KR102597991B1 - Apparatus and method for adjusting label data for artificial intelligence learning - Google Patents

Abstract

The present invention provides an input unit; and loading processed data from a plurality of acquired source data, performing a polygon labeling process on the data to generate label data, and using a folding tool based on user input received through the input unit, to collect the label data. and a processor that modifies and completes the generation of processed data for machine learning, wherein the folding tool includes a plurality of lines and an artificial intelligence learning label having at least one bend at a connection point of the plurality of lines. It is about a data modification device.

Description

Apparatus and method for adjusting label data for artificial intelligence learning {Apparatus and method for adjusting label data for artificial intelligence learning}

The present invention relates to a method and device for modifying label data for artificial intelligence learning.

Technologies utilizing artificial intelligence are being developed and utilized across industries. Machine learning is essential to utilize artificial intelligence. Machine learning requires a large amount of data, and the process of processing the data so that the machine can recognize it is necessary. A lot of research is being done on how to process data for machine learning, and research is needed on efficient methods for the labeling process.

In particular, polygon labeling is performed by stamping a plurality of points on the lines that make up the object, but in order to correct incorrect points, there is the inconvenience of having to stamp each point again one by one.

Registration number 10-2399425

In order to solve the above problems, the purpose of the present invention is to provide a device and method for correcting label data for artificial intelligence learning that can easily correct incorrect points during polygon labeling.

The problems of the present invention are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.

In order to achieve the above task, the label data modification device for artificial intelligence learning according to an embodiment of the present invention modifies label data using a folding tool.

An apparatus for modifying label data for artificial intelligence learning according to an embodiment of the present invention includes an input unit; and loading processed data from a plurality of acquired source data, performing a polygon labeling process on the data to generate label data, and using a folding tool based on user input received through the input unit, to collect the label data. and a processor that modifies and completes the generation of processed data for machine learning, wherein the folding tool includes a plurality of lines and has at least one bend at a connection point of the plurality of lines.

The processor modifies the label data by moving the folding tool while the folding tool is positioned on the line to be corrected.

The folding tool includes a first line portion extending in a first direction; a second line portion extending in a second direction different from the first direction to form a bend at a connection point with the first line portion; and a push unit vertically connected to the first line unit or the second line unit to receive a user input for a push.

If the processor determines that the label data deviates from preset standards, it modifies the label data.

A method of modifying label data for artificial intelligence learning according to an embodiment of the present invention includes steps of a processor acquiring a plurality of source data;

Loading, by a processor, data to be processed from among the plurality of source data;

A processor performing a polygon labeling process on the data to generate label data; Modifying the label data using a folding tool based on user input received through an input unit; And when the modification of the label data is completed, completing the generation of processed data for machine learning, wherein the folding tool includes a plurality of lines, and at least one bend at a connection point of the plurality of lines. have wealth

Specific details of other embodiments are included in the detailed description and drawings.

According to the present invention, one or more of the following effects are achieved.

First, label data productivity is increased by easily correcting incorrect label data using a folding tool.

Second, in polygon labeling for multiple data, it has the effect of reducing the time required to generate label data.

Third, adjusting the folding tool in various ways has the effect of increasing user convenience.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the claims.

1 is a diagram referenced for explaining data labeling according to an embodiment of the present invention.
Figure 2 is a diagram referenced for explaining a label data correction device for artificial intelligence learning according to an embodiment of the present invention.
Figures 3 and 4 are flow charts referenced in explaining a method of modifying label data for artificial intelligence learning according to an embodiment of the present invention.
Figure 5 is a diagram referenced for explaining a folding tool according to an embodiment of the present invention.
Figures 6 and 7 are diagrams referenced for explaining the operation of modifying label data for artificial intelligence learning according to an embodiment of the present invention.

Hereinafter, embodiments disclosed in the present specification will be described in detail with reference to the attached drawings. However, identical or similar components will be assigned the same reference numbers regardless of reference numerals, and duplicate descriptions thereof will be omitted. The suffixes “module” and “part” for components used in the following description are given or used interchangeably only for the ease of preparing the specification, and do not have distinct meanings or roles in themselves. Additionally, in describing the embodiments disclosed in this specification, if it is determined that detailed descriptions of related known technologies may obscure the gist of the embodiments disclosed in this specification, the detailed descriptions will be omitted. In addition, the attached drawings are only for easy understanding of the embodiments disclosed in this specification, and the technical idea disclosed in this specification is not limited by the attached drawings, and all changes included in the spirit and technical scope of the present invention are not limited. , should be understood to include equivalents or substitutes.

Terms containing ordinal numbers, such as first, second, etc., may be used to describe various components, but the components are not limited by the terms. The above terms are used only for the purpose of distinguishing one component from another.

When a component is said to be "connected" or "connected" to another component, it is understood that it may be directly connected to or connected to the other component, but that other components may exist in between. It should be. On the other hand, when it is mentioned that a component is “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between.

Singular expressions include plural expressions unless the context clearly dictates otherwise.

In this application, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are not intended to indicate the presence of one or more other features. It should be understood that this does not exclude in advance the possibility of the existence or addition of elements, numbers, steps, operations, components, parts, or combinations thereof.

1 is a diagram referenced for explaining data labeling according to an embodiment of the present invention.

Referring to the drawing, a large amount of input data (10) is required to generate output (OP) by performing artificial intelligence (AI) machine learning (ML). Input data 10 must be processed so that it meets the purpose of machine learning and can be recognized by the machine.

Data before processing may be referred to as source data (SD), and data after processing may be referred to as processed data.

To generate the processed data 10, a labeling process must be performed on the original data (SD). Data on which a labeling process has been performed on source data (SD) may be named label data. Label data can be used directly as processed data 10, but if the label data deviates from labeling standards, a process of modifying the label data is necessary.

The device 100 and method (S300) according to an embodiment of the present invention can generate processed data by modifying label data.

Meanwhile, data labeling can be done through bounding box, keypoint, polygon, polyline, semantic segmentation, OCR (Optical Character Recognition), etc.

The device 100 and method (S300) according to an embodiment of the present invention can be used in the case of polygon labeling and polyline labeling.

Figure 2 is a diagram referenced for explaining a label data correction device for artificial intelligence learning according to an embodiment of the present invention.

Referring to the drawings, the device 100 according to an embodiment of the present invention can modify label data for artificial intelligence learning through a method (S300) described later.

The device 100 may include an input unit 110, a communication unit 120, a memory 140, an output unit 150, a processor 170, and a power supply unit 190.

The input unit 110 can convert an input signal into an electrical signal. The input unit 110 may include at least one device for signal conversion. The input unit 110 may include at least one of a keyboard, mouse, touch screen, and microphone.

The input unit 110 can receive user input. The input unit 110 may receive user input for obtaining a plurality of source data, loading data, performing a labeling process, modifying label data, etc.

Meanwhile, the user may be named labeler.

The communication unit 120 may exchange signals with an external device wired or wirelessly using a preset communication protocol.

The communication unit 120 may receive a plurality of source data from an external device (eg, a server or a terminal).

The communication unit 120 may transmit processed data to an external device (eg, a server or a terminal).

The memory 140 can store basic data for the unit, control data for controlling the operation of the unit, and input/output data. In terms of hardware, the memory 140 may be a variety of storage devices such as ROM, RAM, EPROM, flash drive, hard drive, etc.

The memory 140 may store a plurality of source data.

The memory 140 can store label data.

The memory 140 can store processed data.

The memory 140 may store labeling standard information.

The output unit 150 can convert electrical signals into visual or auditory signals. The output unit 150 may include at least one device for signal conversion. The output unit 150 may include at least one of a display and a speaker.

The output unit 150 may output loaded source data.

The output unit 150 can output the labeling process. The output unit 150 may output label data after completing the labeling process.

The output unit 150 may output labeling standard information together with the label data.

The output unit 150 may output the modification process of label data. The output unit 150 may output processed data after the label data has been modified.

The processor 170 may be electrically connected to the input unit 110, communication unit 120, memory 140, output unit 150, interface unit 180, and power supply unit 190.

The processor 170 includes application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, and controllers. It may be implemented using at least one of controllers, micro-controllers, microprocessors, and other electrical units for performing functions.

The processor 170 may be the subject of an operation that performs a method (S300) described later. The processor 170 may generate processed data 10 by performing method S300.

The power supply unit 190 can supply power necessary for the operation of each component of the device 100. The power supply unit 190 can receive power from a commercial power source or a battery and provide it to each component of the device 100.

Figures 3 and 4 are flow charts referenced in explaining a method of modifying label data for artificial intelligence learning according to an embodiment of the present invention.

Referring to the drawings, the method (S300) according to an embodiment of the present invention can modify label data for artificial intelligence learning. The method (S300) may include acquiring a plurality of source data (S310) and generating processed data (S320).

Referring to FIG. 3, the processor 170 may acquire a plurality of source data (S310). The processor 170 may obtain a plurality of source data from an external device through the communication unit 120. The processor 170 may store the acquired source data in the memory 140. The processor 170 may obtain a plurality of source data based on user input through the input unit 110.

The processor 170 may process source data to generate processed data (S320). The processor 170 may generate processed data according to labeling standards created based on the field of use of the processed data. The processor 170 may generate processed data by performing a labeling process. The processor 170 may generate processed data by modifying label data.

The processor 170 may generate processed data based on user input through the input unit 110.

The step of generating processed data (S320) includes loading the data to be processed (S330), generating label data (S340), determining whether the label data deviates from the labeling standard (S350), and labeling the data. It may include a step of modifying (S360) and a step of completing the creation of processed data (S370).

The processor 170 may load processing target data from a plurality of source data stored in the memory 140 (S330). Data to be processed may be determined depending on the purpose of machine learning. For example, when machine learning is for the purpose of implementing a specific function of autonomous driving, the processor 170 may load any one of the vehicle exterior image data as data to be processed.

The processor 170 may load data to be processed based on user input through the input unit 110.

The processor 170 may generate label data by performing a labeling process on data to be processed (S340). The labeling process can be performed through at least one of the following methods: bounding box, keypoint, polygon, polyline, semantic segmentation, and OCR (Optical Character Recognition). there is. For example, the processor 170 may generate label data by performing a polygon labeling process on data to be processed.

Meanwhile, selection of a usage method among multiple labeling methods may be determined depending on the purpose or target of machine learning.

The step of generating label data (S340) may include performing a labeling process in which the processor 170 marks a plurality of points along the border of a two-dimensional object in an image based on data to be processed. Neighboring points can be connected with lines to create a final area with an area. This type of labeling process can be described as a polygon labeling process.

Label data can be described as the result of a labeling process performed on data to be processed.

The processor 170 may generate label data by performing a labeling process based on user input through the input unit 110.

The processor 170 may determine whether the label data deviates from preset labeling standards (S350). Labeling criteria can be set according to the purpose or target of machine learning. For example, when a labeling process is performed on CCTV images for the purpose of detecting dangerous substances held by passers-by, the labeling standard can be set to passers-by exposed to more than 70% and all items held by passers-by.

Meanwhile, depending on the embodiment, if it is determined that the label data deviates from the preset labeling standards, the processor 170 may output an indicator (visual or auditory) through the output unit 150.

If it is determined that the label data deviates from the labeling standards, the processor 170 may modify the label data using a folding tool (S360).

A folding tool can be defined as a tool that modifies label data by moving the position of a point created in the label data according to user input.

The folding tool may include a plurality of lines and have at least one bend at a connection point of the plurality of lines.

In the modifying step (S360), the processor 170 may modify the label data by moving the folding tool while the folding tool is positioned on the line to be modified.

The processor 170 may modify label data based on user input through the input unit 110.

Referring to Figure 4, the step of modifying the label data (S360) includes the step of calling the folding tool (S410), the step of adjusting the folding tool (S420), the step of moving the folding tool (S430), and the modification complete state. It may include a step of determining whether the method is recognized (S440) and a step of removing the folding tool (S450).

The processor 170 may call a folding tool (S410). The processor 170 may call a folding tool from the memory 140 based on user input through the input unit 110. The processor 170 may call a folding tool on the image while the image based on label data is displayed on the display. The processor 170 may display a folding tool on the outer line of the object to be labeled in polygon labeling.

The processor 170 can adjust the folding tool (S420). The adjustment operation for the folding tool will be described with reference to FIG. 5.

The processor 170 can move the folding tool (S430). The processor 170 may push the folding tool based on user input through the input unit 110.

The processor 170 may move the points by pushing a plurality of points formed along the border with a folding tool in the polygon labeling process.

When label data modification is completed (S440), the processor 170 may remove the folding tool (S450).

If label data modification is not completed (S440), step S430 can be repeated.

Referring to FIG. 3 again, when label data modification is completed (S360), the processor 170 may complete generating processed data for machine learning (S370).

The processor 170 may store processed data in the memory 140 .

The processor 170 may transmit processed data to an external device through the communication unit 120.

Figure 5 is a diagram referenced for explaining a folding tool according to an embodiment of the present invention.

Referring to the drawing, the processor 170 may call the folding tool 510. The processor 170 may call the folding tool 510 on the label data image that needs to be modified.

The folding tool 510 can be used when there is a bend in the line constituting the two-dimensional object.

The folding tool 510 may include a first line portion 511, a second line portion 512, a bend portion 513, and a push portion 514.

The first line portion 511 and the second line portion 512 may be defined as being disposed on a plurality of points created on a line to be modified to recognize a plurality of points to be modified.

The first line portion 511 may extend in the first direction.

The second line portion 512 may extend in a second direction that is different from the first direction to form a bent portion 513 at a connection point with the first line portion 511.

The push unit 514 may be defined as being vertically connected to the first line unit 511 to receive a user input for pushing.

The processor 170 may position an indicator (e.g., mouse point) on the push unit 514 while the first line unit 511 and the second line unit 512 are located on the line to be modified. . In this case, the bent portion 513 is located in the line to be corrected. With the indicator located on the push unit 514, when the indicator is moved while at least one button is pressed through the input unit 110, the first line unit 511 and the second line unit 512 move and a plurality of lines are formed. The point to be modified can also be moved together.

The processor 170, through the first line portion 511 and the second line portion 512, overlaps the first line portion 511 and the second line portion 512 or is located within a preset distance. Points subject to correction can be recognized.

The processor 170 may receive a push input through a user input through the input unit 110 while the indicator is located on the push unit 514.

The processor 170 can adjust the folding tool 510. The processor 170 may adjust the folding tool 510 based on user input through the input unit 110.

The processor 170 may adjust the angle of the interior angle formed by the first line portion 511 and the second line portion 512. The processor 170 adjusts the interior angle to an obtuse angle, as illustrated by reference numeral 510, adjusts the interior angle to a right angle, as illustrated by reference symbol 520, or adjusts the interior angle to an acute angle, as illustrated by reference symbol 530. can be adjusted.

The processor 170 may add a third line portion 515 to the folding tool 510, as illustrated by reference numeral 540. If the two-dimensional object subject to labeling has two bent parts, the folding tool 540 including the first line part 511, the second line part 512, and the third line part 515 is used. It can be.

The processor 170 may adjust the folding tool 510 (530) by increasing or decreasing the length of at least one of the first line portion 511 and the second line portion 512. By increasing or decreasing the length of at least one of the first line portion 511 and the second line portion 512, a correction operation can be performed in response to various sizes of the object.

The processor 170 adjusts the folding tool 510 so that the posture of at least one of the first line portion 511 and the second line portion 512 displayed on the output portion 150 is changed (520). 560) You can. By changing the posture of at least one of the first line portion 511 and the second line portion 512, a correction operation can be performed in response to various shapes of the object.

The processor 170 may change the position of the push unit 514 in a relative relationship with the first line unit 511 (540). The processor 170 may adjust the folding tool 510 (570) by increasing or decreasing the area of the push unit 514.

The location of the push unit 514 may indicate a part of high importance in labeling. The portion of the first line portion 511 where the push portion 514 is located may be classified as a portion of high importance. When modifying, the processor 170 may additionally create a point at the portion where the push unit 514 is located.

When the area of the push unit 514 increases and the portion of the first line portion 511 where the push portion 514 is located increases, the processor 170 may generate additional points corresponding to the increased portion.

Figures 6 and 7 are diagrams referenced for explaining the operation of modifying label data for artificial intelligence learning according to an embodiment of the present invention.

Referring to FIG. 6, the processor 170 may load data to be processed and display an image based on the data on a display.

The processor 170 may generate label data by performing a labeling 620 process based on user input through the input unit 110.

The processor 170 may display an image 610 based on label data on the display.

The processor 170 may generate the folding tool 630 based on user input. The folding tool 630 may be displayed on the image 610 based on label data.

The processor 170 can adjust the folding tool 630. The processor 170 may adjust the folding tool 630 based on user input through the input unit 110.

Meanwhile, the processor 170 may change the position of the folding tool 630. The processor 170 may change the position of the folding tool 630 based on user input through the input unit 110. The processor 170 may change the position of the folding tool 630 toward the point to be modified.

Meanwhile, the labeling 620 process may be performed through the process of stamping a plurality of points. In this case, multiple points may be connected with lines. These connected lines can function to distinguish between objects and backgrounds in the image.

Referring to FIG. 7 , the processor 170 may move the folding tool 630 toward the point to be corrected. The processor 170 may push the folding tool 630 while the folding tool 630 is in contact with the point to be modified. The processor 170 may move the point to be modified together with the folding tool 630 by a specific displacement in the direction in which the folding tool 630 is pushed. As the point to be modified is moved, the length of the line connecting the moved point and the non-moved point increases.

In a state in which the folding tool 630 is pushed, the processor 170 may drag the folding tool 630 in a direction 680 that is different from the pushing direction 670 of the folding tool 630. The processor 170 may move the point to be modified together with the folding tool 630 in the direction in which the folding tool 630 is dragged. As the point to be modified is moved, the length of the line between the moved points to get closer to the previously moved point becomes shorter, and the length of the line between the moved point and the non-moved point increases.

The processor 170 may move the folding tool 630 to another point after pushing and dragging. The processor 170 may push the folding tool 630 again while the folding tool 630 is in contact with another point to be modified.

The above-described present invention can be implemented as computer-readable code on a program-recorded medium. Computer-readable media includes all types of recording devices that store data that can be read by a computer system. Examples of computer-readable media include HDD (Hard Disk Drive), SSD (Solid State Disk), SDD (Silicon Disk Drive), ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. There is. Additionally, the computer may include a processor or control unit. Accordingly, the above detailed description should not be construed as restrictive in all respects and should be considered illustrative. The scope of the present invention should be determined by reasonable interpretation of the appended claims, and all changes within the equivalent scope of the present invention are included in the scope of the present invention.

100: Label data correction device for artificial intelligence learning
S300: How to modify label data for artificial intelligence learning

Claims (5)

input unit; and
Loading processed data from a plurality of acquired source data,
Generate label data by performing a polygon labeling process on the data,
Based on the user input received through the input unit, the label data is modified using a folding tool,
It includes a processor that completes the generation of processed data for machine learning,
The folding tool is,
It includes a plurality of lines and has at least one bend at a connection point of the plurality of lines,
a first line portion extending in a first direction;
a second line portion extending in a second direction different from the first direction to form a bend at a connection point with the first line portion; and
It includes a push unit vertically connected to the first line unit or the second line unit to receive a user input for a push,
The processor,
Modify the label data by moving a plurality of points,
Adjusting the folding tool by increasing or decreasing the length of at least one of the first line portion and the second line portion,
With the folding tool located on the line to be modified, the label data is modified by moving the folding tool,
With the folding tool located on the line to be corrected and the indicator positioned on the push unit, the indicator is moved while at least one button is pressed through the input unit, thereby making the first line unit and the second A label data correction device for artificial intelligence learning that moves multiple correction target points together as the line part moves. delete delete According to clause 1,
The processor,
A label data correction device for artificial intelligence learning that corrects label data when it is judged to be outside of preset standards. Obtaining, by a processor, a plurality of source data;
Loading, by a processor, data to be processed from among the plurality of source data;
A processor performing a polygon labeling process on the data to generate label data;
Modifying the label data using a folding tool based on user input received through an input unit; and
When the modification of the label data is completed, completing the generation of processed data for machine learning,
The folding tool is,
It includes a plurality of lines and has at least one bend at a connection point of the plurality of lines,
a first line portion extending in a first direction;
a second line portion extending in a second direction different from the first direction to form a bend at a connection point with the first line portion; and
It includes a push unit vertically connected to the first line unit or the second line unit to receive a user input for a push,
The above modifying steps are:
Modify the label data by moving a plurality of points,
Adjusting the folding tool by increasing or decreasing the length of at least one of the first line portion and the second line portion,
With the folding tool located on the line to be modified, the label data is modified by moving the folding tool,
With the folding tool located on the line to be corrected and the indicator positioned on the push unit, the indicator is moved while at least one button is pressed through the input unit, thereby making the first line unit and the second A method of modifying label data for artificial intelligence learning in which multiple correction target points are moved together as the line part is moved.