CN115796272B

CN115796272B - Model training method based on deep learning platform, data processing method and device

Info

Publication number: CN115796272B
Application number: CN202211496874.1A
Authority: CN
Inventors: 杨烨华; 刘爽侨; 张一超; 韩超; 赵乔; 毕然
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2022-11-24
Filing date: 2022-11-24
Publication date: 2024-03-12
Anticipated expiration: 2042-11-24
Also published as: CN115796272A

Abstract

The disclosure provides a model training method based on a deep learning platform, relates to the technical field of artificial intelligence, and particularly relates to the technical fields of deep learning, edge computing, distributed computing and the like. The specific implementation scheme is as follows: in response to receiving first operation information from a target user, displaying model data of a first target deep learning model in at least one preset deep learning model; responsive to receiving second operational information from the target user, presenting target scene data associated with the first target deep learning model; responding to the received third operation information from the target user, and obtaining a second target deep learning model according to the target scene data and the first target deep learning model; and training a second target deep learning model according to the sample data related to the target scene. The disclosure also provides a data processing method and device based on the deep learning platform, electronic equipment and a storage medium.

Description

Model training method based on deep learning platform, data processing method and device

Technical Field

The disclosure relates to the technical field of artificial intelligence, in particular to the technical fields of deep learning, edge computing, distributed computing and the like. More specifically, the present disclosure provides a model training method, a data processing method, an apparatus, an electronic device, and a storage medium based on a deep learning platform.

Background

With the development of artificial intelligence technology, the application scenario of deep learning technology is increasing. The user may select a model in the deep learning platform and make parameter adjustments in conjunction with the relevant scene to apply the model to the relevant scene.

Disclosure of Invention

The disclosure provides a model training method, a data processing method, a device, electronic equipment and a storage medium based on a deep learning platform.

According to an aspect of the present disclosure, there is provided a model training method based on a deep learning platform, the deep learning platform corresponding to at least one preset deep learning model, the method comprising: in response to receiving first operation information from a target user, model data of a first target deep learning model in at least one preset deep learning model is displayed, wherein the first target deep learning model corresponds to the first operation information; responsive to receiving second operational information from the target user, displaying target scene data associated with the first target deep learning model, wherein the target scene data corresponds to a target scene, the second operational information corresponds to the target scene, and the target scene is from at least one preset scene associated with the first target deep learning model; responding to the received third operation information from the target user, and obtaining a second target deep learning model according to the target scene data and the first target deep learning model; and training a second target deep learning model according to sample data related to the target scene, wherein the target scene is related to at least one target device, and the second target deep learning model is used for carrying out data processing according to target data acquired by the target device.

According to another aspect of the present disclosure, there is provided a data processing method based on a deep learning platform, the method including: according to target scene data related to a second target deep learning model, deploying the second target deep learning model to target equipment related to a target scene, inputting the target data into the second target deep learning model to obtain a processing result, wherein the target data is acquired by the target equipment, wherein the second target deep learning model is trained by using the training method provided by the disclosure, wherein the target scene data comprises target model deployment data related to the target scene, the target model deployment data comprises first equipment parameters related to the second target deep learning model, and the equipment parameters of the target equipment are consistent with the first equipment parameters.

According to another aspect of the present disclosure, there is provided a model training apparatus based on a deep learning platform corresponding to at least one preset deep learning model, the apparatus comprising: the first display module is used for displaying model data of a first target deep learning model in at least one preset deep learning model in response to receiving first operation information from a target user, wherein the first target deep learning model corresponds to the first operation information; the second display module is used for displaying target scene data related to the first target deep learning model in response to receiving second operation information from a target user, wherein the target scene data corresponds to a target scene, the second operation information corresponds to the target scene, and the target scene is from at least one preset scene related to the first target deep learning model; the first obtaining module is used for responding to the received third operation information from the target user, and obtaining a second target deep learning model according to the target scene data and the first target deep learning model; and the training module is used for training a second target deep learning model according to sample data related to a target scene, wherein the target scene is related to at least one target device, and the second target deep learning model is used for carrying out data processing according to target data acquired by the target device.

According to another aspect of the present disclosure, there is provided a data processing apparatus based on a deep learning platform, the apparatus comprising: the deployment module is used for deploying the second target deep learning model to target equipment related to the target scene according to target scene data related to the second target deep learning model, the second obtaining module is used for inputting the target data into the second target deep learning model to obtain a processing result, the target data are obtained by the target equipment, the second target deep learning model is trained by using the training device provided by the disclosure, the target scene data comprise target model deployment data related to the target scene, the target model deployment data comprise first equipment parameters related to the second target deep learning model, and the equipment parameters of the target equipment are consistent with the first equipment parameters.

According to another aspect of the present disclosure, there is provided an electronic device including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method provided in accordance with the present disclosure.

According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform a method provided according to the present disclosure.

According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements a method provided according to the present disclosure.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the disclosure, nor is it intended to be used to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following specification.

Drawings

The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is an exemplary system architecture to which a deep learning platform based model training method and/or data processing method may be applied, in accordance with an embodiment of the present disclosure;

FIG. 2 is a flow chart of a deep learning platform based model training method according to one embodiment of the present disclosure;

FIG. 3 is a flow chart of a model training method according to another embodiment of the present disclosure;

FIG. 4A is a schematic diagram showing a preset deep learning model according to one embodiment of the present disclosure;

FIG. 4B is a schematic diagram of presentation model data according to one embodiment of the present disclosure;

5A-5C are schematic diagrams showing target scene data according to one embodiment of the present disclosure;

FIG. 6 is a flow chart of a deep learning platform based data processing method according to one embodiment of the present disclosure;

fig. 7A-7B are schematic diagrams showing target scene data according to one embodiment of the present disclosure;

8A-8B are schematic illustrations of an initial interaction page associated with a first target deep learning model according to another embodiment of the present disclosure;

FIG. 9 is a block diagram of a deep learning platform based model training apparatus according to one embodiment of the present disclosure;

FIG. 10 is a block diagram of a deep learning platform based data processing apparatus according to another embodiment of the present disclosure; and

fig. 11 is a block diagram of an electronic device to which a deep learning platform based model training method and/or a deep learning platform based data processing method may be applied, according to one embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

With the development of deep learning technology, the application scenes of the related models are increasing. In some embodiments, a dataset related to a target language (e.g., chinese) may be absent when the model is applied to the target scene. For example, the Hugging Face platform provides about 10 pre-trained models and 1 play data set. However, these data sets are mostly English data sets, and there are few Chinese data sets. Furthermore, these models have uneven performance and are difficult to apply to a specific scenario.

When applied to an actual business scenario, parameters or structure of the model may be adjusted so that the model adapts to hardware devices in the relevant scenario. Various deep learning platforms, while providing a large number of models, have few models that can be applied to actual business scenarios. In addition, these models have high requirements on the programming ability of the user during use. For example, models in tensor flow (TensorFlow) platforms have undergone longer optimization with stronger reasoning capabilities. It is difficult for a user to determine whether the model matches the business scenario before downloading the model. In addition, the programming paradigm of the model of the tensor flow platform is a static diagram, and the programming capability requirement of a user is high. Also for example, a high performance model that can replicate that disclosed in academic papers is included in the Pytorch platform. However, the models are focused and scientific in scenes, and are difficult to deploy to actual business scenes.

FIG. 1 is an exemplary system architecture to which a deep learning platform based model training method and/or data processing method may be applied, in accordance with an embodiment of the present disclosure.

It should be noted that fig. 1 is only an example of a system architecture to which embodiments of the present disclosure may be applied to assist those skilled in the art in understanding the technical content of the present disclosure, but does not mean that embodiments of the present disclosure may not be used in other devices, systems, environments, or scenarios.

As shown in fig. 1, a system architecture 100 according to this embodiment may include a terminal device 101, a network 102, and a server cluster 103. Network 102 is the medium used to provide communication links between terminal devices 101 and server cluster 103. Network 102 may also be used to provide a medium for communication links within server cluster 103. Network 102 may include various connection types, such as wired and/or wireless communication links, and the like.

A user may interact with the server cluster 103 via the network 102 using the terminal device 101 to receive or send messages or the like. For example, terminal device 101 may send a request to server cluster 103 over network 102 for training a deep learning model.

Various communication client applications may be installed on the terminal device 101, such as a knowledge reading class application, a web browser application, a search class application, an instant messaging tool, a mailbox client and/or social platform software, to name a few.

The terminal device 101 may be a variety of electronic devices having a display screen and supporting web browsing, including but not limited to smartphones, tablets, laptop and desktop computers, and the like.

The server cluster 103 may be a server providing various services, such as a background management server (for example only) providing support for requests sent by users with the terminal device 101.

The server cluster 103 may be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so as to solve the defects of large management difficulty and weak service expansibility in the traditional physical hosts and VPS service ("virtual privateserver" or simply "VPS"). The server may also be a server of a distributed system or a server that incorporates a blockchain.

The model training method and/or the data processing method may be applied to the server cluster 103. The server cluster 103 comprises a plurality of server nodes 1031, 1032, 1033, 1034, each comprising a plurality of hardware units.

The deep learning model may be trained using a hardware unit in any one of the server nodes in the server cluster 103. The deep learning model may also be trained according to a distributed strategy using a plurality of hardware units in a plurality of server nodes in the server cluster 103 to improve the training efficiency of the deep learning model. After training is complete, the deep learning model may be deployed to devices associated with the target scene such that the deep learning model may process data from the target scene.

It should be understood that the number of server nodes in the terminal device, network and server cluster in fig. 1 is merely illustrative. There may be any number of terminal devices, networks and server nodes, as desired for implementation.

It should be noted that the sequence numbers of the respective operations in the following methods are merely representative of the operations for the purpose of description, and should not be construed as representing the order of execution of the respective operations. The method need not be performed in the exact order shown unless explicitly stated.

It will be appreciated that while the deep learning platform of the present disclosure is described above, a model training method based on the deep learning platform will be described below in connection with the related embodiments.

Fig. 2 is a flow chart of a model training method based on a deep learning platform according to one embodiment of the present disclosure.

In embodiments of the present disclosure, the deep learning platform may be associated with at least one preset deep learning model. For example, a preset deep learning model may be used to perform various tasks such as image classification tasks, object detection tasks, and the like.

As shown in fig. 2, the method 200 may include operations S210 to S240.

In response to receiving the first operation information from the target user, model data of a first target deep learning model of the at least one preset deep learning model is presented in operation S210.

In the disclosed embodiments, the target user may be a user of the deep learning platform.

In an embodiment of the present disclosure, the first target deep learning model may correspond to the first operation information. For example, the first operation information may indicate a preset deep learning model selected by the user. According to the first operation information, a preset deep learning model selected by the user can be used as a first target deep learning model.

In embodiments of the present disclosure, the model data may include various data related to the first target deep learning model, which is not limited herein. For example, the model data may include at least one preset scene data associated with the first target deep learning model. The preset scene data may correspond to a preset scene.

In response to receiving the second operation information from the target user, target scene data related to the first target deep learning model is presented in operation S220.

In an embodiment of the present disclosure, the second operation information may correspond to a target scene. For example, the second operation information may indicate a preset scene selected by the user. According to the second operation information, the user can select the preset scene as the target scene.

In the embodiment of the present disclosure, the target scene data corresponds to a target scene. The target scene is from at least one preset scene associated with the first target deep learning model. For example, taking the example that the first target deep learning model is a YOLOv2 (You LookOnlyOncev, generation 2 is seen only once) model, the YOLOv2 model can be applied to the following preset scenarios: fire/smoke detection scenarios, helmet detection scenarios, etc. According to the second operation information, a fire/smoke detection scene may be taken as a target scene. It will be appreciated that in a fire/smoke detection scenario, object detection may be performed based on the YOLOv2 model to detect whether there is a flame or smoke in the input image.

In the disclosed embodiments, the target scene data may include various data related to the target scene. For example, the target scene data may include model training data related to the target scene. The model training data may include sample pre-processing data. The sample pre-processing data may indicate the manner in which the original sample data was processed. For another example, the sample pre-processing data may indicate that negative samples and negative labels are added to the original sample data to reduce false detection of the model. In one example, the raw sample data is an image with a flame and the negative sample may be an image of the evening primrose. The labels of the two may not be identical. The labels of the original sample data are positive labels, and the labels of the negative samples are negative labels.

In response to receiving the third operation information from the target user, a second target deep learning model is obtained from the target scene data and the first target deep learning model in operation S230.

In the embodiment of the disclosure, the target user may send third operation information for deploying the first target deep learning model to the target scene. This information may be used as third operation information.

In embodiments of the present disclosure, the structure or parameters of the first target deep learning model may be determined using the target scene data. For example, the target scene data may include some parameters of the YOLOv2 model applied to the fire/smoke detection scene, and the parameters are used as initial parameters of the first target deep learning model, so that a second target deep learning model may be obtained.

In operation S240, a second target deep learning model is trained from sample data related to the target scene.

In an embodiment of the present disclosure, the second target deep learning model may be trained based on at least one of a supervised training approach and an unsupervised training approach. For example, taking the example that the target scene is a fire/smoke detection scene, the user may upload a sample image related to the scene as sample data. After the sample image is input into the second target deep learning model, a detection frame of the sample image can be obtained. Based on the supervised training mode, loss information can be determined according to the position of the detection frame and the label of the sample image, and parameters of the second target deep learning model are adjusted by using the loss information.

In the embodiment of the disclosure, the target scene is related to at least one target device, and the second target deep learning model is used for performing data processing according to target data acquired by the target device. For example, taking the example that the target scene is a fire/smoke detection scene, the target device may be a camera. The second deep learning model can perform data processing according to video data acquired by the camera.

It will be appreciated that the second target deep learning model may be deployed to each target device, or may be deployed to a server separately connected to at least one target device, as this disclosure is not limited in this regard. For example, taking the example that the target scene is a fire/smoke detection scene, a second target deep learning model may be deployed to the camera.

According to the embodiment of the invention, the deep learning model can be quickly constructed according to the operation information from the user, and the model to be deployed in the relevant scene is trained, so that the workload of developers is reduced, the development efficiency is improved, and the user experience is improved. Furthermore, with the disclosed embodiments, the second deep learning model may process data from edge devices (e.g., cameras), reducing the usage threshold of edge computing techniques.

It will be appreciated that the method of the present disclosure is described above in connection with fig. 2. The method of the present disclosure will be described in further detail below in conjunction with fig. 3.

Fig. 3 is a flow chart of a model training method according to another embodiment of the present disclosure.

As shown in fig. 3, the method 300 may demonstrate model data of a first target deep learning model of the at least one preset deep learning model. The following will explain in conjunction with operations S311 to S313.

In operation S311, a preset deep learning model is displayed.

In some embodiments, controls associated with a preset deep learning model may be presented.

In the disclosed embodiments, the controls may include task controls related to a preset deep learning model. For example, the task control may be multiple. The plurality of task controls are respectively associated with a plurality of tasks. The plurality of tasks may include an image classification task, a target object detection task, an image recognition task, a keypoint detection task, a multi-target tracking task, a semantic analysis task, a video classification task, a speech recognition task, a speech synthesis task, a natural language processing task, and so forth.

In embodiments of the present disclosure, the controls may also include scene controls associated with a preset deep learning model. For example, the scene control may be multiple. The scene controls are respectively related to the preset scenes. The plurality of scenes may include a lightweight license plate recognition scene, an in-elevator battery car detection scene, a road surface garbage detection scene, a helmet detection scene, a fall detection scene, an industrial pointer type meter/meter reading scene, a form recognition scene, a fire/smoke detection scene, a person image segmentation scene, and the like.

In embodiments of the present disclosure, the controls may also include dataset controls related to a preset deep learning model. For example, the dataset controls may be multiple. The plurality of data set controllers are respectively associated with the plurality of sample data sets. The plurality of sample data sets may include a common audio (common) data set, an expression (emotion) data set, and so forth.

In embodiments of the present disclosure, the controls may also include developer controls related to a preset deep learning model. For example, the developer control may be multiple. The plurality of developer controls are respectively associated with a plurality of common developers. Such developers may include well known scholars in the art of artificial intelligence, technical experts, active users of open source communities, and the like.

In the disclosed embodiments, the controls may also include search controls. For example, the search control may obtain model keywords from the target user and invoke a related search algorithm to return a preset deep learning model corresponding to the model keywords. By the embodiment of the disclosure, various types of controls are displayed, so that a user can accurately select a model, and the efficiency of model selection is improved.

In an embodiment of the present disclosure, the first operation information indicates a first target deep learning model corresponding to at least one of a target task, a target scene, a target sample data set, a target developer, and a target model keyword. For example, the user may trigger a control such that the deep learning platform receives the first operational information. For example, the user may trigger a task control corresponding to the target detection task such that the deep learning platform exposes at least one model control. These model controls correspond to preset deep learning models that can perform target detection tasks. In one example, the YOLOv2 model, YOLOv5 (You Look Only Once v, passage 5 only once) model may perform the target detection task. For another example, the user may trigger a model control corresponding to the YOLOv2 model such that the deep learning platform takes the YOLOv2 model as the first target deep learning model. It can be understood that the information triggering the generation of the task control corresponding to the target detection task and the information triggering the generation of the model control corresponding to the YOLOv2 model can be used as the first operation information.

In operation S312, model data is presented.

For example, model data of the YOLOv2 model may be presented.

In an embodiment of the present disclosure, the model data includes at least one of model training data, model deployment data, model performance data, the model performance data including at least one of model structure data and historical task performance data.

In embodiments of the present disclosure, the model training data may include model supervised training data and model unsupervised training data. For example, model training data may indicate a training manner of the deep learning model. For another example, the model training data may also include training effects of different training patterns.

In the disclosed embodiments, the model deployment data may include device parameters related to the model.

In the disclosed embodiments, the model performance data may include model structure data and historical task execution data. For example, the model structure data may indicate the network structure of the model, the size of the input data, the volume of the model, and so forth. As another example, historical task execution data may include tasks that a model may execute. By the embodiment of the invention, various data related to the model are displayed, so that the global cognition of a user to the model can be improved, and the user can be helped to determine whether to use the model.

Next, the method 300 may also enable the user to experience performance of the first target deep learning model, as will be described below in connection with operation S313.

In operation S313, an initial output result is displayed.

In an embodiment of the present disclosure, an initial output result of a first target deep learning model is presented based on initial sample data from a target user.

In embodiments of the present disclosure, initial sample data from a target user may be obtained. For example, an initial processing function is determined based on a first target deep learning model. And determining an initial interaction page related to the first target deep learning model according to the initial processing function. And acquiring initial sample data by using the initial interaction page. The type of the initial sample data is consistent with the preset sample data type. As another example, the initial processing function may be a calling function. Based on the initial processing function, a first target deep learning model may be invoked. For another example, the initial interaction page may be generated based on an initial processing function, a preset sample data type, and a preset output result type. For another example, the initial interaction page may include a data acquisition control. After the user triggers the data acquisition control, the initial sample data can be uploaded, so that the deep learning platform acquires the initial sample data. In one example, as described above, the first target deep learning model may be a YOLOv2 model and the initial sample data may be Image1.

In embodiments of the present disclosure, initial sample data may be presented using an initial interaction page. For example, the Image1 described above may be presented using the initial interaction page.

In the embodiment of the disclosure, the initial sample data may be processed by using the first target deep learning model to obtain an initial output result. In one example, the YOLOv2 model may output a detection box of the object in Image1.

In the embodiments of the present disclosure, an initial output result may be presented. For example, the initial output results may be presented using an initial interaction page. For another example, the detection box of the object in Image1 may be presented using the initial interaction page. According to the embodiment of the invention, the sample data and the output result of the model are displayed on the initial interaction page, so that a user can quickly and intuitively experience the function of the model, and the user can be helped to further determine whether the model is suitable for executing related tasks.

Next, the method 300 may perform operation S323.

In operation S323, the model is matched with the scene.

It is understood that operation S323 may include operations corresponding to operation S220 and operation 230 described above.

In an embodiment of the present disclosure, target scene data associated with a first target deep learning model is presented in response to receiving second operational information from a target user. The second operation information may indicate a preset scene selected by the user. According to the second operation information, the user can select the preset scene as the target scene. Taking the example that the first target deep learning model is a YOLOv2 model, the YOLOv2 model can be applied to the following preset scenarios: in-elevator storage battery car detection scene, road surface garbage detection scene, safety helmet detection scene and fire/smoke detection scene. According to the second operation information, a fire/smoke detection scene may be taken as a target scene.

In an embodiment of the disclosure, in response to receiving third operation information from the target user, a second target deep learning model is obtained from the target scene data and the first target deep learning model. For example, as described above, the target user may send third operational information deploying the first target deep learning model to the target scene. This information may be used as third operation information. The target scene data may include some parameters of the YOLOv2 model applied to the fire/smoke detection scene, and the parameters are used as initial parameters of the first target deep learning model, so that a second target deep learning model may be obtained.

Next, the method 300 may further include operation S340.

In operation S340, a second target deep learning model is trained.

For example, taking the example that the target scene is a fire/smoke detection scene, the user may upload a sample image related to the scene as sample data. After the sample image is input into the second target deep learning model, a detection frame of the sample image can be obtained. Based on the supervised training mode, loss information can be determined according to the position of the detection frame and the label of the sample image, and parameters of the second target deep learning model are adjusted by using the loss information.

In embodiments of the present disclosure, the second target deep learning model may be trained using a cloud training platform. Further, in the disclosed embodiments, the code of the second target deep learning model may be adjusted using a cloud compiling tool. In embodiments of the present disclosure, relevant code for the second target deep learning model may also be downloaded to the user's device in order to train the second target deep learning model.

Further, in another embodiment of the present disclosure, in response to receiving fourth operation information from the target user, operation S311 is returned to. For example, the target user may transmit information to reselect the deep learning model or scene, which may be the fourth operation information.

It will be appreciated that while the method of the present disclosure is described above, some implementations of model data exhibiting a predetermined deep learning model will be further described below in connection with related embodiments.

Fig. 4A is a schematic diagram showing a preset deep learning model according to one embodiment of the present disclosure.

In some implementations of operation S311 described above, a control related to a preset deep learning model may be presented using page 410 in embodiments of the present disclosure. As shown in FIG. 4A, page 410 may expose a plurality of task controls, a plurality of scene controls, a plurality of developer controls, and a plurality of dataset controls. For example, the task control 411 may correspond to a target detection task. The plurality of other task controls may correspond to an image classification task, an image recognition task, and a speech synthesis task, respectively. As another example, scene control 412 may correspond to a fire/smoke detection scene. The plurality of other scene controls may correspond to a fall detection scene, a form recognition scene, respectively. For another example, the plurality of developer controls may correspond to developer AA, developer BB, and developer CC, respectively. For another example, the dataset controls may include a generic speech dataset control, an expression dataset control, and a generic language understanding assessment dataset control. In one example, after the user triggers the target detection task control 411, a plurality of model controls corresponding to target detection tasks and fire/smoke detection scenarios may be presented. The plurality of model controls may correspond to a YOLOv2 model, a YOLOv5 model, and an SSD (Single Shot Multi-Box Detector, single-shot multi-Box Detector) model, respectively. For example, model control 413 may correspond to a YOLOv2 model.

Next, if the model control 413 is detected to be triggered, information for triggering the model control 413 is used as first operation information. Next, in response to receiving the first operational information, model data may be presented.

Fig. 4B is a schematic diagram showing model data according to one embodiment of the present disclosure.

In some implementations of operation S312 described above, the model data may be presented using page 420 in embodiments of the present disclosure. As shown in fig. 4B, page 420 may present model structure data "YOLOv2 model includes a convolutional layer and a batch normalization (Batch Normalization, BN) layer. It will be appreciated that the model structure data shown in fig. 4B is merely an example, and that the model structure data of the YOLOv2 model may also include other content, which is not limited by the present disclosure.

Further, in some implementations of operation S313 described above, the page 420 may be utilized to present the initial output result in embodiments of the present disclosure. As shown in FIG. 4B, page 420 includes an online experience module. For example, according to the YOLOv2 model, the initial processing function may be determined. An online experience module associated with the first target deep learning model may be determined based on the initial processing function, based on which an initial interaction page may be determined. Initial sample data may be obtained using the upload control of the online experience module in page 420. For example, after the upload control is triggered, initial sample data 421 uploaded by the user may be obtained and presented. After the run control is triggered, initial sample data 421 is processed using the YOLOv2 model, and initial output results 422 can be obtained and presented. As shown in fig. 4B, the initial output result 422 may include "the upper left vertex coordinates of the detection frame are (a, B), the width is w, and the height is h". It will be appreciated that the page 420 may be used as an initial interaction page, the upload control may be used as the data acquisition control described above, and the initial sample data may be the image described above.

It will be appreciated that some implementations of the presentation model data are described above and some of the ways in which the model matches the scene will be described below in connection with the relevant examples.

Fig. 5A-5C are schematic diagrams showing target scene data according to one embodiment of the present disclosure.

In embodiments of the present disclosure, the target scene data may include scene demand feature data related to the target scene. As shown in fig. 5A and 5B, taking an example in which the target scene is a fire/smoke detection scene, the scene demand feature data may include scene difficulty data: the model needs to respond in time in a fire/smoke scene, has a plurality of objects and smoke approaching (lamplight and evening) and is easy to cause false detection, and the model needs to be deployed to the edge side of a camera. The scene demand feature data may also include model required capability data: the model reasoning speed is high, and the recall rate is high and the false detection rate is low.

In embodiments of the present disclosure, the target scene data may further include at least one preset data set related to the target scene. Two sample data in the preset dataset are shown in fig. 5A.

In embodiments of the present disclosure, the target scene data may also include model effect data related to the target scene. As shown in fig. 5B, the model effect may include: the performance of the YOLOv2 model is advanced, and the precision is better than that of the YOLOv5 model.

In embodiments of the present disclosure, the target scene data may also include model training strategy data related to the target scene. As shown in fig. 5C, the model training strategy data may include model training preset strategies and corresponding model training effect data. The model training preset strategy may correspond to a "model training code train. Py", which may be run directly. The model training effect data corresponding to the model training preset strategy is as follows: the recall rate is high, but the false detection rate is high, the recall rate can reach 95.1%, and the false detection rate is 23.2%. As shown in fig. 5C, the model training strategy data may also include model training optimization strategies and corresponding model training effect data. Model training optimization strategies may include: "data enhancement", "adding multiple pre-trained models" and "adding negative samples and corresponding negative labels that are prone to false detection". The model training effect data corresponding to the model training optimization strategy is as follows: the recall rate can reach 96%, and the false detection rate is 2.2%. It will be appreciated that the sample data may be enhanced in a variety of ways.

Next, a second target deep learning model may be derived from the target scene data and the first target deep learning model based on the received third operational information indicating deployment of the model to the target scene. For example, the target scene data may include: the YOLOv2 model is applied to some parameters of a fire/smoke detection scenario. And taking the parameters as initial parameters of the first target deep learning model, and obtaining a second target deep learning model. Next, a second target deep learning model may be trained in accordance with the preset data set and training optimization strategy described above.

In an embodiment of the disclosure, the target device is deployed with a second target deep learning model. For example, taking the example that the target scene is a fire/smoke detection scene, a trained second target deep learning model may be deployed to the camera.

It will be appreciated that while the training method of the present disclosure is described above, the data processing method of the present disclosure will be described below in connection with the related embodiments.

Fig. 6 is a flowchart of a deep learning platform based data processing method according to one embodiment of the present disclosure.

As shown in fig. 6, the method 600 may include operations S610 to S620.

In operation S610, a second target deep learning model is deployed to a target device associated with the target scene according to target scene data associated with the second target deep learning model.

In an embodiment of the disclosure, the target scene data includes model deployment data associated with the target scene, the model deployment data including first device parameters associated with the second target deep learning model, the device parameters of the target device being consistent with the first device parameters. For example, the first device parameter may correspond to a model number and a number of graphics processing units. The target device may have a corresponding model and number of graphics processing units. As another example, taking the example that the target scene is a fire/smoke detection scene, the target device may be a camera with a lightweight graphics processing unit. A second target deep learning model may be deployed to the lightweight graphics processing unit.

In operation S620, the target data is input into the second target deep learning model, and the processing result is obtained.

In the disclosed embodiments, the target data may be acquired by the target device. For example, the target data may be image data acquired by a camera.

In the disclosed embodiments, the processing results may indicate whether a flame or smoke is present in the target data.

In an embodiment of the present disclosure, the second deep learning model may be trained by the method 200 described above.

It will be appreciated that while the data processing method of the present disclosure has been described above, the target scene data will be further described below in connection with the related embodiments

Fig. 7A-7B are schematic diagrams showing target scene data according to one embodiment of the present disclosure.

In embodiments of the present disclosure, the target scenario data may also include model deployment data related to the target scenario. As shown in fig. 7A, taking the example that the target scenario is a fire/smoke detection scenario, the model deployment data may include deployment environment data: "end-side deployment", "small model volume" and "high real-time processing performance". The model deployment data may also include hardware and predictive library data: "lightweight graphics processing unit" and "Paddle information prediction library". In one example, the lightweight graphics processing units may be various lightweight graphics processing units. The Paddle information prediction library can be a prediction library preset by the flying oar platform, and can provide configuration parameters of a model. These configuration parameters help to improve data processing efficiency.

In the disclosed embodiments, the target scene data may include preprocessing policy data related to the target scene. Inputting the target data into a second target deep learning model, and obtaining a processing result comprises: and adjusting the target data according to the preprocessing strategy data to obtain adjusted target data. And inputting the adjusted target data into a second target deep learning model to obtain a processing result. As shown in fig. 7B, the preprocessing policy data may indicate that the target device accesses the business process. For example, based on the preprocessing strategy data, the video stream acquired by the access target device may be read. For another example, an image in a video stream may be taken as target data. And adjusting the size of the image according to the preprocessing strategy data. And inputting the adjusted image into a second target deep learning model.

In an embodiment of the present disclosure, the target scene data further includes visualization policy data related to the target scene. Inputting target data acquired by target equipment into a second target deep learning model, and obtaining a processing result comprises the following steps: and adjusting the processing result according to the visualization strategy data to obtain a visualization result. And displaying the visual result on a visual interface. For example, the visual interface is generated from the visualization policy data. As shown in fig. 7B, the processing result may be adjusted according to the visualization strategy data. For another example, an image corresponding to a result of processing for indicating the occurrence of a fire may be used as the visualized result. And displaying the visual result on a visual interface.

According to the embodiment of the disclosure, as shown in fig. 5A to 5C and fig. 7A to 7B, the model can be rapidly deployed to the target scene, and the whole processes of model selection, scene and model adaptation, model training and model deployment can be assisted by a user according to user requirements, so that the use threshold of the model in the relevant scene is greatly reduced, the experience of the user in using the deep learning platform is improved, and the application scene of the deep learning model is further expanded.

Fig. 8A-8B are schematic diagrams of initial interaction pages related to a first target deep learning model according to another embodiment of the present disclosure.

As in fig. 8A, an initial processing function may be determined based on the model for performing the image recognition task. Based on the initial processing function, an online experience module associated with the model that performs the image recognition task may be determined, based on which an initial interaction page may be determined. Initial sample data may be obtained using an upload control of the online experience module. For example, after the upload control is triggered, the initial sample data 823 uploaded by the user may be obtained and presented. After the run control is triggered, initial output results 824 may be obtained and presented using the model processing initial sample data 823. As shown in FIG. 8A, the initial output result 824 may include "a white puppy running past the grass".

As shown in fig. 8B, an initial processing function may be determined based on a model for performing text classification tasks. Based on the initial processing function, an online experience module associated with the model that performs the text classification task may be determined, based on which an initial interaction page may be determined. Initial sample data 825 may be obtained using the upload control of the online experience module. For example, after the upload control is triggered, the initial sample data 825 uploaded by the user may be obtained and presented. The initial sample data 825 may be the text "<" > dream of the red blood cell "is a long novel authored by the sake of the composer Cao Xueqin. After the run control is triggered, initial sample data 825 is processed using the model, and initial output results 826 may be obtained and presented. As shown in fig. 8B, initial output result 826 may include a category label for a word in text. For example, the initial output result 826 may include a category label "person name" of the word "Cao Xueqin", a category label "book name" of the word "dream of the red blood cell".

Fig. 9 is a block diagram of a deep learning platform based model training apparatus according to one embodiment of the present disclosure.

In an embodiment of the present disclosure, the deep learning platform is associated with at least one preset deep learning model.

Model training apparatus 900 may include a first presentation module 910, a second presentation module 920, a first acquisition module 930, and a training module 940.

The first presentation module 910 is configured to present model data of a first target deep learning model in the at least one preset deep learning model in response to receiving first operation information from the target user. For example, the first target deep learning model corresponds to the first operation information.

A second presentation module 920 configured to present the target scene data related to the first target deep learning model in response to receiving the second operation information from the target user. For example, the target scene data corresponds to a target scene, the second operation information corresponds to the target scene, and the target scene is from at least one preset scene related to the first target deep learning model.

The first obtaining module 930 is configured to obtain, in response to receiving the third operation information from the target user, a second target deep learning model according to the target scene data and the first target deep learning model.

A training module 940 for training a second target deep learning model based on sample data related to the target scene. For example, the target scene is associated with at least one target device, and the second target deep learning model is used for data processing according to target data acquired by the target device.

In some embodiments, the first operational information indicates a first target deep learning model corresponding to at least one of a target task, a target scene, a target sample data set, a target developer, a target model keyword. The model data includes at least one of model training data, model deployment data, model performance data including at least one of model structure data and historical task execution data.

In some embodiments, the first display module further comprises: and the first display sub-module is used for displaying an initial output result of the first target deep learning model according to initial sample data from the target user.

In some embodiments, the first presentation submodule includes: an acquisition unit configured to acquire initial sample data from a target user; the processing unit is used for processing the initial sample data by utilizing the first target deep learning model to obtain an initial output result; and a display unit for displaying the initial output result.

In some embodiments, the acquisition unit comprises: the first determining subunit is used for determining an initial processing function according to the first target deep learning model; the second determining subunit is used for determining an initial interaction page related to the first target deep learning model according to the initial processing function; and the acquisition subunit is used for acquiring initial sample data by utilizing the initial interaction page.

In some embodiments, the acquisition unit further comprises: and the first display subunit is used for displaying the initial sample data by using the initial interaction page.

In some embodiments, the display unit comprises: and the second display subunit is used for displaying the initial output result by using the initial interaction page.

In some embodiments, the target scene data includes at least one of: scene demand feature data associated with the target scene, at least one preset data set, model training strategy data.

In some embodiments, the target device is deployed with the second target deep learning model.

In some embodiments, the target tasks include an image processing task, and the image processing task includes a target object detection task and an image recognition task.

Fig. 10 is a block diagram of a deep learning platform based data processing apparatus according to one embodiment of the present disclosure.

As shown in fig. 10, the apparatus 1000 may include a deployment module 1010 and a second acquisition module 1020.

A deployment module 1010 for deploying the second target deep learning model to a target device associated with the target scene based on target scene data associated with the second target deep learning model.

And a second obtaining module 1020, configured to input the target data into a second target deep learning model, to obtain a processing result.

In the disclosed embodiment, the target data is acquired by the target device,

in an embodiment of the present disclosure, the second target deep learning model is trained using the training apparatus provided by the present disclosure,

in an embodiment of the disclosure, the target scene data includes target model deployment data related to the target scene, the target model deployment data including first device parameters related to the second target deep learning model, the device parameters of the target device being consistent with the first device parameters.

In some embodiments, the target scene data further includes preprocessing strategy data associated with the target scene. The second obtaining module includes: and the first adjusting sub-module is used for adjusting the target data according to the preprocessing strategy data to obtain the adjusted target data. The first obtaining sub-module is used for inputting the adjusted target data into the second target deep learning model to obtain a processing result.

In some embodiments, the target scene data further includes visualization policy data related to the target scene. The second obtaining module includes: and the second adjustment sub-module is used for adjusting the processing result according to the visualization strategy data to obtain a visualization result. And the second obtaining sub-module is used for displaying the visual result on a visual interface, wherein the visual interface is generated according to the visual strategy data.

In the technical scheme of the disclosure, the related processes of collecting, storing, using, processing, transmitting, providing, disclosing and the like of the personal information of the user accord with the regulations of related laws and regulations, and do not violate the common-order fortunes.

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

Fig. 11 illustrates a schematic block diagram of an example electronic device 1100 that can be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 11, the apparatus 1100 includes a computing unit 1101 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 1102 or a computer program loaded from a storage unit 1108 into a Random Access Memory (RAM) 1103. In the RAM 1103, various programs and data required for the operation of the device 1100 can also be stored. The computing unit 1101, ROM 1102, and RAM 1103 are connected to each other by a bus 1104. An input/output (I/O) interface 1105 is also connected to bus 1104.

Various components in device 1100 are connected to I/O interface 1105, including: an input unit 1106 such as a keyboard, a mouse, etc.; an output unit 1107 such as various types of displays, speakers, and the like; a storage unit 1108, such as a magnetic disk, optical disk, etc.; and a communication unit 1109 such as a network card, modem, wireless communication transceiver, or the like. The communication unit 1109 allows the device 1100 to exchange information/data with other devices through a computer network such as the internet and/or various telecommunication networks.

The computing unit 1101 may be a variety of general purpose and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 1101 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1101 performs the respective methods and processes described above, such as a model training method based on a deep learning platform and/or a data processing method based on a deep learning platform. For example, in some embodiments, the deep learning platform based model training method and/or the deep learning platform based data processing method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 1108. In some embodiments, some or all of the computer programs may be loaded and/or installed onto device 1100 via ROM 1102 and/or communication unit 1109. When the computer program is loaded into the RAM 1103 and executed by the computing unit 1101, one or more steps of the deep learning platform based model training method and/or the deep learning platform based data processing method described above may be performed. Alternatively, in other embodiments, the computing unit 1101 may be configured to perform the deep learning platform based model training method and/or the deep learning platform based data processing method in any other suitable manner (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

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

The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the internet.

The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel, sequentially, or in a different order, provided that the desired results of the disclosed aspects are achieved, and are not limited herein.

The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.

Claims

1. A model training method based on a deep learning platform, the deep learning platform being associated with at least one preset deep learning model, comprising:

responding to receiving first operation information from a target user, and displaying model data of a first target deep learning model in at least one preset deep learning model by using an initial interaction page, wherein the first target deep learning model corresponds to the first operation information;

responsive to receiving second operational information from the target user, exhibiting target scene data associated with the first target deep learning model, wherein the target scene data corresponds to a target scene from at least one preset scene associated with the first target deep learning model, the second operational information being used to indicate the preset scene selected by the target user, the target scene data comprising at least one preset dataset corresponding to the target scene, initial parameters of the first target deep learning model applied to the target scene, and model training strategy data;

Responding to the received third operation information from the target user, and obtaining a second target deep learning model according to the initial parameters of the first target deep learning model applied to the target scene in the target scene data and the first target deep learning model; and

training the second target deep learning model based on sample data associated with the target scene, wherein the target scene is associated with at least one target device, the trained second target deep learning model being deployed to at least one of the target devices for data processing based on target data acquired by the target device, the sample data associated with the target scene including the at least one preset data set corresponding to the target scene,

the displaying the model data of the first target deep learning model in the at least one preset deep learning model by using the initial interaction page further includes:

determining a calling function for calling the first target deep learning model according to the first target deep learning model;

determining an online experience module related to the first target deep learning model according to the calling function;

Determining an initial interaction page based on the online experience module;

acquiring initial sample data uploaded by the target user by utilizing the initial interaction page, wherein the initial sample data is related to the target scene;

the first target deep learning model is called to process the initial sample data by utilizing the calling function, and an initial output result is obtained; and

the initial output result is presented by the initial interaction page,

wherein said training said second target deep learning model comprises:

according to the model training strategy data, adding a negative sample and a negative label which are easy to cause false detection into the preset data set;

and training the second target deep learning model according to a preset data set added with the negative sample and the negative label.

2. The method of claim 1, wherein the first operational information indicates the first target deep learning model corresponding to at least one of a target task, a target sample data set, a target developer, a target model keyword,

the model data includes at least one of model training data, model deployment data, model performance data including at least one of model structure data and historical task execution data.

3. The method of claim 1, wherein the obtaining initial sample data from the target user upload using the initial interaction page further comprises:

and displaying the initial sample data by using the initial interaction page.

4. The method of claim 1, wherein the target scene data further comprises scene demand feature data related to the target scene.

5. The method of claim 1, wherein the target device is deployed with the second target deep learning model.

6. The method of claim 2, wherein the target task comprises an image processing task comprising a target object detection task and an image recognition task.

7. A data processing method based on a deep learning platform comprises the following steps:

deploying the second target deep learning model to target equipment related to a target scene according to target scene data related to the second target deep learning model; and

inputting the target data into the second target deep learning model to obtain a processing result,

wherein the target data is acquired by the target device,

wherein the second target deep learning model is trained using the training method of any one of claims 1 to 6,

Wherein the target scene data comprises target model deployment data related to the target scene, the target model deployment data comprising first device parameters related to the second target deep learning model, the device parameters of the target device being consistent with the first device parameters.

8. The method of claim 7, wherein the target scene data further comprises preprocessing strategy data associated with the target scene,

inputting the target data into the second target deep learning model, and obtaining a processing result comprises:

according to the preprocessing strategy data, adjusting the target data to obtain adjusted target data; and

and inputting the adjusted target data into the second target deep learning model to obtain the processing result.

9. The method of claim 7, wherein the target scene data further comprises visualization strategy data associated with the target scene,

according to the visualization strategy data, the processing result is adjusted to obtain a visualization result; and

And displaying the visual result on a visual interface, wherein the visual interface is generated according to the visual strategy data.

10. A model training apparatus based on a deep learning platform, the deep learning platform being associated with at least one preset deep learning model, comprising:

the first display module is used for displaying model data of a first target deep learning model in at least one preset deep learning model by utilizing an initial interaction page in response to receiving first operation information from a target user, wherein the first target deep learning model corresponds to the first operation information;

a second display module, configured to display target scene data related to the first target deep learning model in response to receiving second operation information from the target user, where the target scene data corresponds to a target scene, the target scene being from at least one preset scene related to the first target deep learning model, the second operation information being used to indicate a preset scene selected by the target user, the target scene data including at least one preset data set corresponding to the target scene, initial parameters of the first target deep learning model applied to the target scene, and model training policy data;

The first obtaining module is used for responding to the received third operation information from the target user, and obtaining a second target deep learning model according to the initial parameters of the first target deep learning model applied to the target scene in the target scene data and the first target deep learning model; and

a training module for training the second target deep learning model according to sample data related to the target scene, wherein the target scene is related to at least one target device, the trained second target deep learning model is deployed to at least one target device so as to perform data processing according to target data acquired by the target device, the sample data related to the target scene comprises the at least one preset data set corresponding to the target scene,

wherein, the first display module further includes:

a first determining subunit, configured to determine, according to the first target deep learning model, a calling function for calling the first target deep learning model;

the second determining subunit is used for determining an online experience module related to the first target deep learning model according to the calling function; determining an initial interaction page based on the online experience module;

The acquisition subunit is used for acquiring initial sample data uploaded by the target user by utilizing the initial interaction page, wherein the initial sample data is related to the target scene;

the processing unit is used for calling the first target deep learning model to process the initial sample data by using the calling function so as to obtain an initial output result; and

a second display subunit, configured to display the initial output result by using the initial interaction page,

wherein, training module is still used for: according to the model training strategy data, adding a negative sample and a negative label which are easy to cause false detection into the preset data set; and training the second target deep learning model according to a preset data set added with the negative sample and the negative label.

11. The apparatus of claim 10, wherein the first operational information indicates the first target deep learning model corresponding to at least one of a target task, a target sample data set, a target developer, a target model keyword,

12. The apparatus of claim 10, wherein the first display module further comprises:

and the first display subunit is used for displaying the initial sample data by using the initial interaction page.

13. The apparatus of claim 10, wherein the target scene data further comprises scene demand feature data related to the target scene.

14. The apparatus of claim 10, wherein the target device is deployed with the second target deep learning model.

15. The apparatus of claim 11, wherein the target task comprises an image processing task comprising a target object detection task and an image recognition task.

16. A deep learning platform based data processing apparatus comprising:

the deployment module is used for deploying the second target deep learning model to target equipment related to a target scene according to target scene data related to the second target deep learning model;

a second obtaining module for inputting the target data into the second target deep learning model to obtain a processing result,

wherein the target data is acquired by the target device,

Wherein the second target deep learning model is trained using the training apparatus of any one of claims 10 to 15,

17. The apparatus of claim 16, wherein the target scene data further comprises preprocessing strategy data associated with the target scene,

the second obtaining module includes:

the first adjustment sub-module is used for adjusting the target data according to the preprocessing strategy data to obtain adjusted target data; and

the first obtaining sub-module is used for inputting the adjusted target data into the second target deep learning model to obtain the processing result.

18. The apparatus of claim 16, wherein the target scene data further comprises visualization policy data associated with the target scene,

the second obtaining module includes:

the second adjustment sub-module is used for adjusting the processing result according to the visualization strategy data to obtain a visualization result; and

And the second obtaining sub-module is used for displaying the visual result on a visual interface, wherein the visual interface is generated according to the visual strategy data.

19. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 9.

20. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1 to 9.