CN116051618A - Image annotation quality auditing system and method - Google Patents

Image annotation quality auditing system and method Download PDF

Info

Publication number
CN116051618A
CN116051618A CN202211389669.5A CN202211389669A CN116051618A CN 116051618 A CN116051618 A CN 116051618A CN 202211389669 A CN202211389669 A CN 202211389669A CN 116051618 A CN116051618 A CN 116051618A
Authority
CN
China
Prior art keywords
auditing
image
quality
labeling
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202211389669.5A
Other languages
Chinese (zh)
Inventor
钱基业
钮益峰
梁琰
陈咏涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Electric Power Research Institute of State Grid Chongqing Electric Power Co Ltd
State Grid Corp of China SGCC
Original Assignee
Electric Power Research Institute of State Grid Chongqing Electric Power Co Ltd
State Grid Corp of China SGCC
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Electric Power Research Institute of State Grid Chongqing Electric Power Co Ltd, State Grid Corp of China SGCC filed Critical Electric Power Research Institute of State Grid Chongqing Electric Power Co Ltd
Priority to CN202211389669.5A priority Critical patent/CN116051618A/en
Publication of CN116051618A publication Critical patent/CN116051618A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/10—File systems; File servers
    • G06F16/16—File or folder operations, e.g. details of user interfaces specifically adapted to file systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30168—Image quality inspection
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Human Computer Interaction (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Geometry (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides an image annotation quality auditing system and method, belongs to the technical field of computer vision, and solves the problems that in the prior art, auditing efficiency is low and a special auditing system is lacked in annotation auditing work; the system comprises a data generation module and a quality auditing module, wherein the data generation module is used for customizing auditing data, the auditing data is an object for image labeling quality auditing in the quality auditing module, and the quality auditing module is used for displaying labeling data to be audited for a user and simultaneously providing operation for the user for performing image labeling quality auditing on the labeling data to be audited; by focusing on target information, shielding irrelevant content, optimizing data display and auditing modes and other measures, the invention provides a new solution for improving the auditing process of the image labeling quality, thereby improving the auditing efficiency and auditing quality of the image labeling quality as a whole.

Description

Image annotation quality auditing system and method
Technical Field
The invention belongs to the technical field of computer vision, and is applied to the process of auditing the quality of image labels, in particular to an auditing system and method for the quality of the image labels.
Background
In the field of computer vision, a large number of images are required to be marked for training a model for both supervised target identification and target segmentation; labeling high-quality images is important to realizing a high-performance supervision method; therefore, quality audit is an important link for controlling the labeling quality of the image.
In the prior art, a plurality of image marking systems, such as LabelMe, label Studio and the like, are provided, but the auditing system of the image marking quality is lacked; the conventional mode directly uses the image annotation system for auditing, or customizes some auditing functions based on the image annotation system; this results in insufficient auditing efficiency and difficulty in customizing the auditing scheme as desired.
When the unmanned aerial vehicle power inspection image is marked, small pins with hundreds of pixels are marked from the large-size image with tens of millions of pixels, the marking is checked, the image is required to be continuously moved and scaled, and the efficiency is extremely low; meanwhile, when the unmanned aerial vehicle power inspection image is marked, hundreds of targets are marked according to the national grid company specifications, and the marks are inspected, so that one user cannot grasp all inspection standards of all types, and the quality of inspection work is also difficult to guarantee; the worker needs to make good use of his own device, so that a special system is designed for the verification process of the image marking quality, thereby improving the verification method of the image marking quality, and being an important requirement in the technical field of computer vision.
Disclosure of Invention
The invention designs a special image marking quality auditing system and an auditing method matched with the system, and provides a new solution for improving the auditing process of the image marking quality; by focusing on target information, shielding irrelevant content, optimizing data display and auditing modes and other measures, the auditing efficiency and auditing quality of the image labeling quality are improved as a whole.
The invention adopts the following technical scheme to achieve the purpose:
an image annotation quality auditing system comprises a data generation module and a quality auditing module; the data generation module is used for customizing auditing data, wherein the auditing data is an object for image annotation quality auditing in the quality auditing module; the quality auditing module is used for displaying the labeling data to be audited for the user and providing the operation of image labeling quality auditing for the labeling data to be audited for the user.
Further, the data generation module comprises a labeling data acquisition module and an auditing data generation module; the annotation data acquisition module is used for acquiring the annotation data of the image; the auditing data generation module is used for customizing the labeling data of the image into the auditing data according to an auditing scheme.
Further, in the annotation data acquisition module, the annotation data comprises an image file and an annotation file corresponding to the image file; the annotation data acquisition module is used for respectively corresponding the image file and the annotation file to form a plurality of image file-annotation file data pairs; the labeling file comprises labeling area information and labeling label information.
The method comprises the steps that the annotation files are generated by annotating the image files, one image file corresponds to only one annotation file, and the image files correspond to the annotation files one by one to form the image file-annotation file data pair; the labeling file mainly comprises two types of information to be checked, namely a labeling area and a labeling label: the labeling area determines the position of the target in the image, and the labeling label determines the attribute of the target in the image.
Further, the audit data generation module is used for decomposing the image file-annotation file data pair into image area-target information data pairs one by one according to the annotated targets, and determining the image area and target information of each target; the auditing data generation module is also used for pushing the image area-target information data pairs to the quality auditing module one by one according to different auditing schemes.
The verification data generation module is used for customizing the marking data of the image into marking data capable of verifying marking quality according to objects one by one, in the process of decomposing the image file-marking file data pair, the marked objects are decomposed into image areas and object information corresponding to the image areas one by one according to the objects, namely the image area-object information data pair is formed, one image area corresponds to one item of marking information, and the image area only comprises one object.
Further, the image area is: in the image file-annotation file data pair, a subgraph containing a target in an annotation file is intercepted from the image file according to the annotation region of the target; the image area is larger than the labeling area of the corresponding target in the labeling file; the target information includes: and in the image region-target information data pair, region information and label information of the corresponding target are extracted from the annotation file.
Further, in the auditing data generation module, in order to realize a strategy of pushing a plurality of image area-target information data pairs to the quality auditing module one by one according to target auditing marking quality one by one, an auditing scheme designed by the auditing module comprises an image priority scheme and a target priority scheme; the image priority scheme firstly examines the labeling quality of one image and then examines the labeling quality of the next image; the target priority scheme firstly examines the labeling quality of the targets of the same class, and then examines the labeling quality of the targets of the next class.
Further, the quality auditing module comprises a labeling data display module, a labeling area auditing module and a labeling label auditing module; the labeling data display module is used for displaying labeling data for user auditing, the labeling area auditing module is used for providing operation for labeling quality of area information in the user auditing target information, and the labeling label auditing module is used for providing operation for labeling quality of label information in the user auditing target information.
Specifically, the functions of the annotation data display module include: and determining a display mode according to the information such as the image area, user setting and the like, and displaying the image area-target information data pair in the display area.
Specifically, the labeling area auditing module is used for providing an area labeling quality option and an interaction mode for judging the labeling quality of the area information for a user, wherein the area labeling quality option comprises over labeling, under labeling and wrong labeling; the label auditing module is used for providing label labeling quality options and interaction modes for judging label information labeling quality of users, wherein the label labeling quality options comprise wrong labels and missed labels.
The invention also provides an image annotation quality auditing method based on the image annotation quality auditing system, which comprises the following steps:
s1, an annotation data acquisition module acquires an image file and an annotation file in annotation data, and the image file and the annotation file are in one-to-one correspondence to form an image file-annotation file data pair;
s2, the audit data generation module decomposes the image file-annotation file data pairs into corresponding image area-target information data pairs one by one according to the annotated targets, each target corresponds to one image area-target information data pair, and the image area and the target information of each target are determined according to the image area-target information data pairs;
s3, an auditing data generation module selects an auditing scheme, and pushes the image area-target information data pair to a quality auditing module for auditing according to a data pair pushing strategy in the auditing scheme;
s4, after the quality checking module receives the pushed image area-target information data pair, the image area-target information data pair is displayed in a display area;
s5, a user uses a labeling area auditing module to label quality of the area information in the displayed image area-target information data pair, and selects a corresponding area labeling quality option to complete area labeling quality auditing;
s6, a user uses a labeling label auditing module to label quality of label information in the displayed image region-target information data pair, and selects a corresponding label labeling quality option to complete label labeling quality auditing.
Wherein the auditing scheme comprises an image priority scheme and a target priority scheme;
the image priority scheme is as follows: firstly checking the labeling quality of one image, and then checking the labeling quality of the next image; the data pair pushing strategy corresponding to the image priority scheme is as follows: firstly pushing all image area-target information data pairs of one image to a quality checking module one by one and checking, and then pushing all image area-target information data pairs of the next image to the quality checking module one by one;
the target priority scheme is as follows: firstly, checking all labeling quality of the same type of targets, and then checking all labeling quality of the next type of targets; the data pair push strategy corresponding to the target priority scheme is: firstly, pushing all image area-target information data pairs of the same type of targets of a plurality of images to a quality auditing module one by one, and after auditing is completed, pushing all image area-target information data pairs of the next type of targets one by one to the quality auditing module for auditing.
In summary, by adopting the technical scheme, the invention has the following beneficial effects:
according to the method, the image area is intercepted, the labeling information and the neighborhood background of the target are focused, redundant information irrelevant to verification is shielded, the display mode and the verification mode of the labeling data are optimized, and the verification efficiency and the verification quality are improved.
According to the method, the auditing process is driven by a data pushing strategy, a traditional auditing mode of auditing one by one according to the images is upgraded to auditing one by one according to the targets, the business flow of auditing of the image labeling quality is remodeled, and according to different application scenes, an auditing scheme customized according to the needs can be provided for a user so as to obtain the optimal auditing efficiency, and the auditing quality is improved, and particularly, the method has an excellent application prospect for auditing of small target labeling in a large-size image.
Drawings
FIG. 1 is a schematic diagram of the system of the present invention;
FIG. 2 is a schematic diagram of an image file-annotation file data pair;
FIG. 3 is a schematic diagram of an image region-target information data pair;
fig. 4 is a schematic view of an application effect of the quality audit module.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1, an image labeling quality auditing system includes a data generation module and a quality auditing module; the data generation module is used for customizing audit data, wherein the audit data is an object for image annotation quality audit in the quality audit module; the quality auditing module is used for displaying the labeling data to be audited for the user and providing the operation of image labeling quality auditing for the labeling data to be audited for the user.
The data generation module comprises a labeling data acquisition module and an auditing data generation module; the annotation data acquisition module is used for acquiring annotation data of the image; the auditing data generation module is used for customizing the labeling data of the image into auditing data according to an auditing scheme.
The quality auditing module comprises a labeling data display module, a labeling area auditing module and a labeling label auditing module; the labeling data display module is used for displaying labeling data for user auditing, the labeling area auditing module is used for providing operation for the labeling quality of the area information in the user auditing target information, and the labeling label auditing module is used for providing operation for the labeling quality of the label information in the user auditing target information.
The embodiment shows the operation mode and the checking method of the image marking quality checking system by the marking quality checking process of the two images.
The two image files employed in this embodiment are denoted as I 1 、I 2 The corresponding markup document is denoted as L 1 、L 2 . Image file I 1 Marked with O 1 、O 2 Two kinds of targets, the corresponding marked area is marked as R 1, 、R 1, The corresponding label is marked as T 1 、T 2 The method comprises the steps of carrying out a first treatment on the surface of the Image file I 2 Also marked with O 1 、O 2 Two kinds of targets, the corresponding marked area is marked as R 2,1 、R 2, The corresponding label is also T 1 、T 2 . The labeling area determines the position of the target in the image, in this embodiment a rectangular area containing the labeling target. The labeling tag determines the attribute of the target in the image, in this embodiment, the category name of the category to which the target belongs.
The marking data acquisition module is used for acquiring marking data, wherein the marking data comprises an image file and a marking file corresponding to the image file; the annotation data acquisition module is used for respectively corresponding the image files and the annotation files to form a plurality of image file-annotation file data pairs, and can be illustrated by referring to fig. 2; the labeling file comprises labeling area information and labeling label information.
The method comprises the steps that the annotation files are generated by annotating the image files, one image file corresponds to only one annotation file, and the image files and the annotation files are in one-to-one correspondence to form image file-annotation file data pairs; the labeling file mainly comprises two types of information to be checked, namely a labeling area and a labeling label: the labeling area determines the position of the target in the image, and the labeling label determines the attribute of the target in the image.
In this embodiment, an image file-annotation file data pair is obtained to obtain I 1 -L 1 And I 2 -L 2 Two data pairs. Annotation file L 1 The file content comprises two kinds of information to be checked, namely a labeling area and a labeling label, and can be expressed as [ O ] 1 (R 1,1 ,T 1 )O 2 (R 1,2 ,T 2 )]The method comprises the steps of carrying out a first treatment on the surface of the Annotation file L 2 The file content can be expressed as O 1 (R 2,1 ,T 1 )O 2 (R 2,2 ,T 2 )]。
Next, the audit data generation module is used for decomposing the image file-annotation file data pairs into image area-object information data pairs one by one according to the annotated objects to determine image area and object information, wherein the image area-object information data pairs can be illustrated by referring to fig. 3, and when the audit data generation module is actually applied, one image file-annotation file data pair contains a plurality of objects and can be decomposed into a plurality of image area-object information data pairs; the auditing data generation module is also used for pushing the image area-target information data pairs to the quality auditing module one by one according to different auditing schemes.
The verification data generation module is used for customizing the marking data of the image into marking data capable of verifying marking quality according to objects one by one, in the process of decomposing the image file-marking file data pair, the marked objects are decomposed into image areas and object information corresponding to the image areas one by one according to the objects one by one, namely, an image area-object information data pair is formed, one image area corresponds to one item of marking information, and the image area only comprises one object.
In this embodiment, the processing of the above content by the audit data generation module specifically includes:
determining an image area: and intercepting a subgraph which comprises the labeling area and is 4 times of the labeling area from the image file to serve as an image area. Marking region R 1，1 Is recorded as the image area of (1)
Figure BDA0003931481760000071
Marking region R 1，2 Is marked as +.>
Figure BDA0003931481760000072
Marking region R 2，1 Is marked as +.>
Figure BDA0003931481760000073
Marking region R 2，2 Is marked as +.>
Figure BDA0003931481760000074
Determining target information: extracting target information from the annotation file; marking region R 1，1 Target information of (2) is O 1 (R 1，1 ，T 1 ) Marking region R 1，2 Target information of (2) is O 2 (R 1，2 ，T 2 ) Marking region R 2，1 Target information of (2) is O 1 (R 2，1 ，T 1 ) Marking region R 2，2 Target information of (2) is O 2 (R 2，2 ，T 2 )。
Decomposing the image file-annotation file data pair of the annotation data into an image area-target information data pair: pair I 1 -L 1 Decomposing the data pairs to obtain two data pairs
Figure BDA0003931481760000075
Pair I 2 -L 2 Decomposing the data pairs to obtain two data pairs +.>
Figure BDA0003931481760000076
In the above decomposition process, the image area is: in the image file-annotation file data pair, according to the annotation region of the target in the annotation file, intercepting a subgraph containing the target from the image file; the image area is larger than the labeling area of the corresponding target in the labeling file; the target information includes: in the image file-annotation file data pair, the region information and the label information of the corresponding target are extracted from the annotation file.
After the decomposition is completed and a plurality of image area-target information data pairs are obtained, the audit data generation module pushes the plurality of image area-target information data pairs to a strategy of the quality audit module one by one in order to audit the labeling quality according to the target by one, and correspondingly selects an audit scheme; the auditing scheme comprises an image priority scheme and a target priority scheme, wherein the image priority scheme firstly examines the labeling quality of one image and then examines the labeling quality of the next image; the target priority scheme firstly examines the labeling quality of the targets of the same class, and then examines the labeling quality of the targets of the next class.
When the image priority scheme is selected to be adopted, specific pushing strategies of the 4 image area-target information data pairs in the embodiment are as follows: first, the image file I is used for 1 Is a target information data pair
Figure BDA0003931481760000077
Pushing the image files one by one to a quality checking module after checking 2 Is +.>
Figure BDA0003931481760000078
Figure BDA0003931481760000079
And checking by a quality checking module.
A target priority scheme may also be selected, where the specific pushing policy of the 4 image area-target information data pairs in this embodiment is: first, target O 1 Is a target information data pair
Figure BDA0003931481760000081
Pushing the target O one by one to a quality auditing module after auditing 2 Is +.>
Figure BDA0003931481760000082
Figure BDA0003931481760000083
And checking by a quality checking module.
Every time an image area-target information data pair is pushed to the quality checking module, the quality checking module starts to perform corresponding checking processing, and the working modes of the quality checking module are the same for each image area-target information data pair, in this embodimentDescription only of
Figure BDA0003931481760000084
Is a verification process of the system.
The interface schematic of the quality audit module can be seen in fig. 4, wherein the functions of the annotation data display module specifically include: and determining a display mode according to the information such as the image area, user setting and the like, and displaying the image area-target information data pair in the display area.
The labeling area auditing module is used for providing an area labeling quality option and an interaction mode for judging the labeling quality of the area information for a user, wherein the area labeling quality option comprises over labeling, under labeling and wrong labeling; the labeling label auditing module is used for providing label labeling quality options and interaction modes for judging the labeling quality of the label information for the user, wherein the label labeling quality options comprise wrong labels and missed labels.
In the present embodiment, the image region-target information data pair is displayed
Figure BDA0003931481760000085
The display mode set by the user is self-adaption and coloring. The user can conveniently read the labeling information of the target without dragging and zooming, which is beneficial to improving the auditing efficiency of the user and guaranteeing the quality of auditing work. Specifically, the label T is first attached 1 The color of the brush is set as C 1 The marker is used for marking the region R 1，1 Drawing to image area +.>
Figure BDA0003931481760000086
Finally, the image area is->
Figure BDA0003931481760000087
And (5) scaling to the maximum area which can be displayed by the annotation data display module.
Subsequently, in the labeling area auditing module, the user reviews the image area by auditing the image area
Figure BDA0003931481760000088
Marked area R 1，1 Positional relationship with the targetInformation is obtained, whether the quality of the marked area meets the standard or not is judged, and whether the marked area is marked by mistake, under mark or wrong mark is judged; and if the quality of the marked area is not up to standard, selecting the type with the quality not up to standard through a mouse.
In the process of judging the quality of the marked area, the over-marked meaning is as follows: a sub-region that is much smaller than the labeling region can still contain the entire target; the under-labeled meaning is: the labeling area only contains part of the targets; the mislabeling means: the labeling area has no overlapping portion with the target.
Similarly, in the labeling tag auditing module, the user reviews the image region by auditing
Figure BDA0003931481760000089
Marked area R 1,1 Judging whether the quality of the labeling label is qualified or whether the labeling label is wrong or missing; if the quality of the labeling label is unqualified, the quality unqualified type can be selected through a mouse.
In the label quality judging process, the meaning of the wrong label is as follows: labels inconsistent with the real labels of the targets exist in the labeling labels of the targets; the meaning of the missed label is: the labeling label of the object lacks a label that should be assigned to the object.
The operation process of the image annotation quality auditing system provided by the embodiment is integrated, and the image annotation quality auditing method when the system is used is summarized at the same time in the embodiment, and comprises the following steps:
s1, an annotation data acquisition module acquires an image file and an annotation file in annotation data, and the image file and the annotation file are in one-to-one correspondence to form an image file-annotation file data pair;
s2, the audit data generation module decomposes the image file-annotation file data pairs into corresponding image area-target information data pairs one by one according to the annotated targets, each target corresponds to one image area-target information data pair, and the image area and the target information of each target are determined according to the image area-target information data pairs;
s3, the auditing data generation module selects an auditing scheme, and pushes the image area-target information data pair to the quality auditing module for auditing according to a data pair pushing strategy in the auditing scheme;
s4, after the quality checking module receives the pushed image area-target information data pair, the image area-target information data pair is displayed in a display area;
s5, a user uses a labeling area auditing module to label quality of the area information in the displayed image area-target information data pair, and selects a corresponding area labeling quality option to complete area labeling quality auditing;
s6, a user uses a labeling label auditing module to label quality of label information in the displayed image region-target information data pair, and selects a corresponding label labeling quality option to complete label labeling quality auditing.
Also, in the auditing scheme selection process of step S3, the auditing scheme includes an image priority scheme and a target priority scheme.
The image priority scheme is as follows: firstly checking the labeling quality of one image, and then checking the labeling quality of the next image; the data pair pushing strategy corresponding to the image priority scheme is as follows: firstly pushing all image area-target information data pairs of one image to a quality checking module one by one and checking, and then pushing all image area-target information data pairs of the next image to the quality checking module one by one;
the target priority scheme is as follows: firstly, checking the labeling quality of the same type of targets, and then checking the labeling quality of the next type of targets; the data pair push strategy corresponding to the target priority scheme is: firstly, pushing all image area-target information data pairs of the same type of targets of a plurality of images to a quality auditing module one by one, and after auditing is completed, pushing all image area-target information data pairs of the next type of targets one by one to the quality auditing module for auditing.

Claims (10)

1. An image annotation quality auditing system is characterized in that: the system comprises a data generation module and a quality auditing module; the data generation module is used for customizing auditing data, wherein the auditing data is an object for image annotation quality auditing in the quality auditing module; the quality auditing module is used for displaying the labeling data to be audited for the user and providing the operation of image labeling quality auditing for the labeling data to be audited for the user.
2. The image annotation quality auditing system of claim 1, wherein: the data generation module comprises a labeling data acquisition module and an auditing data generation module; the annotation data acquisition module is used for acquiring the annotation data of the image; the auditing data generation module is used for customizing the labeling data of the image into the auditing data according to an auditing scheme.
3. The image annotation quality auditing system of claim 2, wherein: the annotation data acquisition module is used for acquiring annotation data, wherein the annotation data comprises an image file and an annotation file corresponding to the image file; the annotation data acquisition module is used for respectively corresponding the image file and the annotation file to form a plurality of image file-annotation file data pairs; the labeling file comprises labeling area information and labeling label information.
4. The image annotation quality review system of claim 3, wherein: the auditing data generation module is used for decomposing the image file-annotation file data pairs into image area-target information data pairs one by one according to the annotated targets, and determining the image area and target information of each target; the auditing data generation module is also used for pushing the image area-target information data pairs to the quality auditing module one by one according to different auditing schemes.
5. The image annotation quality review system of claim 4, wherein the image area is: in the image file-annotation file data pair, a subgraph containing a target in an annotation file is intercepted from the image file according to the annotation region of the target; the image area is larger than the labeling area of the corresponding target in the labeling file; the target information includes: and in the image file-annotation file data pair, extracting the region information and the label information of the corresponding target from the annotation file.
6. The image annotation quality auditing system of claim 4, wherein: the auditing scheme includes an image priority scheme and a target priority scheme.
7. The image annotation quality auditing system of claim 4, wherein: the quality auditing module comprises a labeling data display module, a labeling area auditing module and a labeling label auditing module; the labeling data display module is used for displaying labeling data for user auditing, the labeling area auditing module is used for providing operation for labeling quality of area information in the user auditing target information, and the labeling label auditing module is used for providing operation for labeling quality of label information in the user auditing target information.
8. The image annotation quality auditing system of claim 7, wherein: the labeling area auditing module is used for judging the labeling quality of the area information for the user and providing an area labeling quality option and an interaction mode; the labeling label auditing module is used for judging the labeling quality of label information for a user and providing label labeling quality options and interaction modes.
9. An image marking quality auditing method, characterized in that the image marking quality auditing system according to claims 1 to 8 is used, and the auditing method comprises the following steps:
s1, an annotation data acquisition module acquires an image file and an annotation file in annotation data, and the image file and the annotation file are in one-to-one correspondence to form an image file-annotation file data pair;
s2, the audit data generation module decomposes the image file-annotation file data pairs into corresponding image area-target information data pairs one by one according to the annotated targets, each target corresponds to one image area-target information data pair, and the image area and the target information of each target are determined according to the image area-target information data pairs;
s3, an auditing data generation module selects an auditing scheme, and pushes the image area-target information data pair to a quality auditing module for auditing according to a data pair pushing strategy in the auditing scheme;
s4, after the quality checking module receives the pushed image area-target information data pair, the image area-target information data pair is displayed in a display area;
s5, a user uses a labeling area auditing module to label quality of the area information in the displayed image area-target information data pair, and selects a corresponding area labeling quality option to complete area labeling quality auditing;
s6, a user uses a labeling label auditing module to label quality of label information in the displayed image region-target information data pair, and selects a corresponding label labeling quality option to complete label labeling quality auditing.
10. The image annotation quality auditing method according to claim 9, characterized in that: the auditing scheme comprises an image priority scheme and a target priority scheme;
the image priority scheme is as follows: firstly checking the labeling quality of one image, and then checking the labeling quality of the next image; the data pair pushing strategy corresponding to the image priority scheme is as follows: firstly pushing all image area-target information data pairs of one image to a quality checking module one by one and checking, and then pushing all image area-target information data pairs of the next image to the quality checking module one by one;
the target priority scheme is as follows: firstly, checking the labeling quality of the same type of targets, and then checking the labeling quality of the next type of targets; the data pair push strategy corresponding to the target priority scheme is: firstly, pushing all image area-target information data pairs of the same type of targets of a plurality of images to a quality auditing module one by one, and after auditing is completed, pushing all image area-target information data pairs of the next type of targets one by one to the quality auditing module for auditing.
CN202211389669.5A 2022-11-08 2022-11-08 Image annotation quality auditing system and method Pending CN116051618A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211389669.5A CN116051618A (en) 2022-11-08 2022-11-08 Image annotation quality auditing system and method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211389669.5A CN116051618A (en) 2022-11-08 2022-11-08 Image annotation quality auditing system and method

Publications (1)

Publication Number Publication Date
CN116051618A true CN116051618A (en) 2023-05-02

Family

ID=86122502

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211389669.5A Pending CN116051618A (en) 2022-11-08 2022-11-08 Image annotation quality auditing system and method

Country Status (1)

Country Link
CN (1) CN116051618A (en)

Similar Documents

Publication Publication Date Title
US11868717B2 (en) Multi-page document recognition in document capture
CN111079387B (en) Automatic questionnaire generation method and device, computer equipment and storage medium
CN100465945C (en) Method and apparatus for populating electronic forms from scanned documents
CN112100742B (en) Method and device for processing map based on three-dimensional model of hull structure and storage medium
CN112102443B (en) Labeling system and labeling method suitable for substation equipment inspection image
CN105469513A (en) Self-service all-in-one machine based on face detection and character recognition and using method thereof
US6970267B1 (en) Gray scale optical mark reader
CN113220920A (en) Satellite remote sensing image sample labeling system and method based on micro-service architecture
CN114998905A (en) Method, device and equipment for verifying complex structured document content
CN113345052B (en) Classification data multi-view visualization coloring method and system based on similarity significance
CN116051618A (en) Image annotation quality auditing system and method
JP2009086952A (en) Information processing system and information processing program
CN100485693C (en) Method and device for positioning layout error object, device for displaying inspection result
JP2008003739A (en) Input correction method, postscript information processing method, postscript information processor, and program
CN116757886A (en) Data analysis method and analysis device
CN111696012A (en) Medical image remote teaching method, device, equipment and storage medium
JP5766438B2 (en) Method and system for click-through function in electronic media
CN109919153A (en) Automatic document entry system and method based on handwriting recognition artificial intelligence technology
CN113468355B (en) Display mother board data management method and device
CN114663975A (en) Model training method and device, electronic equipment and storage medium
CN113989268A (en) Method for identifying GIS sleeve of designated equipment in infrared image and storage medium
CN104112287B (en) Method and device for segmenting characters in picture
CN117634426B (en) Defect mark generation method of circuit design, electronic equipment and storage medium
CN111368929A (en) Picture labeling method
CN115202542B (en) Automatic link and skip method for circuit ports in electronic drawing based on OCR technology

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