CN111414907A - Data set labeling method, data set labeling device and computer-readable storage medium - Google Patents
Data set labeling method, data set labeling device and computer-readable storage medium Download PDFInfo
- Publication number
- CN111414907A CN111414907A CN202010168698.3A CN202010168698A CN111414907A CN 111414907 A CN111414907 A CN 111414907A CN 202010168698 A CN202010168698 A CN 202010168698A CN 111414907 A CN111414907 A CN 111414907A
- Authority
- CN
- China
- Prior art keywords
- data
- coordinate
- task
- marked
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000002372 labelling Methods 0.000 title claims abstract description 112
- 238000000034 method Methods 0.000 claims abstract description 81
- 230000011218 segmentation Effects 0.000 claims abstract description 55
- 238000012216 screening Methods 0.000 claims abstract description 50
- 238000001514 detection method Methods 0.000 claims abstract description 29
- 238000013135 deep learning Methods 0.000 claims abstract description 9
- 230000004044 response Effects 0.000 claims abstract description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 238000004422 calculation algorithm Methods 0.000 description 50
- 238000012545 processing Methods 0.000 description 21
- 230000008569 process Effects 0.000 description 17
- 230000003993 interaction Effects 0.000 description 8
- 238000012549 training Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 4
- 230000018109 developmental process Effects 0.000 description 4
- 230000010339 dilation Effects 0.000 description 3
- 238000003708 edge detection Methods 0.000 description 3
- 230000000877 morphologic effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
- G06T5/30—Erosion or dilatation, e.g. thinning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a data set labeling method, a data set labeling device and a computer readable storage medium. The data set labeling method comprises the following steps: receiving data to be marked, analyzing the data to be marked to determine the task type of the data to be marked, and storing the data to be marked to a local storage area according to the task type; marking the data to be marked in response to the marking instruction so as to determine a contour coordinate set of the data to be marked; determining coordinate screening conditions corresponding to the task categories, screening coordinate points in the contour coordinate set according to the coordinate screening conditions, and determining a target coordinate set; wherein the task types include: and target segmentation and target detection based on the deep learning task. The method is used for carrying out auxiliary labeling on the data set of the deep learning data labeling task, different screening conditions are set for different data to be labeled, the labeling speed and quality of the data set are improved, the working difficulty of labeling personnel is reduced, and the labeling efficiency is improved.
Description
Technical Field
The invention belongs to the technical field of data annotation, and particularly relates to a data set annotation method, a data set annotation device and a computer-readable storage medium.
Background
In many image data set annotation tools and image data annotation platforms based on deep learning tasks, the traditional image data set annotation form is generally manual annotation, and the annotation tools and modes of different image data sets of tasks are generally different according to deep learning application. The target detection comprises a target type and a target positioning at the same time, and the labeling form is the position of the target in the marked image and the type label of the target at the position; the target segmentation is similar to the target segmentation in that the target position and the target category need to be labeled, and the difference is that the target is segmented into an object outer contour label at a pixel level, and the target detection is a rectangular contour label.
In a data crowdsourcing annotation platform, visual related tasks often need to label image data, and how to enable data annotation personnel to efficiently and accurately complete an image annotation task becomes a problem which needs to be solved urgently.
Disclosure of Invention
The present invention is directed to solving one of the technical problems of the prior art or the related art.
To this end, the first aspect of the invention provides a data set annotation method.
A second aspect of the present invention provides a data set annotation apparatus.
A third aspect of the invention provides a computer-readable storage medium.
In view of the above, a data set annotation method according to a first aspect of the present invention is provided, including: receiving data to be marked, analyzing the data to be marked to determine the task type of the data to be marked, and storing the data to be marked to a local storage area according to the task type; marking the data to be marked in response to the marking instruction so as to determine a contour coordinate set of the data to be marked; determining coordinate screening conditions corresponding to the task categories, screening coordinate points in the contour coordinate set according to the coordinate screening conditions, and determining a target coordinate set; wherein the task types include: and target segmentation and target detection based on the deep learning task.
The invention provides a data set labeling method, which comprises the steps of receiving data to be labeled, configuring the server, namely, the data to be marked is uploaded to a server, the operation to be executed on the data to be marked is determined to determine the task category when the data to be marked is uploaded, the task category is stored in the data to be marked in a label form, the server analyzes the data to be marked and determines the task category of the data to be marked, the data to be marked is stored in a local storage area according to the difference of the task categories, wherein the task category comprises a target segmentation task and a target detection task, the local storage area of the server also comprises a target segmentation task storage area and a target detection task storage area, and the data to be labeled of different task categories are respectively stored in the two storage areas according to the task category of the data to be labeled, so that the subsequent data to be labeled can be conveniently searched and used. The server receives a labeling instruction, labels data to be labeled to obtain a contour coordinate set of the data to be labeled, wherein the labeling instruction comprises a task starting instruction, a task selecting instruction, a task executing instruction and the like, the server responds to the labeling instruction to select the data to be labeled and labels the data, the specific labeling mode is that the data to be labeled are labeled through coordinate points, and the set of the coordinate points used for labeling the picture is the contour coordinate set. The obtained contour coordinate set needs to be screened to determine a target coordinate set, different screening conditions are selected for the data to be labeled of different task categories to be screened, and coordinate points in the contour coordinate set are screened according to the screening conditions corresponding to the task categories to determine a final target coordinate set. The target contour is extracted through the traditional image processing segmentation algorithm and manual interaction, and is applied to the auxiliary data annotation in the data annotation task of the data crowdsourcing platform, by classifying the data to be labeled and selecting different screening conditions for the data to be labeled of different types, the labeling task of the data to be labeled of different types can be completed even if common data labeling personnel with insufficient algorithm speciality in a crowdsourcing platform, because different screening conditions are set for different data to be labeled, two types of data labeling tasks can be completed based on the method, compared with the traditional image processing segmentation algorithm application corresponding method, the labeling speed and quality of the data set are improved, compared with the data pre-labeled by using a deep learning model in the related technology, the method does not need professional algorithm personnel to support the training model, and saves the labeling time. And moreover, the convenient interaction between the labeling personnel and the labeling tool is supported, so that the working difficulty of the labeling personnel is reduced, the labeling efficiency is improved, and the development time of an algorithm model is saved.
It is understood that the data to be labeled may be picture data or video data. When the data to be marked is video data, the video data is decomposed into video frame images.
The data set labeling method is applied to a data crowdsourcing platform, the executed labeling tasks are target segmentation tasks or target detection tasks based on a deep learning task, different auxiliary labeling tools are selected according to the task types of the received data to be labeled, and the auxiliary labeling tools label the data to be labeled according to a labeling instruction sent manually, so that the auxiliary labeling of the tasks based on the crowdsourcing platform is realized.
In addition, according to the data set labeling method in the above technical solution provided by the present invention, the following additional technical features may also be provided:
in the above technical solution, the step of labeling the data to be labeled in response to the labeling instruction specifically includes: selecting a corresponding auxiliary marking tool according to the task category of the data to be marked; and controlling the auxiliary marking tool to mark the data to be marked according to the marking instruction.
In the technical scheme, after a marking instruction of data to be marked is received, a corresponding auxiliary marking tool is selected according to the task category of the data to be marked, and the auxiliary marking tool is controlled to mark the data to be marked according to the marking instruction, wherein the task category to be marked also comprises the data to be marked of a simple background and the data to be marked of a conventional background. For the data to be marked of the simple background, a segmentation method based on a gray threshold or a segmentation method based on edge detection is used. For the conventional background, the picture to be marked contains a complex background and a plurality of targets, more complex segmentation methods are adopted, specifically, two types of segmentation methods are adopted, and one type of segmentation method is based on an image region and comprises a watershed algorithm, a seed region growing algorithm and the like. Another category of graph theory-based methods are most used by the GrabCut algorithm (picture segmentation algorithm). Different auxiliary marking tools are selected for different data to be marked, so that the marking accuracy of the auxiliary marking tools is guaranteed, and the working difficulty of marking personnel is further reduced.
It can be understood that the two types of algorithms assist in obtaining the contour coordinate set of the target to be marked in the picture. Meanwhile, in the specific execution process, some detailed processing is performed, for example, in order to ensure that the labeling frame can completely contain the object edge in the subsequent processing, morphological dilation processing can be performed on the image contour before the target contour coordinate set is extracted, and the image edge is expanded outwards.
In any of the above technical solutions, the step of screening the coordinate points in the contour coordinate set according to the coordinate screening condition, where the task type is target detection, specifically includes: determining a rectangular surrounding frame of an area where the outline coordinate set is located, and collecting all coordinate points in the rectangular surrounding frame; and screening all coordinate points in the rectangular surrounding frame, and extracting the coordinate point with the minimum coordinate value and the coordinate point with the maximum coordinate value as a target coordinate set. When the task type is a target detection task, useful coordinate points in the outline coordinate set are screened.
In the technical scheme, when the task type is target detection, according to each coordinate point in the outline coordinate set of the target data, a corresponding rectangular surrounding frame is generated for the coordinate points in the outline coordinate set, the coordinate points of the rectangular surrounding frame are collected, namely all the coordinate points forming the rectangular surrounding frame, two points with the minimum coordinate value and the maximum coordinate value in all the coordinate points are determined, namely the upper left-corner coordinate point and the lower right-corner coordinate point of the rectangular surrounding frame, and the set of the two coordinate points is used as a target coordinate set.
It can be understood that when the task is target detection, only whether a target area in the data to be labeled is consistent with a target image needs to be judged, so that two most representative coordinate points are selected as a target coordinate set, and the memory occupation amount of the target coordinate set is reduced.
In any of the above technical solutions, the step of performing target segmentation on the task type and screening the coordinate points in the contour coordinate set according to the coordinate screening condition specifically includes: establishing grid line coordinates for data to be marked; and screening target coordinate points which are overlapped with the grid points of the grid lines in the contour coordinate set, and determining the target coordinate set according to the target coordinate points.
In the technical scheme, when the task type is target segmentation, grid line coordinates are established for data to be marked with contour coordinate points, coordinate points coincident with the grid line coordinates are screened out, and a set of the coordinate points is used as a target coordinate set. When the task type is the target segmentation task, the screening of useful coordinate points in the outline coordinate set is realized.
In any of the above technical solutions, the data set labeling method further includes: selecting at least one target coordinate point as a set coordinate point; and determining the coordinate points within the set range of the set coordinate points as redundant coordinate points, and removing the redundant coordinate points in the target coordinate set.
In the technical scheme, when the coordinate points are excessive, at least one target coordinate point is selected as a set coordinate point, all the coordinate points within a set range from the set coordinate point are determined as redundant coordinate points, and the redundant coordinate points are removed from the target coordinate set, so that the effect of the target coordinate set is guaranteed, the data volume of the target coordinate set is reduced, and the occupied amount of a system space is reduced.
It should be noted that, if two or more coordinate points are selected as the set coordinate points, the distance between two adjacent set coordinate points is larger than the diameter of the set range, so as to avoid determining the set coordinate points as the redundant coordinate points when screening the redundant coordinate points.
In any of the above technical solutions, the step of storing the data to be labeled in the local storage area according to the task type specifically includes: determining task information of data to be marked; storing the data to be marked and corresponding task information in a local storage area according to task types; the task information includes, but is not limited to, task tags and data sources to be annotated.
In the technical scheme, the task information of the data to be marked is determined, the data to be marked and the task information corresponding to the data to be marked are stored in the local storage area together, the data to be marked and the task information need to be stored according to task categories when being stored, the task information comprises but is not limited to task tags and sources of representative data, the data to be marked are stored in a classified mode, the data to be marked can be traced, and the relevant task information of the data to be marked can be directly searched from the local storage area.
In any of the above technical solutions, the data set labeling method further includes: and generating a label file according to the target coordinate set and the corresponding task information, and storing the label file in a local storage area in a set format.
According to the technical scheme, after a target coordinate set is determined, a label file is generated according to the target coordinate set and corresponding task information, and meanwhile, the label file is stored in a local storage area in a set format to wait for calling of other tasks. The markup file format is generally an xml file or a json file.
In any of the above technical solutions, before the step of storing the markup file in the local storage area in the set format, the method further includes: and determining that the annotation file has an annotation error, and correcting the annotation file.
In the technical scheme, the annotation file is detected, whether the annotation file has the problem of wrong annotation is determined, if the annotation file has the problem of wrong annotation, the annotation file is corrected, and if the annotation file does not have the problem, the annotation file is directly stored in a local storage area of the server.
It can be appreciated that manual checking can also be used to ensure the accuracy of the training data.
According to a second aspect of the present invention, there is provided a data set annotation apparatus, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor; when executed by a processor, the computer program implements the data set annotation method according to any of the above-mentioned technical solutions, so that the computer program has all the beneficial effects of the data set annotation method according to any of the above-mentioned technical solutions, and will not be described in detail herein.
According to a third aspect of the present invention, a computer-readable storage medium is provided, where a data set annotation program is stored on the computer-readable storage medium, and when the data set annotation program is executed by a processor, the steps of the data set annotation method in any of the above technical solutions are implemented, so that all beneficial effects of the data set annotation method in any of the above technical solutions are achieved, and redundant description is omitted here.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
The above and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a flow diagram illustrating a data set annotation process according to an embodiment of the invention;
FIG. 2 is a flow chart illustrating a data set annotation process according to another embodiment of the invention;
FIG. 3 is a flow chart diagram illustrating a data set annotation process according to yet another embodiment of the invention;
FIG. 4 is a flow chart diagram illustrating a data set annotation process according to yet another embodiment of the invention;
FIG. 5 is a flow diagram illustrating a data set annotation process in accordance with a complete embodiment of the present invention;
FIG. 6 is a diagram illustrating a target coordinate set extraction gridding algorithm demonstration in a data set annotation method according to an embodiment of the present invention
FIG. 7 shows a schematic block diagram of a data set annotation device of one embodiment of the present invention.
Detailed Description
In order that the above objects, features and advantages of the present invention can be more clearly understood, a more particular description of the invention will be rendered by reference to the appended drawings. It should be noted that the embodiments and features of the embodiments of the present application may be combined with each other without conflict.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, however, the present invention may be practiced in other ways than those specifically described herein, and therefore the scope of the present invention is not limited by the specific embodiments disclosed below.
A data set annotation method, a data set annotation apparatus, and a computer-readable storage medium according to some embodiments of the invention are described below with reference to fig. 1 to 7.
The first embodiment is as follows:
as shown in fig. 1, in an embodiment of the present invention, a method for annotating a data set includes:
step S102, receiving data to be marked;
step S104, analyzing the data to be labeled to determine the task type of the data to be labeled, and storing the data to be labeled to a local storage area according to the task type;
step S106, marking the data to be marked in response to the marking instruction so as to determine the contour coordinate set of the data to be marked;
and S108, determining coordinate screening conditions corresponding to the task categories, screening coordinate points in the contour coordinate set according to the coordinate screening conditions, and determining a target coordinate set.
Wherein the task types include: and target segmentation and target detection based on the deep learning task.
In the embodiment, the data set annotation method comprises the steps of receiving data to be annotated, configuring the server, namely, the data to be marked is uploaded to a server, the operation to be executed on the data to be marked is determined to determine the task category when the data to be marked is uploaded, the task category is stored in the data to be marked in a label form, the server analyzes the data to be marked and determines the task category of the data to be marked, the data to be marked is stored in a local storage area according to the difference of the task categories, wherein the task category comprises a target segmentation task and a target detection task, the local storage area of the server also comprises a target segmentation task storage area and a target detection task storage area, and the data to be labeled of different task categories are respectively stored in the two storage areas according to the task category of the data to be labeled, so that the subsequent data to be labeled can be conveniently searched and used. The server receives a labeling instruction, labels data to be labeled to obtain a contour coordinate set of the data to be labeled, wherein the labeling instruction comprises a task starting instruction, a task selecting instruction, a task executing instruction and the like, the server responds to the labeling instruction to select the data to be labeled and labels the data, the specific labeling mode is that the data to be labeled are labeled through coordinate points, and the set of the coordinate points used for labeling the picture is the contour coordinate set. The obtained contour coordinate set needs to be screened to determine a target coordinate set, different screening conditions are selected for the data to be labeled of different task categories to be screened, and coordinate points in the contour coordinate set are screened according to the screening conditions corresponding to the task categories to determine a final target coordinate set. The target contour is extracted through the traditional image processing segmentation algorithm and manual interaction, and is applied to the auxiliary data annotation in the data annotation task of the data crowdsourcing platform, by classifying the data to be labeled and selecting different screening conditions for the data to be labeled of different types, the labeling task of the data to be labeled of different types can be completed even if common data labeling personnel with insufficient algorithm speciality in a crowdsourcing platform, because different screening conditions are set for different data to be labeled, two types of data labeling tasks can be completed based on the method, compared with the traditional image processing segmentation algorithm application corresponding method, the labeling speed and quality of the data set are improved, compared with the data pre-labeled by using a deep learning model in the related technology, the method does not need professional algorithm personnel to support the training model, and saves the labeling time. And moreover, the convenient interaction between the labeling personnel and the labeling tool is supported, so that the working difficulty of the labeling personnel is reduced, the labeling efficiency is improved, and the development time of an algorithm model is saved.
It is understood that the data to be labeled may be picture data or video data. When the data to be marked is video data, the video data is decomposed into video frame images.
In the embodiment, the data set labeling method is applied to a data crowdsourcing labeling platform, the executed labeling tasks are all target segmentation tasks or target detection tasks based on a deep learning task, different auxiliary labeling tools are selected according to the task categories of the received data to be labeled, and the auxiliary labeling tools label the data to be labeled according to a labeling instruction sent manually, so that the auxiliary labeling of the tasks based on the crowdsourcing platform is realized.
Example two:
in another embodiment of the present invention, as shown in fig. 2, a method for annotating a data set includes:
step S202, receiving data to be marked;
step S204, analyzing the data to be labeled to determine the task category of the data to be labeled;
step S206, determining task information of the data to be annotated, and storing the data to be annotated and the corresponding task information in a local storage area according to task types;
s208, selecting a corresponding auxiliary marking tool according to the task category of the data to be marked;
step S210, controlling an auxiliary marking tool to mark data to be marked according to a marking instruction;
step S212, determining coordinate screening conditions corresponding to task categories, screening coordinate points in the contour coordinate set according to the coordinate screening conditions, and determining a target coordinate set;
step S214, generating a label file according to the target coordinate set and the corresponding task information, and storing the label file in a local storage area in a set format.
The task information includes, but is not limited to, a task tag and a source of data to be labeled.
In the embodiment, when the data to be annotated is uploaded, it is required to determine that the data to be annotated performs operation to determine a task category, the task category is stored in the data to be annotated in a tag form, task information of the data to be annotated is determined, the data to be annotated and the task information corresponding to the data to be annotated are stored in a local storage area together, when the data to be annotated and the task information are stored, the data to be annotated and the task information are stored according to the task category, the task information includes but is not limited to the source of the task tag and the representative data, the data to be annotated is stored in a classified manner, the source of the data to be annotated can be traced, and the relevant task information of the data to be annotated can be directly searched from.
After receiving a marking instruction of data to be marked, selecting a corresponding auxiliary marking tool according to the task category of the data to be marked, and controlling the auxiliary marking tool to mark the data to be marked according to the marking instruction, wherein the task category to be marked also comprises the data to be marked of a simple background and the data to be marked of a conventional background. For the data to be marked of the simple background, a segmentation method based on a gray threshold or a segmentation method based on edge detection is used. For the conventional background, the picture to be marked contains a complex background and a plurality of targets, more complex segmentation methods are adopted, specifically, two types of segmentation methods are adopted, and one type of segmentation method is based on an image region and comprises a watershed algorithm, a seed region growing algorithm and the like. The other type is a graph theory-based method which uses GrabCut algorithm most, and the two types of algorithms assist in obtaining a contour coordinate set of the target to be marked in the picture. Meanwhile, in the specific execution process, some detailed processing is performed, for example, in order to ensure that the labeling frame can completely contain the object edge in the subsequent processing, morphological dilation processing can be performed on the image contour before the target contour coordinate set is extracted, and the image edge is expanded outwards. Different auxiliary marking tools are selected for different data to be marked, so that the marking accuracy of the auxiliary marking tools is guaranteed, and the working difficulty of marking personnel is further reduced.
After the target coordinate set is determined, generating a label file according to the target coordinate set and corresponding task information, and meanwhile, storing the label file in a local storage area in a set format to wait for calling of other tasks. The markup file format is generally an xml file or a json file.
In the above embodiment, before the step of storing the markup file in the local storage area in the set format, the method further includes: and determining that the annotation file has an annotation error, and correcting the annotation file.
In the embodiment, the annotation file is detected, whether the annotation file has the problem of annotation errors is determined, if the annotation file has the problem of annotation errors, the annotation file is corrected, and if the annotation file does not have the problem, the annotation file is directly stored in a local storage area of the server. And manual checking can be used to ensure the accuracy of the training data.
Example three:
in another embodiment of the present invention, as shown in fig. 3, a method for annotating a data set includes:
step S302, receiving data to be marked;
step S304, analyzing the data to be labeled to determine the task type of the data to be labeled, and storing the data to be labeled into a local storage area according to the task type;
step S306, marking the data to be marked in response to the marking instruction so as to determine a contour coordinate set of the data to be marked;
step S308, determining the task type as target detection, determining a rectangular surrounding frame of the area where the outline coordinate set is located, and collecting all coordinate points in the rectangular surrounding frame;
step S310, all coordinate points in the rectangular surrounding frame are screened, and the coordinate point with the minimum coordinate value and the coordinate point with the maximum coordinate value are extracted as a target coordinate set.
In the embodiment, the data set annotation method comprises the steps of receiving data to be annotated, configuring the server, namely, the data to be marked is uploaded to a server, the operation to be executed on the data to be marked is determined to determine the task category when the data to be marked is uploaded, the task category is stored in the data to be marked in a label form, the server analyzes the data to be marked and determines the task category of the data to be marked, the data to be marked is stored in a local storage area according to the difference of the task categories, wherein the task category comprises a target segmentation task and a target detection task, the local storage area of the server also comprises a target segmentation task storage area and a target detection task storage area, and the data to be labeled of different task categories are respectively stored in the two storage areas according to the task category of the data to be labeled, so that the subsequent data to be labeled can be conveniently searched and used. The server receives a labeling instruction, labels data to be labeled to obtain a contour coordinate set of the data to be labeled, wherein the labeling instruction comprises a task starting instruction, a task selecting instruction, a task executing instruction and the like, the server responds to the labeling instruction to select the data to be labeled and labels the data, the specific labeling mode is that the data to be labeled are labeled through coordinate points, and the set of the coordinate points used for labeling the picture is the contour coordinate set. The obtained contour coordinate set needs to be screened to determine a target coordinate set, different screening conditions are selected for the data to be labeled of different task categories to be screened, and coordinate points in the contour coordinate set are screened according to the screening conditions corresponding to the task categories to determine a final target coordinate set.
When the task type is target detection, according to each coordinate point in the outline coordinate set of the target data, generating a corresponding rectangular surrounding frame for the coordinate points in the outline coordinate set, collecting the coordinate points of the rectangular surrounding frame, namely forming all the coordinate points of the rectangular surrounding frame, determining two points with the minimum coordinate value and the maximum coordinate value in all the coordinate points, namely an upper left-corner coordinate point and a lower right-corner coordinate point of the rectangular surrounding frame, and taking the set of the two coordinate points as a target coordinate set. When the task is target detection, only whether a target area in the data to be labeled is consistent with a target image needs to be judged, so that two most representative coordinate points are selected as a target coordinate set, and the memory occupation amount of the target coordinate set is reduced.
The target contour is extracted through the traditional image processing segmentation algorithm and manual interaction, and is applied to the auxiliary data annotation in the data annotation task of the data crowdsourcing platform, by classifying the data to be labeled and selecting different screening conditions for the data to be labeled of different types, the labeling task of the data to be labeled of different types can be completed even if common data labeling personnel with insufficient algorithm speciality in a crowdsourcing platform, because different screening conditions are set for different data to be labeled, two types of data labeling tasks can be completed based on the method, compared with the traditional image processing segmentation algorithm application corresponding method, the labeling speed and quality of the data set are improved, compared with the method for pre-labeling data by using a deep learning model in the related technology, the method does not need professional algorithm personnel to support the training model, and saves the labeling time. And moreover, the convenient interaction between the labeling personnel and the labeling tool is supported, so that the working difficulty of the labeling personnel is reduced, the labeling efficiency is improved, and the development time of an algorithm model is saved.
Example four:
in another embodiment of the present invention, as shown in fig. 4, a method for labeling a data set includes:
step S402, receiving data to be marked;
step S404, analyzing the data to be labeled to determine the task type of the data to be labeled, and storing the data to be labeled to a local storage area according to the task type;
step S406, marking the data to be marked in response to the marking instruction so as to determine a contour coordinate set of the data to be marked;
step S408, determining the task category as target segmentation, and establishing grid line coordinates for data to be annotated;
and S410, screening target coordinate points which are overlapped with grid points of the grid lines in the contour coordinate set, and determining the target coordinate set according to the target coordinate points.
In the embodiment, the data set annotation method comprises the steps of receiving data to be annotated, configuring the server, namely, the data to be marked is uploaded to a server, the operation to be executed on the data to be marked is determined to determine the task category when the data to be marked is uploaded, the task category is stored in the data to be marked in a label form, the server analyzes the data to be marked and determines the task category of the data to be marked, the data to be marked is stored in a local storage area according to the difference of the task categories, wherein the task category comprises a target segmentation task and a target detection task, the local storage area of the server also comprises a target segmentation task storage area and a target detection task storage area, and the data to be labeled of different task categories are respectively stored in the two storage areas according to the task category of the data to be labeled, so that the subsequent data to be labeled can be conveniently searched and used. The server receives a labeling instruction, labels data to be labeled to obtain a contour coordinate set of the data to be labeled, wherein the labeling instruction comprises a task starting instruction, a task selecting instruction, a task executing instruction and the like, the server responds to the labeling instruction to select the data to be labeled and labels the data, the specific labeling mode is that the data to be labeled are labeled through coordinate points, and the set of the coordinate points used for labeling the picture is the contour coordinate set. The obtained contour coordinate set needs to be screened to determine a target coordinate set, different screening conditions are selected for the data to be labeled of different task categories to be screened, and coordinate points in the contour coordinate set are screened according to the screening conditions corresponding to the task categories to determine a final target coordinate set.
And when the task type is target segmentation, establishing grid line coordinates for the data to be labeled with the contour coordinate points, screening out coordinate points which are coincident with the grid line coordinates, and taking a set of the coordinate points as a target coordinate set. When the task type is the target segmentation task, the screening of useful coordinate points in the outline coordinate set is realized
The target contour is extracted through the traditional image processing segmentation algorithm and manual interaction, and is applied to the auxiliary data annotation in the data annotation task of the data crowdsourcing platform, by classifying the data to be labeled and selecting different screening conditions for the data to be labeled of different types, the labeling task of the data to be labeled of different types can be completed even if common data labeling personnel with insufficient algorithm speciality in a crowdsourcing platform, because different screening conditions are set for different data to be labeled, two types of data labeling tasks can be completed based on the method, compared with the traditional image processing segmentation algorithm application corresponding method, the labeling speed and quality of the data set are improved, compared with the data pre-labeled by using a deep learning model in the related technology, the method does not need professional algorithm personnel to support the training model, and saves the labeling time. And moreover, the convenient interaction between the labeling personnel and the labeling tool is supported, so that the working difficulty of the labeling personnel is reduced, the labeling efficiency is improved, and the development time of an algorithm model is saved.
In the above embodiment, at least one target coordinate point is selected as the set coordinate point; and determining the coordinate points within the set range of the set coordinate points as redundant coordinate points, and removing the redundant coordinate points in the target coordinate set.
In the embodiment, when the number of coordinate points is too many, at least one target coordinate point is selected as a set coordinate point, all coordinate points within a set range from the set coordinate point are determined as redundant coordinate points, and the redundant coordinate points are removed from the target coordinate set, so that the effect of ensuring the target coordinate set is realized, the data volume of the target coordinate set is reduced, and the occupied amount of a system space is reduced.
It should be noted that, if two or more coordinate points are selected as the set coordinate points, the distance between two adjacent set coordinate points is larger than the diameter of the set range, so as to avoid determining the set coordinate points as the redundant coordinate points when screening the redundant coordinate points.
As shown in fig. 6, if the annotation data is the target segmentation task data, the contour coordinate set is optimized by using a contour coordinate set optimization algorithm according to the contour coordinates in the contour coordinate set of the target, and the number of the contour coordinates is reduced and saved as the target coordinate set. In the algorithm process, a gridding segmentation intersection point algorithm is used for finding a labeling coordinate set.
The target in fig. 6 is an actual object to be labeled, the edge of the object is an edge of the object to be labeled, and the coordinate point of the edge is labeled and extracted by using an auxiliary labeling tool, then the target coordinate point is obtained by setting a grid coordinate system on a pixel in the image, the target coordinate point is obtained by setting a distance d between grids of grid coordinates, assuming that the coordinate value of each point in the object contour coordinate set is (X, Y), and when X-nd is equal to 0 and Y-nd is equal to 0, (X, Y) is determined as the target coordinate point.
It is understood that 0. ltoreq. nd.ltoreq.W and 0. ltoreq. nd.ltoreq.H.
Where W denotes the width of the image being labeled, H denotes the height of the image being labeled, and n may take values of 0 and a positive integer.
It is worth noting that when the grid lines and the object contour lines are just overlapped when the target coordinate points are extracted in a grid mode, a large number of coordinate sets meeting conditions are generated, at the moment, the coordinate sets with similar distances need to be further filtered, the filtering method is a searching algorithm, whether the target coordinate points exist in a certain range of straight line distance D near the points or not is searched, and if the target coordinate points exist, the target coordinate points are deleted from the marked points.
Example five:
as shown in fig. 5, a complete embodiment of the present invention provides a method for annotating a data set, including:
step S502, uploading data;
step S504, selecting the type of the labeling task;
step S506, determining a target segmentation task;
step S508, determining a target detection task;
step S510, extracting a contour coordinate set;
s512, extracting a contour labeling coordinate set;
step S514, extracting target rectangular contour points;
step S516, labels are marked manually;
step S518, segmenting the annotation file by the target;
step S520, detecting a label file by a target;
in step S522, the data is finally verified and saved.
In the embodiment, data is uploaded and an annotation task category is selected; the process mainly includes the steps of creating a task, uploading data to be labeled to a database server of a crowdsourcing labeling platform, and in the step, selecting the type of the task, and setting information such as a label name of a target to be labeled and a task data source description according to project requirements.
The image annotation task is only related to an image annotation task, and the image annotation task generally comprises three types of target detection and target segmentation.
And carrying out auxiliary labeling on the picture to extract a target contour coordinate set. The image processing algorithm generally involved in the process is mainly a target segmentation algorithm in the traditional image processing algorithm, and also comprises other image processing algorithms, such as commonly used operation algorithms of image enhancement, image filling algorithm, image contour extraction, expansion corrosion in image morphology processing and the like.
In the process of extracting the contour coordinate set, the following object segmentation methods can be used in the general implementation process of the algorithm:
(1) for an image to be annotated with a simple background, a segmentation method based on a gray threshold or a segmentation method based on edge detection is generally used.
(2) In general, the more common scenes are that the image to be marked contains a complex background and a plurality of objects, so that more complex segmentation methods are adopted, and one type is a segmentation method based on image regions, including a watershed algorithm, a seed region growing algorithm and the like. Another category of graph theory-based methods are the GrabCut algorithms that are most used.
And obtaining an outer contour coordinate set of the target to be marked in the picture by the aid of the two algorithms. Meanwhile, in the specific execution process, some detailed processing is performed, for example, in order to ensure that the labeling frame can completely contain the object edge in the subsequent processing, morphological dilation processing can be performed on the image contour before the target contour coordinate set is extracted, and the image edge is expanded outwards.
And if the target detection task data is the target detection task data, extracting the target rectangular outer contour according to each contour coordinate set of the target, and searching and recording target coordinates of the upper left corner and the lower right corner of the target rectangular outer contour by using an algorithm. In the process, the maximum rectangular enclosure frame of the area where the outline coordinate set is located needs to be calculated, and the coordinates of the rectangular enclosure frame are recorded, a common coordinate calculation mode is to calculate the minimum values Xmin and Ymin and the maximum values Xmax and Ymax of the values of all the coordinates in the coordinate set, and then two coordinate points of the maximum rectangular enclosure frame are recorded as follows: upper left corner (Xmin, Ymin), lower right corner (Xmax, Ymax).
As shown in fig. 6, if the target segmentation task is performed, the contour coordinate set is optimized by using a contour coordinate set optimization algorithm according to the contour coordinates in the contour coordinate set of the target, and the number of the contour coordinates is reduced and stored as the target coordinate set. In the algorithm process, a gridding segmentation intersection point algorithm is used for finding a labeling coordinate set.
The target in fig. 6 is an actual object to be labeled, the edge of the object is an edge of the object to be labeled, and the coordinate point of the edge is labeled and extracted by using an auxiliary labeling tool, then the target coordinate point is obtained by setting a grid coordinate system on a pixel in the image, the target coordinate point is obtained by setting a distance d between grids of grid coordinates, assuming that the coordinate value of each point in the object contour coordinate set is (X, Y), and when X-nd is equal to 0 and Y-nd is equal to 0, (X, Y) is determined as the target coordinate point.
It is understood that 0. ltoreq. nd.ltoreq.W and 0. ltoreq. nd.ltoreq.H.
Where W denotes the width of the image being labeled, H denotes the height of the image being labeled, and n may take values of 0 and a positive integer.
It is worth noting that when the grid lines and the object contour lines are just overlapped when the target coordinate points are extracted in a grid mode, a large number of coordinate sets meeting conditions are generated, at the moment, the coordinate sets with similar distances need to be further filtered, the filtering method is a searching algorithm, whether the target coordinate points exist in a certain range of straight line distance D near the points or not is searched, and if the target coordinate points exist, the target coordinate points are deleted from the marked points.
After the annotation point is extracted, the annotation point target needs to be selected to manually select a label according to the label name, and then the annotation file is stored, wherein the format of the annotation file for target detection and target segmentation is generally an xml (file format) file or a json (file format) file. The process of generating the annotation file is generally automatically generated after the background of the intelligent crowdsourcing platform automatically reads the annotation information and the picture information.
And finally correcting the marked file, and storing the marked file to a crowdsourcing data platform background database. Manual checking is a necessary condition for ensuring the accuracy of training data in a general data labeling process. In the step, the marked file needs to be manually checked whether the marked file is correct or not, if the marked file is correct, the marked file is stored in a background database, and then the marking of the next picture is continued. And if the labeling position is inaccurate or the label is wrong, correcting the labeling information on the labeling page of the labeling platform and then storing the labeling information.
Example six:
as shown in FIG. 7, in one embodiment of the present invention, a data set annotation apparatus 100 is provided, which includes: memory 110, processor 120, and computer programs stored on memory 110 and executable on processor 120; when executed by a processor, the computer program implements the data set annotation method according to any of the above technical solutions, thereby having all the beneficial effects of the data set annotation method in any of the above embodiments, and not being described in detail herein.
Example seven:
in another embodiment of the present invention, a computer-readable storage medium is provided, in which a control program for labeling a data set is stored, and when the data set labeling program is executed by a processor, the steps of the control method for labeling a valve body data set as in the first embodiment are implemented. Therefore, all the beneficial effects of the data set labeling method proposed in any embodiment of the present invention are achieved, and are not described herein again.
In the present invention, the term "plurality" means two or more unless explicitly defined otherwise. The terms "mounted," "connected," "fixed," and the like are to be construed broadly, and for example, "connected" may be a fixed connection, a removable connection, or an integral connection; "coupled" may be direct or indirect through an intermediary. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
In the description herein, the description of the terms "one embodiment," "some embodiments," "specific embodiments," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A method for annotating a data set, comprising:
receiving data to be marked;
analyzing the data to be marked to determine the task type of the data to be marked, and storing the data to be marked to a local storage area according to the task type;
marking the data to be marked in response to a marking instruction so as to determine a contour coordinate set of the data to be marked;
determining coordinate screening conditions corresponding to the task categories, screening coordinate points in the contour coordinate set according to the coordinate screening conditions, and determining a target coordinate set;
wherein the task types include: and target segmentation and target detection based on the deep learning task.
2. The data set annotation method of claim 1, wherein the step of annotating the data to be annotated in response to an annotation instruction specifically comprises:
selecting a corresponding auxiliary marking tool according to the task category of the data to be marked;
and controlling the auxiliary marking tool to mark the data to be marked according to the marking instruction.
3. The data set labeling method according to claim 2, wherein the task type is target detection, and the step of screening the coordinate points in the contour coordinate set according to the coordinate screening condition specifically includes:
determining a rectangular surrounding frame of an area where the outline coordinate set is located, and collecting all coordinate points in the rectangular surrounding frame;
screening all coordinate points in the rectangular surrounding frame, and extracting the coordinate point with the minimum coordinate value and the coordinate point with the maximum coordinate value as the target coordinate set.
4. The data set labeling method according to claim 3, wherein the task type is target segmentation, and the step of screening the coordinate points in the contour coordinate set according to the coordinate screening condition specifically comprises:
establishing grid line coordinates for the data to be marked;
and screening target coordinate points which are coincident with grid points of the grid lines in the contour coordinate set, and determining the target coordinate set according to the target coordinate points.
5. The data set annotation method of claim 4, further comprising:
selecting at least one target coordinate point as a set coordinate point;
and determining the coordinate points within the set range of the set coordinate points as redundant coordinate points, and removing the redundant coordinate points in the target coordinate set.
6. The data set annotation method according to any one of claims 1 to 5, wherein the step of storing the data to be annotated to a local storage area according to the task category specifically includes:
determining task information of the data to be marked;
storing the data to be marked and the corresponding task information in a local storage area according to the task type;
the task information includes but is not limited to a task tag and a source of data to be annotated.
7. The data set annotation method of claim 6, further comprising:
and generating a label file according to the target coordinate set and the corresponding task information, and storing the label file in a local storage area in a set format.
8. The data set annotation method of claim 7, wherein prior to the step of storing the annotation file in a set format in a local storage area, further comprising:
and determining that the annotation file has an annotation error, and correcting the annotation file.
9. A data set annotation device, comprising:
a memory, a processor, and a computer program stored on the memory and executable on the processor;
the computer program, when executed by the processor, implements a data set annotation method as claimed in any one of claims 1 to 8.
10. A computer-readable storage medium, having stored thereon a data set annotation program which, when executed by a processor, carries out the steps of the data set annotation method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010168698.3A CN111414907A (en) | 2020-03-12 | 2020-03-12 | Data set labeling method, data set labeling device and computer-readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010168698.3A CN111414907A (en) | 2020-03-12 | 2020-03-12 | Data set labeling method, data set labeling device and computer-readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111414907A true CN111414907A (en) | 2020-07-14 |
Family
ID=71494250
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010168698.3A Pending CN111414907A (en) | 2020-03-12 | 2020-03-12 | Data set labeling method, data set labeling device and computer-readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111414907A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112434660A (en) * | 2020-12-11 | 2021-03-02 | 宁夏回族自治区自然资源信息中心 | High-resolution remote sensing image land class data set manufacturing method based on segmentation algorithm |
| CN112732949A (en) * | 2021-01-19 | 2021-04-30 | 广州虎牙科技有限公司 | Service data labeling method and device, computer equipment and storage medium |
| CN115018975A (en) * | 2022-05-20 | 2022-09-06 | 北京火山引擎科技有限公司 | Data set generation method and device, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108241662A (en) * | 2016-12-23 | 2018-07-03 | 北京国双科技有限公司 | The optimization method and device of data mark |
| CN109492686A (en) * | 2018-11-01 | 2019-03-19 | 郑州云海信息技术有限公司 | A kind of picture mask method and system |
| CN109523563A (en) * | 2018-11-01 | 2019-03-26 | 华东师范大学 | A method of improving micro-image annotating efficiency |
| CN110321450A (en) * | 2019-05-05 | 2019-10-11 | 苏宁易购集团股份有限公司 | A kind of data auxiliary mask method, apparatus and system for target detection |
| WO2020015077A1 (en) * | 2018-07-18 | 2020-01-23 | 平安科技（深圳）有限公司 | Method and apparatus for labeling and displaying electronic map, terminal device, and storage medium |
-
2020
- 2020-03-12 CN CN202010168698.3A patent/CN111414907A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108241662A (en) * | 2016-12-23 | 2018-07-03 | 北京国双科技有限公司 | The optimization method and device of data mark |
| WO2020015077A1 (en) * | 2018-07-18 | 2020-01-23 | 平安科技（深圳）有限公司 | Method and apparatus for labeling and displaying electronic map, terminal device, and storage medium |
| CN109492686A (en) * | 2018-11-01 | 2019-03-19 | 郑州云海信息技术有限公司 | A kind of picture mask method and system |
| CN109523563A (en) * | 2018-11-01 | 2019-03-26 | 华东师范大学 | A method of improving micro-image annotating efficiency |
| CN110321450A (en) * | 2019-05-05 | 2019-10-11 | 苏宁易购集团股份有限公司 | A kind of data auxiliary mask method, apparatus and system for target detection |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112434660A (en) * | 2020-12-11 | 2021-03-02 | 宁夏回族自治区自然资源信息中心 | High-resolution remote sensing image land class data set manufacturing method based on segmentation algorithm |
| CN112434660B (en) * | 2020-12-11 | 2023-08-22 | 宁夏回族自治区自然资源信息中心 | High-resolution remote sensing image ground data set manufacturing method based on segmentation algorithm |
| CN112732949A (en) * | 2021-01-19 | 2021-04-30 | 广州虎牙科技有限公司 | Service data labeling method and device, computer equipment and storage medium |
| CN112732949B (en) * | 2021-01-19 | 2023-10-17 | 广州虎牙科技有限公司 | Service data labeling method and device, computer equipment and storage medium |
| CN115018975A (en) * | 2022-05-20 | 2022-09-06 | 北京火山引擎科技有限公司 | Data set generation method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111414907A (en) | Data set labeling method, data set labeling device and computer-readable storage medium | |
| CN109522900B (en) | Natural scene character recognition method and device | |
| CN110502985B (en) | Form identification method and device and form identification equipment | |
| CN110751682B (en) | Method, device, terminal equipment and storage medium for extracting and identifying image | |
| CN110838105A (en) | Business process model image identification and reconstruction method | |
| CN110263794B (en) | Training method of target recognition model based on data enhancement | |
| CN110503705B (en) | Image labeling method and device | |
| CN111415364B (en) | Conversion method, system and storage medium for image segmentation sample in computer vision | |
| CN112883795B (en) | Rapid and automatic table extraction method based on deep neural network | |
| CN115063618B (en) | Defect positioning method, system, equipment and medium based on template matching | |
| CN110598698A (en) | Natural scene text detection method and system based on adaptive regional suggestion network | |
| CN114419647A (en) | Table information extraction method and system | |
| CN115909059A (en) | Natural resource sample library establishing method and device | |
| CN111612886B (en) | Indoor three-dimensional model generation method and system | |
| CN112633118A (en) | Text information extraction method, equipment and storage medium | |
| CN110135407B (en) | Sample labeling method and computer storage medium | |
| CN111626145A (en) | Simple and effective incomplete form identification and page-crossing splicing method | |
| CN112733855B (en) | Table structuring method, table recovering device and device with storage function | |
| CN117612195A (en) | Graph model generation method and device based on main wiring graph recognition technology | |
| CN115984796A (en) | Image annotation method and system | |
| Herold et al. | Automatically georeferenced maps as a source for high resolution urban growth analyses | |
| CN114240734A (en) | Image data augmentation method, image data augmentation device, electronic apparatus, and storage medium | |
| CN111507287A (en) | Method and system for extracting road zebra crossing corner points in aerial image | |
| CN110598697A (en) | Container number positioning method based on thickness character positioning | |
| CN110298287A (en) | A kind of power equipment automatic identifying method based on deep learning |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200714 |