039] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0040] The present disclosure provides a model training method. Figure 1 As shown, including:

[0041] Step S11: generating a plurality of first images including map elements according to map original data from multiple data sources and data source information corresponding to the map original data; the map original data includes a plurality of map elements;

[0042] Step S12: generating a plurality of second images for training a target model to be trained according to the plurality of first images;

[0043] Step S13: training a target model to be trained based on the plurality of second images to obtain a trained target model, wherein the target model is used to obtain the confidence of the input image based on the input image, and the confidence is used to indicate the probability that the image is correct.

[0044] In this embodiment, the original map data from multiple data sources may be original map data from multiple different data sources, for example, original map data collected by multiple different devices.

[0045] In a specific implementation, the data source may correspond to different map data collection devices, or the original map data sender. The map data collection device may include an ordinary vehicle, a special map data collection vehicle, a mobile device dedicated to map data collection, and the like.

[0046] The original map data may be data containing map elements, such as sensor data collected by ordinary vehicles, radar point cloud data collected by special map data collection vehicles, measurement data collected by special map data collection equipment, etc. In addition, the original map data may also be data in any format that can generate an electronic map.

[0047] In a possible implementation, map elements may include points, lines, and surfaces in map data. Map elements may be the basic content that constitutes a map. From a classification perspective, they may also include mathematical elements, geographical elements, and auxiliary elements.

[0048] In one possible implementation, map elements may also include elements specifically specified according to the purpose of map data updating. For example, map elements may be specified to include elements that may exist in a high-precision map and may affect vehicle driving, such as lane lines, traffic signs, median strips, other road markings, etc.

[0049] In another possible implementation, map elements may also include road-related elements that may affect the travel of autonomous vehicles or unmanned vehicles, such as lane lines, traffic signs, median strips, road arrows, other road markings, restricted traffic signs, control notices, accident cones, construction cones, etc.

[0050] The data source information corresponding to the original map data may include data source information corresponding to different parts of the original map data. When the original map data includes multiple map elements, the data source information may include data source information corresponding to each map element.

[0051] In a possible implementation, the data source information may include information related to the confidence or reliability of the original map data. For example, the data source information may include the device model corresponding to the data source, the signal condition when the data source collects the original map data, the road environment condition when the data source collects the original map data, the time when the data source collects the original map data, etc.

[0052] Generating a plurality of first images containing map elements according to the map original data from multiple data sources may be generating images with map elements corresponding to the map original data according to the map original data as the first images.

[0053] In a possible implementation, the original map data may include vector data of various map elements. The data source information may include information about the map data acquisition device collecting the original map data, such as the acquisition time, acquisition location, signal strength at the time of acquisition, weather at the time of acquisition, road environment at the time of acquisition, and traffic flow record data at the time of acquisition.

[0054] Generating a plurality of second images for training the model to be trained based on the plurality of first images may include determining a reference image based on the plurality of first images, and generating the second image based on the reference image, wherein the reference image may be an image in the first image whose confidence level is higher than a certain threshold.

[0055] In this embodiment, the target model may be a Generative Adversarial Networks (GAN) model.

[0056] In one possible implementation, a target model is trained based on multiple second images. When the target model is a discriminator of a generative adversarial network, the method may include simultaneously training a generator to be trained and a discriminator to be trained based on the multiple second images, so that the generator to be trained and the discriminator to be trained are synchronously optimized.

[0057] In this embodiment, the target model can also be a graph neural network model.

[0058] Due to objective factors such as equipment, weather, and environment when collecting raw map data, the raw map data may have certain errors compared to the actual situation. For example, in the case of occlusion, map elements in the actual road environment may be missing or broken; in the case of poor signal, failure of the map data collection device to focus, or poor adjustment of the map data collection device, the acquired raw map data may reflect incorrect map elements such as distortion, blur, and blur. Therefore, the disclosed embodiment uses confidence to determine the probability of an image being correct, that is, the probability that the image is consistent with the real environment.

[0059] In this embodiment, a first image containing map elements is generated based on the original map data and the corresponding data source information, so that a target model can be trained based on the first image, so that the target model can judge the confidence of the image containing the map elements, which is conducive to screening out images that meet the confidence requirements to update the map data.

[0060] In one embodiment, Figure 2 As shown, based on the multiple first images, multiple second images for training the model to be trained are generated, including:

[0061] Step S21: determining the confidence level of each first image according to the information of the data source;

[0062] Step S22: taking the first image whose confidence level is higher than a set threshold as the third image;

[0063] Step S23: obtaining a second image according to the third image.

[0064] In this embodiment, different first images may be first images for different areas.

[0065] Using the first image with a confidence level higher than a set threshold as the third image may include selecting the first image with a high confidence level as the second image.

[0066] In a possible implementation, the confidence level corresponding to the first image can be determined based on the data source information of the map element in the first image. The data source information may include the information flow index and the collection environment index when the original map data corresponding to the map element is collected. The information flow index, which may also be the communication flow index, can be reflected through positioning information, signal strength, etc. The collection environment index can be determined based on the traffic volume on the road obtained by each sensor and image acquisition device, and the existence of obstacles on the road. The aforementioned obstacles may be static elements on the road that block the map elements.

[0067] For example, for a piece of original map data, if the corresponding signal strength is very low and the GPS positioning is unclear, the confidence of the first image converted from the original map data is low.

[0068] Obtaining the second image based on the third image may include using the third image as a reference image or a standard image, mixing a non-reference image or a non-standard image therein, and using the obtained image set as the second image.

[0069] In another possible implementation, obtaining the second image based on the third image may include using the third image as the second image, so that when the second image is used to train the target model, the target model can learn information about images with high confidence, and thus learn to filter out images with high confidence.

[0070] In this embodiment, a second image for training a target model can be obtained based on multiple first images whose confidence levels in the first image are higher than a set threshold, so that the target model trained with the second image has the function of recognizing images with high confidence levels.

[0071] In one embodiment, obtaining the second image according to the third image includes:

[0072] Adding disturbance information to the third image to obtain a fourth image having a lower confidence level than the third image;

[0073] The fourth image and the third image are taken as the second image.

[0074] In this embodiment, the disturbance information may include information for reducing the confidence of the third image. Adding the disturbance information to the third image may include adding randomly generated disturbance information to the third image.

[0075] In this embodiment, the confidence level of the fourth image is lower than that of the third image.

[0076] In a specific implementation, the disturbance information may include a randomly generated layer. Alternatively, the disturbance information may include a content erasing instruction, a mosaic processing instruction, a distortion instruction, a blurring instruction, a smearing instruction, etc., so that after the disturbance information is added to the third image, a content erasing editing operation, a mosaic operation, a distortion operation, an image content blurring operation, an image content smearing operation, etc. will be performed on the area of ​​the third image to which the disturbance information is added, so that the fourth image has the disturbance information.

[0077] In this embodiment, the disturbance information may be at least one of a plurality of pre-set information. For example, the preset disturbance information may be a given icon content, a content erasing instruction, a mosaic processing instruction, a distortion instruction, a blurring instruction, a smearing instruction, etc. Before adding the disturbance information to the third image, the disturbance information may be generated in advance according to the set information.

[0078] In this embodiment, disturbance information is added to the third image with higher confidence to obtain a fourth image with lower confidence than the third image. The fourth image and the third image are used to train the target model, so that more training samples can be obtained using limited map original data, and the model can also learn how to distinguish between images with high confidence and images with low confidence.

[0079] In one embodiment, Figure 3 As shown, disturbance information is added to the third image to obtain a fourth image, including:

[0080] Step S31: determining an occluded area in the second image; the size information of the occluded area belongs to a preset size information range;

[0081] Step S32: determining the occlusion content in the occlusion area;

[0082] Step S33: When there are map elements in the occluded area, the occluded content is used as disturbance information to cover the occluded area in the second image to obtain a fourth image.

[0083] In this embodiment, the size information of the occluded area may include the pixel size of the occluded area and the proportion of the occluded area in the second image.

[0084] In this embodiment, different disturbance information may be generated for each second image, and different positions in the image may also be determined as positions that need to be blocked.

[0085] Determining the occluded area in the second image may include randomly selecting a specific size from a preset size information range as the size of the occluded area of ​​the second image that currently needs to be disturbed; selecting a random position in the second image to generate a closed curve; and adjusting the size of the area surrounded by the closed curve to obtain the occluded area.

[0086] In another possible implementation, the shape and size of the occlusion area can be preset to corresponding options or ranges. For example, the shape of the occlusion area can be preset to be a square, rectangle, triangle, circle, ellipse, body shape, and other known geometric shapes, or any irregular shape. Then, a shape can be randomly selected from known geometric shapes and irregular shapes such as square, rectangle, triangle, circle, ellipse, body shape, etc. as the shape of the occlusion area. A size can be randomly selected from a pre-set size range as the size of the occlusion area.

[0087] In another possible implementation, the occluded area and the occluded content in the occluded area may be determined in any order. The occluded area and the occluded content in the occluded area may also be determined at the same time. The occluded content may also be determined when there are map elements in the occluded area.

[0088] In another possible implementation, the occlusion area may be determined first, and when it is further determined that there is a map element in the occlusion area, the occlusion content in the occlusion area may be determined. Alternatively, the occlusion content may be determined first, and then the occlusion area containing the map element may be determined from the second image.

[0089] In this embodiment, the occlusion content can be selected from a pre-selected occlusion content repository. The occlusion content repository can be constructed according to the occlusion conditions that may exist in the actual road, such as obstacle occlusion, pedestrian occlusion, dust occlusion, rain and snow occlusion, sunlight reflection occlusion, etc. When determining the occlusion content, the pre-stored occlusion content is randomly selected from the occlusion content repository.

[0090] In this embodiment, when it is determined that there are map elements in the occluded area, disturbance information of the third image can be generated, so that a fourth image with low confidence can be obtained based on the third image with high confidence, making full use of limited original map data and image resources to obtain more target model training samples.

[0091] In one embodiment, when the map original data includes multiple map elements, generating multiple first images including map elements according to the map original data from multiple data sources and the data source information corresponding to the map original data includes:

[0092] In the case where there are repeated map elements among the multiple map elements, selecting a target map element from the repeated multiple map elements according to data source information corresponding to the map elements;

[0093] A first image is obtained according to the target map element.

[0094] In this embodiment, the repetition of map elements in vector format may include the case where the same map element is collected multiple times by using different map data collection devices and the original map data corresponding to the same map element is obtained multiple times. It may also include the case where the same map element is collected multiple times by using the same map element collection device and the original map data collected by the same map element collection device is obtained multiple times.

[0095] Selecting a target map element from a plurality of repeated map elements according to data source information corresponding to the map element may include selecting the target map element from the plurality of repeated map elements according to data source information corresponding to each of the plurality of repeated map elements.

[0096] Selecting a target map element from a plurality of repeated map elements according to data source information corresponding to the map element may also include, for each map element in the plurality of repeated map elements, determining a confidence level of the map element according to corresponding data source information; and selecting a target map element from the plurality of repeated map elements according to the confidence level of each map element in the plurality of repeated map elements.

[0097] In this embodiment, obtaining the first image according to the target map element may include determining the background content of the first image according to the existing electronic map data corresponding to the target map element, and obtaining the first image according to the background content and the target map element. Specifically, the target map element may be added to the background content.

[0098] In another possible implementation, a first image containing only map elements may be generated, so that the confidence judgment of the target model on the image is more targeted.

[0099] A first image is obtained according to the target map element, and a first image is obtained according to the target map element and other map elements at the same location.

[0100] In this embodiment, the first image can be obtained according to the target map element, so that the original map data can be converted into image data, so that the model can recognize and determine the confidence.

[0101] In one implementation, when the target map element is a map element in a vector format, obtaining a first image according to the target map element includes:

[0102] The target map elements are rasterized to obtain a first image.

[0103] In this embodiment, rasterizing the target map element to obtain the first image may include rasterizing the target map element and other map elements related to the target map element to obtain the first image.

[0104] Rasterizing target map features can convert target map features in other formats into raster format images.

[0105] In this embodiment, a rasterized image is obtained, thereby obtaining data in a format that can be recognized by the model.

[0106] In one implementation, training a target model to be trained according to the plurality of second images to obtain a trained target model includes:

[0107] Training the generative adversarial network model to be trained to obtain a trained generative adversarial network model; the generative adversarial network model includes a generator and a discriminator, the generator is used to generate a restored image according to the second image, and the discriminator is used to generate a confidence level corresponding to the second image according to the second image;

[0108] The discriminator of the trained generative adversarial network model is used as the target model.

[0109] The generative adversarial network model can be a multi-network adversarial generative model, which can be used in almost all scenarios where deep learning needs to be generated (especially in graphics generation tasks).

[0110] In this embodiment, the generator and the discriminator of the generative adversarial network model can be trained and optimized simultaneously. The generator can determine whether the second image needs to be repaired based on the input second image and output the repaired image of the second image if it needs to be repaired. If it is determined that the second image does not need to be repaired, the second image is output as the repaired image.

[0111] In another possible implementation, the input image of the generator of the generative adversarial network model may be an image with low confidence in the second image.

[0112] The discriminator of the generative adversarial network model can generate and output the corresponding confidence level based on the second image and the repaired image output by the generator.

[0113] In this embodiment, the discriminator of the generative adversarial network model is used as the target model to improve the effect of the target model in discriminating the confidence of the image.

[0114] The present disclosure also provides a method for processing map data. Figure 4 As shown, including:

[0115] Step S41: generating a plurality of fifth images including map elements according to the original map data from multiple data sources;

[0116] Step S42: Use the target model to process the fifth image to obtain the confidence of the fifth image; the target model is a trained target model provided by any embodiment of the present disclosure.

[0117] In this embodiment, generating multiple fifth images including map elements based on map original data from multiple data sources may include converting all map original data into fifth images including map elements, that is, the same map element may correspond to multiple fifth images.

[0118] In the disclosed embodiment, the fifth image generated from the original map data can be analyzed for confidence, which is beneficial for using a high-confidence image to update and generate the electronic map data when generating the electronic map data, thereby improving the quality of the electronic map data and reducing manual operations. The requirement for timely updating of the electronic map data can be met in an efficient and simple operation manner.

[0119] The method provided by the disclosed embodiment can be applied to confidence assessment of images converted from crowdsourced map data. Crowdsourced map data refers to real-time updated data of static road environment and dynamic information obtained through crowdsourcing, which is a high-precision map that meets the timeliness requirements of autonomous driving. Crowdsourcing collection has very different characteristics from traditional collection. Specifically: crowdsourced map data coverage is complex: for example, a road is only partially covered, and a road is covered multiple times; at the same time, crowdsourced map data errors are unknown: the data quality of different sources varies greatly, and the error model is unknown; in addition, crowdsourced map data observation conflicts: the same area has multiple sources of observations, and the observation values ​​may be different; therefore, map vendors are required to provide the confidence of elements based on the results of multi-source fusion for use by downstream modules.

[0120] In the example disclosed in this disclosure, the target model can be the discriminator of the GAN model. A discriminator network capable of calculating confidence is trained through the GAN framework. The overall framework is as follows: Figure 5 As shown in the figure, the G network is the generator network and the D network is the discriminator network.

[0121] Still refer to Figure 5 In the disclosed example, the model training method includes the following stages:

[0122] Rasterization stage. Because the input of the GAN network is an image, in this example, the crowdsourced map can be first rasterized (each raster has multiple attributes, such as color, color confidence, different locations of data sources, whether the positioning status is good, positioning confidence, lane line equation, lane line confidence, if none, it is assumed to be the default value) and processed into an image, hereinafter referred to as the crowdsourced map image.

[0123] Still refer to Figure 5 ,The model training method also includes a training phase of the target model.

[0124] like Figure 6 As shown, during the target model training phase, the following may be included:

[0125] Step S61: Data set construction.

[0126] In the above steps, crowdsourced map images with high confidence can be selected and added to the Figure 5 The crowdsourced map dataset shown.

[0127] Step S62: Add random perturbations. Perturb the original high-confidence crowdsourced map image to make it a low-confidence crowdsourced map image.

[0128] Step S63: training the generator network. In this example, the input data of the generator network is the low-confidence crowdsourced map image after random perturbation, and the output is the high-confidence crowdsourced map image restored by the network.

[0129] Step S64: train the discriminator network. The network input is the original high-confidence crowdsourced map image and the high-confidence crowdsourced map image restored by the generator network (i.e., the repaired image in the aforementioned embodiment), and the output is whether the image is real or fake. The discriminator in this example can also be a discriminator.

[0130] The loss value during the training phase is determined at least based on whether the discriminator classifies correctly.

[0131] During the training phase, the generator (G network) and the discriminator (D network) can be trained and optimized with each other. Based on the restored image output by the G network, the discrimination ability of the D network can be tested or trained. The loss value (loss) can be calculated through the confidence output by the D network. By back-propagating the loss value to the D network and the G network, the D network and the G network can be optimized respectively. By optimizing the GAN framework, the generator network and the discriminator network can be trained through the GAN framework to reach the Nash equilibrium point.

[0132] During the model deployment phase, the trained discriminator network can be deployed to the application environment, and the map elements corresponding to different areas of the crowdsourcing map (i.e., the original map data in the aforementioned embodiment) are traversed. After being rasterized and converted into images, they are sent to the discriminator network to output a number between [0, 1], i.e., the confidence level of the crowdsourcing map of the corresponding geographic area.

[0133] The target model obtained by the model training method provided by the embodiment of the present disclosure has stronger generalization ability. By making various random perturbations to the crowdsourced map images, negative samples are generated to enhance the generalization ability of the trained target model. In addition, the upgrade cost of the target model is low. The embodiment of the present disclosure adopts a data-driven approach. Whether it is upgrading the model or finding more training data, the pattern is relatively fixed and the cost is low. At the same time, the map data processing method of the embodiment of the present disclosure adopts a target model to process the original map data from multiple data sources, with low computational loss. In each area, the discriminator network only needs to be processed once, without the need for traversal operations.

[0134] The present disclosure also provides a model training device, such as Figure 7 As shown, including:

[0135] The first image module 71 is used to generate a plurality of first images containing map elements according to the map original data from multiple data sources and the data source information corresponding to the map original data; the map original data includes a plurality of map elements;

[0136] A second image module 72, used to generate a plurality of second images for training a target model to be trained according to the plurality of first images;

[0137] The training module 73 is used to train the target model to be trained based on the multiple second images to obtain the trained target model. The target model is used to obtain the confidence of the input image based on the input image, and the confidence is used to indicate the correctness of the image.

[0138] In one embodiment, Figure 8 As shown, the second image module includes:

[0139] A confidence unit 81, used to determine the confidence of each first image according to information of the data source;

[0140] A third image unit 82, configured to use the first image with a confidence level higher than a set threshold as the third image;

[0141] The third image processing unit 83 is configured to obtain a second image according to the third image.

[0142] In one implementation, the third image processing unit is further configured to:

[0143] Adding disturbance information to the third image to obtain a fourth image;

[0144] The fourth image and the third image are taken as the second image.

[0145] In one implementation, the third image processing unit is further configured to:

[0146] Determine an occluded area in the second image; the size information of the occluded area belongs to a preset size information range;

[0147] In the case where there are map elements in the random occlusion area, occlusion content is generated;

[0148] The occluded content is used as disturbance information to cover the occluded area in the second image to obtain a fourth image.

[0149] In one embodiment, Fig. 9 As shown, in the case where the map original data includes a plurality of map elements, the first image module includes:

[0150] A target map element selection unit 91 is used to select a target map element from the plurality of repeated map elements according to data source information corresponding to the map elements when there are repeated map elements in the plurality of map elements;

[0151] The target map element processing unit 92 is used to obtain the first image according to the target map element.

[0152] In one embodiment, when the target map element is a map element in a vector format, the target map element processing unit is further configured to:

[0153] The map elements in vector format are rasterized to obtain a first image.

[0154] In one embodiment, Fig.10 As shown, the training module includes:

[0155] A generative adversarial model training unit 101 is used to train a generative adversarial network model to be trained to obtain a trained generative adversarial network model; the generative adversarial network model includes a generator and a discriminator, the generator is used to generate a repaired image according to a first input image, and the discriminator is used to generate a confidence level corresponding to the second input image according to a second input image;

[0156] The generative adversarial model processing unit 102 is used to use the discriminator of the trained generative adversarial network model as the target model.

[0157] The present disclosure also provides a map data processing device, such as Fig.11 As shown, including:

[0158] The fifth image module 111 is used to generate a plurality of fifth images including map elements according to the original map data from multiple data sources;

[0159] The fifth image processing module 112 is used to process the fifth image using a target model to obtain the confidence of the fifth image; the target model is a trained target model provided by any embodiment of the present disclosure.

[0160] In the technical solution disclosed herein, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0161] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0162] Fig.12A schematic block diagram of an example electronic device 120 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0163] like Fig.12 As shown, the device 120 includes a computing unit 121, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 122 or a computer program loaded from a storage unit 128 into a random access memory (RAM) 123. In the RAM 123, various programs and data required for the operation of the device 120 can also be stored. The computing unit 121, the ROM 122, and the RAM 123 are connected to each other via a bus 124. An input / output (I / O) interface 125 is also connected to the bus 124.

[0164] A number of components in the device 120 are connected to the I / O interface 125, including: an input unit 126, such as a keyboard, a mouse, etc.; an output unit 127, such as various types of displays, speakers, etc.; a storage unit 128, such as a disk, an optical disk, etc.; and a communication unit 129, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 129 allows the device 120 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0165] The computing unit 121 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 121 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 121 performs the various methods and processes described above, such as the model training method. For example, in some embodiments, the model training method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 128. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 120 via ROM 122 and / or communication unit 129. When the computer program is loaded into RAM 123 and executed by the computing unit 121, one or more steps of the model training method described above may be performed. Alternatively, in other embodiments, the computing unit 121 may be configured to perform the model training method in any other appropriate manner (e.g., by means of firmware).

[0166] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0167] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0168] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0169] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0170] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0171] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0172] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0173] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A model training method, comprising: Generate a plurality of first images containing map elements according to map original data from multiple data sources and data source information corresponding to the map original data; The map original data includes a plurality of map elements; Generating a plurality of second images for training a target model to be trained according to the plurality of first images; According to the plurality of second images, a generative adversarial network model to be trained is trained to obtain a trained generative adversarial model; the generative adversarial network model includes a generator and a discriminator, the generator is used to generate a restored image according to the second image, and the discriminator is used to generate a confidence level corresponding to the second image according to the second image; The discriminator of the trained generative adversarial network model is used as the target model, and the target model is used to obtain the confidence of the input image according to the input image, and the confidence is used to represent the probability that the image is correct.

2. The method according to claim 1, wherein: The step of generating a plurality of second images for training the model to be trained according to the plurality of first images comprises: Determining the confidence level of each of the first images according to the information of the data source; The first image whose confidence level is higher than a set threshold is used as the third image; The second image is obtained according to the third image.

3. The method according to claim 2, wherein: The step of obtaining the second image according to the third image comprises: adding disturbance information to the third image to obtain a fourth image having a lower confidence level than the third image; The fourth image and the third image are used as the second image.

4. The method according to claim 3, wherein: The adding disturbance information to the third image to obtain a fourth image includes: Determine an occluded area in the second image; the size information of the occluded area belongs to a preset size information range; Determining the occlusion content in the occlusion area; In the case where there are map elements in the occluded area, the occluded content is used as the disturbance information and covered on the occluded area in the second image to obtain a fourth image.

5. The method according to any one of claims 1 to 4, wherein: In the case where the map original data includes a plurality of map elements, generating a plurality of first images including map elements according to the map original data from multiple data sources and the data source information corresponding to the map original data includes: In the case where there are repeated map elements among the plurality of map elements, selecting a target map element from the repeated plurality of map elements according to data source information corresponding to the map elements; The first image is obtained according to the target map element.

6. The method according to claim 5, wherein: In the case where the target map element is a map element in a vector format, obtaining the first image according to the target map element includes: The target map element is rasterized to obtain the first image.

7. A map data processing method, comprising: generating a plurality of fifth images including map elements according to the original map data from multiple data sources; The fifth image is processed using a target model to obtain the confidence of the fifth image; the target model is the trained target model described in any one of claims 1-6.

8. A model training device, comprising: A first image module, used to generate a plurality of first images containing map elements according to map original data from multiple data sources and data source information corresponding to the map original data; The map original data includes a plurality of map elements; A second image module, used to generate a plurality of second images for training a target model to be trained according to the plurality of first images; A training module, used for training the target model to be trained according to the plurality of second images to obtain a trained target model, wherein the target model is used for obtaining a confidence of the input image according to the input image, and the confidence is used for indicating a probability that the image is correct; Wherein, the training module includes: A generative adversarial model training unit, used to train a generative adversarial network model to be trained to obtain a trained generative adversarial network model; the generative adversarial network model includes a generator and a discriminator, the generator is used to generate a restored image according to the second image, and the discriminator is used to generate a confidence level corresponding to the second image according to the second image; A generative adversarial model processing unit is used to use the discriminator of the trained generative adversarial network model as the target model.

9. The device according to claim 8, wherein: The second image module comprises: a confidence unit, configured to determine the confidence of each of the first images according to the information of the data source; A third image unit, used to use the first image with a confidence level higher than a set threshold as the third image; The third image processing unit is used to obtain the second image according to the third image.

10. The device according to claim 9, wherein: The third image processing unit is further used for: adding disturbance information to the third image to obtain a fourth image having a lower confidence level than the third image; The fourth image and the third image are used as the second image.

11. The device according to claim 10, wherein: The third image processing unit is further used for: Determine an occluded area in the second image; the size information of the occluded area belongs to a preset size information range; Determining the occlusion content in the occlusion area; In the case where there are map elements in the occluded area, the occluded content is used as the disturbance information and covered on the occluded area in the second image to obtain a fourth image.

12. The device according to any one of claims 8 to 11, wherein: In the case where the map original data includes a plurality of map elements, the first image module includes: a target map element selection unit, configured to select a target map element from the plurality of repeated map elements according to data source information corresponding to the map elements when there are repeated map elements among the plurality of map elements; The target map element processing unit is used to obtain the first image according to the target map element.

13. The device according to claim 12, wherein: In the case where the target map element is a map element in a vector format, the target map element processing unit is further used for: The map elements in the vector format are rasterized to obtain the first image.

14. A map data processing device, comprising: A fifth image module, used to generate a plurality of fifth images containing map elements according to original map data from multiple data sources; A fifth image processing module is used to process the fifth image using a target model to obtain the confidence of the fifth image; the target model is the trained target model described in any one of claims 1-6.

15. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

16. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

17. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images