CN108573307A

CN108573307A - A kind of method and terminal of processing neural network model file

Info

Publication number: CN108573307A
Application number: CN201810179892.4A
Authority: CN
Inventors: 况辉
Original assignee: Vivo Mobile Communication Co Ltd
Current assignee: Vivo Mobile Communication Co Ltd
Priority date: 2018-03-05
Filing date: 2018-03-05
Publication date: 2018-09-25

Abstract

The embodiment of the present invention provides a kind of method and terminal of processing neural network model file, is applied to field of communication technology, can solve the problems, such as that the process of terminal acquisition neural network model file takes.This method includes：In the case where terminal shows that interface is arranged in neural network model, terminal receives the first input of user, and the first input is the operation that user triggers the first model data of input, and the first model data includes at least the first species parameter；In response to the first input, terminal obtains the first nerves network model file of the first species parameter instruction according to the first model data；Wherein, the first species parameter is used to indicate the kind of document of first nerves network model file, and this document type includes the classification of the classification of image recognition, the classification of speech recognition and text classification.

Description

A kind of method and terminal of processing neural network model file

Technical field

The present embodiments relate to field of communication technology more particularly to it is a kind of processing neural network model file method and Terminal.

Background technology

With the development of the communication technology, neural network (Neural Networks, NNs), such as convolutional neural networks The algorithm of (Convolutional Neural Networks, CNN) is such as applied to image recognition, voice using more and more extensive In the scenes such as identification and text classification.

In the prior art, when user needs using a neural network algorithm, it usually needs specific program code is write, The neural network model file of specific function can just be obtained.

The problem is that the process of user written program code is usually relatively complex, so that terminal obtains the journey Sequence code corresponds to neural network model file processes and takes.

Invention content

The embodiment of the present invention provides a kind of method and terminal of processing neural network model file, and god is obtained to solve terminal The time-consuming problem of process through network model file.

In order to solve the above-mentioned technical problem, the embodiment of the present invention is realized in：

In a first aspect, providing a kind of method of processing neural network model file, this method includes：Nerve is shown in terminal In the case that interface is arranged in network model, terminal receives the first input of user, and the first input is that user triggers the first mould of input The operation of type data, the first model data include at least the first species parameter；In response to the first input, terminal is according to the first model Data obtain the first nerves network model file of the first species parameter instruction；Wherein, the first species parameter is used to indicate first The kind of document of neural network model file, kind of document include the classification of image recognition, the classification of speech recognition and text The classification of classification.

Second aspect, the embodiment of the present invention additionally provide a kind of method of processing neural network model file, this method packet It includes：The first model data that server receiving terminal is sent, the first model data include at least the first species parameter；Server root According to the first model data, the first nerves network model file of the first species parameter instruction is obtained；Server sends the to terminal One neural network model file；Wherein, the first species parameter is used to indicate the kind of document of first nerves network model file, should Kind of document includes the classification of the classification of image recognition, the classification of speech recognition and text classification.

The third aspect, the embodiment of the present invention additionally provide a kind of terminal, which includes receiving module and acquisition module；It connects Module is received, the first input in the case where terminal shows that interface is arranged in neural network model, receiving user, the first input The operation of the first model data of input is triggered for user, the first model data includes at least the first species parameter；Acquisition module is used The first god of the first species parameter instruction is obtained according to the first model data in the first input received in response to receiving module Through network model file；Wherein, the first species parameter is used to indicate the kind of document of first nerves network model file, file kind Class includes the classification of the classification of image recognition, the classification of speech recognition and text classification.

Fourth aspect, the embodiment of the present invention additionally provide a kind of server, which includes：Receiving module obtains mould Block and sending module；Receiving module, the first model data for receiving terminal transmission, the first model data include at least first Species parameter；Acquisition module, for the first model data for being received according to receiving module, obtain that the first species parameter indicates the One neural network model file；Sending module, for sending the first nerves network model file that acquisition module obtains to terminal； Wherein, the first species parameter is used to indicate the kind of document of first nerves network model file, and kind of document includes image recognition Classification, the classification of speech recognition and the classification of text classification.

5th aspect an embodiment of the present invention provides a kind of terminal, including processor, memory and is stored in the memory Computer program that is upper and can running on the processor, realizes such as first aspect when which is executed by the processor Processing neural network model file method the step of.

6th aspect an embodiment of the present invention provides a kind of server, including processor, memory and is stored in the storage On device and the computer program that can run on the processor, such as second party is realized when which is executed by the processor The step of method of the processing neural network model file in face.

7th aspect, an embodiment of the present invention provides a kind of computer readable storage medium, the computer-readable storage mediums Computer program is stored in matter, and the processing neural network model such as first aspect is realized when which is executed by processor The step of method of file.

Eighth aspect, an embodiment of the present invention provides a kind of computer readable storage medium, the computer-readable storage mediums Computer program is stored in matter, and the processing neural network model such as second aspect is realized when which is executed by processor The step of method of file.

In the embodiment of the present invention, terminal can provide a user neural network model setting interface, to support user to input The model data of the neural network model file of demand so that terminal can obtain the neural network model according to the model data File.That is, user need not write complicated program code can be so that terminal obtains the neural network mould of demand Type file.In this way, can solve the problems, such as that the process of the neural network model file of terminal acquisition demand takes.Also, it reduces Operation of the user when using terminal obtains the neural network model file of demand, can improve the user in the operating process Experience.

Description of the drawings

Fig. 1 is a kind of configuration diagram of possible Android operation system provided in an embodiment of the present invention；

Fig. 2 is a kind of flow diagram of the method for processing neural network model file provided in an embodiment of the present invention；

Fig. 3 is the flow diagram of another method for handling neural network model file provided in an embodiment of the present invention；

Fig. 4 is the flow diagram of another method for handling neural network model file provided in an embodiment of the present invention；

Fig. 5 is the flow diagram of another method for handling neural network model file provided in an embodiment of the present invention；

Fig. 6 is a kind of structural schematic diagram of possible terminal provided in an embodiment of the present invention；

Fig. 7 is the structural schematic diagram of alternatively possible terminal provided in an embodiment of the present invention；

Fig. 8 is a kind of structural schematic diagram of possible server provided in an embodiment of the present invention；

Fig. 9 is a kind of hardware architecture diagram of terminal provided in an embodiment of the present invention；

Figure 10 is a kind of hardware architecture diagram of server provided in an embodiment of the present invention.

Specific implementation mode

Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation describes, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair Embodiment in bright, the every other implementation that those of ordinary skill in the art are obtained without creative efforts Example, shall fall within the protection scope of the present invention.

It should be noted that "/" herein indicate or the meaning, for example, A/B can indicate A or B；Herein "and/or" is only a kind of incidence relation of description affiliated partner, indicates may exist three kinds of relationships, for example, A and/or B, it can To indicate：Individualism A exists simultaneously A and B, these three situations of individualism B." multiple " refer to two or more.

It should be noted that in the embodiment of the present invention, " illustrative " or " such as " etc. words make example, example for indicating Card or explanation.Be described as in the embodiment of the present invention " illustrative " or " such as " any embodiment or design scheme do not answer It is interpreted than other embodiments or design scheme more preferably or more advantage.Specifically, " illustrative " or " example are used Such as " word is intended to that related notion is presented in specific ways.

The method and terminal of processing neural network model file provided in an embodiment of the present invention, user need not write complexity Program code, but interface and terminal interaction are arranged by the neural network model of terminal, the nerve of demand can be obtained Network model file.In this way, can solve the problems, such as that the process of the neural network model file of terminal acquisition demand takes.

Terminal in the embodiment of the present invention can be the terminal with operating system.The operating system can be Android (Android) operating system can be ios operating systems, can also be other possible operating systems, the embodiment of the present invention is not Make specific limit.

Below by taking Android operation system as an example, processing neural network model file provided in an embodiment of the present invention is introduced The software environment applied of method.

As shown in Figure 1, being a kind of configuration diagram of possible Android operation system provided in an embodiment of the present invention.Scheming In 1, the framework of Android operation system includes 4 layers, respectively：Application layer, application framework layer, system operation library layer and Inner nuclear layer (is specifically as follows Linux inner core).

Wherein, application layer includes each application program (including system application and in Android operation system Tripartite's application program).

Application framework layer is the frame of application program, and developer can be in the exploitation of the frame in accordance with application program In the case of principle, some application programs are developed based on application framework layer.

System operation library layer includes library (also referred to as system library) and Android operation system running environment.Library is mainly Android behaviour All kinds of resources needed for it are provided as system.Android operation system running environment is used to provide software loop for Android operation system Border.

Inner nuclear layer is the operating system layer of Android operation system, belongs to the bottom of Android operation system software level.It is interior Stratum nucleare provides core system service and hardware-related driver based on linux kernel for Android operation system.

By taking Android operation system as an example, in the embodiment of the present invention, developer can be based on above-mentioned Android as shown in Figure 1 The software of the method for processing neural network model file provided in an embodiment of the present invention is realized in the system architecture of operating system, exploitation Program, so that the method for the processing neural network model file can be run based on Android operation system as shown in Figure 1. I.e. processor or terminal device can realize that the embodiment of the present invention is carried by running the software program in Android operation system The method of the processing neural network model file of confession.

In a kind of specific embodiment of the present invention, pass through the stream of the method for processing neural network model file shown in Fig. 2 The method of processing neural network model file provided in an embodiment of the present invention is described in detail in journey figure.Wherein, although in side Shown in method flow chart it is provided in an embodiment of the present invention processing neural network model file method logical order, but It in some cases, can be with the steps shown or described are performed in an order that is different from the one herein.For example, processing shown in Figure 2 The method of neural network model file may include S201 and S202：

S201, in the case where terminal shows that interface is arranged in neural network model, terminal receives the first input of user, the One input is the operation that user triggers the first model data of input, and the first model data includes at least the first species parameter.

Wherein, the first species parameter is used to indicate the kind of document of first nerves network model file, this document type packet Include the classification of the classification of image recognition, the classification of speech recognition and text classification.That is, provided in an embodiment of the present invention Neural network model file can identify the neural network algorithm of the classifications such as class, speech recognition class or text classification with correspondence image.

Optionally, above-mentioned image recognition class can specifically include character image identification, animal painting identification, the knowledge of still life image Not, hand-written script identification (such as Handwritten Digit Recognition, handwritten Kanji recognition or hand-written English letter recognition).

It is understood that the neural network model setting interface that above-mentioned terminal is shown can be applied by one in terminal Program provides, alternatively, what neural network model setting interface can be provided by a webpage in the browser of terminal, this hair Bright embodiment is not especially limited this.

Wherein, above-mentioned neural network model setting interface may include the region that the first input is carried out for user, for example, Option, control or the input frame etc. of the first input are carried out for user.Specifically, above-mentioned neural network model setting interface can be with It include the region that the first model data is inputted for user.

In general, there is terminal touch screen, the touch screen can be used for receiving the input of user, and in response to the input To execute the operation of input instruction.

Optionally, it is above-mentioned first input be user terminal neural network model be arranged interface on touch-screen input or Fingerprint inputs.Wherein, above-mentioned touch-screen input be user the pressing input of the touch screen of terminal, long-press are inputted, slidably inputed, point Hit input and the input (input of the user near touch screen) etc. that suspends.Fingerprint input is Fingerprint Identification Unit of the user to terminal Sliding fingerprint, long-press fingerprint click fingerprint and double-click fingerprint etc..Gravity input is shaking of the user to terminal specific direction, spy Determine the inputs such as the shaking of number.Key-press input corresponds to user's clicking to buttons such as the power keys, volume key, Home key of terminal The inputs such as input, long-press input and combination button input are double-clicked in input.

Wherein, terminal provided in an embodiment of the present invention receives the of user on the neural network model setting interface of terminal The mode of one input can be one kind in above-mentioned multiple input modes, and the embodiment of the present invention is not construed as limiting this.

S202, it is inputted in response to first, terminal obtains the first god of the first species parameter instruction according to the first model data Through network model file.

Wherein, above-mentioned first model data is the corresponding model data of first nerves network model file that user needs.

It is understood that a neural network model file can be a neural network algorithm corresponding program generation Code, the program code can execute in terminal, realize the corresponding function of the program code.

Illustratively, the first nerves network model file of the first species parameter instruction in the first model data, can be with For the neural network model file of image recognition type.Wherein, the corresponding program code of neural network model file is in terminal When upper execution, the identification to image, such as identification character image or identification animal painting may be implemented in terminal.

It should be noted that the method for processing neural network model file provided in an embodiment of the present invention, terminal can be to User provides neural network model and interface is arranged, to support user to input the model data of the neural network model file of demand, Terminal is allow to obtain the neural network model file according to the model data.That is, user need not write complexity Program code can so that terminal obtain demand neural network model file.In this way, can solve terminal obtains demand Neural network model file the time-consuming problem of process.Also, reduce the nerve net that user obtains demand in using terminal Operation when network model file can improve the user experience in the operating process.

In one possible implementation, the method for processing neural network model file provided in an embodiment of the present invention, It can also include the server for the neural network model that may be used to provide user demand.Specifically, as shown in figure 3, being this hair The flow diagram of the method for another processing neural network model file that bright embodiment provides.In method shown in Fig. 3, on It states S202 and could alternatively be S202a and S202b：

S202a, terminal to server send the first model data, and the first model data refers to for obtaining the first species parameter The first nerves network model file shown.

Wherein, when providing the interface of the neural network model setting human-computer interaction that interface is realized in above-mentioned terminal, terminal connects After receiving the first model data input by user, first nerves network model can be obtained from server side according to the model data File.

S202b, terminal receive the first nerves network model file that server is sent.

Wherein, before above-mentioned server sends the first nerves network model file, can server side obtain this One neural network model file.

It should be noted that the method for processing neural network model file provided in an embodiment of the present invention, terminal can root According to model data input by user, the corresponding neural network model file of the model data is obtained from server.In this way, being conducive to Solve the problems, such as that the neural network model file processes of terminal acquisition user demand take.

In one possible implementation, above-mentioned first nerves network model file includes the of at least three network layers One information, the first information of a network layer include the second species parameter of a network layer, the first function of network layer The sequence identification of parameter and a network layer.

Optionally, second species parameter is used to indicate the functional type of network layer, which includes input layer, convolution Layer, pond layer, full articulamentum and output layer.

In general, a neural network model file generally include at least three network layers be an input layer, one Output layer and at least one functional layer, a functional layer can be any one of above-mentioned convolutional layer, pond layer or full articulamentum.

Illustratively, the second species parameter of a neural network model file (being denoted as neural network model file 1) is Handwritten Digit Recognition in above-mentioned image recognition class.Neural network model file 1 includes 6 network layers.Wherein, first network Second species parameter in the first information of layer (being denoted as network layer 1) is " input layer ", the first functional parameter is " 32*32 ", suitable Sequence is identified as " 1 "；That is the pending image of the input layer of the neural network model file 1 is the picture element matrix that size is 32*32. Second species parameter in the first information of second network layer (being denoted as network layer 2) is " convolutional layer ", the first functional parameter is 6*5*5, sequence identification are " 2 "；I.e. network layer 2 includes the convolution kernel that 6 sizes are 5*5.Third network layer (is denoted as network layer 3) the second species parameter in the first information is " pond layer ", the first functional parameter is 6*2*2, and sequence identification is " 3 "；That is net Network layers 2 include the matrix with pond function that 6 specifications are 2*2.The first information of 4th network layer (being denoted as network layer 4) In second species parameter be " full articulamentum ", the first functional parameter be 120, sequence identification is " 4 "；That is the matrix of network layer 4 Size be 120 (i.e. 1*120) pixels matrix.Second in the first information of 5th network layer (being denoted as network layer 5) Class parameter is " full articulamentum ", the first functional parameter is 100, and sequence identification is " 5 "；I.e. the size of the matrix of network layer 5 is 100 The matrix (also referred to as operator) of (i.e. 1*100) pixel.Second in the first information of 6th network layer (being denoted as network layer 6) Class parameter is " output layer ", the first functional parameter is 10, and sequence identification is " 6 "；I.e. the size of the matrix of network layer 6 is 10 (i.e. 1*10) the matrix of pixel.

It is understood that convolutional layer, pond layer, the full articulamentum in a neural network model file all can be more It is a.

It should be noted that the input layer user in a neural network model file receives outside to the neural network mould The input of type file.Convolutional layer in one neural network model file can be by one or more of convolution kernels to preceding The output result (picture element matrix of such as one image) of one network layer is filtered, noise reduction process.One neural network model text Pond layer in part can reduce previous network layer output as a result, as an image picture element matrix size, to reduce Parameter in the neural network model file, to facilitate terminal to execute the neural network model file.One neural network model All nodes of all nodes and previous network layer in full articulamentum in file are connected with each other, and be can be used for completing classification and are appointed Business.Output layer in one neural network model file is used to export the handling result of the neural network model file.

In one possible implementation, the method for processing neural network model file provided in an embodiment of the present invention, First model data further includes the second information of at least one network layer at least three network layers, the instruction of at least three network layers Practice the test data set of data set and at least three network layers；Wherein, the second information of a network layer include it is following at least It is a kind of：The sequence identification of the second functional parameter and a network layer of the second species parameter of one network layer, network layer.

It should be noted that in the method for processing neural network model file provided in an embodiment of the present invention, user can be with One or more for generating neural network model file is inputted in the above-mentioned neural network model setting interface that terminal provides A model data.

Illustratively, for may include in the first model of input each network layer at least three network layers sequence Mark, the sequence identification in the second information of first network layer is " 1 ", the sequence mark in the second information of second network layer It is " 2 " to know, the sequence identification in the second information of third network layer is " 3 ", suitable in the second information of the 4th network layer Sequence be identified as " 4 ", the 5th network layer the second information in sequence identification be " 5 ", in the second information of the 6th network layer Sequence identification be " 6 ".In addition, the second species parameter that user can also input in the second information of second network layer is " convolutional layer ", the second functional parameter are 6*3*3, i.e. second network layer of user's instruction includes the convolution that 6 sizes are 3*3 Core.

It is understood that first nerves network model file provided in an embodiment of the present invention can be by some training Data training simultaneously passes through what some test datas were tested.

Illustratively, the neural network model file 1 obtained in conjunction with above-mentioned terminal, neural network model of the user in terminal It is 6*3*3 that the second functional parameter in the second information of above-mentioned second network layer of interface input, which is arranged,.Above-mentioned server can be with Second functional parameter of second network layer of user's acquisition is trained and is tested, the first of second network layer is obtained Functional parameter, such as the first functional parameter of above-mentioned network layer 2 are 6*5*5.

Optionally, the method for processing neural network model file provided in an embodiment of the present invention, above-mentioned neural network model It may include first area, second area, third region, the fourth region, the 5th region and the 6th region to be arranged in interface；Its In, first area is for inputting the first species parameter, and second area is for inputting each network layer at least three network layers Second species parameter, third region are used to input the second functional parameter of each network layer at least three network layers, the 4th area Domain is used to input the sequence identification of each network layer at least three network layers, and the 5th region is for inputting at least three network layers Training data and the 6th region be used to input the test datas of at least three network layers.Wherein, user can be above-mentioned each A region inputs corresponding model data.

Optionally, neural network model setting provided in an embodiment of the present invention interface can provide a user at least three nets Multiple preset model data of network layers.

Specifically, each above-mentioned region in neural network model setting provided in an embodiment of the present invention interface can wrap Include the controls such as one or more options or input frame.For example, second species parameter in the second information of a network layer, first Functional parameter, sequence identification can correspond to an option respectively.Wherein, which can be used for the number that the manual the input phase of user is answered According to alternatively, selecting a preset model number in one or more preset model data that the option is provided for user from terminal According to.

It should be noted that the method for processing neural network model file provided in an embodiment of the present invention, user can be Multiple model datas for generating neural network model file are inputted in the neural network model setting interface that terminal provides.Such as This so that terminal is according to the neural network model file that the neural network model file that the model data obtains is user demand. It is thus possible to improve user experience when user's using terminal obtains neural network model file.

In one possible implementation, as shown in figure 4, processing neural network model provided in an embodiment of the present invention is literary The method of part can also include S203 and S204 after above-mentioned S202：

It is corresponding to generate first nerves network model file according to the first information of at least three network layers for S203, terminal Three-dimensional visualization model.

Illustratively, terminal can generate one three according to each model parameter in above-mentioned neural network model file 1 Tie up Visualization Model (being denoted as three-dimensional visualization model 1).Wherein, three-dimensional visualization model 1 includes 6 network layers, i.e., above-mentioned net The three-dimensional rendering result of network layers 1- network layers 6.

S204, terminal show three-dimensional visualization model.

Illustratively, an image, the image (i.e. input matrix) are inputted to the network layer 1 of neural network model file 1 After the processing of network layer 1, the matrix (being denoted as matrix 1) of 32*32 pixels composition is obtained；The matrix 1 in network layer 2 by wrapping 6 sizes included are the convolution algorithm of the convolution kernel of 5*5, obtain the matrix (note of 6 28*28 (32-5+1=28) pixels composition For matrix 2)；6 matrixes 2 obtain 6 14*14 after the filter process that 6 sizes that network layer 3 includes are 2*2 The matrix (being denoted as matrix 3) of (28/2=14) pixel composition；It is 1*120 that 6 matrixes 3, which include size by network layer 4, The full connection operation of operator obtains the matrix (being denoted as matrix 4) of 1*120 pixels composition；The matrix 4 passes through network layer 5 Size is the full connection operation of the operator of 1*100, obtains the matrix (being denoted as matrix 5) of 1*100 pixels composition；The matrix 5 passes through The operation of the operator for the 1*10 that network layer 6 includes obtains the matrix of 1*10 pixels composition, i.e. output matrix.Wherein, of the invention For convenience, " matrix " that the functional parameter of some network layers indicates is described as " operator " here for embodiment, essence It is identical.

Wherein, the three-dimensional visualization model 1 that terminal is shown can show each net in above-mentioned neural network model file 1 Connection relation between the structure of network layers and each network layer.Optionally, each matrix node between each network layer Connection can lead to lines expression.Each network layer is each in above-mentioned neural network model file 1 in the three-dimensional visualization model 1 The three-dimensional rendering result of a network layer.

It should be noted that the three-dimensional visualization model that above-mentioned terminal is shown can be dragged by user and be rotated so that User more clearly know each network layer of the corresponding neural network model file of the three-dimensional visualization model structure and Output result after each network layer handles.

Optionally, terminal is after getting the corresponding three-dimensional visualization model of above-mentioned first nerves network model file, Terminal can also receive file to be judged input by user, such as an image for including handwritten numeral " 4 ".Then, terminal can be with It is sent to server to the file to be judged so that the file to be judged is input to first nerves network model file by server In, it obtains the band and judges file by treated output as a result, the number " 4 " gone out such as a terminal recognition.

It is emphasized that the neural network model setting interface that terminal provides can also include waiting judging text for inputting The region of part, to support user to input file to be judged.

Further, server can be by above-mentioned using the first nerves network model file process pending file processes In, the output result after each network layer handles is sent to terminal so that each network layer that terminal is sent according to server Export the output result of result adjustment three-dimensional visualization model.

Further, if user inputs another file to be judged again, terminal can be interacted with server again, to adjust The output result of whole three-dimensional visualization model.

Further, if user inputs a group model data again, terminal can be interacted with server again, be somebody's turn to do with obtaining The neural network model file indicated by second species parameter that model data includes.

Optionally, terminal can also generate and show the corresponding two-dimensional visualization model of neural network model file.It is similar , the description of two-dimensional visualization model is referred to retouch three-dimensional visualization in above-described embodiment in the embodiment of the present invention It states, the embodiment of the present invention is not described in detail this.

Optionally, three-dimensional visualization model is used to show the neural network model file of image recognition class, two-dimensional visualization Model is for showing the other kinds of neural network model file in addition to image recognition class.

It should be noted that in the method for processing neural network model file provided in an embodiment of the present invention, terminal can be with According to the demand of user, such as different model datas input by user, the three-dimensional visualization model that dynamic adjustment terminal is shown.Such as This so that user can intuitively know the structure of each network layer of neural network model file of user demand and this is each Network layer handles are wait the output for judging file as a result, being conducive to improve when user's using terminal handles neural network model file User experience.

In one possible implementation, as shown in figure 5, processing neural network model provided in an embodiment of the present invention is literary The method of part can also include S205-S207 before above-mentioned S202a and S202b：

The first model data that S205, server receiving terminal are sent, the first model data include at least first nerves net First species parameter of network model.

It should be noted that the neural network algorithm that can preserve various species in server is corresponding for generating god Template file through network model file, wherein the preset model data at least three networks that a template file includes.

S206, server obtain the first nerves network model text of the first species parameter instruction according to the first model data Part.

Optionally, it in server can also include the corresponding neural network model text of predefined each first species parameter Part.

Illustratively, the first species parameter (being denoted as species parameter 1) input by user is the hand in above-mentioned image recognition class Digital identification is write, above-mentioned neural network model file 1 is the predefined neural network model file preserved in server.This When, server can get the neural network model file 1 according to the species parameter 1.

S207, server send first nerves network model file to terminal.

It should be noted that the method for processing neural network model file provided in an embodiment of the present invention, server can be with Obtain the model data that user inputs in end side so that server can get the god of user's needs according to the model data Through network model file.To which server can be interacted with terminal so that terminal can obtain user from server and need Neural network model file.That is, the program code that user need not write complexity can be from the god of the demand of acquisition Through network model file.In this way, can solve the problems, such as that the process of the neural network model file of terminal acquisition demand takes.And And reduce operation of the user when using terminal obtains the neural network model file of demand, the operating process can be improved In user experience.

Optionally, server can also be according to the first model data input by user, and it includes above-mentioned at least three net to generate The first nerves network model file of the first information of network layers, the first information of a network layer include the second of a network layer The sequence identification of species parameter, the first functional parameter of network layer and a network layer.

In one possible implementation, the method for processing neural network model file provided in an embodiment of the present invention, Above-mentioned S206 specifically includes S206a and S206b：

It is corresponding other to obtain the first species parameter preserved in server according to the first model data for S206a, server The third information of network layer, other network layers are the network layer in addition at least one network layer at least three network layers.

Illustratively, the first model file input by user is the part mould for generating first nerves network model file Type data, at this time server can obtain packet in the said one template file (such as template file 1) that is preserved in this server The preset model data of at least three networks included.Wherein, above-mentioned first species parameter corresponds to the template file 1, and one above-mentioned The third information of other network layers can be a default template data in the template file 1.

S206b, server generate first nerves network mould according to the third information of the first model data and other network layers Type file.

Wherein, the third information of a network layer includes following at least one：The second species parameter of one network layer, one The sequence identification of the third functional parameter of a network layer, network layer.

It is understood that the third functional parameter of a network can be that server is obtaining the net in above-described embodiment Without the functional parameter of training and test before first functional parameter of network layers.

It should be noted that even if user cannot input neural network model text for a neural network model file The corresponding all model datas of part, the server can also obtain the corresponding template file preserved in the server and the template Preset model data in file.In this way, server can input the part mould of the neural network model file needed in user When type data so that terminal can obtain the neural network model file of needs.In this way, being conducive to improve at user's using terminal Manage user experience when neural network model file.

In one possible implementation, the method for processing neural network model file provided in an embodiment of the present invention, Above-mentioned S206 specifically includes S206c and S206d：

It is corresponding at least to obtain the first species parameter preserved in server according to the first model data for S206c, server The third information of three network layers.

Illustratively, the first model data input by user is all moulds for generating first nerves network model file Type data, server can obtain the said one template file (such as template file 1) preserved in this server at this time, and make The preset model data that template file 1 includes at least three networks are replaced with model data used input by user.Wherein, on It states the first species parameter and corresponds to the template file 1, the third information of an above-mentioned other network layers can be in the template file 1 A default template data.

S206d, server generate first nerves net according to the third information of the first model data and at least three network layers Network model file.

Wherein, the third information of a network layer includes following at least one：The second species parameter of one network layer, one The sequence identification of the third functional parameter of a network layer and a network layer.

It should be noted that user can input the corresponding all model datas of neural network model file of a needs In the case of, server can obtain the neural network model file of the needs according to all model datas.In this way, being conducive to Server obtains the neural network model file for meeting user demand, is conducive to improve user's using terminal processing neural network mould User experience when type file.

In a kind of specific embodiment of the present invention, as shown in fig. 6, being a kind of possible end provided in an embodiment of the present invention The structural schematic diagram at end.Terminal 60 shown in Fig. 6 includes receiving module 601 and acquisition module 602；

Wherein, receiving module 601, in the case where terminal 60 shows that interface is arranged in neural network model, receiving and using First input at family, the first input are the operation that user triggers the first model data of input, and the first model data includes at least the One species parameter；Acquisition module 602, first for being received in response to receiving module 601 inputs, according to the first model data, Obtain the first nerves network model file of the first species parameter instruction；Wherein, the first species parameter is used to indicate first nerves The kind of document of network model file, this document type include the classification of image recognition, the classification of speech recognition and text point The classification of class.

In one possible implementation, acquisition module 602 are specifically used for sending the first model data to server, First model data is used to obtain the first nerves network model file of the first species parameter instruction；Receive server is sent the One neural network model file.

In one possible implementation, above-mentioned first nerves network model file includes the of at least three network layers One information, the first information of a network layer include the second species parameter of a network layer, the first function of network layer The sequence identification of parameter and a network layer；Wherein, second species parameter is used to indicate the functional type of network layer, the function Type includes input layer, convolutional layer, pond layer, full articulamentum and output layer.

In one possible implementation, above-mentioned first model data further includes at least one at least three network layers The test data set of second information of network layer, the training dataset of at least three network layers and at least three network layers；Its In, the second information of a network layer includes following at least one：The second species parameter of one network layer, network layer The sequence identification of second functional parameter and a network layer.

In one possible implementation, as shown in fig. 7, terminal 60 further includes generation module 603 and display module 604；Wherein, generation module 603, for, according to the first model data, obtaining the instruction of the first species parameter in acquisition module 602 After first nerves network model file, according to the first information of at least three network layers, first nerves network model text is generated The corresponding three-dimensional visualization model of part；Display module 604, the three-dimensional visualization model generated for showing generation module 603.

In one possible implementation, above-mentioned neural network model setting interface includes first area, the secondth area Domain, third region, the fourth region, the 5th region and the 6th region；Wherein, first area is used to input the first species parameter, the Two regions are used to input the second species parameter of each network layer at least three network layers, and third region is for inputting at least three Second functional parameter of each network layer in a network layer, the fourth region is for inputting each network layer at least three network layers Sequence identification, the 5th region be used for input at least three network layers training data and the 6th region for input at least The test data of three network layers.

Terminal 60 provided in an embodiment of the present invention can realize each process that terminal is realized in above method embodiment, be It avoids repeating, which is not described herein again.

Terminal provided in an embodiment of the present invention, which can provide a user neural network model setting interface, with branch Hold the model data that user inputs the neural network model file of demand so that terminal can obtain the god according to the model data Through network model file.That is, the program code that user need not write complexity can be so that terminal obtains demand Neural network model file.In this way, can solve the problems, such as that the process of the neural network model file of terminal acquisition demand takes. Also, reduces operation of the user when using terminal obtains the neural network model file of demand, the operation can be improved User experience in journey.

In a kind of specific embodiment of the present invention, as shown in figure 8, being a kind of possible clothes provided in an embodiment of the present invention The structural schematic diagram of business device.Server 80 shown in Fig. 8 includes receiving module 801, acquisition module 802 and sending module 803；It connects Module 801 is received, the first model data for receiving terminal transmission, the first model data includes at least the first species parameter；It obtains Modulus block 802, the first model data for being received according to receiving module 801 obtain the first god of the first species parameter instruction Through network model file；Sending module 803, for sending the first nerves network model text that acquisition module 802 obtains to terminal Part；Wherein, the first species parameter is used to indicate the kind of document of first nerves network model file, and this document type includes image The classification of the classification of identification, the classification of speech recognition and text classification.

In a kind of specific embodiment of the present invention, first nerves network model file includes the of at least three network layers One information, the first information of a network layer include the second species parameter of a network layer, the first function of network layer The sequence identification of parameter and a network layer；Wherein, second species parameter is used to indicate the functional type of network layer, the function Type includes input layer, convolutional layer, pond layer, full articulamentum and output layer.

In one possible implementation, the first model data further includes at least one network at least three network layers The second information, the training dataset of at least three network layers and the test data set of at least three network layers of layer；Wherein, one Second information of a network layer includes following at least one：The second of the second species parameter of one network layer, network layer The sequence identification of functional parameter and a network layer.

In one possible implementation, acquisition module 802 are specifically used for, according to the first model data, obtaining service The third information of the corresponding other network layers of the first species parameter preserved in device 80, other network layers are at least three network layers In network layer in addition at least one network layer；According to the third information of the first model data and other network layers, the is generated One neural network model file；Wherein, the third information of a network layer includes following at least one：The second of one network layer The sequence identification of species parameter, the third functional parameter of network layer, network layer.

In one possible implementation, acquisition module 802 are specifically used for, according to the first model data, obtaining service The third information of corresponding at least three network layer of the first species parameter preserved in device；According to the first model data and at least three The third information of a network layer generates first nerves network model file；Wherein, the third information of a network layer includes following It is at least one：The sequence mark of the second species parameter of one network layer, the third functional parameter of network layer, network layer Know.

Server 80 provided in an embodiment of the present invention can realize each mistake that server is realized in above method embodiment Journey, to avoid repeating, which is not described herein again.

Server provided in an embodiment of the present invention, the server can obtain the model data that user inputs in end side, Server is allow to get the neural network model file of user's needs according to the model data.To which, server can be with It is interacted with terminal so that terminal can obtain the neural network model file that user needs from server.That is, with Family need not write complicated program code can be from the neural network model file of the demand of acquisition.In this way, end can be solved The problem for holding the process of the neural network model file of acquisition demand time-consuming.Also, reduce user needs in using terminal acquisition Operation when the neural network model file asked, can improve the user experience in the operating process.Of the invention a kind of specific Embodiment in, a kind of hardware architecture diagram of Fig. 9 terminals that embodiment provides to realize the present invention, which includes But it is not limited to：Radio frequency unit 101, network module 102, audio output unit 103, input unit 104, sensor 105, display are single The components such as member 106, user input unit 107, interface unit 108, memory 109, processor 110 and power supply 111.Ability Field technique personnel are appreciated that the restriction of the not structure paired terminal of terminal structure shown in Fig. 9, terminal may include than diagram More or fewer components either combine certain components or different components arrangement.In embodiments of the present invention, terminal packet Include but be not limited to mobile phone, tablet computer, laptop, palm PC, car-mounted terminal, wearable device and pedometer etc..

Wherein, user input unit 107, in the case where terminal shows that interface is arranged in neural network model, receiving The first input of user, the first input are the operation that user triggers the first model data of input, and the first model data includes at least First species parameter of first nerves network model；Processor 110, first for being received in response to user input unit 107 Input obtains the first nerves network model file of the first species parameter instruction according to the first model data.

Terminal provided in an embodiment of the present invention, the user which can provide a user input the neural network mould of demand Interface is arranged in the neural network model of the model data of type file so that terminal can obtain the nerve net according to the model data Network model file.That is, user need not write complicated program code can be so that terminal obtains the nerve of demand Network model file.In this way, can solve the problems, such as that the process of the neural network model file of terminal acquisition demand takes.And And reduce operation of the user when using terminal obtains the neural network model file of demand, the operating process can be improved In user experience.

It should be understood that the embodiment of the present invention in, radio frequency unit 101 can be used for receiving and sending messages or communication process in, signal Send and receive, specifically, by from base station downlink data receive after, to processor 110 handle；In addition, by uplink Data are sent to base station.In general, radio frequency unit 101 includes but not limited to antenna, at least one amplifier, transceiver, coupling Device, low-noise amplifier, duplexer etc..In addition, radio frequency unit 101 can also by radio communication system and network and other set Standby communication.

Terminal has provided wireless broadband internet to the user by network module 102 and has accessed, and such as user is helped to receive and dispatch electricity Sub- mail, browsing webpage and access streaming video etc..

It is that audio output unit 103 can receive radio frequency unit 101 or network module 102 or in memory 109 The audio data of storage is converted into audio signal and exports to be sound.Moreover, audio output unit 103 can also provide and end The relevant audio output of specific function (for example, call signal receives sound, message sink sound etc.) that end 100 executes.Sound Frequency output unit 103 includes loud speaker, buzzer and receiver etc..

Input unit 104 is for receiving audio or video signal.Input unit 104 may include graphics processor (Graphics Processing Unit, GPU) 1041 and microphone 1042, graphics processor 1041 is in video acquisition mode Or the image data of the static images or video obtained by image capture apparatus (such as camera) in image capture mode carries out Reason.Treated, and picture frame may be displayed on display unit 106.Through graphics processor 1041, treated that picture frame can be deposited Storage is sent in memory 109 (or other storage mediums) or via radio frequency unit 101 or network module 102.Mike Wind 1042 can receive sound, and can be audio data by such acoustic processing.Treated audio data can be The format output of mobile communication base station can be sent to via radio frequency unit 101 by being converted in the case of telephone calling model.

Terminal 100 further includes at least one sensor 105, such as optical sensor, motion sensor and other sensors. Specifically, optical sensor includes ambient light sensor and proximity sensor, wherein ambient light sensor can be according to ambient light Light and shade adjusts the brightness of display panel 1061, and proximity sensor can close display panel when terminal 100 is moved in one's ear 1061 and/or backlight.As a kind of motion sensor, accelerometer sensor can detect in all directions (generally three axis) and add The size of speed can detect that size and the direction of gravity when static, can be used to identify terminal posture (such as horizontal/vertical screen switching, Dependent game, magnetometer pose calibrating), Vibration identification correlation function (such as pedometer, tap) etc.；Sensor 105 can be with Including fingerprint sensor, pressure sensor, iris sensor, molecule sensor, gyroscope, barometer, hygrometer, thermometer, Infrared sensor etc., details are not described herein.

Display unit 106 is for showing information input by user or being supplied to the information of user.Display unit 106 can wrap Display panel 1061 is included, liquid crystal display (Liquid Crystal Display, LCD), Organic Light Emitting Diode may be used Forms such as (Organic Light-Emitting Diode, OLED) configure display panel 1061.

User input unit 107 can be used for receiving the number or character information of input, and generates and set with the user of terminal It sets and the related key signals of function control inputs.Specifically, user input unit 107 include touch panel 1071 and other Input equipment 1072.Touch panel 1071, also referred to as touch screen, collect user on it or neighbouring touch operation (such as User is using any suitable objects or attachment such as finger, stylus on touch panel 1071 or near touch panel 1071 Operation).Touch panel 1071 may include both touch detecting apparatus and touch controller.Wherein, touch detecting apparatus is examined The touch orientation of user is surveyed, and detects the signal that touch operation is brought, transmits a signal to touch controller；Touch controller from Touch information is received on touch detecting apparatus, and is converted into contact coordinate, then gives processor 110, receives processor 110 The order sent simultaneously is executed.Furthermore, it is possible to using multiple types such as resistance-type, condenser type, infrared ray and surface acoustic waves Realize touch panel 1071.In addition to touch panel 1071, user input unit 107 can also include other input equipments 1072. Specifically, other input equipments 1072 can include but is not limited to physical keyboard, function key (such as volume control button, switch Button etc.), trace ball, mouse, operating lever, details are not described herein.

Further, touch panel 1071 can be covered on display panel 1061, when touch panel 1071 is detected at it On or near touch operation after, send processor 110 to determine the type of touch event, be followed by subsequent processing device 110 according to touch The type for touching event provides corresponding visual output on display panel 1061.Although in fig.9, touch panel 1071 and display Panel 1061 is to realize the function that outputs and inputs of terminal as two independent components, but in certain embodiments, can The function that outputs and inputs of terminal is realized so that touch panel 1071 and display panel 1061 is integrated, is not limited herein specifically It is fixed.

Interface unit 108 is the interface that external device (ED) is connect with terminal 100.For example, external device (ED) may include it is wired or Wireless head-band earphone port, external power supply (or battery charger) port, wired or wireless data port, memory card port, For connecting the port of device with identification module, the port audio input/output (I/O), video i/o port, ear port Etc..Interface unit 108 can be used for receiving the input (for example, data information, electric power etc.) from external device (ED) and will One or more elements that the input received is transferred in terminal 100 or can be used for terminal 100 and external device (ED) it Between transmission data.

Memory 109 can be used for storing software program and various data.Memory 109 can include mainly storing program area And storage data field, wherein storing program area can storage program area, application program (such as the sound needed at least one function Sound playing function, image player function etc.) etc.；Storage data field can store according to mobile phone use created data (such as Audio data, phone directory etc.) etc..In addition, memory 109 may include high-speed random access memory, can also include non-easy The property lost memory, a for example, at least disk memory, flush memory device or other volatile solid-state parts.

Processor 110 is the control centre of terminal, using the various pieces of various interfaces and the entire terminal of connection, is led to It crosses operation or executes the software program and/or module being stored in memory 109, and call and be stored in memory 109 Data execute the various functions and processing data of terminal, to carry out integral monitoring to terminal.Processor 110 may include one Or multiple processing units；Preferably, processor 110 can integrate application processor and modem processor, wherein application processing The main processing operation system of device, user interface and application program etc., modem processor mainly handles wireless communication.It can manage Solution, above-mentioned modem processor can not also be integrated into processor 110.

Terminal 100 can also include the power supply 111 (such as battery) powered to all parts, it is preferred that power supply 111 can be with It is logically contiguous by power-supply management system and processor 110, to by power-supply management system realize management charging, electric discharge, with And the functions such as power managed.

In addition, terminal 100 includes some unshowned function modules, details are not described herein.

Optionally, the embodiment of the present invention also provides a kind of terminal, including processor 110, and memory 109 is stored in storage It is real when which is executed by processor 110 on device 109 and the computer program that can be run on the processor 110 Each process of the embodiment of the method for existing above-mentioned processing neural network model file, and identical technique effect can be reached, to keep away Exempt to repeat, which is not described herein again.

In a kind of specific embodiment of the present invention, a kind of Figure 10 servers that embodiment provides to realize the present invention it is hard Part structural schematic diagram.Server 2600 shown in Figure 10 includes：Processor 2601, transceiver 2602, memory 2603, Yong Hujie Mouth 2604 and bus interface.

Wherein, transceiver 2602, the first model data for receiving terminal transmission, the first model data include at least the First species parameter of one neural network model；Processor 2601, the first model data for being received according to transceiver 2602, Obtain the first nerves network model file of the first species parameter instruction；Transceiver 2602, this is used to send processor to terminal The 2601 first nerves network model files obtained.

Server provided in an embodiment of the present invention, the server can obtain the model data that user inputs in end side, Server is allow to get the neural network model file of user's needs according to the model data.To which, server can be with It is interacted with terminal so that terminal can obtain the neural network model file that user needs from server.That is, with Family need not write complicated program code can be from the neural network model file of the demand of acquisition.In this way, end can be solved The problem for holding the process of the neural network model file of acquisition demand time-consuming.Also, reduce user needs in using terminal acquisition Operation when the neural network model file asked, can improve the user experience in the operating process.In the embodiment of the present invention, In Figure 10, bus architecture may include the bus and bridge of any number of interconnection, specifically represented by processor 2601 one or The various circuits for the memory that multiple processors and memory 2603 represent link together.Bus architecture can also will be such as outer Various other circuits of peripheral equipment, voltage-stablizer and management circuit or the like link together, these are all that this field institute is public Know, therefore, it will not be further described herein.Bus interface provides interface.Transceiver 2602 can be multiple members Part includes transmitter and receiver, provides the unit for being communicated over a transmission medium with various other devices.For difference Server, user interface 2604 can also be can the external inscribed interface for needing equipment, the equipment of connection includes but unlimited In keypad, display, loud speaker, microphone, control stick etc..Processor 2601 is responsible for bus architecture and common place Reason, memory 2603 can store the used data when executing operation of processor 2601.

In addition, server 2600 further includes some unshowned function modules, details are not described herein.

The embodiment of the present invention also provides a kind of computer readable storage medium, and meter is stored on computer readable storage medium Calculation machine program, the computer program realize the side for the processing neural network model file that above-mentioned terminal executes when being executed by processor Each process of method embodiment, and identical technique effect can be reached, to avoid repeating, which is not described herein again.Wherein, described Computer readable storage medium, such as read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disc or CD etc..

The embodiment of the present invention also provides a kind of computer readable storage medium, and meter is stored on computer readable storage medium Calculation machine program, the computer program realize the processing neural network model file that above-mentioned server executes when being executed by processor Each process of embodiment of the method, and identical technique effect can be reached, to avoid repeating, which is not described herein again.Wherein, described Computer readable storage medium, such as ROM, RAM, magnetic disc or CD.

It should be noted that herein, the terms "include", "comprise" or its any other variant are intended to non-row His property includes, so that process, method, article or device including a series of elements include not only those elements, and And further include other elements that are not explicitly listed, or further include for this process, method, article or device institute it is intrinsic Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including this There is also other identical elements in the process of element, method, article or device.

Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side Method can add the mode of required general hardware platform to realize by software, naturally it is also possible to by hardware, but in many cases The former is more preferably embodiment.Based on this understanding, technical scheme of the present invention substantially in other words does the prior art Going out the part of contribution can be expressed in the form of software products, which is stored in a storage medium In (such as ROM/RAM, magnetic disc, CD), including some instructions are used so that a station terminal (can be mobile phone, computer, service Device, air conditioner or network equipment etc.) execute method described in each embodiment of the present invention.

The embodiment of the present invention is described with above attached drawing, but the invention is not limited in above-mentioned specific Embodiment, the above mentioned embodiment is only schematical, rather than restrictive, those skilled in the art Under the inspiration of the present invention, without breaking away from the scope protected by the purposes and claims of the present invention, it can also make very much Form belongs within the protection of the present invention.

Claims

1. a kind of method of processing neural network model file, which is characterized in that including：

In the case where terminal shows that interface is arranged in neural network model, the terminal receives the first input of user, and described the One input is the operation that user triggers the first model data of input, and first model data includes at least the first species parameter；

In response to first input, the terminal obtains the first species parameter instruction according to first model data First nerves network model file；

Wherein, first species parameter is used to indicate the kind of document of the first nerves network model file, the file Type includes the classification of the classification of image recognition, the classification of speech recognition and text classification.

2. according to the method described in claim 1, it is characterized in that, the terminal obtains institute according to first model data The first nerves network model file of the first species parameter instruction is stated, including：

The terminal to server sends first model data, and first model data is for obtaining first type The first nerves network model file of parameter instruction；

The terminal receives the first nerves network model file that the server is sent.

3. method according to claim 1 or 2, which is characterized in that the first nerves network model file includes at least The first information of the first information of three network layers, a network layer includes the second species parameter of a network layer, a net The sequence identification of first functional parameter of network layers and a network layer；

Wherein, the second species parameter is used to indicate the functional type of network layer, and the functional type includes input layer, convolution Layer, pond layer, full articulamentum and output layer.

4. according to the method described in claim 3, it is characterized in that, first model data further includes at least three net Second information of at least one network layer, the training dataset of at least three network layer and described at least three in network layers The test data set of network layer；

Wherein, the second information of a network layer includes following at least one：

The sequence identification of the second functional parameter and a network layer of the second species parameter of one network layer, network layer.

5. according to the method described in claim 3, it is characterized in that, being obtained according to first model data in the terminal After the first nerves network model file of the first species parameter instruction, further include：

The terminal generates the first nerves network model file and corresponds to according to the first information of at least three network layer Three-dimensional visualization model；

The terminal shows the three-dimensional visualization model.

6. according to the method described in claim 5, it is characterized in that, neural network model setting interface includes the firstth area Domain, second area, third region, the fourth region, the 5th region and the 6th region；

Wherein, the first area is for inputting first species parameter, and the second area is for inputting described at least three The second species parameter of each network layer in a network layer, the third region are every at least three network layer for inputting Second functional parameter of a network layer, the fourth region are used to input the suitable of each network layer at least three network layer Sequence identifies, and the 5th region is used to input the training data of at least three network layer and the 6th region is used for Input the test data of at least three network layer.

7. a kind of terminal, which is characterized in that including receiving module and acquisition module；

The receiving module, in the case where the terminal shows that interface is arranged in neural network model, receiving the of user One input, first input are the operation that user triggers the first model data of input, and first model data includes at least First species parameter；

The acquisition module, first input for being received in response to the receiving module, according to first pattern number According to the first nerves network model file of acquisition the first species parameter instruction；

8. terminal according to claim 7, which is characterized in that

The acquisition module is specifically used for sending first model data to server, and first model data is for obtaining Take the first nerves network model file of the first species parameter instruction；

Receive the first nerves network model file that the server is sent.

9. terminal according to claim 7 or 8, which is characterized in that the first nerves network model file includes at least The first information of the first information of three network layers, a network layer includes the second species parameter of a network layer, a net The sequence identification of first functional parameter of network layers and a network layer；

10. terminal according to claim 9, which is characterized in that first model data further includes described at least three Second information of at least one network layer, the training dataset of at least three network layer and described at least three in network layer The test data set of a network layer；

11. terminal according to claim 9, which is characterized in that the terminal further includes generation module and display module；

The generation module, for, according to first model data, obtaining first species parameter in the acquisition module After the first nerves network model file of instruction, according to the first information of at least three network layer, described first is generated The corresponding three-dimensional visualization model of neural network model file；

The display module, the three-dimensional visualization model generated for showing the generation module.

12. a kind of terminal, which is characterized in that including processor, memory and be stored on the memory and can be at the place The computer program run on reason device is realized when the computer program is executed by the processor as appointed in claim 1 to 6 The step of method of processing neural network model file described in one.