CN109739644B - Data picture labeling method, system and device based on computer - Google Patents
Data picture labeling method, system and device based on computer Download PDFInfo
- Publication number
- CN109739644B CN109739644B CN201811551963.5A CN201811551963A CN109739644B CN 109739644 B CN109739644 B CN 109739644B CN 201811551963 A CN201811551963 A CN 201811551963A CN 109739644 B CN109739644 B CN 109739644B
- Authority
- CN
- China
- Prior art keywords
- pictures
- classification
- classified
- execution end
- task
- 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.)
- Active
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to the field of data processing, in particular to a data picture labeling method, system and device based on a computer. It mainly comprises: acquiring a plurality of pictures to be examined which need to be subjected to data annotation; sequentially coding a plurality of pictures to be examined to form a slice task queue, and performing a slice task on the pictures to be examined according to the coding sequence to obtain a plurality of small pictures to be classified; sequentially coding the small pictures to be classified to form a classification task queue, performing classification tasks on the small pictures to be classified according to the coding sequence, and outputting a classification result; auditing the classification result output by the classification task; and outputting the pictures for data labeling and the corresponding labeling information. By the data picture labeling method, the data picture labeling system and the data picture labeling device, a large number of data pictures can be labeled accurately and efficiently.
Description
[ technical field ] A method for producing a semiconductor device
The invention relates to the field of data processing, in particular to a data picture labeling method, system and device based on a computer.
[ technical background ] A method for producing a semiconductor device
With the continuous development of internet technology in the scientific and technological field, the collection, arrangement, analysis and classification of data information become important.
For example, in the field of unmanned retail, data images of a large number of commodities sold by others need to be collected and labeled in the early stage so as to serve for later-stage sales. The data picture marking of the commodity for sale is the most troublesome and important link, and the traditional marking is to slice and classify the data picture of the commodity for sale manually. When the data volume needing to be labeled and processed is large, the traditional labeling mode cannot meet the labeling requirement of the huge data pictures. The traditional marking method not only reduces the marking efficiency, but also is not beneficial to performance assessment and marking effect examination of marking personnel.
Therefore, in order to solve the existing technical problems, a new data picture labeling method, system and device are urgently needed.
[ summary of the invention ]
In order to solve the technical problems of the traditional data picture marking, the invention provides a data picture marking method, system and device based on a computer.
The invention provides a data picture marking method based on a computer, which solves the technical problem and comprises the following steps:
step S1: acquiring a plurality of pictures to be examined which need to be subjected to data annotation;
step S2: sequentially coding a plurality of pictures to be examined to form a slice task queue, receiving a slice task request, and performing slice task allocation on the pictures to be examined according to the coding sequence, wherein at the moment, the pictures to be examined corresponding to the codes are locked and sliced to obtain a plurality of small pictures to be classified;
step S3: sequentially coding the small pictures to be classified to form a classification task queue, receiving a classification task request, distributing classification tasks to the small pictures to be classified according to the coding sequence, locking and classifying the small pictures to be classified corresponding to the coding, and outputting a classification result;
step S4: auditing the classification result output by the classification task; and
step S5: outputting the pictures for data labeling and the corresponding labeling information;
the slice task formed in step S2 and the classification task queue formed in step S3 are encoded in a manner that is sorted based on the time of picture storage, the picture size, and the picture color to obtain a corresponding encoding order.
Preferably, the step S2 includes the following steps:
step S21: sequentially encoding a plurality of pictures to be examined to form a slice task queue;
step S22: after receiving a slice task request sent by at least one first execution end and/or at least one second execution end, traversing the pictures to be examined in the slice task queue based on sequential coding and distributing the pictures to be examined to the first execution end and/or the second execution end; and
step S23: and after the picture to be checked is distributed, locking the picture to be checked, setting a locking period, and if the information of completing the slicing task returned by the first execution end and/or the second execution end is not received in the locking period, releasing the lock of the locked picture to be checked.
Preferably, the step S3 includes the following steps:
step S31: sequentially coding a plurality of small pictures to be classified to form a classification task queue;
step S32: after receiving a classification task request sent by at least one first execution end and/or at least one second execution end, traversing small pictures to be classified in a classification task queue based on sequential coding and distributing the small pictures to be classified to the first execution end and/or the second execution end; and
step S33: and after the small pictures to be classified are distributed, locking the small pictures to be classified, setting a locking period, and if the information of finishing the classification task returned by the first execution end and/or the second execution end is not received in the locking period, performing lock releasing operation on the locked small pictures to be classified.
The step S33 further includes:
step S331, a first execution end receives the small pictures to be classified, the small pictures to be classified which need to be classified are locked, a locking time limit is set, and the small pictures to be classified which are set with the locking time limit are in a locking state;
step S332 determines whether the first execution end completes the classification task within the locking deadline; if the completion, go to step S333, if not, go to step S334;
step S333, outputting a classification result;
step S334 is performed to release the lock of the locked thumbnail to be classified, and when a new classification task request is received, the thumbnail to be classified after releasing the lock is re-allocated to the corresponding first execution end.
Preferably, after step S33, the method further includes outputting a first classification result of the first execution-side classification; outputting a second classification result of the second execution end classification; the checking in step S4 includes comparing the first classification result with the second classification result, and obtaining a comparison result; if the comparison results are the same, outputting the pictures for data annotation and the corresponding annotation information; if the comparison result is different, the first execution end is redistributed to carry out secondary classification.
The invention provides a data picture marking system based on a computer, which comprises: the receiving module is used for acquiring a plurality of pictures to be examined which need to be subjected to data annotation; the slicing module is used for sequentially coding a plurality of pictures to be classified to form a slicing task queue, receiving a slicing task request, and carrying out slicing task allocation on the pictures to be classified according to the coding sequence, wherein at the moment, the small pictures to be classified corresponding to the codes are locked and classified to obtain a plurality of small pictures to be classified; the classification module is used for sequentially coding the small pictures to be classified to form a classification task queue, receiving a classification task request, performing classification task allocation on the small pictures to be classified according to the coding sequence, locking and classifying the small pictures to be classified corresponding to the coding at the moment, and outputting a classification result; the auditing module is used for auditing the result output by the classification task; the output module is used for outputting the pictures for data annotation and the corresponding annotation information; the coding mode is based on the time for storing pictures, the size of the pictures and the colors of the pictures to be sequenced so as to obtain a corresponding coding sequence.
Preferably, the slicing module comprises: the slice task queue storage module is used for storing a plurality of pictures to be examined and codes corresponding to the pictures to be examined; the slicing task allocation module is used for traversing codes corresponding to the pictures to be examined in the slicing task queue and allocating the pictures to be examined to the first execution end and/or the second execution end after receiving the slicing task request sent by the first execution end and/or the second execution end; and the first locking module is used for locking and releasing the pending pictures which are distributed to the first execution end to execute the slicing task.
Preferably, the classification module comprises: the classification task queue storage module is used for storing a plurality of small pictures to be classified and codes corresponding to the small pictures; the classification task allocation module is used for traversing codes corresponding to the small pictures to be classified in the classification task queue and allocating the small pictures to be classified to the first execution end and/or the second execution end after receiving the classification task request sent by the first execution end and/or the second execution end; and the second locking module is used for locking and releasing the small pictures to be classified which are distributed to the first execution end to execute the classification task.
Preferably, the auditing module comprises a classification result extractor and a comparator; the classification result extractor is used for extracting a first classification result and a second classification result; and the comparator is used for comparing the first classification result and the second classification result extracted by the classification result extractor and outputting a difference result.
The invention provides a data picture labeling device based on a computer, which comprises a storage unit and a processing unit, wherein the storage unit is used for storing a computer program, and the processing unit is used for executing steps in a data picture labeling method through the computer program stored in the storage unit.
Compared with the prior art, the data picture labeling method based on the computer provided by the invention has the advantages that compared with the traditional labeling method, the slicing task and the classification task are divided into two tasks which are independently carried out, so that parallelization and pipelining of massive data picture labeling can be realized, the labeling efficiency is improved, and the labeling task of the data picture can be accurately and efficiently completed by encoding the picture and forming the corresponding slicing task sequence and the classification task sequence.
Firstly, a plurality of pictures to be examined are sequentially coded to form a slice task queue, so that the pictures to be examined are controlled, and the state information of the pictures to be examined is conveniently tracked and recorded; in addition, in the slice task allocation process, the slices are traversed one by one according to the coding sequence, omission does not occur, and the slice accuracy is improved.
Secondly, sequentially coding a plurality of small pictures to be classified obtained after the slicing task is completed to form a classification task queue, so that the small pictures to be classified are controlled, and the state information of the pictures to be examined is conveniently tracked and recorded; in the process of distributing the classification tasks, the encoding sequence is traversed one by one, omission does not occur, and the classification accuracy is improved.
Finally, the classification output result is audited through an auditing process, so that the labeling accuracy is improved; furthermore, by calling coding information in the slice task queue and the classification task queue, the state of the picture corresponding to the coding can be checked, so that the marking workload and the marking accuracy of each person in the manual operation process can be checked, and performance assessment can be performed.
According to the data picture labeling method provided by the invention, manual operation and/or artificial intelligence can be selected to complete the data picture labeling task in the picture labeling process, so that the working flexibility is increased, and the manual workload is further reduced; standardized data acquisition can be realized, and the data can be conveniently used for training in a picture recognition algorithm.
In the manual operation process, the first execution end actively preempts the work task, so that the problem that in the whole labeling task process, when the labeling task is overstocked too much, manual processing is quickly collected without re-distributing tasks already distributed in the system is solved, the flexibility is further improved, and the occurrence of manual intervention is reduced.
In the picture processing process, the processed picture is locked and the locking time is set, so that the problem that a plurality of execution ends operate the same picture at the same time is avoided, and the picture is locked by the same execution end for a long time is solved.
The invention also provides a data picture marking system and device based on the computer, which have the same beneficial effects as the data picture marking method and can accurately and efficiently complete the marking task of the data picture; particularly, parallelization and pipelining of massive data picture labeling are realized, and the labeling efficiency is improved.
[ description of the drawings ]
FIG. 1 is a flow chart of a method for annotating data pictures based on a computer according to a first embodiment of the present invention;
FIG. 2 is a flowchart illustrating a slicing task of a computer-based data picture labeling method according to a first embodiment of the present invention;
FIG. 3 is a flowchart illustrating a process of locking a pending picture in a slicing task according to a first embodiment of the present invention;
FIG. 4 is a flowchart illustrating a classification task of a method for providing computer-based annotation of data pictures according to a first embodiment of the present invention;
FIG. 5 is a flowchart illustrating a process of locking small pictures to be classified in a classification task of the computer-based data picture labeling method according to the first embodiment of the present invention;
FIG. 6 is a schematic diagram illustrating a search encoding process of a computer-based data picture labeling method according to a first embodiment of the present invention;
FIG. 7 is a schematic diagram of a frame structure of a computer-based data picture annotation system according to a second embodiment of the present invention;
FIG. 8 is a block diagram of a slicing module of the computer-based data picture labeling system according to a second embodiment of the present invention;
FIG. 9 is a block diagram of a classification module of the computer-based data image annotation system according to a second embodiment of the present invention;
FIG. 10 is a block diagram of a framework of an audit module of the system for annotating data pictures based on a computer according to a second embodiment of the present invention;
fig. 11 is a schematic structural diagram of a frame of a computer-based data picture annotation device according to a third embodiment of the invention.
The attached drawings are as follows:
10. a data picture labeling method based on a computer;
21. a slicing task queue; 211. a first slice code; 212. encoding a second slice; 213. encoding a third slice; 214. encoding the Nth slice; 22. classifying the task queue; 221. a first classification code; 222. second classification coding; 223. third classification coding; 224. fourth classification coding; 225. and (4) encoding the Nth classification.
30. A computer-based data picture annotation system; 31. a receiving module; 33. a slicing module; 34. a classification module; 35. an audit module; 36. an output module; 331. a slice task queue storage module; 333. a slice task allocation module; 334. a first locking module; 3341. a first locker; 3342. a first judger; 3343. a first output device; 3344. a first releaser; 341. a classification task queue storage module; 343. a classification task allocation module; 344. a second locking module; 3441. a second locker; 3442. a second determiner; 3443. a second output device; 3444. a second releaser; 351. a classification result extractor; 352. a comparator;
40. a data picture labeling device based on a computer; 41. a storage unit; 42. and a processing unit.
[ detailed description ] embodiments
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, the present invention discloses a data picture labeling method 10 based on a computer, which can be used for rapidly labeling a data picture to be labeled and outputting the data picture and corresponding labeling information. The data picture is a plane medium composed of pictures or images containing specific information.
The computer-based data picture labeling method 10 comprises the following steps:
step S1: acquiring a plurality of pictures to be examined which need to be subjected to data annotation;
step S2: sequentially coding a plurality of pictures to be examined to form a slice task queue, and performing a slice task on the pictures to be examined according to the coding sequence to obtain a plurality of small pictures to be classified;
step S3: sequentially coding the small pictures to be classified to form a classification task queue, performing classification tasks on the small pictures to be classified according to the coding sequence, and outputting a classification result;
step S4: auditing the classification result output by the classification task; and
step S5: and outputting the pictures for data labeling and the corresponding labeling information.
Specifically, in step S1, the multiple pending pictures to be examined that need to be subjected to data annotation are obtained, where the multiple pending pictures include data pictures that are acquired in either manual acquisition or automatic acquisition or a combination of both, and the data pictures are imported into the system.
Step S2 sequentially encodes a plurality of pictures to be examined to form a slice task queue, and performs a slice task on the pictures to be examined according to the encoding sequence to obtain a plurality of small pictures to be classified. The slicing task is to cut the picture to be examined into small pictures through one or more ways of picture frame, mask or region marking. The multiple pictures to be examined are sequentially coded to form a slice task queue, so that the pictures to be examined are controlled, and the state information of the pictures to be examined is conveniently tracked and recorded; in addition, in the slice task allocation process, the slices are traversed one by one according to the coding sequence, omission does not occur, and the slice accuracy is improved. The picture state information includes: processing execution end, picture serial number, picture slicing processing time and the like. And checking the state information of the picture to be checked corresponding to the code so as to check the workload and the marking accuracy of each person for slicing the picture to be checked in the slicing operation process, and performing performance assessment. In addition, in the slicing task allocation process, the slices are sequentially traversed one by one, and omission does not occur, so that the slicing accuracy can be improved, and the problem of repeated slicing cannot occur.
And step S3, sequentially coding the small pictures to be classified to form a classification task queue, performing classification tasks on the small pictures to be classified according to the coding sequence, and outputting a classification result. And the classification task is to perform corresponding classification and marking on the corresponding objects or other characteristic information in the small pictures to be classified obtained after the pictures to be examined are sliced in the step S2. The small pictures to be classified obtained after the slicing task are sequentially coded to form a classification task queue, so that the small pictures to be classified are controlled, and the state information of the small pictures to be classified is conveniently tracked and recorded; in the process of distributing the classification tasks, the encoding sequence is traversed one by one, omission does not occur, and the classification accuracy can be improved.
The status information of the small pictures to be classified comprises: processing execution end, picture serial number, picture classification processing time and the like. And checking the state information of the small pictures to be classified corresponding to the codes so as to check the marking workload and the marking accuracy of each person in the manual operation process and perform performance assessment. And in the process of distributing the classification tasks, the classification tasks are traversed one by one in sequence, so that omission does not occur, and the classification accuracy can be improved.
Based on the limitations of the steps S2 and S3, the flexibility of the data picture marking work can be increased, and the manual work load is further reduced; standardized data acquisition can be realized, and the data can be conveniently used for training in a picture recognition algorithm.
Optionally, in some embodiments of the present invention, the slicing task is completed through an execution end, and the slicing task is further divided into a plurality of second execution ends and a plurality of first execution ends according to whether the slicing task performs algorithm slicing.
Optionally, in the present invention, the first executing end may execute a slicing work or a picture recognition and slicing tool for a manual operation; in some specific embodiments, the second execution end may be an AI algorithm module to implement a frame operation on a picture to be examined.
As shown in fig. 2, the step S2 mentioned above is to sequentially encode a plurality of pictures to be examined to form a slice task queue, and perform a slice task on the pictures to be examined according to the encoding sequence to obtain a plurality of small pictures to be classified, and may specifically include the following steps:
step S21: sequentially encoding a plurality of pictures to be examined to form a slice task queue;
step S22: after receiving a slice task request sent by a first execution end and/or a second execution end, traversing pictures to be examined in a slice task queue based on sequential coding and distributing the pictures to be examined to the first execution end and/or the second execution end; and
step S23: and after the picture to be checked is distributed, locking the picture to be checked, setting a locking period, and if the information of completing the slicing task returned by the first execution end and/or the second execution end is not received in the locking period, performing lock releasing operation on the locked picture to be checked.
And after receiving the corresponding pictures to be examined, the first execution end and/or the second execution end can process the pictures and output a plurality of small pictures to be classified.
And after receiving a new slicing task request, reallocating the lock-released pending pictures to the corresponding first execution end and/or second execution end.
The first execution end and/or the second execution end sends the slicing task request, and the slicing task is distributed based on the slicing task request, so that the problem of excessive overstocking of the slicing task caused by the system distribution problem in the whole slicing task process can be solved. Based on the distribution method, the distributed tasks do not need to be redistributed, so that the personnel number adjustment has greater flexibility, and the occurrence of manual intervention is reduced.
As shown in fig. 3, taking the first execution end executing the picture slicing as an example, the step S23 may specifically include the following steps:
step S231, after the first execution end receives the picture to be checked, locking the picture to be checked, and setting a locking time limit, wherein the picture to be checked with the locking time limit is in a locking state;
step S232, judging whether the first execution end completes the slicing task within the locking period; if the completion is made, go to step S233, if the completion is not made, go to step S234;
step S233, obtaining corresponding small pictures to be classified;
step S234, performing a lock releasing operation on the locked pending picture, and after receiving a new slice task request, re-allocating the locked pending picture to the corresponding first execution end. Therefore, the first execution end can be prevented from operating the same picture at the same time, and the problem that the picture is locked by the same first execution end for a long time is solved.
Alternatively, in step S232, it may be specifically determined whether the first execution end completes the slicing task within the locking period based on the time when the first execution end returns the information of completing the slicing task.
And the classification task is to perform corresponding labeling on the small pictures to be classified. And the classification task is completed by the first execution end and/or the second execution end. Optionally, the classification task may be executed by the second execution end and the plurality of first execution ends simultaneously in parallel. Specifically, the first execution end executes classification work for manual operation; the second execution end executes the same image classification work as the first execution end by utilizing an AI algorithm.
As shown in fig. 4, in the step S3, sequentially encoding the small pictures to be classified to form a classification task queue, and performing a classification task on the small pictures to be classified according to the encoding sequence includes the following steps:
step S31: sequentially coding a plurality of small pictures to be classified to form a classification task queue;
step S32: after receiving a classification task request sent by a first execution end and/or a second execution end, traversing small pictures to be classified in a classification task queue based on sequential coding and distributing the small pictures to be classified to the first execution end and/or the second execution end; and
step S33: and after the small pictures to be classified are distributed, locking the small pictures to be classified, setting a locking period, and if the information of finishing the classification task returned by the first execution end and/or the second execution end is not received in the locking period, performing lock releasing operation on the locked small pictures to be classified. And after receiving a new classification task request, reallocating the small picture to be classified after releasing the lock to the corresponding first execution end and/or second execution end.
After the first execution end and/or the second execution end receive the corresponding small pictures to be classified, the small pictures can be processed and the classification results of the specific pictures can be output.
The first execution end and/or the second execution end sends the classification task request, and the classification task is distributed based on the request, so that the problem that in the whole classification task process, the flexibility of distribution of the classification task is improved, and the occurrence of manual intervention is reduced.
As shown in fig. 5, taking the first execution end to execute the image classification as an example, the step S33 may specifically include the following steps:
step S331, a first execution end receives the small pictures to be classified, the small pictures to be classified which need to be classified are locked, a locking time limit is set, and the small pictures to be classified which are set with the locking time limit are in a locking state;
step S332 determines whether the first execution end completes the classification task within the locking deadline; if the completion, go to step S333, if not, go to step S334;
step S333, outputting a classification result;
step S334 is executed to release the lock of the locked thumbnail to be classified, and when a new classification task request is received, the locked thumbnail to be classified is re-allocated to the corresponding first execution end.
Therefore, the first execution end can be prevented from operating the same picture at the same time, and the problem that the picture is locked by the same first execution end for a long time is solved.
In other embodiments of the present invention, the classification task can be selectively executed by the second execution end or the plurality of first execution ends.
In the present invention, in the above steps S2 and S3, the plurality of pictures are sequentially encoded to form a corresponding task queue, and an encoded data set is formed for the system to recognize. In other embodiments, in different task queues, the encoding manner may also be sorted based on the time for storing the picture, the size of the picture, the main color of the picture, and other information, which is only used for explanation and is not used as a limitation of the present invention.
In the invention, after the system receives the task request, the system linearly searches the codes in the coded data set, calls the pictures corresponding to the codes and sends the pictures to the corresponding first execution end and/or second execution end. Compared with the prior art, the method and the device have the advantages that the picture slicing and classifying processing are divided into two independent tasks, and the parallelization and pipelining of the tasks with different functions can be favorably realized, so that the picture processing operation is more standard and faster.
Specifically, as shown in fig. 6, taking the first execution end as an example, the steps of the above-mentioned data image annotation method 10 based on the computer are specifically described as follows:
after the to-be-examined pictures are sequentially encoded to form the slice task queue 21, when the system receives a first slice task request, the first slice code 211 is preferentially searched according to a linear search principle, and the to-be-examined picture corresponding to the first slice code 211 is sent to the corresponding first execution end, at this time, the to-be-examined picture corresponding to the first slice code 211 is locked, and the system cannot process the to-be-examined picture corresponding to the first slice code 211.
When the first execution end completes the task of slicing the to-be-inspected picture corresponding to the first slice code 211 within the locking period, a to-be-classified small picture is generated, and the to-be-classified small picture enters the classification task queue 22. While the system cannot search for the first slice code 211.
When the first execution end does not slice the pending picture corresponding to the first slice code 211 within the locking period, the pending picture is unlocked, and the system can search the first slice code 211 in the slice task queue 21 again.
After the system receives the second slicing task request, the system linearly searches for the code of the picture to be examined, and if the picture to be examined corresponding to the first slicing code 211 is unlocked at the moment, the system preferably searches for the first slicing code 211; otherwise, the linear search continues for the second slice encoding 212, and so on.
After the small pictures to be classified are sequentially coded to form the classification task queue 22, when the system receives a first classification task request, the first classification code 221 is preferentially searched according to a linear search principle, and the small pictures to be classified corresponding to the first classification code 221 are sent to the corresponding first execution end, and at this time, the small pictures to be classified corresponding to the first classification code 221 are locked.
And when the first execution end completes the task of classifying the small pictures to be classified corresponding to the first classification codes 221 within the locking time limit, outputting a classification result. Meanwhile, the system always locks the small pictures to be classified corresponding to the first classification code 221, and does not distribute the small pictures.
When the first execution end does not classify the to-be-classified small picture corresponding to the first classification code 221 within the locking period, the to-be-classified small picture is unlocked, and the system can obtain the to-be-classified small picture corresponding to the first classification code 221 again in the classification task queue 22.
And after the system receives the second classification task request, linearly searching the small picture codes to be classified, preferably searching the small pictures to be classified matched with the first classification codes 221 if the small pictures to be classified corresponding to the first classification codes 221 are unlocked, otherwise, continuously linearly searching the small pictures to be classified matched with the second classification codes 222, and the like.
In the data picture labeling method 10 based on a computer provided by the present invention, the step S4 can be specifically subdivided into:
the classification result obtained by the classification task comprises a commodity large classification of SKU (Stock Keeping Unit/Stock Keeping Unit) to be examined corresponding to the object in the picture. The first execution end and the second execution end output corresponding first classification results and second classification results.
The checking step comprises the steps of comparing the first classification result with the second classification result to obtain a comparison result; if the comparison results are the same, outputting the pictures for data annotation and the corresponding annotation information; if the comparison result is different, the first execution end is redistributed to carry out secondary classification. The process of checking the difference result can reduce the workload of manual checking and improve the production efficiency.
Referring to fig. 7, a second embodiment of the present invention provides a computer-based data image annotation system 30, which includes a receiving module 31, a slicing module 33, a classifying module 34, an auditing module 35, and an output module 36.
The receiving module 31 is configured to obtain a plurality of pictures to be examined that need to be subjected to data annotation;
the slicing module 33 is configured to sequentially encode a plurality of pictures to be examined to form a slicing task queue, and perform a slicing task on the pictures to be examined according to the encoding sequence to obtain a plurality of small pictures to be classified;
the classification module 34 is configured to sequentially encode the small pictures to be classified to form a classification task queue, perform a classification task on the small pictures to be classified according to the encoding sequence, and output a classification result;
the auditing module 35 audits the result output by the classification task; and
and the output module 36 is configured to output the pictures subjected to data annotation and the corresponding annotation information.
As shown in fig. 8, the slicing module 33 further includes a slicing task queue storage module 331, a slicing task assignment module 333, and a first locking module 334. The specific contents are as follows:
a slice task queue storage module 331, configured to store a plurality of pictures to be examined and their corresponding codes;
the slice task allocation module 333 is configured to traverse the pictures to be examined in the slice task queue based on sequential coding and allocate the pictures to be examined to the first execution end and/or the second execution end after receiving the slice task request sent by the first execution end and/or the second execution end;
the first locking module 334 is configured to lock and release the to-be-examined picture that is allocated to the first execution end to execute the slicing task;
as further shown in fig. 8, the first locking module 334 further includes a first locker 3341, a first determiner 3342, a first outputter 3343, and a first releaser 3344.
The first locker 3341 is configured to lock the to-be-examined picture allocated to the first execution end to execute the slicing task and set a locking term;
a first judger 3342 configured to judge whether the first execution end completes the slicing task within the locking period;
the first output device 3343 outputs the small pictures to be classified, which are obtained by completing the slicing task within the locking period; and
and the first releaser 3344 releases the pending pictures which are not sliced within the locking period, and after receiving a new slice task request, reallocates the pending pictures after releasing the lock to the corresponding first execution end.
Specifically, when the first releaser 3344 does not receive the information that the first execution end returns to complete the slicing task within the specified time limit, the lock releasing operation is performed on the locked pending picture. When the first output device 3343 receives the information that the first execution end returns to the task of completing the slicing within the specified time limit, the thumbnail to be classified is output.
As shown in fig. 9, the classification module 34 includes: a classification task queue storage module 341, a classification task allocation module 343, and a second locking module 344.
A classification task queue storage module 341, configured to store a plurality of to-be-classified small pictures and their corresponding codes;
the classification task allocation module 343 is configured to, after receiving a classification task request sent by the first execution end and/or the second execution end, traverse codes corresponding to the to-be-classified thumbnails in the classification task queue and allocate the to-be-classified thumbnails to the first execution end and/or the second execution end;
and the second locking module 344 is configured to lock and release the to-be-classified thumbnail distributed to the first execution end to execute the classification task.
As further shown in fig. 9, the second locking module 344 further includes a second locker 3441, a second determiner 3442, a second outputter 3443, and a second releaser 3444.
A second locker 3441, configured to lock the to-be-classified thumbnail distributed to the first execution end to execute the classification task and set a locking time limit;
a second determiner 3442, configured to determine whether the first execution end completes the classification task within the lock deadline;
a second output unit 3443 for outputting the classification result obtained by completing the classification task within the lock-up period; and
the second releaser 3444 releases the to-be-classified photos that have not been classified within the locking period, and after receiving a new classification task request, reallocates the to-be-classified photos after releasing the lock to the corresponding first execution end.
Specifically, when the second releaser 3444 does not receive the information that the first execution end returns the completion of the classification task within the specified time limit, the lock releasing operation is performed on the locked thumbnail to be classified. When the second output device 3443 completes receiving the information of completing the classification task returned by the first execution end within the specified time limit, the classification result is output.
In order to obtain a more accurate classification result, the accuracy of the classification of the first execution end and the second execution end is counted and evaluated.
As shown in fig. 10, the auditing module 35 may also further include a classification result extractor 351 and a comparator 352; the classification result extractor 351 is configured to extract a first classification result and a second classification result; the comparator 352 is configured to compare the first classification result and the second classification result extracted by the classification result extractor 351, and output a difference result.
As shown in fig. 11, a third embodiment of the present invention provides a computer-based data picture annotation device 40, which includes a storage unit 41 and a processing unit 42. The storage unit 41 is configured to store a computer program, and the processing unit 42 is configured to execute the steps in the data picture labeling method through the computer program stored in the storage unit 41. By the data picture labeling device 40 based on the computer, the slicing task and the classification task are divided into two tasks which are independently carried out, so that parallelization and pipelining of massive data picture labeling can be realized, the labeling efficiency is improved, and the data picture labeling task can be accurately and efficiently completed by encoding the pictures and forming the corresponding slicing task sequence and the corresponding classification task sequence.
Compared with the prior art, the data picture labeling method based on the computer provided by the invention has the advantages that compared with the traditional labeling method, the slicing task and the classification task are divided into two tasks which are independently carried out, so that parallelization and pipelining of massive data picture labeling can be realized, the labeling efficiency is improved, and the labeling task of the data picture can be accurately and efficiently completed by encoding the picture and forming the corresponding slicing task sequence and the classification task sequence.
Firstly, a plurality of pictures to be examined are sequentially coded to form a slice task queue, so that the pictures to be examined are controlled, and the state information of the pictures to be examined is conveniently tracked and recorded; in addition, in the slice task allocation process, the slices are traversed one by one according to the coding sequence, omission does not occur, and the slice accuracy is improved.
Secondly, sequentially coding a plurality of small pictures to be classified obtained after the slicing task is completed to form a classification task queue, so that the small pictures to be classified are controlled, and the state information of the pictures to be examined is conveniently tracked and recorded; in the process of distributing the classification tasks, the encoding sequence is traversed one by one, omission does not occur, and the classification accuracy is improved.
Finally, the classification output result is audited through an auditing process, so that the labeling accuracy is improved; furthermore, by calling coding information in the slice task queue and the classification task queue, the state of the picture corresponding to the coding can be checked, so that the marking workload and the marking accuracy of each person in the manual operation process can be checked, and performance assessment can be performed.
According to the data picture labeling method provided by the invention, manual operation and/or artificial intelligence can be selected to complete the data picture labeling task in the picture labeling process, so that the working flexibility is increased, and the manual workload is further reduced; standardized data acquisition can be realized, and the data can be conveniently used for training in a picture recognition algorithm.
In the manual operation process, the first execution end actively preempts the work task, so that the problem that in the whole labeling task process, when the labeling task is overstocked too much, manual processing is quickly collected without re-distributing tasks already distributed in the system is solved, the flexibility is further improved, and the occurrence of manual intervention is reduced.
In the picture processing process, the processed picture is locked and the locking time is set, so that the problem that a plurality of execution ends operate the same picture at the same time is avoided, and the picture is locked by the same execution end for a long time is solved.
The invention also provides a data picture marking system and device based on the computer, which have the same beneficial effects as the data picture marking method and can accurately and efficiently complete the marking task of the data picture; particularly, parallelization and pipelining of massive data picture labeling are realized, and the labeling efficiency is improved.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit of the present invention are intended to be included within the scope of the present invention.
Claims (10)
1. A data picture labeling method based on a computer is characterized in that: the method comprises the following steps:
step S1: acquiring a plurality of pictures to be examined which need to be subjected to data annotation;
step S2: sequentially coding a plurality of pictures to be examined to form a slice task queue, receiving a slice task request, and performing slice task allocation on the pictures to be examined according to the coding sequence, wherein at the moment, the pictures to be examined corresponding to the codes are locked and sliced to obtain a plurality of small pictures to be classified;
step S3: sequentially coding the small pictures to be classified to form a classification task queue, receiving a classification task request, distributing classification tasks to the small pictures to be classified according to the coding sequence, locking and classifying the small pictures to be classified corresponding to the coding, and outputting a classification result;
step S4: auditing the classification result output by the classification task; and
step S5: outputting the pictures for data labeling and the corresponding labeling information;
the slice task formed in step S2 and the classification task queue formed in step S3 are encoded in a manner that is sorted based on the time of picture storage, the picture size, and the picture color to obtain a corresponding encoding order.
2. The computer-based data picture annotation process of claim 1, wherein: the step S2 includes the following steps:
step S21: sequentially encoding a plurality of pictures to be examined to form a slice task queue;
step S22: after receiving a slice task request sent by at least one first execution end and/or at least one second execution end, traversing the pictures to be examined in the slice task queue based on sequential coding and distributing the pictures to be examined to the first execution end and/or the second execution end; and
step S23: and after the picture to be checked is distributed, locking the picture to be checked, setting a locking period, and if the information of completing the slicing task returned by the first execution end and/or the second execution end is not received in the locking period, releasing the lock of the locked picture to be checked.
3. The computer-based data picture annotation process of claim 1, wherein: in the step S3, the method includes the following steps:
step S31: sequentially coding a plurality of small pictures to be classified to form a classification task queue;
step S32: after receiving a classification task request sent by at least one first execution end and/or at least one second execution end, traversing small pictures to be classified in a classification task queue based on sequential coding and distributing the small pictures to be classified to the first execution end and/or the second execution end; and
step S33: and after the small pictures to be classified are distributed, locking the small pictures to be classified, setting a locking period, and if the information of finishing the classification task returned by the first execution end and/or the second execution end is not received in the locking period, performing lock releasing operation on the locked small pictures to be classified.
4. The computer-based data picture annotation process of claim 3, wherein: the step S33 further includes:
step S331, a first execution end receives the small pictures to be classified, the small pictures to be classified which need to be classified are locked, a locking time limit is set, and the small pictures to be classified which are set with the locking time limit are in a locking state;
step S332 determines whether the first execution end completes the classification task within the locking deadline; if the completion, go to step S333, if not, go to step S334;
step S333, outputting a classification result;
step S334 is performed to release the lock of the locked thumbnail to be classified, and when a new classification task request is received, the locked thumbnail to be classified after the lock release operation is re-allocated to the corresponding first execution end.
5. The computer-based data picture annotation process of claim 3, wherein: step S33 includes outputting a first classification result of the first execution side classification; outputting a second classification result of the second execution end classification; the checking in step S4 includes comparing the first classification result with the second classification result, and obtaining a comparison result; if the comparison results are the same, outputting the pictures for data annotation and the corresponding annotation information; if the comparison result is different, the first execution end is redistributed to carry out secondary classification.
6. A data picture labeling system based on a computer is characterized in that: the method comprises the following steps:
the receiving module is used for acquiring a plurality of pictures to be examined which need to be subjected to data annotation;
the slicing module is used for sequentially coding a plurality of pictures to be examined to form a slicing task queue, receiving a slicing task request, and carrying out slicing task allocation on the pictures to be examined according to the coding sequence, wherein at the moment, the pictures to be examined corresponding to the coding are locked and sliced to obtain a plurality of small pictures to be classified;
the classification module is used for sequentially coding the small pictures to be classified to form a classification task queue, receiving a classification task request, performing classification task allocation on the small pictures to be classified according to the coding sequence, locking and classifying the small pictures to be classified corresponding to the coding at the moment, and outputting a classification result;
the auditing module is used for auditing the result output by the classification task; and
the output module is used for outputting the pictures for data annotation and the corresponding annotation information;
the coding mode is based on the time for storing pictures, the size of the pictures and the colors of the pictures to be sequenced so as to obtain a corresponding coding sequence.
7. The computer-based data picture annotation system of claim 6, wherein: the slicing module includes:
the slice task queue storage module is used for storing a plurality of pictures to be examined and codes corresponding to the pictures to be examined;
the slicing task allocation module is used for traversing codes corresponding to the pictures to be examined in the slicing task queue and allocating the pictures to be examined to the first execution end and/or the second execution end after receiving the slicing task request sent by the first execution end and/or the second execution end;
and the first locking module is used for locking and releasing the picture to be examined which is distributed to the first execution end and/or the second execution end to execute the slicing task.
8. The computer-based data picture annotation system of claim 6, wherein: the classification module comprises
The classification task queue storage module is used for storing a plurality of small pictures to be classified and codes corresponding to the small pictures;
the classification task allocation module is used for traversing codes corresponding to the small pictures to be classified in the classification task queue and allocating the small pictures to be classified to the first execution end and/or the second execution end after receiving the classification task request sent by the first execution end and/or the second execution end;
and the second locking module is used for locking and releasing the small pictures to be classified which are distributed to the first execution end and/or the second execution end to execute the classification tasks.
9. The computer-based data picture annotation system of claim 8, wherein: the auditing module comprises a classification result extractor and a comparator; the classification result extractor is used for extracting a first classification result and a second classification result; and the comparator is used for comparing the first classification result and the second classification result extracted by the classification result extractor and outputting a difference result.
10. The utility model provides a data picture mark device based on computer which characterized in that: comprising a storage unit for storing a computer program and a processing unit for performing the steps of the computer-based data picture annotation method according to any one of claims 1-5 by means of the computer program stored in the storage unit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811551963.5A CN109739644B (en) | 2018-12-18 | 2018-12-18 | Data picture labeling method, system and device based on computer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811551963.5A CN109739644B (en) | 2018-12-18 | 2018-12-18 | Data picture labeling method, system and device based on computer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109739644A CN109739644A (en) | 2019-05-10 |
| CN109739644B true CN109739644B (en) | 2021-06-11 |
Family
ID=66360428
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811551963.5A Active CN109739644B (en) | 2018-12-18 | 2018-12-18 | Data picture labeling method, system and device based on computer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109739644B (en) |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108287751B (en) * | 2017-01-09 | 2022-02-01 | 阿里巴巴集团控股有限公司 | Task execution method and device and distributed system |
| CN106874684B (en) * | 2017-03-03 | 2019-03-12 | 浙江禾连网络科技有限公司 | A kind of image labeling system and method |
| CN108197658B (en) * | 2018-01-11 | 2020-08-14 | 阿里巴巴集团控股有限公司 | Image annotation information processing method, device, server and system |
| CN108960297B (en) * | 2018-06-15 | 2021-07-30 | 北京金山云网络技术有限公司 | Picture labeling method, labeling device, equipment and storage medium |
| CN108985293A (en) * | 2018-06-22 | 2018-12-11 | 深源恒际科技有限公司 | A kind of image automation mask method and system based on deep learning |
-
2018
- 2018-12-18 CN CN201811551963.5A patent/CN109739644B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| 一种基于SVDD的图像自动标注方法;何希圣等;《计算机应用与软件》;20101031;第27卷(第10期);第1-4页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109739644A (en) | 2019-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP7465914B2 (en) | Systems and methods for biological particle classification - Patents.com | |
| CN106534344B (en) | Cloud platform video processing system and application method thereof | |
| CN105045820B (en) | Method for processing video image information of high-level data and database system | |
| WO2022228176A1 (en) | Order processing method, and device and storage medium | |
| CN107292186A (en) | A kind of model training method and device based on random forest | |
| US8768064B2 (en) | Image feature extraction device | |
| KR20190062848A (en) | System of big data mining using incremental learning and a method thereof | |
| CN110799997A (en) | Industrial data service, data modeling, and data application platform | |
| CN111026870A (en) | ICT system fault analysis method integrating text classification and image recognition | |
| CN113657504A (en) | Image retrieval method, image retrieval device, computer equipment and storage medium | |
| CN109739644B (en) | Data picture labeling method, system and device based on computer | |
| CN114493514A (en) | Data processing method and device applied to human resources | |
| CN117455324B (en) | Large port operation management method and system based on physical model | |
| CN111061779A (en) | Data processing method and device based on big data platform | |
| CN111651410B (en) | Dynamic balance method and system for sample data | |
| CN111737371A (en) | Data flow detection classification method and device capable of dynamically predicting | |
| CN114610234B (en) | Storage system parameter recommendation method and related device | |
| CN113110804B (en) | Duplicate picture deleting method, device, equipment and storage medium | |
| CN113157814B (en) | Query-driven intelligent workload analysis method under relational database | |
| JP2003281358A (en) | Patent analysis system and patent analysis graph preparation method | |
| CN118014466B (en) | Digital full supply chain logistics method and system | |
| JP6317280B2 (en) | Same form file selection device, same form file selection method, and same form file selection program | |
| CN115841312B (en) | Real-time collection and analysis system for enterprise management file data and collection terminal thereof | |
| CN117115289A (en) | Method and system for converting OFG format file into picture based on MapReduce license | |
| Кузьміних et al. | The influence of current results in a eventoriented data collection system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |