Data processing method, device, platform, medium and equipment based on neural network

Abstract

The application discloses a neural network-based data processing method, device, platform, medium and equipment, and belongs to the field of data processing. The method comprises the following steps: data definition is performed on to-be-processed data, and an object instance is created; a parameter interface pre-encapsulated by the object instance is called to set parameters of the object instance, and object parameters are obtained; the object parameters and the object instance are simultaneously input into a neural network by calling a function processing interface, the to-be-processed data is preprocessed in the neural network by using an algorithm reasoning library, preprocessed data is obtained, and the preprocessed data is output by calling a data output interface; the neural network performs optimization processing on the preprocessed data by using an algorithm configuration file and the algorithm reasoning library, and processing data is obtained; and the processing data is post-processed according to user pre-defined parameters, and standard processing data is obtained. The application does not need algorithm conversion, is simple and efficient, and saves the actual deployment time of a vehicle.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a data processing method and device based on a neural network, a platform, a medium and equipment. BACKGROUND

[0002] Generally speaking, embedded devices support the use of C / C++ language to run development, and algorithm development generally tends to use Python language for writing. In the process of delivering, engineering and productizing algorithm scripts based on Python language, the algorithm scripts need to be converted into scripts written in C++ language according to the writing method of Python. However, for engineering / productization personnel, various types of data processing involved in algorithm conversion often need to spend a long time in actual deployment.

[0003] Although the algorithm running on the embedded device can well reduce the vehicle deployment time, its framework based on the old version has some discount in subsequent image processing. Since the algorithm inference library containing multiple algorithm models runs better in the neural network and spends less time, how to develop a Python script conversion to C++ script in the neural network needs to be solved. SUMMARY

[0004] In order to achieve the purpose of running the algorithm in the neural network to make the programming language conversion simple and save the time spent in vehicle deployment, the present application mainly provides a data processing method and device based on a neural network, a platform, a medium and equipment.

[0005] In a first aspect, the present application provides a data processing method based on a neural network, which comprises:

[0006] Defining data for the to-be-processed data, creating an object instance;

[0007] Setting parameters for the object instance by calling the pre-encapsulated parameter interface of the object instance, obtaining object parameters, wherein the object parameters include image object parameters and / or point cloud object parameters;

[0008] Inputting the object parameters and the object instance into the neural network at the same time by calling the function processing interface, pre-processing the to-be-processed data in the neural network by using the algorithm inference library, obtaining pre-processed data, and outputting the pre-processed data by calling the data output interface;

[0009] Optimizing the pre-processed data by using the algorithm configuration file and the algorithm inference library, obtaining processed data;

[0010] Post-processing the processed data according to the user pre-defined parameters, obtaining standard processed data;

[0011] The to-be-processed data is obtained through a sensor, and the to-be-processed data includes image data and / or point cloud data, and the object instance includes an image object instance and / or a point cloud object instance.

[0012] In a second aspect, an embodiment of the present application provides a data processing apparatus based on a neural network, which comprises:

[0013] a data definition module configured to perform data definition on to-be-processed data and create an object instance, wherein the to-be-processed data includes image data and / or point cloud data, and the object instance includes an image object instance and / or a point cloud object instance;

[0014] a parameter setting module configured to perform parameter setting on the object instance by calling a parameter interface pre-encapsulated by the object instance, and obtain object parameters, wherein the object parameters include image object parameters and / or point cloud object parameters;

[0015] a network processing module configured to input the object parameters and the object instance into a neural network at the same time by calling a function processing interface, and perform preprocessing on the to-be-processed data by the neural network to obtain preprocessed data, and output the preprocessed data by calling a data output interface;

[0016] an algorithm running module configured to perform optimization processing on the preprocessed data by the neural network by using an algorithm configuration file and an algorithm inference library, and obtain processed data;

[0017] a data post-processing module configured to perform post-processing on the processed data according to user pre-defined parameters, and obtain standard processed data.

[0018] In a third aspect, an embodiment of the present application provides a general neural network data processing platform, which comprises a data preprocessing layer and a neural network running layer, wherein the data preprocessing layer and the neural network running layer are configured to execute the data processing method based on the neural network in the first aspect.

[0019] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores computer instructions, and the computer instructions are executed to make a computer execute the data processing method based on the neural network in the first aspect.

[0020] In a fifth aspect, an embodiment of the present application provides a computer device, which comprises a processor and a memory, and the memory stores computer instructions, and the computer instructions are executed by the processor to implement the data processing method based on the neural network in the first aspect.

[0021] This application embodiment configures a dedicated layer in the neural network via an interface call to perform data preprocessing; it then uses an algorithm inference library to infer the configured information into the neural network and run the network. This solution requires no algorithm conversion, is simple and efficient, and saves time in actual vehicle deployment. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description exemplarily illustrate some embodiments of this application.

[0023] Figure 1 This is a schematic diagram of a specific implementation of a data processing method based on a neural network according to this application;

[0024] Figure 2 This is a schematic diagram of a specific embodiment of a data processing device based on a neural network according to this application.

[0025] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0026] The preferred embodiments of this application will now be described in detail with reference to the accompanying drawings, so that the advantages and features of this application can be more easily understood by those skilled in the art, thereby providing a clearer and more definite definition of the scope of protection of this application.

[0027] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0028] There are many algorithms for realizing different functions in intelligent driving, assisted driving or autonomous driving vehicles. When delivering a Python version algorithm, it takes a lot of time to convert it into a C++ version algorithm by engineering, because engineers are not very familiar with data processing of graphics, graphics computers, etc. Many problems will be encountered in the process of writing algorithm logic using C++, thereby causing a long time to be spent in the process from algorithm research and development to engineering landing and actual deployment. Since Python language is simple, provides various tools, is more convenient, and is closer to actual development, algorithm researchers are more inclined to use Python development. The main purpose of the present scheme is to run a data processing method in a neural network to reduce the conversion process from algorithm to engineering, avoid time consumption, communication cost and switching cost.

[0029] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. The specific embodiments described below can be combined with each other to form new embodiments. For the same or similar ideas or processes described in one embodiment, they can not be described again in other certain embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0030] Figure 1 A specific embodiment of a neural network-based data processing method of the present application is shown.

[0031] In Figure 1 In the specific embodiment shown, a neural network-based data processing method mainly includes:

[0032] Step S101, data definition is performed on the data to be processed to create an object instance; wherein the data to be processed includes image data and / or point cloud data, and the object instance includes an image object instance and / or a point cloud object instance.

[0033] In the present embodiment, the types of data to be processed collected by different types of sensors will also be different. For image data, the present scheme encapsulates the algorithm regular operation written in C++ into an image processing component, and for point cloud data, the present scheme encapsulates the algorithm regular operation written in C++ into a point cloud processing component. By calling a parameter interface, a user can customize a data object instance, which is simple to use.

[0034] In a specific embodiment of the present application, the data definition of the to-be-processed data creates an object instance, including: in the case that the to-be-processed data is image data, defining one or more of the shape, format, pixel, affine matrix, custom transformation and image storage device type of the source image and / or target image, and setting the image region of interest and padding information, creating an image object instance; in the case that the to-be-processed data is point cloud data, defining one or more of the length, channel and point cloud storage device type of the point cloud data, creating a point cloud object instance.

[0035] In the present embodiment, for image data, the present scheme supports data formats including bgr format, gray format, nv21 format, yv12 format and nv12 format, etc. in the process of creating an image object instance; and supports device types including CPU, GPU and NPU. The defined affine matrix can convert the input source image into a target image; and the format and data of the target image are defined so that the algorithm model can process the image into the image expected by the user. For point cloud data, the present scheme obtains the input point cloud data, actual data length and channel number according to the defined point cloud data and its data length, and creates a point cloud object instance in the process of creating a point cloud object instance. The object instance created by the present scheme can obtain different kinds of information of the to-be-processed data, and can also obtain other parameter information by calling an interface, which is convenient and easy for the user to use, and also provides image definition and format conversion, point cloud definition and length size, etc.

[0036] In Figure 1 In the specific embodiment shown in the specific embodiment, a neural network-based data processing method further includes:

[0037] In step S102, the object instance is parameterized by calling the parameter interface of the pre-encapsulated object instance, and an object parameter is obtained, wherein the object parameter includes an image object parameter and / or a point cloud object parameter.

[0038] In the present embodiment, the present scheme encapsulates the parameter setting process in actual operation into a component in which the user can set parameters, so that the user can set the size of the input source image and the size of the target image expected by the user through the component, and set the information required for converting the source image into the target image, such as normalization information, padding information and image channel format. By calling the parameter interface, the user can set various forms of parameters, and can also obtain information called from other components, which is convenient and flexible to use.

[0039] In one specific embodiment of the present application, the object instance is parameterized by calling the parameter interface pre-encapsulated by the object instance to obtain object parameters, including: when the object instance is an image object instance, the image parameter interface in the parameter interface pre-encapsulated by the image object instance is called to set the image processing parameters of the source image and / or the target image to obtain image object parameters, wherein the image processing parameters include one or more of normalization information, padding information, region of interest processor, size of the source image, size of the target image, channel format, pixel number information, and byte information; when the object instance is a point cloud object instance, the point cloud parameter interface in the parameter interface pre-encapsulated by the point cloud object instance is called to set the point cloud processing parameters of the point cloud data to obtain point cloud object parameters, wherein the point cloud processing parameters include one or more of data length, data type, and channel number of the point cloud data.

[0040] In the present embodiment, the image data and the point cloud data are different, and the parameters set thereby are also different. By calling the parameter interface of the image data or the point cloud data, the parameters set by the user are simple, clear, and easy to understand the design logic.

[0041] It should be noted that the channel format refers to the channel format of the transmitted image. Since the algorithm has specific requirements for the spatial arrangement of the image, the pixel arrangement of the image needs to be set.

[0042] In one specific example of the present application, the affine matrix performs image transformation on the source image to convert the source image into the target image, wherein the image transformation includes image scaling, cropping, flipping, and / or translation. The affine matrix is a 3x3 parameter matrix, and the parameters of the horizontal and vertical coordinate axes can be specified. The set affine matrix can perform image scaling, cropping, flipping, and / or translation, etc. linear transformation to deform the image. The normalization information is set according to the user-defined parameters, which compresses the pixel space of the source image; the image pixel space actually refers to the space range in which the pixel value is located. Normalizing the image makes image data processing more convenient. When the pixel value in the source image is distributed in the space range of 0-127, the user hopes to compress the space range to 0-1, and the algorithm model will process the image data according to the set normalization information, so that the pixel space of the entire source image is compressed.

[0043] In Figure 1 In the specific embodiment shown in the specific embodiment, a neural network-based data processing method further includes:

[0044] In step S103, the object parameter and the object instance are input into the neural network by calling the function processing interface, the algorithm inference library is used to preprocess the to-be-processed data in the neural network, and the preprocessing data is obtained and output by calling the data output interface.

[0045] In the embodiment, in the neural network, the algorithm inference library performs inference on the image object instance or the corresponding point cloud object instance according to the image object parameter or the point cloud object parameter by calling the function processing interface, performs the preprocessing process, and outputs the preprocessing data to other calculation layers in the neural network for other processing by calling the data output interface. The algorithm inference library has better running performance in the neural network.

[0046] In one specific embodiment of the present application, the object parameter and the object instance are input into the neural network by calling the function processing interface, including: the image information or the point cloud information of the object instance is set according to the object parameter by using the function processing component by calling the function processing interface, and the object parameter and the image information or the point cloud information of the object instance are input into the neural network at the same time.

[0047] In the embodiment, the image information or the point cloud information of the corresponding image object instance or point cloud object instance and the transformation matrix are set according to the image object parameter or the point cloud object parameter by using the function processing component by calling the function processing interface, and are input into the neural network at the same time to create a layer of neural network for performing preprocessing of the image object instance or the point cloud object instance. Various parameter interfaces and function processing interfaces are written by algorithm developers using C++ language, and the corresponding interfaces are automatically called by the user to assist the user to complete the custom parameter setting and improve the user experience.

[0048] In one specific embodiment of the present application, the algorithm inference library is used to preprocess the to-be-processed data in the neural network to obtain preprocessing data, including: the neural network is run by using the algorithm inference library by calling the function processing interface, and the to-be-processed data is preprocessed to obtain the preprocessing data.

[0049] In the embodiment, the function processing interface corresponding to the algorithm inference library is called, and the algorithm inference library runs the neural network to complete the preprocessing of the image data or the point cloud data.

[0050] In Figure 1 In the specific embodiment shown in the specific embodiment, a data processing method based on a neural network further includes:

[0051] In step S104, the preprocessing data is processed by using the algorithm configuration file and the algorithm inference library, and the processing data is obtained.

[0052] In this embodiment, the algorithm configuration file contains the relationship between each model and the set processing parameters, which can meet most of the user's functional requirements. The algorithm configuration file is equivalent to an algorithm model, and the algorithm inference library loads and runs the algorithm configuration file to generate each function unit according to the processing parameters in the algorithm configuration file, to process data.

[0053] In one specific embodiment of the present application, the neural network optimizes the preprocessed data using the algorithm configuration file and the algorithm inference library to obtain the processed data, including: setting the processing parameters according to the user's customized functions to create an algorithm configuration file; using one or more algorithm configuration files to set the data processing logic of the algorithm configuration file to obtain an algorithm inference library; calling the algorithm configuration interface, the algorithm inference library runs the algorithm configuration file to optimize the preprocessed data to obtain the processed data.

[0054] In this embodiment, according to the user's customized functions for data, the product requirements are determined, and the algorithm development personnel configure various algorithm models into corresponding algorithm configuration files, and obtain the algorithm inference library according to the data processing logic in the algorithm configuration file, to run the algorithm configuration file to optimize the preprocessed data.

[0055] In Figure 1 In the specific embodiment shown in the specific embodiment, a neural network-based data processing method further includes:

[0056] Step S105, according to the user pre-defined parameters, the processed data is post-processed to obtain standard processing data.

[0057] In this embodiment, the result of the algorithm model processing data is not necessarily the result that the user ultimately expects, and this part still needs to write algorithm logic to calculate the result of the algorithm model processing data, so as to obtain the result that the user ultimately expects.

[0058] In one specific example of the present application, when detecting pedestrians in an image, the user expects the result to be the position coordinate information of the pedestrian, but the result output by the algorithm model after processing is not necessarily in the data format that the user wants. At this time, the data output by the algorithm model, i.e. the processed data, needs to be converted to correct the result and obtain standard processing data.

[0059] The application defines data of sensor collected data to be processed, and creates object instances. For image data, affine matrix, source image data, shape of source image data, target image data, format of target image data, shape of target image, user-defined data and transformation method are defined, and image data is set, which includes height and width of image and bit flag, data format, and image height, width and pixel depth can be obtained, and set region of interest, padding information, affine matrix and image data pool can be obtained, the image data pool contains data in different formats, and an image object instance is created according to input image data; for point cloud data, point cloud data, data length and channel number can be obtained, and a point cloud object instance is created according to input point cloud data. Object parameters are obtained by calling corresponding parameter interfaces to set parameters of object instances. Different parameter settings can be made for image object instances and point cloud object instances through interfaces corresponding to each parameter. Object parameters and object instances are synchronized to a neural network by calling function processing interfaces, preprocessing of data to be processed is performed in the neural network by using an algorithm inference library, preprocessed data is output by calling a data output interface. Image information or point cloud information and conversion matrix of image object instances or point cloud object instances are set by function processing components according to image object parameters or point cloud object parameters by calling function processing interfaces, and the image information or point cloud information and the conversion matrix are input to the neural network at the same time to create a layer of neural network, which is used to preprocess image object instances or point cloud object instances. The neural network optimizes preprocessed data by using an algorithm configuration file and an algorithm inference library to obtain processed data. According to the customized function of data required by the user, product requirements are determined, algorithm developers configure various algorithm models to obtain corresponding algorithm configuration files, and an algorithm inference library is obtained according to algorithm logic in the algorithm configuration file, so as to run the algorithm configuration file to optimize preprocessed data. The processed data is post-processed according to the data parameters set by the user to obtain standard processed data.

[0060] Figure 2 A specific embodiment of a data processing apparatus based on a neural network is shown.

[0061] In Figure 2 In the specific embodiment shown, a data processing apparatus based on a neural network mainly includes:

[0062] A data definition module 201 is configured to define data of data to be processed and create object instances, wherein the data to be processed includes image data and / or point cloud data, and the object instances include image object instances and / or point cloud object instances.

[0063] The parameter setting module 202 is used to set parameters of an object instance by calling the parameter interface pre-encapsulated by the object instance to obtain object parameters, wherein the object parameters include image object parameters and, or point cloud object parameters.

[0064] The network processing module 203 is used to input object parameters and object instances into the neural network simultaneously by calling the function processing interface. The neural network preprocesses the data to be processed, obtains preprocessed data, and outputs the preprocessed data by calling the data output interface.

[0065] The algorithm execution module 204 is used to optimize the preprocessed data using the algorithm configuration file and algorithm inference library, and the neural network to obtain the processed data.

[0066] The data post-processing module 205 is used to post-process the processed data according to user-defined parameters to obtain standard processed data.

[0067] In this embodiment, various types of parameters are set by calling various parameter interfaces, and the user-defined parameters are simultaneously input into the neural network using function processing components to create a new neural network layer for data preprocessing. This includes user-defined parameters that can meet most functional requirements. In the newly created neural network, the preprocessed data is processed according to the set parameters to obtain preprocessed data. By inputting the preprocessed data into other computational layers of the neural network for further processing, the processed data can be obtained. Finally, the processed data is standardized and verified to obtain standard processed data.

[0068] The present application provides a data processing device based on a neural network, which can be used to execute the data processing method based on a neural network described in any of the above embodiments. Its implementation principle and technical effect are similar, and will not be repeated here.

[0069] In one specific embodiment of this application, the functional modules of the neural network-based data processing device of this application may be directly in hardware, in software modules executed by a processor, or in a combination of both.

[0070] Software modules may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in this art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium.

[0071] The processor can be a central processing unit (CPU), a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof. The general purpose processor can be a microprocessor, but in the alternative, the processor can be any conventional processor, controller, microcontroller, or state machine. The processor can also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. The storage medium can be integral with the processor. The processor and the storage medium can reside in an ASIC. The ASIC can reside in a user terminal. In the alternative, the processor and the storage medium can reside as discrete components in a user terminal.

[0072] In another specific embodiment of the present application, a general neural network processing data platform includes a data preprocessing layer and a neural network running layer, wherein the data preprocessing layer and the neural network running layer are configured to perform the neural network-based data processing method of any embodiment.

[0073] In another specific embodiment of the present application, a computer readable storage medium stores computer instructions, which when executed cause a computer to perform the neural network-based data processing method of any embodiment.

[0074] In another specific embodiment of the present application, a computer device includes a processor and a memory, wherein the memory stores computer instructions, which when executed by the processor implement the neural network-based data processing method of any embodiment.

[0075] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. For example, the embodiments of the apparatus described above are merely schematic, and the division of the units is merely logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0076] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0077] The above description is merely some embodiments of the present application, and is not intended to limit the patent scope of the present application. Any equivalent structure transformation made by using the content of the present application specification and drawings, or directly or indirectly applied to other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A data processing method based on neural networks, characterized in that, include: Define the data to be processed and create object instances; By calling the parameter interface pre-encapsulated by the object instance, the parameters of the object instance are set to obtain the object parameters, wherein the object parameters include image object parameters and / or point cloud object parameters; By calling the function processing interface, the object parameters and the object instance are simultaneously input into the neural network. In the neural network, the data to be processed is preprocessed using the algorithm inference library to obtain preprocessed data. The preprocessed data is then output by calling the data output interface. Specifically, the neural network is run by the algorithm inference library to run inference on the image object instance or the corresponding point cloud object instance according to the image object parameters or point cloud object parameters, execute the preprocessing process, and output the preprocessed data to other computing layers in the neural network for further processing by calling the data output interface. Using the algorithm configuration file and the algorithm inference library, the neural network optimizes the preprocessed data to obtain processed data. The algorithm inference library is obtained by using one or more algorithm configuration files and setting the data processing logic of the algorithm configuration files. By calling the algorithm configuration interface, the algorithm inference library runs the algorithm configuration file to optimize the preprocessed data and obtain processed data. The processed data is post-processed according to user-defined parameters to obtain standard processed data; The data to be processed is acquired through sensors, and includes image data and / or point cloud data. The object instance includes image object instance and / or point cloud object instance.

2. The data processing method based on neural networks as described in claim 1, characterized in that, The step of simultaneously inputting the object parameters and the object instance into the neural network by calling the function processing interface includes: By calling the function processing interface, the function processing component sets the image information or point cloud information of the object instance according to the object parameters, and simultaneously inputs the object parameters and the image information or point cloud information of the object instance into the neural network.

3. The data processing method based on neural networks as described in claim 2, characterized in that, The process of preprocessing the data to be processed using an algorithm inference library in the neural network to obtain preprocessed data includes: By calling the function processing interface, the neural network is run using the algorithm inference library to preprocess the data to be processed, and the preprocessed data is obtained.

4. The data processing method based on neural networks as described in claim 1, characterized in that, The neural network optimizes the preprocessed data using the algorithm configuration file and the algorithm inference library to obtain processed data, including: Based on the user's customized functions, the processing parameters are set, and the algorithm configuration file is created; By using one or more of the aforementioned algorithm configuration files, the data processing logic of the algorithm configuration files is set to obtain the algorithm inference library; By calling the algorithm configuration interface, the algorithm inference library runs the algorithm configuration file to optimize the preprocessed data and obtain the processed data.

5. The data processing method based on neural networks as described in claim 1, characterized in that, The process of defining the data to be processed and creating object instances includes: Given that the data to be processed is the image data, define one or more of the following: shape, format, pixels, affine matrix, custom transformation, and image storage device type of the source image and / or target image, and set the region of interest and padding information to create the image object instance; Given that the data to be processed is the point cloud data, define one or more of the following: the length, channels, and point cloud storage device type of the point cloud data, and create the point cloud object instance.

6. The data processing method based on neural networks as described in claim 5, characterized in that, The step of setting parameters for the object instance by calling the pre-encapsulated parameter interface of the object instance to obtain object parameters includes: When the object instance is the image object instance, the image processing parameters of the source image and / or the target image are set by calling the image parameter interface in the parameter interface pre-encapsulated by the image object instance, and the image object parameters are obtained. The image processing parameters include one or more of the following: normalization information, padding information, region of interest processor, size of the source image, size of the target image, channel format, number of pixels, and byte information. When the object instance is the point cloud object instance, the point cloud processing parameters of the point cloud data are set by calling the point cloud parameter interface in the parameter interface pre-encapsulated by the point cloud object instance, and the point cloud object parameters are obtained. The point cloud processing parameters include one or more of the data length, data type, and number of channels of the point cloud data.

7. A data processing device based on a neural network, characterized in that, include: The data definition module is used to define the data to be processed and create object instances, wherein the data to be processed includes image data and / or point cloud data, and the object instances include image object instances and / or point cloud object instances. The parameter setting module is used to set parameters of the object instance by calling the parameter interface pre-encapsulated by the object instance to obtain object parameters, wherein the object parameters include image object parameters and / or point cloud object parameters; The network processing module is used to simultaneously input the object parameters and the object instance into the neural network by calling the function processing interface. The neural network preprocesses the data to be processed to obtain preprocessed data and outputs the preprocessed data by calling the data output interface. The neural network is run by the algorithm inference library to run inference on the image object instance or the corresponding point cloud object instance according to the image object parameters or point cloud object parameters, executes the preprocessing process, and outputs the preprocessed data to other computing layers in the neural network for further processing by calling the data output interface. The algorithm execution module is used to optimize the preprocessed data using the algorithm configuration file and the algorithm inference library to obtain processed data. The algorithm inference library is obtained by using one or more algorithm configuration files and setting the data processing logic of the algorithm configuration files. By calling the algorithm configuration interface, the algorithm inference library runs the algorithm configuration file to optimize the preprocessed data and obtain processed data. The data post-processing module is used to post-process the processed data according to user-defined parameters to obtain standard processed data.

8. A computer-readable storage medium storing computer instructions, characterized in that, When the computer instructions are executed, the computer performs the data processing method based on a neural network as described in any one of claims 1-6.

9. A computer device comprising a processor and a memory, the memory storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the neural network-based data processing method as described in any one of claims 1-6.

Images