Vehicle loss assessment method, combined training method, device, equipment and medium

Abstract

The invention relates to the technical field of vehicle loss assessment, and provides a vehicle loss assessment method, a joint training method, a device, equipment and a storage medium, and the method comprises the steps: obtaining an accident image corresponding to a vehicle; based on an image segmentation large model, carrying out image segmentation on the accident image to obtain a scene image area of the accident image; based on a scene classification model, performing scene classification on the scene image area to obtain a preset scene type corresponding to the scene image area; based on a feature extraction model, performing multi-modal feature extraction on the scene image area to obtain a preset scene feature of the scene image area; based on a visual large model, according to the preset scene type and the preset scene features, carrying out refined scene classification on the scene image area to obtain a target scene type corresponding to the scene image area; and based on the vehicle damage detection model, determining vehicle damage information of the vehicle according to the accident image and the target scene type so as to improve the vehicle damage assessment accuracy.

Description

Technical Field

[0001] The present application relates to the field of vehicle damage assessment technology, and in particular to a vehicle damage assessment method, a joint training method, an apparatus, equipment and a medium. Background Art

[0002] Currently, vehicle damage assessments are required for vehicles involved in accidents. However, these assessments typically only consider the vehicle's physical damage. If a vehicle owner fabricates damage reports, this can lead to inaccurate damage assessments. Summary of the Invention

[0003] The main purpose of this application is to provide a vehicle damage assessment method, joint training method, device, equipment and medium, aiming to improve the accuracy of vehicle damage assessment.

[0004] In a first aspect, the present application provides a vehicle damage assessment method, comprising:

[0005] Obtaining the accident image corresponding to the vehicle;

[0006] Based on the image segmentation model, the accident image is segmented to obtain a scene image region of the accident image;

[0007] Based on the scene classification model, the scene image area is classified to obtain a preset scene type corresponding to the scene image area;

[0008] Based on the feature extraction model, multimodal feature extraction is performed on the scene image area to obtain preset scene features of the scene image area;

[0009] Based on the visual big model, according to the preset scene type and the preset scene features, the scene image area is refined scene classified to obtain the target scene type corresponding to the scene image area;

[0010] Based on a vehicle damage detection model, the vehicle damage information of the vehicle is determined according to the accident image and the target scene type.

[0011] In a second aspect, the present application also provides a joint training method for a vehicle detection model, comprising:

[0012] Acquire a training sample set, the training sample set including accident images corresponding to a plurality of vehicles and vehicle damage information labels corresponding to a plurality of vehicles;

[0013] Based on the image segmentation model, the accident image is segmented to obtain a scene image region of the accident image;

[0014] Based on the scene classification model, the scene image area is classified to obtain a preset scene type corresponding to the scene image area;

[0015] Based on the feature extraction model, multimodal feature extraction is performed on the scene image area to obtain preset scene features of the scene image area;

[0016] Based on the visual big model, according to the preset scene type and the preset scene features, the scene image area is refined scene classified to obtain the target scene type corresponding to the scene image area;

[0017] Determining vehicle damage information corresponding to the accident image based on the accident image and the target scene type based on a vehicle damage detection model;

[0018] According to the vehicle damage information and the vehicle damage information label, adjust the model parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model.

[0019] In a third aspect, the present application further provides a vehicle damage assessment device, the vehicle damage assessment device comprising:

[0020] An acquisition module, used to acquire an accident image corresponding to the vehicle;

[0021] An image segmentation module, configured to perform image segmentation on the accident image based on a large image segmentation model to obtain a scene image region of the accident image;

[0022] A first scene classification module is configured to perform scene classification on the scene image region based on a scene classification model to obtain a preset scene type corresponding to the scene image region;

[0023] A feature extraction module, configured to perform multimodal feature extraction on the scene image region based on a feature extraction model to obtain preset scene features of the scene image region;

[0024] A second scene classification module is configured to perform refined scene classification on the scene image area based on the visual macro model and according to the preset scene type and the preset scene features, to obtain a target scene type corresponding to the scene image area;

[0025] The vehicle damage detection module is used to determine the vehicle damage information corresponding to the accident image based on the vehicle damage detection model and the accident image and the target scene type.

[0026] In a fourth aspect, the present application further provides a computer device, comprising a memory and a processor;

[0027] The memory is used to store computer programs;

[0028] The processor is used to execute the computer program and implement the steps of the vehicle damage assessment method as described above, or implement the steps of the joint training method of the vehicle damage detection model as described above when executing the computer program.

[0029] In a fifth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the vehicle damage assessment method as described above are implemented, or the steps of the joint training method of the vehicle damage detection model as described above are implemented.

[0030] The present application provides a vehicle damage assessment method, joint training method, apparatus, device, and medium. The vehicle damage assessment method includes: obtaining an accident image corresponding to a vehicle; performing image segmentation on the accident image based on a large image segmentation model to obtain a scene image region of the accident image; performing scene classification on the scene image region based on a scene classification model to obtain a preset scene type corresponding to the scene image region; performing multimodal feature extraction on the scene image region based on a feature extraction model to obtain preset scene features of the scene image region; performing refined scene classification on the scene image region based on the preset scene type and preset scene features based on a large visual model to obtain a target scene type corresponding to the scene image region; and determining vehicle damage information based on the accident image and the target scene type based on a vehicle damage detection model to improve the accuracy of vehicle damage assessment. For example, this method can be used in the auto insurance business in the financial sector. For auto insurance businesses, when damage assessment is required for a vehicle in an accident scenario, an accident image corresponding to the vehicle can be obtained and the vehicle damage information can be determined using the vehicle damage assessment method. Accordingly, business personnel can promote auto insurance business based on the vehicle damage information, which is conducive to improving the efficiency of auto insurance business promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0032] Figure 1 This is a flow chart of a vehicle damage assessment method provided in an embodiment of the present application;

[0033] Figure 2 This is a flow chart of a vehicle damage assessment method according to an embodiment of the present application;

[0034] Figure 3This is a flow chart of a joint training method for a vehicle damage detection model provided in an embodiment of the present application;

[0035] Figure 4 This is a schematic block diagram of a vehicle damage assessment device provided in an embodiment of the present application;

[0036] Figure 5 This is a schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0038] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0039] The embodiments of the present application provide a vehicle damage assessment method, a joint training method, an apparatus, a device, and a storage medium. The vehicle damage assessment method and the joint training method of the vehicle damage detection model can be applied to a computer device, which can be a mobile phone, a tablet computer, a laptop computer, a desktop computer, or the like. The method can also be applied to a server, which can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0040] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments may be combined with each other.

[0041] See also Figure 1 , Figure 1This is a flow chart of a vehicle damage assessment method provided in an embodiment of the present application. It should be noted that the vehicle damage assessment method provided in an embodiment of the present application can be used on a computer device, and of course, can also be used on a server. For example, the server can obtain an accident image corresponding to the vehicle from the computer device, and process the accident image accordingly according to the vehicle damage assessment method to obtain vehicle damage information. For example, the server can send the vehicle damage information to the computer device; of course, this is not limited to this and is not limited here.

[0042] In specific implementation, computer devices include but are not limited to: any one of mobile phones, tablet computers, laptops, and desktop computers; the server can be a single server or a server cluster, or a cloud server that provides cloud computing services.

[0043] like Figure 1 As shown, the vehicle damage assessment method includes steps S101 to S106.

[0044] Step S101: Acquire an accident image corresponding to a vehicle.

[0045] When damage assessment is required for a vehicle involved in an accident, the computer device can acquire accident images corresponding to the vehicle. The accident images can be used for subsequent damage assessment.

[0046] For example, in the auto insurance business, when applying for an insurance claim, a car owner can take a photo of the vehicle to obtain the corresponding accident image of the vehicle. Of course, this is not limited to this. When handling an insurance claim request, the insurance agent can also take a photo of the car owner's vehicle to obtain the corresponding accident image of the vehicle. This is not limited here.

[0047] In some embodiments, the computer device may output a vehicle photography prompt to prompt at least one of the vehicle owner and the salesperson to photograph the vehicle and the accident scene in which the vehicle is located, thereby capturing an accident image corresponding to the vehicle. The accident image may include the vehicle image area and the scene image area excluding the vehicle image area. The vehicle photography prompt may include text prompts, image prompts, video prompts, animation prompts, and the like, without limitation.

[0048] For example, when an accident image corresponding to a vehicle is acquired, the computer device may perform image reshooting and recognition on the accident image.

[0049] In some embodiments, based on an image remake recognition model, it is determined whether the accident image is a remake image; when it is determined that the accident image is a remake image, the accident image is filtered.

[0050] For example, the computer device can input the accident image into the image copy recognition model. The image copy recognition model can be a pre-trained model with image copy recognition function. The image copy recognition model can determine whether the accident image is a copy image based on the acquired accident image. For example, the image copy recognition model can determine whether the accident image is a copy of the vehicle image on paper. For another example, the image copy recognition model can determine whether the accident image is a copy of the vehicle image displayed on an electronic screen. There is no limitation here.

[0051] Accordingly, the image remake recognition model can output a determination result on whether the accident image is a remake to the computer device. If the accident image is determined to be a remake, the computer device can determine that there is no need to continue with the vehicle damage assessment and can filter the accident image to remove it. If the accident image is determined not to be a remake, the computer device can determine that there is a need to continue with the vehicle damage assessment and can retain the accident image. For example, the computer device can subsequently input the accident image that is not a remake into the large image segmentation model for subsequent vehicle damage assessment.

[0052] In this way, when the accident image corresponding to the vehicle is obtained, by performing re-photographed image recognition on the accident image to filter out the accident images that are re-photographed images, the time for assessing the damage to the vehicle based on the accident images that are re-photographed images can be saved, which is conducive to improving the efficiency of vehicle damage assessment.

[0053] Step S102: Based on the image segmentation model, the accident image is segmented to obtain the scene image area of ​​the accident image.

[0054] For example, when assessing vehicle damage, a computer device may assess the impact of the accident scenario on the vehicle's damage information. For example, when assessing vehicle damage, the computer device may need to determine whether the accident scenario involved in the vehicle caused the vehicle's damage information. Based on this, the computer device may obtain the accident scenario involved in the vehicle.

[0055] When an accident image is acquired, the computer device may determine the scene image area included in the accident image based on the accident image, so as to acquire the accident scene where the vehicle is located.

[0056] For example, a computer device may input an accident image into an image segmentation model. The image segmentation model may be a pre-trained model with image segmentation capabilities. The image segmentation model may segment the acquired accident image to obtain a scene image region of the accident image.

[0057] For example, an accident image includes a vehicle image region and a scene image region. The image segmentation model can segment the accident image into the vehicle image region and the scene image region based on image segmentation. Accordingly, the image segmentation model can determine the scene image region, allowing the computer to determine the accident scene in which the vehicle was located, and further, the computer can assess the impact of the accident scene on the vehicle's damage information.

[0058] In this way, when the accident image is segmented based on the image segmentation model and the scene image area of ​​the accident image is obtained, the scene image area can be used to determine the accident scene in which the vehicle is located. The computer equipment can then integrate the accident scene in which the vehicle is located and evaluate the impact of the accident scene in which the vehicle is located on the vehicle damage assessment, which is conducive to subsequently improving the accuracy of the vehicle damage assessment.

[0059] In some embodiments, a graph convolutional network is used to perform correlation identification on each preset scene object in the accident image to obtain scene correlation information corresponding to the accident image; based on the scene correlation information, the accident image is segmented to obtain the scene image area of ​​the accident image.

[0060] For example, the large image segmentation model can be provided with a graph convolutional network. When the large image segmentation model obtains an accident image, it can use the graph convolutional network to perform correlation identification on each preset scene object in the accident image to obtain corresponding scene correlation information. For example, the preset scene objects in the accident image may include buildings, road lines, traffic lights, trees, signs, etc. The graph convolutional network can perform correlation identification on buildings with preset scene objects such as road lines, traffic lights, trees, and signs, and obtain first correlation information corresponding to the buildings. Correspondingly, the graph convolutional network can also perform correlation identification on road lines with preset scene objects such as buildings, traffic lights, trees, and signs, and obtain second correlation information corresponding to the road lines. Similarly, the graph convolutional network can integrate the correlation information corresponding to all preset scene objects to determine the scene correlation information corresponding to the accident image.

[0061] Once the scene association information is determined, the image segmentation model can segment the accident image based on the scene association information. For example, the image segmentation model can retain image regions associated with pre-set scene objects in the accident image and remove image regions not associated with the pre-set scene objects based on the scene association information, thereby obtaining the scene image region.

[0062] In this way, in the process of the image segmentation model determining the scene image area of ​​the accident image, the scene association information corresponding to the accident image is determined through the graph convolutional network. The image segmentation model can then obtain the scene image area from the accident image based on the scene association relationship, which is conducive to improving the convenience and accuracy of obtaining the scene image area.

[0063] Step S103: Based on the scene classification model, the scene image area is classified to obtain a preset scene type corresponding to the scene image area.

[0064] In some embodiments, the likelihood of a traffic accident occurring in different accident scenarios may be different, and the vehicle damage information for different accident scenarios may be different. For example, if the accident scenario is indoors, indoor scenarios include, for example, parking lots, residential areas, etc. Traffic accidents occurring in indoor scenarios may include vehicles hitting pedestrians, vehicles scraping against walls, etc., which are likely to cause damage to a smaller area of ​​the vehicle. If the accident scenario is outdoor, outdoor scenarios include, for example, highways, streets, etc. Traffic accidents occurring in outdoor scenarios may include vehicles hitting pedestrians, vehicles hitting other vehicles, vehicles hitting fixed objects such as trees, utility poles, buildings, etc., vehicles rolling over, etc., which are likely to cause damage to a larger area of ​​the vehicle. Based on this, when evaluating the impact of the accident scenario in which the vehicle is located on the vehicle's damage information, the computer device can determine the scenario type of the vehicle's accident scenario, so that the vehicle can be subsequently assessed for damage based on the scenario type.

[0065] For example, the computer device may input the acquired scene image region into a scene classification model. The scene classification model may be a pre-trained model with scene classification capabilities. The scene classification model may perform scene classification on the acquired scene image region based on the acquired scene image region to obtain a preset scene type corresponding to the scene image.

[0066] In an exemplary embodiment, the preset scene type may include either indoor or outdoor. For example, if the accident scene included in the scene image region is an indoor scene, the scene classification model determines that the preset scene type corresponding to the scene image region may be indoor. Accordingly, if the accident scene included in the scene image region is an outdoor scene, the scene classification model determines that the preset scene type corresponding to the scene image region is outdoor.

[0067] In this way, when a computer device can determine the preset scene type corresponding to a scene image region using a scene classification model, this facilitates improved convenience in determining the preset scene type corresponding to the scene image region. Accordingly, the computer device can subsequently assess vehicle damage based on the preset scene type, thereby assessing the impact of the preset scene type on the vehicle damage information, thereby facilitating improved accuracy in subsequent vehicle damage assessments.

[0068] Step S104: Based on the feature extraction model, multimodal feature extraction is performed on the scene image area to obtain preset scene features of the scene image area.

[0069] Exemplarily, the computer device may extract preset scene features of the scene image area so that the computer device can evaluate the impact of the accident scene in which the vehicle is located on the vehicle damage information.

[0070] For example, the computer device may input a scene image region into a feature extraction model. The feature extraction model may be a pre-trained model with multimodal feature extraction capabilities. The feature extraction model may perform multimodal feature extraction on the scene image region based on the acquired scene image region to obtain predetermined scene features for the scene image region.

[0071] In some embodiments, image feature extraction is performed on the scene image area to obtain scene image features of the scene image area; semantic feature extraction is performed on the scene image area to obtain scene semantic features of the scene image area; and the scene image features and the scene semantic features are fused to obtain preset scene features of the scene image area.

[0072] The multimodal feature extraction performed by the feature extraction model on the scene image region may include image feature extraction and semantic feature extraction.

[0073] For example, the scene image region may include scene image sub-regions corresponding to respective preset scene objects such as buildings, road lines, traffic lights, trees, and signboards. The feature extraction model may determine the scene image features of the scene image region based on the scene image sub-regions corresponding to respective preset scene objects in the scene image region. For example, the feature extraction model may perform image feature extraction on the scene image region to extract scene image sub-features of the scene image sub-regions corresponding to respective preset scene objects in the scene image region. In the case where the preset scene objects included in different scene image sub-regions may be different, the scene image sub-features of the different scene image sub-regions may also be different. The feature extraction model may then determine the scene image features of the scene image region by integrating the scene image sub-features of all scene image sub-regions.

[0074] For example, the scene image sub-region corresponding to the preset scene object included in the scene image region may contain text. For example, a speed limit value may be set on the road line, such as a maximum speed limit value, a minimum speed limit value, and so on. For another example, a location indication text may be set on a sign, such as the distance from a certain place or destination. The sign may also be provided with a speed limit value, service facility text, special area text, and so on, which are not limited here. The feature extraction model may determine the scene semantic features of the scene image region based on the scene image sub-region containing text. In an exemplary embodiment, the feature extraction model may perform semantic feature extraction on the scene image region to extract the scene semantic sub-features of the scene image sub-region containing text in the scene image region. In the case where the texts included in different scene image sub-regions may be different, the scene semantic sub-features of different scene image sub-regions may also be different. The feature extraction model may synthesize the scene semantic sub-features of all scene image sub-regions to determine the scene semantic features of the scene image region.

[0075] When determining scene image features and scene semantic features of a scene image region, since the scene image features and scene semantic features are features of different modalities, the feature extraction model can establish an association between the scene image features and the scene semantic features to determine preset scene features. The preset scene features can be used by the computer device to determine the accident scene in which the vehicle is located.

[0076] For example, the feature extraction model can fuse the scene image features and the scene semantic features to establish an association between the scene image features and the scene semantic features, thereby obtaining the preset scene features of the scene image area.

[0077] In this way, by performing multimodal feature extraction on the scene image area based on the feature extraction model, preset scene features of the scene image area are obtained. The preset scene features can be used by the computer device to determine the accident scene in which the vehicle is located, and then can be used by the computer device to evaluate the impact of the accident scene in which the vehicle is located on the vehicle damage information when subsequently determining the vehicle damage information, thereby facilitating the subsequent improvement of the accuracy of vehicle damage assessment.

[0078] In some embodiments, semantic features of a scene image region are extracted through a natural language network to obtain preset semantic features of the scene image region; based on the preset semantic features, scene semantic features associated with the preset semantic features are determined from a preset scene knowledge graph.

[0079] For example, the feature extraction model can be provided with a natural language network and a scene knowledge graph. The natural language network can have the function of processing and understanding human language. For example, the natural language network can include a natural language processing (NLP) architecture. The scene knowledge graph is equivalent to a scene semantic network. The scene knowledge graph can be used to enrich semantic context information. The scene knowledge graph can identify the scene entities corresponding to the preset semantic features based on the preset semantic features extracted by the natural language network, and determine the scene semantic features associated with the preset semantic features based on the scene entity.

[0080] Taking the example of a scene image area including a traffic light, the feature extraction model can extract semantic features of the traffic light through a natural language network to obtain preset semantic features corresponding to the traffic light. When the feature extraction model determines the preset semantic features corresponding to the traffic light, it inputs the preset semantic features corresponding to the traffic light into the preset scene knowledge graph. The scene knowledge graph can then use the preset semantic features to determine that the scene entity corresponding to the preset semantic features is a traffic light. Accordingly, the scene knowledge network can determine the scene semantic features associated with the traffic light based on the scene entity being a traffic light, thereby determining the scene semantic features of the scene image area. Of course, the scene entities in the scene knowledge graph are not limited to traffic lights. For example, they also include streets, pedestrians, trees, vehicles, etc., which are not limited here.

[0081] In this way, the feature extraction model can integrate the natural language network and the scene knowledge graph to determine the scene semantic features of the scene image area, which is conducive to improving the convenience and accuracy of determining the scene semantic features of the scene image area.

[0082] In some embodiments, the scene image features and the scene semantic features are fused through an attention mechanism layer to obtain preset scene features.

[0083] For example, the feature extraction model can be provided with an attention mechanism layer. During the process of the feature extraction model determining the scene image features and scene semantic features of a scene image region and fusing the scene image features and scene semantic features, the feature extraction model can focus on the scene semantic features related to the image content indicated by the scene image features through the attention mechanism layer to determine the preset scene features.

[0084] In this way, the feature extraction model can integrate the attention mechanism layer to the preset scene features, which is conducive to improving the convenience of integrating scene image features and scene semantic features, thereby helping to improve the accuracy of determining the preset scene features of the scene image area.

[0085] Step S105 : Based on the visual macro model, the scene image area is refined into scene classification according to the preset scene type and preset scene features to obtain the target scene type corresponding to the scene image area.

[0086] In some embodiments, the preset scene type may include either indoor or outdoor. If the impact of the accident scene on the vehicle's damage information is evaluated based solely on whether the accident scene in which the vehicle is located is either an indoor scene or an outdoor scene, it is easy to fail to effectively utilize the accident scene and scene features related to the accident scene to evaluate the impact of the accident scene on the vehicle's damage information. Based on this, when determining the preset scene type and the preset scene features, the computer device can use the preset scene type and the preset scene features to refine the scene classification of the scene image area to obtain the target scene type corresponding to the scene image area. The target scene type is equivalent to a refined type of the preset scene type. By refining the preset scene type, the accuracy of the computer device's subsequent evaluation of the impact of the accident scene in which the vehicle is located on the vehicle's damage information based on the target scene type can be improved, thereby helping to improve the accuracy of vehicle damage assessment.

[0087] For example, when determining the preset scene type and the preset scene features, the computer device can input the preset scene type and the preset scene features into the visual big model, and the visual big model can then perform a refined scene classification on the scene image area based on the preset scene type and the preset scene features to obtain the target scene type of the scene image area. The target scene type may include at least one of the scene types such as road type, environment type, and weather type. For example, the road type may include any one of highways, provincial roads, and the like. The environment type may include any one of urban, rural, mountainous, and other environments. The weather type may include at least one of sunny, rainy, foggy, and other weather types.

[0088] In this way, by integrating preset scene types and features based on the visual macro model, the target scene type corresponding to the scene image area is determined, which helps improve the convenience and accuracy of determining the target scene type. The target scene type can be used by the computer to subsequently assess the impact of the accident scene on the vehicle's damage information, and further determine the vehicle's damage information, which helps improve the accuracy of subsequent vehicle damage assessments.

[0089] Step S106: Based on the vehicle damage detection model, the vehicle damage information is determined according to the accident image and the target scene type.

[0090] In some embodiments, the computer device may input the acquired accident image and target scene type into a vehicle damage detection model, so that the vehicle damage detection model determines the vehicle damage information based on the accident image and the target scene type.

[0091] For example, the vehicle damage detection model can determine the type of vehicle damage based on the vehicle image area in the accident image. For example, the vehicle damage type can include high-level scratches, low-level scratches, etc., without limitation.

[0092] Accordingly, the damage detection model can assess the likelihood of an accident scenario causing the corresponding type of damage to the vehicle based on the target scene type. For example, consider an open field as the target scene type and a high-level scratch as the vehicle's damage type. Given a low probability of high-level scratches occurring in an open field, the damage detection model can determine that the damage type is highly likely to be fabricated, thereby determining the vehicle's damage information. For example, the damage detection model can determine the vehicle's damage information based on a combination of the vehicle's damage type and the likelihood of the accident scenario causing that type of damage. Once the computer device has acquired the vehicle's damage information, it can complete the damage assessment.

[0093] For auto insurance business in the financial field, when handling auto insurance claims, business personnel can judge the possibility of the car owner falsifying auto insurance based on the type of damage indicated by the vehicle damage information and the possibility of the accident scene causing that type of damage, thereby improving the accuracy of vehicle damage assessment.

[0094] In an exemplary embodiment, Figure 2 As shown, when a computer device acquires an accident image corresponding to a vehicle, it can determine whether the accident image is a reprinted image. If it is determined that the accident image is a reprinted image, the reprinted image is filtered. If it is determined that the accident image is not a reprinted image, that is, a real image, the real image is retained. The computer device can input the real image into a large image segmentation model so that the large image segmentation model determines the scene image region of the real image. When the computer device acquires the scene image region of the real image, it can preliminarily determine a preset scene type for the scene image region. Accordingly, the computer device can perform image feature extraction on the scene image region to obtain scene image features, perform semantic feature extraction on the scene image region to obtain scene semantic features, and fuse the scene image features and scene semantic features to obtain preset scene features. Based on this, the computer device can refine the preset scene type by combining the preset scene type and preset scene features to obtain a target scene type. The target scene type can then be used by the computer device to determine vehicle damage information based on the accident image.

[0095] The vehicle damage assessment method provided in the above embodiment obtains an accident image corresponding to a vehicle; performs image segmentation on the accident image based on a large image segmentation model to obtain a scene image region of the accident image; performs scene classification on the scene image region based on a scene classification model to obtain a preset scene type corresponding to the scene image region; performs multimodal feature extraction on the scene image region based on a feature extraction model to obtain preset scene features of the scene image region; performs refined scene classification on the scene image region based on the preset scene type and preset scene features based on the large visual model to obtain a target scene type corresponding to the scene image region; and determines vehicle damage information based on the accident image and the target scene type based on a vehicle damage detection model to improve the accuracy of vehicle damage assessment. For example, this method can be used in the automotive insurance business in the financial sector. For automotive insurance businesses, when damage assessment of a vehicle in an accident scene is required, an accident image corresponding to the vehicle can be obtained and the vehicle damage information can be determined using the vehicle damage assessment method. Accordingly, business personnel can promote automotive insurance business based on vehicle damage information, which helps improve the efficiency of automotive insurance business.

[0096] See also Figure 3 , Figure 3 It is a flow chart of a joint training method for a vehicle damage detection model provided in an embodiment of the present application. It should be noted that the joint training method for a vehicle damage detection model provided in an embodiment of the present application can be used in a computer device, and of course can also be used in a server. For example, the server can obtain a training sample set from a computer device, and process the training sample set according to the joint training method for the vehicle damage detection model to adjust the model parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model, and the vehicle damage detection model, thereby obtaining an optimized image segmentation model, an optimized scene classification model, an optimized feature extraction model, an optimized visual model, and an optimized vehicle damage detection model. For example, the server can send an optimized image segmentation model, an optimized scene classification model, an optimized feature extraction model, an optimized visual model, and an optimized vehicle damage detection model to a computer device, so that when the computer device obtains an accident image corresponding to the vehicle, it can determine the vehicle damage information through the optimized image segmentation model, the optimized scene classification model, the optimized feature extraction model, the optimized visual model, and the optimized vehicle damage detection model; of course, it is not limited to this and is not restricted here.

[0097] In specific implementation, computer devices include but are not limited to: any one of mobile phones, tablet computers, laptops, and desktop computers; the server can be a single server or a server cluster, or a cloud server that provides cloud computing services.

[0098] like Figure 3 As shown, the joint training method of the vehicle damage detection model includes steps S201 to S207.

[0099] Step S201: Obtain a training sample set, where the training sample set includes accident images corresponding to a plurality of vehicles and vehicle damage information labels corresponding to a plurality of vehicles.

[0100] For example, when jointly training a large image segmentation model, a scene classification model, a feature extraction model, a large visual model, and a vehicle damage detection model, a training sample set can be obtained. The training sample set can include accident images corresponding to multiple vehicles. Each vehicle in the training sample set can be provided with a corresponding vehicle damage information label. The training sample set can be used for subsequent joint training of the large image segmentation model, the scene classification model, the feature extraction model, the large visual model, and the vehicle damage detection model. Specifically, before jointly training the large image segmentation model, the scene classification model, the feature extraction model, the large visual model, and the vehicle damage detection model, the computer device can separately pre-train the large image segmentation model, the scene classification model, the feature extraction model, the large visual model, and the vehicle damage detection model, and then integrate the pre-trained large image segmentation model, the pre-trained scene classification model, the pre-trained feature extraction model, the pre-trained large visual model, and the pre-trained vehicle damage detection model for joint training.

[0101] In some embodiments, when an accident image corresponding to a vehicle is obtained, the vehicle damage information label corresponding to the vehicle can be determined by a preset labeling method, which is not limited here.

[0102] In this way, when the accident image corresponding to the vehicle and the vehicle damage information label corresponding to the vehicle are obtained, the image segmentation model, scene classification model, feature extraction model, visual model and vehicle damage detection model can be jointly trained subsequently.

[0103] Step S202: Based on the image segmentation model, the accident image is segmented to obtain the scene image area of ​​the accident image.

[0104] For example, when jointly training the image segmentation model, scene classification model, feature extraction model, visual model, and vehicle damage detection model, it is necessary to learn how to determine the scene image region of the accident image. Based on this, the computer device can input the accident image into the image segmentation model, which then performs image segmentation on the accident image to determine the scene image region of the accident image.

[0105] In this way, when the scene image area of ​​the accident image is determined, the computer device can optimize the model parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model based on the scene image area of ​​the accident image to improve the training effect of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model.

[0106] Step S203: Based on the scene classification model, the scene image area is classified to obtain a preset scene type corresponding to the scene image area.

[0107] For example, when jointly training a large image segmentation model, a scene classification model, a feature extraction model, a large visual model, and a vehicle damage detection model, it is necessary to learn how to determine the preset scene type corresponding to a scene image region. Based on this, the scene image region can be input into the scene classification model, which then classifies the scene image region and determines the preset scene type corresponding to the scene image region.

[0108] In this way, when determining the preset scene type corresponding to the scene image area, the computer device can optimize the model parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model based on the preset scene type corresponding to the scene image area, so as to improve the training effect of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model.

[0109] Step S204: Based on the feature extraction model, multimodal feature extraction is performed on the scene image area to obtain preset scene features of the scene image area.

[0110] For example, when jointly training a large image segmentation model, a scene classification model, a feature extraction model, a large visual model, and a vehicle damage detection model, it is necessary to learn how to determine the preset scene features of a scene image region. Based on this, the computer device can input the scene image region into the feature extraction model, which then performs multimodal feature extraction on the scene image region to obtain the preset scene features of the scene image region.

[0111] In this way, when determining the preset scene features of the scene image area, the computer device can optimize the model parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model based on the preset scene features to improve the training effect of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model.

[0112] Step S205 : Based on the visual macro model, the scene image area is refined into scene classification according to the preset scene type and preset scene features to obtain the target scene type corresponding to the scene image area.

[0113] For example, when jointly training an image segmentation model, a scene classification model, a feature extraction model, a visual model, and a vehicle damage detection model, it is necessary to learn how to determine the target scene type corresponding to a scene image region. Based on this, the computer device can input preset scene types and preset scene features into the visual model, allowing the visual model to perform detailed scene classification on the scene image region and determine the target scene type corresponding to the scene image region.

[0114] In this way, when determining the target scene type corresponding to the scene image area, the computer device can optimize the model parameters of at least one of the image segmentation model, scene classification model, feature extraction model, visual model and vehicle damage detection model based on the target scene type to improve the training effect of the image segmentation model, scene classification model, feature extraction model, visual model and vehicle damage detection model.

[0115] Step S206: Based on the vehicle damage detection model, the vehicle damage information is determined according to the accident image and the target scene type.

[0116] For example, when jointly training a large image segmentation model, scene classification model, feature extraction model, large visual model, and vehicle damage detection model, it is necessary to learn how to determine vehicle damage information. Based on this, the computer device can input the accident image and target scene type into the vehicle damage detection model, which then determines the vehicle damage and obtains the vehicle damage information.

[0117] In this way, when the vehicle damage information is determined, the computer device can optimize the model parameters of at least one of the image segmentation model, scene classification model, feature extraction model, visual model and vehicle damage detection model based on the vehicle damage information to improve the training effect of the image segmentation model, scene classification model, feature extraction model, visual model and vehicle damage detection model.

[0118] Step S207: Adjust the model parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model, and the vehicle damage detection model according to the vehicle damage information and the vehicle damage information label.

[0119] For example, the computer device can determine the model optimization parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model based on the vehicle damage information and the vehicle damage information label.

[0120] In some embodiments, the vehicle damage information obtained after processing the accident image using the large image segmentation model, scene classification model, feature extraction model, large visual model, and vehicle damage detection model is equivalent to the predicted vehicle damage result. The vehicle damage information label corresponding to the vehicle is equivalent to the actual vehicle damage result. Based on the vehicle damage information and the damage information label, a computer device can determine the loss value between the actual vehicle damage result and the predicted vehicle damage result when determining the vehicle damage information using the large image segmentation model, scene classification model, feature extraction model, large visual model, and vehicle damage detection model.

[0121] Accordingly, the computer device can determine model optimization parameters for at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model, and the vehicle damage detection model based on the loss value. The model optimization parameters can be used by the computer device to subsequently adjust the model parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model, and the vehicle damage detection model, so that the optimized image segmentation model, the scene classification model, the feature extraction model, the visual model, and the vehicle damage detection model can more accurately align the predicted vehicle damage results with the actual vehicle damage results, thereby achieving the purpose of training.

[0122] By adjusting the model parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model according to the vehicle damage information and the vehicle damage information label, the optimized image segmentation model, the optimized scene classification model, the optimized feature extraction model, the optimized visual model and the optimized vehicle damage detection model can more accurately bring the predicted vehicle damage results of the vehicle damage information closer to the actual vehicle damage results, so as to achieve the purpose of training, thereby helping to improve the accuracy of computer equipment in determining the vehicle damage information.

[0123] The joint training method for the vehicle damage detection model provided in the above embodiment obtains a training sample set, wherein the training sample set includes accident images corresponding to multiple vehicles and vehicle damage information labels corresponding to multiple vehicles; based on the image segmentation large model, performs image segmentation on the accident image to obtain a scene image region of the accident image; based on the scene classification model, performs scene classification on the scene image region to obtain a preset scene type corresponding to the scene image region; based on the feature extraction model, performs multimodal feature extraction on the scene image region to obtain a preset scene feature of the scene image region; based on the visual large model, performs refined scene classification on the scene image region according to the preset scene type and the preset scene feature to obtain a target scene type corresponding to the scene image region; based on the vehicle damage detection model, determines vehicle damage information according to the accident image and the target scene type; and based on the vehicle damage detection model, adjusts model parameters of at least one of the image segmentation large model, the scene classification model, the feature extraction model, the visual large model, and the vehicle damage detection model according to the vehicle damage information and the vehicle damage information label to improve the accuracy of determining the vehicle damage information, thereby improving the joint training effect of the image segmentation large model, the scene classification model, the feature extraction model, the visual large model, and the vehicle damage detection model. Subsequently, large image segmentation models, scene classification models, feature extraction models, large visual models, and vehicle damage detection models can be applied to vehicle damage assessment, which will help improve the accuracy of vehicle damage assessments. For example, this can be used in the auto insurance business in the financial sector. For accident images corresponding to vehicles requiring damage assessment, the trained large image segmentation model, trained scene classification model, trained feature extraction model, trained large visual model, and trained vehicle damage detection model can be used to detect and determine vehicle damage information, which will help improve the accuracy of vehicle damage assessments. Furthermore, determining vehicle damage information without consuming extensive manpower and time will help accelerate the advancement of auto insurance business. Accordingly, the impact of the accident scene on the vehicle's damage information can be assessed, which will help improve the accuracy of vehicle damage assessments and, therefore, the effectiveness of auto insurance business.

[0124] See also Figure 4 , Figure 4 This is a schematic block diagram of a vehicle damage assessment device provided in an embodiment of the present application. The vehicle damage assessment device can be configured in a server or electronic device to execute the aforementioned vehicle damage assessment method.

[0125] like Figure 4 As shown, the vehicle damage assessment device includes: an acquisition module 110, an image segmentation module 120, a first scene classification module 130, a feature extraction module 140, a second scene classification module 150 and a vehicle damage detection module 160.

[0126] The acquisition module 110 is used to acquire accident images.

[0127] The image segmentation module 120 is used to perform image segmentation on the accident image based on the image segmentation model to obtain the scene image area of ​​the accident image.

[0128] The first scene classification module 130 is configured to perform scene classification on the scene image region based on a scene classification model to obtain a preset scene type corresponding to the scene image region.

[0129] The feature extraction module 140 is configured to perform multimodal feature extraction on the scene image region based on a feature extraction model to obtain preset scene features of the scene image region.

[0130] The second scene classification module 150 is used to perform refined scene classification on the scene image area based on the visual macro model and according to the preset scene type and the preset scene features, so as to obtain the target scene type corresponding to the scene image area.

[0131] The vehicle damage detection module 160 is configured to determine the vehicle damage information corresponding to the accident image based on the accident image and the target scene type based on a vehicle damage detection model.

[0132] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0133] The method of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0134] Illustratively, the above-mentioned method and apparatus may be implemented in the form of a computer program, which may be run on a computer device.

[0135] See also Figure 5 , Figure 5 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server or an electronic device.

[0136] like Figure 5 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a storage medium and an internal memory.

[0137] The storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, may cause the processor to execute any vehicle damage assessment method or any joint training method for a vehicle damage detection model.

[0138] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0139] The internal memory provides an environment for the operation of the computer program in the storage medium. When the computer program is executed by the processor, the processor can execute any vehicle damage assessment method, or can execute any joint training method of the vehicle damage detection model.

[0140] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0141] It should be understood that the processor may be a central processing unit (CPU), 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. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0142] In one embodiment, the processor is configured to execute a computer program and implement the following steps when executing the computer program:

[0143] Obtaining the accident image corresponding to the vehicle;

[0144] Based on the image segmentation model, the accident image is segmented to obtain a scene image region of the accident image;

[0145] Based on the scene classification model, the scene image area is classified to obtain a preset scene type corresponding to the scene image area;

[0146] Based on the feature extraction model, multimodal feature extraction is performed on the scene image area to obtain preset scene features of the scene image area;

[0147] Based on the visual big model, according to the preset scene type and the preset scene features, the scene image area is refined scene classified to obtain the target scene type corresponding to the scene image area;

[0148] Based on a vehicle damage detection model, the vehicle damage information of the vehicle is determined according to the accident image and the target scene type.

[0149] It should be noted that, those skilled in the art will clearly understand that, for the sake of convenience and brevity of description, the above-described specific working process of vehicle damage detection may refer to the corresponding process in the aforementioned vehicle damage assessment method embodiment, and will not be repeated here.

[0150] In another embodiment, the processor is configured to execute a computer program and implement the following steps when executing the computer program:

[0151] Acquire a training sample set, the training sample set including accident images corresponding to a plurality of vehicles and vehicle damage information labels corresponding to a plurality of vehicles;

[0152] Based on the image segmentation model, the accident image is segmented to obtain a scene image region of the accident image;

[0153] Based on the scene classification model, the scene image area is classified to obtain a preset scene type corresponding to the scene image area;

[0154] Based on the feature extraction model, multimodal feature extraction is performed on the scene image area to obtain preset scene features of the scene image area;

[0155] Based on the visual big model, according to the preset scene type and the preset scene features, the scene image area is refined scene classified to obtain the target scene type corresponding to the scene image area;

[0156] Determining vehicle damage information of the vehicle based on the accident image and the target scene type based on a vehicle damage detection model;

[0157] According to the vehicle damage information and the vehicle damage information label, adjust the model parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model.

[0158] It should be noted that, those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the joint training of the vehicle damage detection model described above can refer to the corresponding process in the aforementioned embodiment of the joint training method of the vehicle damage detection model, and will not be repeated here.

[0159] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. The method implemented when the computer program is executed by a processor may refer to the various embodiments of the vehicle damage assessment method of the present application, or may refer to the various embodiments of the joint training method of the vehicle damage detection model of the present application.

[0160] The computer-readable storage medium may be an internal storage unit of the computer device described in the aforementioned embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a flash memory card, etc., equipped on the computer device.

[0161] It should be understood that the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0162] It should be understood that the non-Company's software tools or components that appear in the embodiments of this application are merely examples and do not represent actual use.

[0163] It should also be understood that the term "and / or" used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements that are inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system that includes the element.

[0164] The serial numbers of the embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments. The above description is only a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A vehicle damage assessment method, characterized in that: include: Obtain an accident image corresponding to the vehicle; Based on the image segmentation model, the accident image is segmented to obtain a scene image region of the accident image; Based on the scene classification model, the scene image area is classified to obtain a preset scene type corresponding to the scene image area; Based on the feature extraction model, multimodal feature extraction is performed on the scene image region to obtain preset scene features of the scene image region; Based on the visual big model, according to the preset scene type and the preset scene features, the scene image area is refined for scene classification to obtain a target scene type corresponding to the scene image area; Based on the vehicle damage detection model, the vehicle damage information of the vehicle is determined according to the accident image and the target scene type.

2. The vehicle damage assessment method according to claim 1, characterized in that: The image segmentation based on the large image segmentation model is performed on the accident image to obtain the scene image area of ​​the accident image, including: By using a graph convolutional network, the association of each preset scene object in the accident image is identified to obtain scene association information corresponding to the accident image; The accident image is segmented according to the scene association information to obtain a scene image region of the accident image.

3. The vehicle damage assessment method according to claim 1, characterized in that: The step of performing multimodal feature extraction on the scene image region based on the feature extraction model to obtain preset scene features of the scene image region includes: Performing image feature extraction on the scene image region to obtain scene image features of the scene image region; Extracting semantic features from the scene image region to obtain scene semantic features of the scene image region; The scene image features and the scene semantic features are fused to obtain preset scene features of the scene image area.

4. The vehicle damage assessment method according to claim 3, characterized in that: The extracting semantic features of the scene image region to obtain scene semantic features of the scene image region includes: Extracting semantic features of the scene image region through a natural language network to obtain preset semantic features of the scene image region; According to the preset semantic feature, a scene semantic feature associated with the preset semantic feature is determined from a preset scene knowledge graph.

5. The vehicle damage assessment method according to claim 3, characterized in that: The fusing process of the scene image feature and the scene semantic feature to obtain the preset scene feature of the scene image area includes: Through the attention mechanism layer, the scene image features and the scene semantic features are fused to obtain the preset scene features.

6. The vehicle damage assessment method according to any one of claims 1 to 5, characterized in that: Before performing image segmentation on the accident image based on the image segmentation large model to obtain the scene image region of the accident image, the method further includes: Based on the image remake recognition model, determining whether the accident image is a remake image; When it is determined that the accident image is a retaken image, filtering processing is performed on the accident image.

7. A joint training method for a vehicle damage detection model, characterized in that: include: Acquire a training sample set, wherein the training sample set includes accident images corresponding to a plurality of vehicles and vehicle damage information labels corresponding to a plurality of vehicles; Based on the image segmentation model, the accident image is segmented to obtain a scene image region of the accident image; Based on the scene classification model, the scene image area is classified to obtain a preset scene type corresponding to the scene image area; Based on the feature extraction model, multimodal feature extraction is performed on the scene image region to obtain preset scene features of the scene image region; Based on the visual big model, according to the preset scene type and the preset scene features, the scene image area is refined for scene classification to obtain a target scene type corresponding to the scene image area; Based on the vehicle damage detection model, determining the vehicle damage information of the vehicle according to the accident image and the target scene type; According to the vehicle damage information and the vehicle damage information label, the model parameters of at least one of the image segmentation model, the scene classification model, the feature extraction model, the visual model and the vehicle damage detection model are adjusted.

8. A vehicle damage assessment device, characterized in that: The vehicle damage assessment device comprises: An acquisition module, used to acquire an accident image corresponding to the vehicle; An image segmentation module, used to perform image segmentation on the accident image based on the image segmentation model to obtain a scene image region of the accident image; A first scene classification module, configured to perform scene classification on the scene image region based on a scene classification model to obtain a preset scene type corresponding to the scene image region; A feature extraction module, used to perform multimodal feature extraction on the scene image region based on a feature extraction model to obtain preset scene features of the scene image region; A second scene classification module is used to perform detailed scene classification on the scene image area based on the visual macro model according to the preset scene type and the preset scene features, so as to obtain a target scene type corresponding to the scene image area; The vehicle damage detection module is used to determine the vehicle damage information corresponding to the accident image based on the vehicle damage detection model, according to the accident image and the target scene type.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the vehicle damage assessment method as described in any one of claims 1 to 6 when executing the computer program, or implement the joint training method of the vehicle damage detection model as described in claim 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the vehicle damage assessment method as described in any one of claims 1 to 7 are implemented, or the steps of the joint training method of the vehicle damage detection model as described in claim 7 are implemented.

Images