CN110363222A - Picture mask method, device, computer equipment and storage medium for model training - Google Patents
Picture mask method, device, computer equipment and storage medium for model training Download PDFInfo
- Publication number
- CN110363222A CN110363222A CN201910524478.7A CN201910524478A CN110363222A CN 110363222 A CN110363222 A CN 110363222A CN 201910524478 A CN201910524478 A CN 201910524478A CN 110363222 A CN110363222 A CN 110363222A
- Authority
- CN
- China
- Prior art keywords
- marked
- samples pictures
- task
- data
- task data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The invention discloses picture mask method, device, computer equipment and the storage mediums for model training, task form are established according to samples pictures to be marked, in order to be managed collectively to labeled data;The labeled data for obtaining target object in samples pictures, is updated, and audit to the task data for having marked samples pictures, to guarantee the accuracy of labeled data according to labeled data pair and the associated task data of samples pictures to be marked;The task list of task data that audit is passed through and corresponding samples pictures generate training data, in order to according to training data to data set model training, to improve the precision of model training.
Description
Technical field
The present invention relates to technical field of image processing, more particularly, to picture mask method, device, the meter of model training
Calculate machine equipment and storage medium.
Background technique
It is mainly in the common technological means of picture process field at present: using large-scale artificial mark sample as instruction
Practice picture, neural network model is trained, in order to utilize the neural network model after training to picture number to be processed
According to being handled, to improve treatment effeciency.The quantity and mark of the training result of neural network model and artificial mark sample
Accuracy is directly related.The method for obtaining the artificial labeled data of picture at present mainly passes through special mark personnel, outsourcing or crowd
The mode of packet, is manually labeled picture.But since picture mark is a uninteresting, simple duplicate labour, for a long time
Uninteresting mark be easy to appear marking error, thus influence image training accuracy.
Summary of the invention
Being easy to appear mistake existing artificial mark sample has error, now provides one kind and is intended to can be improved
Mark picture mask method, device, computer equipment and storage medium that accuracy is used for model training.
To achieve the above object, the present invention provides a kind of picture mask method for model training, provides a storage list
Member includes the following steps: for storing an at least samples pictures to be marked
S1. the task form that samples pictures to be marked are established is established according to an at least width, it is every in the task form
The corresponding task data of identification number of every samples pictures to be marked of one task data association one;
S2. the associated identification number of the task data according to the task form extracts and institute from the storage unit
State the corresponding samples pictures to be marked of identification number;
S3. identification acquires target object in the samples pictures to be marked, and extracting the corresponding information of the target object will
The labeled data that the information is converted to is closed according to according to the labeled data pair and the identification number of the samples pictures to be marked
The task data of connection is updated;
S4. the task data for having marked samples pictures is audited;
S5. the task list for the task data that audit passes through and the corresponding samples pictures that marked are generated into instruction
Practice data.
Preferably, the associated identification number of step S2 task data according to the task form is from the storage
The samples pictures to be marked corresponding with the identification number are extracted in unit, further includes:
After receiving the samples pictures to be marked, the received samples pictures to be marked are pre-processed, it will
The size adjusting of the samples pictures to be marked is pre-set dimension.
Preferably, the step S3 identifies target object in the samples pictures to be marked, extracts the target object pair
The information answered converts that information to labeled data, the mark according to the labeled data pair and the samples pictures to be marked
Number associated task data is updated, comprising:
Identification number according to the samples pictures to be marked obtains the marking types of the samples pictures to be marked, Mei Yisuo
The corresponding marking types of identification number are stated, each marking types correspond to corresponding image recognition region, described image identification region pair
The target object to be identified answered;
It is to be marked described in the corresponding image recognition region recognition of the samples pictures to be marked according to the marking types
Target object in samples pictures;
It extracts the corresponding information of the target object and converts that information to labeled data;
It is carried out according to the labeled data pair and the associated task data of identification number of the samples pictures to be marked
It updates.
Preferably, the step S1 is being executed according to before at least a samples pictures to be marked are established in task form
Also, comprising:
Several samples pictures to be marked are converted video data to, several described samples pictures to be marked are stored in described
In storage unit, each samples pictures to be marked and an identification number are corresponded
A task form is established by described, each task data in the task form is associated with a sample to be marked
The identification number of picture establishes task form according to an at least samples pictures to be marked.
Preferably, further includes:
A. the task data that the audit fails is marked, the task data after modification is labeled returns
Execute step S4.
Preferably, further includes:
B. by the audit fails it is corresponding with the task data marked samples pictures and be marked, modification with through marking
Described after note has marked the corresponding task data of samples pictures, returns to step S4.
Preferably, the step S4 carries out audit to the task data for having marked samples pictures and receives the sample to be marked
Picture, comprising:
The task data for having marked samples pictures is audited using scene text detection, judgement is described to have marked sample
Whether the task data of picture and the task data obtained through scene text detection are consistent.
To achieve the above object, the present invention also provides a kind of devices that the picture for model training marks, comprising:
Unit is established, for establishing the task form that samples pictures to be marked are established, the task according to an at least width
The corresponding task data of identification number of every samples pictures to be marked of each task data association one in list;
Extraction unit is used for the associated identification number of the task data according to the task form from the storage unit
It is middle to extract the reception samples pictures to be marked corresponding with the identification number;
Unit is marked, target object in the samples pictures to be marked, it is corresponding to extract the target object for identification
The labeled data that information converts that information to, the identification number according to the labeled data pair and the samples pictures to be marked
The associated task data is updated;
Unit is audited, for auditing to the task data for having marked samples pictures;
Generation unit, the task list of the task data for passing through audit and corresponding described has marked sample
Picture generates training data.
To achieve the above object, the present invention also provides a kind of computer equipment, the computer equipment, including memory,
Processor and storage on a memory and the computer program that can run on a processor, the processor execution calculating
The step of above method is realized when machine program.
To achieve the above object, the present invention also provides a kind of computer readable storage mediums, are stored thereon with computer journey
Sequence, when the computer program is executed by processor the step of the realization above method.
Above-mentioned technical proposal the utility model has the advantages that
In the technical program, provided by the present invention for the picture mask method of model training, device, computer equipment and
Storage medium establishes task form according to samples pictures to be marked, in order to be managed collectively to labeled data;Obtain sample
The labeled data of target object in picture carries out more according to labeled data pair and the associated task data of samples pictures to be marked
Newly, and to the task data for having marked samples pictures it audits, to guarantee the accuracy of labeled data;Audit is passed through
The task list of task data and corresponding samples pictures generate training data, in order to according to training data to data set model
Training, to improve the precision of model training.
Detailed description of the invention
Fig. 1 is a kind of flow chart of embodiment of the picture mask method of the present invention for model training;
Fig. 2 is the flow chart of another embodiment of the picture mask method of the present invention for model training;
Fig. 3 is a kind of module map of the embodiment for the device that the picture of the present invention for model training marks;
Fig. 4 is the hardware structure schematic diagram of one embodiment of computer equipment of the present invention.
Specific embodiment
It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understood
The application is further elaborated.It should be appreciated that specific embodiment described herein is only to explain the application, not
For limiting the application.Based on the embodiment in the application, those of ordinary skill in the art are not before making creative work
Every other embodiment obtained is put, shall fall in the protection scope of this application.
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the present invention can phase
Mutually combination.
Provided by the present invention for picture mask method, device, computer equipment and the storage medium of model training, it is applicable in
In insurance business field, a kind of task pipe that accuracy rate and working efficiency are marked with statistics mark personnel is provided for mark personnel
Reason method.The present invention establishes task form according to samples pictures to be marked, in order to be managed collectively to labeled data;It obtains
The labeled data of target object in samples pictures is carried out according to labeled data pair and the associated task data of samples pictures to be marked
It updates, and the task data for having marked samples pictures is audited, to guarantee the accuracy of labeled data;Audit is passed through
Task data task list and corresponding samples pictures generate training data, in order to according to training data to data set mould
Type training, to improve the precision of model training.
Embodiment one
Referring to Fig. 1, a kind of picture mask method for model training of the present embodiment, provides a storage unit, use
In a storage at least samples pictures to be marked, include the following steps:
S1. the task form that samples pictures to be marked are established is established according to an at least width, it is every in the task form
The corresponding task data of identification number of every samples pictures to be marked of one task data association one;
Specifically, the process of a task form of foundation are as follows:
The task form is default template, and the default template includes a plurality of task data, one by one by task data with
The identification number of each samples of text to be marked in storage unit is associated, to generate task form.
In practical applications, the samples pictures to be marked (such as: 1000) in storage unit are read by management server
Establish a task form, wherein every width picture corresponding task data in the task form, each width picture corresponding one is only
One identification number (identification number can be with the ID number of picture), which is associated with task data.
It should be understood that if sample to be marked is different from preset format, further according to conversion after can first formatting
Picture afterwards generates task form.Such as: sample to be marked is video data, need to convert video data to sample graph to be marked
Piece, then samples pictures to be marked are based on using management server and establish an associated task form.
S2. the associated identification number of the task data according to the task form extracts and institute from the storage unit
State the corresponding samples pictures to be marked of identification number;
Further, after receiving samples pictures to be marked, the received samples pictures to be marked can be carried out pre-
The size adjusting of the samples pictures to be marked is pre-set dimension by processing.
Pre-set dimension is regular length and width, such as 5cm × 8cm.
In this step, in order to avoid the samples pictures to be marked received have exception, such as: format does not meet default want
Ask size too big or format mismatching can not load, the low content that can not clearly observe in picture of clarity etc., can be used and adopt
Samples pictures to be marked are pre-processed with the mode of noise reduction, so that can be reached with normal load, clarity can be just for picture
The requirement often observed.
S3. target object in the samples pictures to be marked is identified, extracting the corresponding information of the target object will be described
The labeled data that information is converted to, it is associated described according to the labeled data pair and the identification number of the samples pictures to be marked
Task data is updated;
Specifically, step S3 can include:
S31. the identification number according to the samples pictures to be marked obtains the marking types of the samples pictures to be marked, often
The corresponding marking types of one identification number, each marking types correspond to corresponding image recognition region, described image cog region
The corresponding target object to be identified in domain;
S32. according to the marking types described in the corresponding image recognition region recognition of the samples pictures to be marked to
Mark target object in samples pictures;
Wherein, the marking types can include: certificate class and/or license class and/or object type and/or vehicle class, often
One marking types correspond to corresponding location information and identification information.
Wherein, certificate class can include: identity card, bank card, the driving quality certification, driver's license or invoice etc.;License class can be with
Be: business license is (such as: self-employed worker's business license, partnership business's business license, individual proprietorship enterprise's business license, Limited Liability
Corporate licence and outside share part Co., Ltd business license etc.);Object type may is that personage, animal or subject
Deng;Vehicle class is the image of vehicle class.It is the band of position of its mark of the samples pictures to be marked of certificate class for marking types
Relatively fixed, i.e., typesetting format is fixed, and corresponding content information is shown in fixed region.
For example and without limitation, when marking types are identity card, corresponding location information (fixed area) be can be
The name band of position, the address band of position, citizenship number region, date of birth region, gender region in identity card and
Regional national;Corresponding identification information are as follows: and the corresponding name in the name band of position, family corresponding with the address band of position live
The corresponding identification card number in location and citizenship number region, the date of birth corresponding with date of birth region, with gender region pair
The gender answered, nationality corresponding with regional national, it should be understood that the format of identification information can be text, number or
Letter etc..
When marking types are invoice, corresponding location information (fixed area) can be cargo or dutiable service, service
Name area and valence tax add up to region etc.;Corresponding identification information are as follows: cargo or dutiable service, the corresponding name in service name region
Claim, valence tax adds up to the corresponding price numerical value in region.
When marking types are business license, corresponding location information (fixed area) can be unified social credibility code
Region, Name area;Corresponding identification information are as follows: code corresponding with unified social credibility code region and Name area pair
The title answered.
When marking types are personage: corresponding location information is the region of the face head portrait of personage, corresponding mark letter
Breath are as follows: face mark;When marking types are animal: corresponding location information is cat face region, corresponding identification information are as follows: cat
Face mark;
When marking types are vehicle class: corresponding location information is the region of vehicle, corresponding identification information are as follows: vehicle
Mark.
S33. it extracts the corresponding information of the target object and converts that information to labeled data;
S34. the associated task data of identification number according to the labeled data pair and the samples pictures to be marked
It is updated
Wherein, the task data can include: the location information of target object in marking types, samples pictures to be marked,
Identification information.
In practical applications, the labeled data of target object in samples pictures to be marked is acquired using mark client,
Such as: the location information of target object in samples pictures to be marked, and pass through the mark corresponding identification information of client typing.
It should be understood that can amplify, reduce to samples pictures to be marked in order to facilitate mark when being labeled
And rotation etc., in order to watch the target object that samples pictures to be marked need to mark.
In this step, by marking the samples pictures to be marked of client scan, according to the samples pictures to be marked
Identification number identifies corresponding marking types, is based on the marking types corresponding image recognition area in the samples pictures to be marked
(such as: the position coordinates of name, address, the corresponding position frame of citizen ID certificate number in identity card) is identified in domain, to obtain
The information of target object, converts this information into labeled data, according to the labeled data pair and the samples pictures to be marked
The associated task data of identification number be updated so that the content in task data and the samples pictures corresponds.
If marking types are type of credential when being object type, corresponding mark class is identified according to the identification number of the samples pictures to be marked
Type, based on the marking types, (such as: face frame is in sample to be marked in corresponding image recognition region in the samples pictures to be marked
Position coordinates in this picture), the identification information of " face head portrait " is obtained, labeled data is converted this information into, according to described in
Labeled data pair and the associated task data of identification number of the samples pictures to be marked are updated, so that task data
It is corresponded with the content in the samples pictures.
When being labeled to samples pictures to be marked, labeled data directly enters corresponding with samples pictures to be marked
In data of being engaged in, in order to the management to the task data in task form.
S4. the task data for having marked samples pictures is audited;
Specifically, the step S4 can include:
The task data for having marked samples pictures is audited using scene text detection, judgement is described to have marked sample
Whether the task data of picture and the task data obtained through scene text detection are consistent.
In this step, in order to which the accuracy for improving the data of mark needs after completing mark to having marked sample graph
The task data of piece is audited.
In practical applications, after to the mark of samples pictures to be marked, which can further be audited, is sentenced
Whether disconnected labeled data corresponds to the corresponding target object marked in samples pictures, if so, audit passes through, if it is not, then
The audit fails, which is back in the mark task of former mark personnel, and former mark personnel is prompted to mark again.
S5. the task list for the task data that audit passes through and the corresponding samples pictures that marked are generated into instruction
Practice data.
Specifically, all task datas audited and passed through are extracted in task form, a task list are generated, one by one by institute
It states each task data in task list and the corresponding samples pictures that marked of identification number corresponding with the task data is closed
Connection generates training data, so that subsequent photographic model is trained.
As shown in Fig. 2, need to judge the task data for having marked samples pictures and through scene before executing step S5
Whether the task data that text detection obtains is consistent, if so, executing step S5;If it is not, executing step A;
A. the task data that the audit fails is marked, the task data after modification is labeled returns
Execute step S4.
In the present embodiment, if task data does not pass through audit, the task data can be marked in auditor, and
The labeled data is back in the mark task of former mark personnel, is marked again by former mark personnel, after completing mark, then it is right
Task data is audited again, to guarantee the accuracy of task data.
The task data that the audit fails may also include that
B. by the audit fails it is corresponding with the task data marked samples pictures and be marked, modification with through marking
Described after note has marked the corresponding task data of samples pictures, returns to step S4.
In this step, if task data does not pass through audit, auditor can mark to corresponding with the task data
Note samples pictures are marked, and the labeled data is back in the mark task of former mark personnel, by former mark personnel weight
New mark, after completing mark, then audits task data, again to guarantee the accuracy of task data.
The present invention establishes task form according to samples pictures to be marked, in order to be managed collectively to labeled data;It obtains
The labeled data for taking target object in samples pictures, according to labeled data pair and the associated task data of samples pictures to be marked into
Row updates, and audits to the task data for having marked samples pictures, to guarantee the accuracy of labeled data;Audit is logical
The task list for the task data crossed and corresponding samples pictures generate training data, in order to according to training data to data set
Model training, to improve the precision of model training.
Embodiment two
Referring to Fig. 3, a kind of picture for model training of the present embodiment mark fill 1, comprising: establish unit 11,
Extraction unit 13, mark unit 12, audit unit 14 and generation unit 15;
Unit 11 is established, for establishing the task form that samples pictures to be marked are established according to an at least width, described
The corresponding task data of identification number of the every samples pictures to be marked of each task data association one being engaged in list;
In practical applications, the samples pictures to be marked (such as: 1000) in storage unit are read by management server
Establish a task form, wherein every width picture corresponding task data in the task form, each width picture corresponding one is only
One identification number (identification number can be with the ID number of picture), which is associated with task data.
Extraction unit 13, it is single from the storage for the associated identification number of the task data according to the task form
The samples pictures to be marked corresponding with the identification number are extracted in member;
Further, after receiving samples pictures to be marked, the received samples pictures to be marked can be carried out pre-
The size adjusting of the samples pictures to be marked is pre-set dimension by processing.
In order to avoid the samples pictures to be marked received have exception, such as: it is too big that format does not meet preset requirement size
Or format mismatching can not load, the low content that can not clearly observe in picture of clarity etc., and the side using noise reduction can be used
Formula pre-processes samples pictures to be marked, so that picture can be reached with normal load, clarity to be wanted with normal observation
It asks.
Unit 12 is marked, for identification target object in the samples pictures to be marked, it is corresponding to extract the target object
The labeled data that converts that information to of information, the mark according to the labeled data pair and the samples pictures to be marked
Number associated task data is updated;
In practical applications, the labeled data of target object in samples pictures to be marked is acquired using mark client,
Such as: the location information of target object in samples pictures to be marked, and pass through the mark corresponding identification information of client typing.
It should be understood that can amplify, reduce to samples pictures to be marked in order to facilitate mark when being labeled
And rotation etc., in order to watch the target object that samples pictures to be marked need to mark.
By marking the samples pictures to be marked of client scan, according to the identification pair of the identification number of the samples pictures to be marked
The marking types answered, based on the marking types, corresponding image recognition region is (such as: identity card in the samples pictures to be marked
Middle name, address, the corresponding position frame of citizen ID certificate number position coordinates) identified, to obtain the letter of target object
Breath, converts this information into labeled data, is associated with according to the labeled data pair with the identification number of the samples pictures to be marked
The task data be updated so that the content in task data and the samples pictures corresponds.If marking types are
When type of credential is object type, corresponding marking types are identified according to the identification number of the samples pictures to be marked, are based on the mark
Corresponding image recognition region is (such as: position of the face frame in samples pictures to be marked in the samples pictures to be marked for type
Set coordinate), obtain " face head portrait " identification information, convert this information into labeled data, according to the labeled data pair with
The associated task data of the identification number of the samples pictures to be marked is updated, so that task data and the samples pictures
In content correspond
When being labeled to samples pictures to be marked, labeled data directly enters corresponding with samples pictures to be marked
In data of being engaged in, in order to the management to the task data in task form.
Unit 14 is audited, for auditing to the task data for having marked samples pictures;
It is needed after completing mark to the number of tasks for having marked samples pictures to improve the accuracy of the data of mark
According to being audited.In practical applications, after to the mark of samples pictures to be marked, which can further be examined
Core, judges whether labeled data corresponds to the corresponding target object marked in samples pictures, if so, audit passes through, if
No, then the audit fails, which is back in the mark task of former mark personnel, and former mark personnel is prompted to mark again
Note.
Generation unit 15, the task list of the task data for passing through audit and corresponding described has marked sample
This picture generates training data.
Specifically, all task datas audited and passed through are extracted in task form, a task list are generated, one by one by institute
It states each task data in task list and the corresponding samples pictures that marked of identification number corresponding with the task data is closed
Connection generates training data, so that subsequent photographic model is trained.
The present invention establishes task form according to samples pictures to be marked, in order to be managed collectively to labeled data;It obtains
The labeled data for taking target object in samples pictures, according to labeled data pair and the associated task data of samples pictures to be marked into
Row updates, and audits to the task data for having marked samples pictures, to guarantee the accuracy of labeled data;Audit is logical
The task list for the task data crossed and corresponding samples pictures generate training data, in order to according to training data to data set
Model training, to improve the precision of model training.
Embodiment three:
To achieve the above object, the present invention also provides a kind of computer equipment 2, which includes multiple calculating
The component part of machine equipment 2, the device 1 that the picture for model training of embodiment two marks is dispersed in different computers
In equipment 2, computer equipment 2 can be smart phone, tablet computer, laptop, desktop computer, the machine for executing program
Rack server, blade server, tower server or Cabinet-type server (including independent server or multiple clothes
Server cluster composed by business device) etc..The computer equipment 2 of the present embodiment includes, but is not limited to: can be total by system
Line is in communication with each other the device 1 that memory 21, processor 23, network interface 22 and the picture for model training of connection mark
(referring to Fig. 4).It should be pointed out that Fig. 4 illustrates only the computer equipment 2 with component-, it should be understood that not
It is required that implement all components shown, the implementation that can be substituted is more or less component.
In the present embodiment, the memory 21 includes at least a type of computer readable storage medium, described readable
Storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory etc.), random access storage device
(RAM), static random-access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory
(EEPROM), programmable read only memory (PROM), magnetic storage, disk, CD etc..In some embodiments, memory
21 can be the internal storage unit of computer equipment 2, such as the hard disk or memory of the computer equipment 2.In other implementations
In example, memory 21 is also possible to the grafting being equipped on the External memory equipment of computer equipment 2, such as the computer equipment 2
Formula hard disk, intelligent memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card, flash card
(Flash Card) etc..Certainly, the memory 21 can also both including computer equipment 2 internal storage unit and also including it
External memory equipment.In the present embodiment, memory 21 is installed on the operating system of computer equipment 2 and all kinds of commonly used in storage
Application software, for example, embodiment one the picture mask method for model training program code etc..In addition, memory 21 is also
It can be used for temporarily storing the Various types of data that has exported or will export.
The processor 23 can be in some embodiments central processing unit (Central Processing Unit,
CPU), controller, microcontroller, microprocessor or other data processing chips.The processor 23 is commonly used in control computer
The overall operation of equipment 2 for example executes control relevant to the computer equipment 2 progress data interaction or communication and processing
Deng.In the present embodiment, the processor 23 is for running the program code stored in the memory 21 or processing data, example
The device 1 that the picture for model training as described in running marks.
The network interface 22 may include radio network interface or wired network interface, which is commonly used in
Communication connection is established between the computer equipment 2 and other computer equipments 2.For example, the network interface 22 is for passing through
The computer equipment 2 is connected by network with exterior terminal, establishes data between the computer equipment 2 and exterior terminal
Transmission channel and communication connection etc..The network can be intranet (Intranet), internet (Internet), the whole world
Mobile communcations system (Global System of Mobile communication, GSM), wideband code division multiple access (Wideband
Code Division Multiple Access, WCDMA), 4G network, 5G network, bluetooth (Bluetooth), the nothings such as Wi-Fi
Line or cable network.
It should be pointed out that Fig. 4 illustrates only the computer equipment 2 with component 21-23, it should be understood that simultaneously
All components shown realistic are not applied, the implementation that can be substituted is more or less component.
In the present embodiment, the device 1 that the picture for model training being stored in memory 21 marks may be used also
To be divided into one or more program module, one or more of program modules are stored in memory 21, and
It is performed by one or more processors (the present embodiment is processor 23), to complete the present invention.
Example IV:
To achieve the above object, the present invention also provides a kind of computer readable storage mediums comprising multiple storage mediums,
Such as flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory), random access storage device (RAM), static state
Random access storage device (SRAM), electrically erasable programmable read-only memory (EEPROM), may be programmed read-only memory (ROM)
Read-only memory (PROM), magnetic storage, disk, CD, server, App are stored thereon with computer using store etc.
Program, program realize corresponding function when being executed by processor 23.The computer readable storage medium of the present embodiment is used for storing
In the device 1 that the picture of model training marks, the picture for model training of embodiment one is realized when being executed by processor 23
Mask method.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (10)
1. a kind of picture mask method for model training, which is characterized in that a storage unit is provided, for storing at least one
Samples pictures to be marked, include the following steps:
S1. a task form is established, each task data in the task form is associated with a samples pictures to be marked
Identification number;
S2. the associated identification number of the task data according to the task form extracts and the mark from the storage unit
Know number corresponding samples pictures to be marked;
S3. it identifies target object in the samples pictures to be marked, extracts the corresponding information of the target object for the information
Labeled data is converted to, the associated number of tasks of identification number according to the labeled data pair and the samples pictures to be marked
According to being updated;
S4. the task data for having marked samples pictures is audited;
S5. the task list for the task data that audit passes through and the corresponding samples pictures that marked are generated into training number
According to.
2. the picture mask method according to claim 1 for model training, which is characterized in that the step S2 according to
The associated identification number of task data described in the task form extracts corresponding with the identification number from the storage unit
The samples pictures to be marked, further includes:
After receiving the samples pictures to be marked, the received samples pictures to be marked are pre-processed, it will be described
The size adjusting of samples pictures to be marked is pre-set dimension.
3. the picture mask method according to claim 1 for model training, which is characterized in that the step S3 identification
Target object in the samples pictures to be marked extracts the corresponding information of the target object and converts that information to mark number
According to, it is updated according to the labeled data pair and the associated task data of identification number of the samples pictures to be marked,
Include:
Identification number according to the samples pictures to be marked obtains the marking types of the samples pictures to be marked, each mark
Know a number corresponding marking types, each marking types correspond to corresponding image recognition region, and described image identification region is corresponding
Target object to be identified;
According to marking types sample to be marked described in the corresponding image recognition region recognition of the samples pictures to be marked
Target object in picture;
It extracts the corresponding information of the target object and converts that information to labeled data;
It is updated according to the labeled data pair and the associated task data of identification number of the samples pictures to be marked.
4. the picture mask method according to claim 1 for model training, which is characterized in that executing the step
Before S1 also, comprising:
Several samples pictures to be marked are converted video data to, several described samples pictures to be marked are stored in the storage
In unit, each samples pictures to be marked and an identification number are corresponded
A task form is established by described, each task data in the task form is associated with a samples pictures to be marked
Identification number.
5. the picture mask method according to claim 1 for model training, which is characterized in that further include:
A. the task data that the audit fails is marked, the task data after modification is labeled is returned and executed
Step S4.
6. the picture mask method according to claim 1 for model training, which is characterized in that further include:
B. by the audit fails it is corresponding with the task data marked samples pictures and be marked, modification with it is labeled after
It is described marked the corresponding task data of samples pictures, return to step S4.
7. the picture mask method according to claim 1 for model training, which is characterized in that the step S4 is to
The task data of mark samples pictures is audited, comprising:
The task data for having marked samples pictures is audited using scene text detection, judgement is described to have marked samples pictures
Task data and through scene text detection obtain task data it is whether consistent.
8. a kind of device that the picture for model training marks characterized by comprising
Unit is established, for establishing the task form that samples pictures to be marked are established, the task form according to an at least width
In one every samples pictures to be marked of each task data association the corresponding task data of identification number;
Extraction unit is mentioned from the storage unit for the associated identification number of the task data according to the task form
Take the reception samples pictures to be marked corresponding with the identification number;
Unit is marked, target object in the samples pictures to be marked, extracts the corresponding information of the target object for identification
The labeled data converted that information to is associated with according to the labeled data pair with the identification number of the samples pictures to be marked
The task data be updated;
Unit is audited, for auditing to the task data for having marked samples pictures;
Generation unit, the task list of the task data for passing through audit and corresponding described has marked samples pictures
Generate training data.
9. a kind of computer equipment, the computer equipment, including memory, processor and storage are on a memory and can be
The computer program run on processor, the processor realize any one of claim 1 to 7 when executing the computer program
The step of the method.
10. a kind of computer readable storage medium, is stored thereon with computer program, it is characterised in that: the computer program
The step of any one of claim 1 to 7 the method is realized when being executed by processor.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910524478.7A CN110363222A (en) | 2019-06-18 | 2019-06-18 | Picture mask method, device, computer equipment and storage medium for model training |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910524478.7A CN110363222A (en) | 2019-06-18 | 2019-06-18 | Picture mask method, device, computer equipment and storage medium for model training |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110363222A true CN110363222A (en) | 2019-10-22 |
Family
ID=68216709
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910524478.7A Pending CN110363222A (en) | 2019-06-18 | 2019-06-18 | Picture mask method, device, computer equipment and storage medium for model training |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110363222A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112084939A (en) * | 2020-09-08 | 2020-12-15 | 深圳市润腾智慧科技有限公司 | Image feature data management method and device, computer equipment and storage medium |
CN112580735A (en) * | 2020-12-25 | 2021-03-30 | 南方电网深圳数字电网研究院有限公司 | Picture online labeling method and device and computer readable storage medium |
CN113178079A (en) * | 2021-04-06 | 2021-07-27 | 青岛以萨数据技术有限公司 | Marking system, method and storage medium for signal lamp and lane line |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106649610A (en) * | 2016-11-29 | 2017-05-10 | 北京智能管家科技有限公司 | Image labeling method and apparatus |
CN107247700A (en) * | 2017-04-27 | 2017-10-13 | 北京捷通华声科技股份有限公司 | A kind of method and device for adding text marking |
CN108241741A (en) * | 2017-12-29 | 2018-07-03 | 深圳市金立通信设备有限公司 | A kind of file classification method, server and computer readable storage medium |
CN109635838A (en) * | 2018-11-12 | 2019-04-16 | 平安科技(深圳)有限公司 | Face samples pictures mask method, device, computer equipment and storage medium |
-
2019
- 2019-06-18 CN CN201910524478.7A patent/CN110363222A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106649610A (en) * | 2016-11-29 | 2017-05-10 | 北京智能管家科技有限公司 | Image labeling method and apparatus |
CN107247700A (en) * | 2017-04-27 | 2017-10-13 | 北京捷通华声科技股份有限公司 | A kind of method and device for adding text marking |
CN108241741A (en) * | 2017-12-29 | 2018-07-03 | 深圳市金立通信设备有限公司 | A kind of file classification method, server and computer readable storage medium |
CN109635838A (en) * | 2018-11-12 | 2019-04-16 | 平安科技(深圳)有限公司 | Face samples pictures mask method, device, computer equipment and storage medium |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112084939A (en) * | 2020-09-08 | 2020-12-15 | 深圳市润腾智慧科技有限公司 | Image feature data management method and device, computer equipment and storage medium |
CN112580735A (en) * | 2020-12-25 | 2021-03-30 | 南方电网深圳数字电网研究院有限公司 | Picture online labeling method and device and computer readable storage medium |
CN113178079A (en) * | 2021-04-06 | 2021-07-27 | 青岛以萨数据技术有限公司 | Marking system, method and storage medium for signal lamp and lane line |
CN113178079B (en) * | 2021-04-06 | 2022-08-23 | 青岛以萨数据技术有限公司 | Marking system, method and storage medium for signal lamp and lane line |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2022134588A1 (en) | Method for constructing information review classification model, and information review method | |
CN105740402B (en) | The acquisition methods and device of the semantic label of digital picture | |
CA3017647C (en) | Optical character recognition in structured documents | |
CN110363222A (en) | Picture mask method, device, computer equipment and storage medium for model training | |
US20160098804A1 (en) | Method and system for using interchangeable analytics modules to provide tax return preparation systems | |
CN109902248A (en) | Page display method, device, computer equipment and readable storage medium storing program for executing | |
CN105306495B (en) | user identification method and device | |
CN110362822A (en) | Text marking method, apparatus, computer equipment and storage medium for model training | |
CN107844634A (en) | Polynary universal model platform modeling method, electronic equipment and computer-readable recording medium | |
CN107911227B (en) | Breakpoint data follow-up method, electronic device and computer-readable storage medium | |
CN109543073A (en) | Enterprise's supply and marketing relation map generation method, device and computer equipment | |
CN108021621A (en) | Database data acquisition method, application server and computer-readable recording medium | |
CN107729549A (en) | A kind of robot client service method and system comprising elements recognition | |
CN108241529A (en) | Wages computational methods, application server and computer readable storage medium | |
CN115758451A (en) | Data labeling method, device, equipment and storage medium based on artificial intelligence | |
CN112116325A (en) | Examination and approval form control method and device, electronic equipment and readable storage medium | |
CN108171699A (en) | Setting loss Claims Resolution method, server and computer readable storage medium | |
CN111143434A (en) | Intelligent data checking method, device, equipment and storage medium | |
CN110187895A (en) | Allocating operating system method, apparatus, equipment and computer readable storage medium | |
CN112035676B (en) | User operation behavior knowledge graph construction method and device | |
CN110019076A (en) | Construction method, device, equipment and the readable storage medium storing program for executing of multisystem daily record data | |
CN109345063B (en) | Data processing method and device for wind control system and storage medium | |
CN116681045A (en) | Report generation method, report generation device, computer equipment and storage medium | |
CN109582834B (en) | Data risk prediction method and device | |
CN112750043A (en) | Business data pushing method and device and server |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |