CN110070533A - A kind of evaluating method of object detection results, device, equipment and storage medium - Google Patents

A kind of evaluating method of object detection results, device, equipment and storage medium Download PDF

Info

Publication number
CN110070533A
CN110070533A CN201910328909.2A CN201910328909A CN110070533A CN 110070533 A CN110070533 A CN 110070533A CN 201910328909 A CN201910328909 A CN 201910328909A CN 110070533 A CN110070533 A CN 110070533A
Authority
CN
China
Prior art keywords
testing result
target
degree
detection
annotation results
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.)
Granted
Application number
CN201910328909.2A
Other languages
Chinese (zh)
Other versions
CN110070533B (en
Inventor
徐文君
潘松
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.)
iFlytek Co Ltd
Original Assignee
iFlytek Co Ltd
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 iFlytek Co Ltd filed Critical iFlytek Co Ltd
Priority to CN201910328909.2A priority Critical patent/CN110070533B/en
Publication of CN110070533A publication Critical patent/CN110070533A/en
Application granted granted Critical
Publication of CN110070533B publication Critical patent/CN110070533B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G06T7/001—Industrial image inspection using an image reference approach
    • 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)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application proposes evaluating method, device, equipment and the storage medium of a kind of object detection results, this method comprises: obtaining the testing result for corresponding to region locating for goal-selling and the annotation results corresponding to region locating for the target respectively；It determines whole departure degree of the testing result relative to the annotation results, and/or determines the testing result to the detection integrity degree of the target；The evaluation result to the testing result is determined to the detection integrity degree of the target relative to the whole departure degree of the annotation results and/or the testing result according to the testing result.This method can be realized the evaluation and test of the automation to object detection results, so that it is determined that the superiority and inferiority of testing result.Also, this method can combine the evaluation and test of multiple dimensions to testing result, and realization evaluates and tests testing result from multiple dimensions, so that more objective to the evaluation and test of testing result, more comprehensively.

Description

A kind of evaluating method of object detection results, device, equipment and storage medium
Technical field
This application involves target detection technique field more particularly to a kind of evaluating method of object detection results, device, set Standby and storage medium.
Background technique
Target detection refers to and detects object content from preset content, is the important prerequisite for identifying object content.
The automatic detection of target may be implemented in algorithm of target detection there are many existing, substantially increases the automatic of target detection Change horizontal.For a user, it is also necessary to object detection results be evaluated and tested, to determine the superiority and inferiority of testing result.
Summary of the invention
Based on the demand, the application proposes that evaluating method, device, equipment and the storage of a kind of object detection results are situated between The evaluation and test to object detection results may be implemented in matter, and the technical solution is specific as follows:
A kind of evaluating method of object detection results, comprising:
It obtains respectively and corresponds to the testing result in region locating for goal-selling and corresponding to region locating for the target Annotation results；
It determines whole departure degree of the testing result relative to the annotation results, and/or determines the detection knot Detection integrity degree of the fruit to the target；
Whole departure degree and/or the testing result pair according to the testing result relative to the annotation results The detection integrity degree of the target determines the evaluation result to the testing result.
Optionally, whole departure degree of the determination testing result relative to the annotation results, comprising:
According to central point and the outside of the position and the annotation results of the central point of the testing result and outer boundary The position on boundary determines whole departure degree of the testing result relative to the annotation results.
Optionally, described according to the central point of the testing result and the position of outer boundary and the annotation results The position of central point and outer boundary determines whole departure degree of the testing result relative to the annotation results, comprising:
According to the position of the position of the outer boundary of the testing result and the outer boundary of the annotation results, determine to include institute State the height and width of the minimum circumscribed rectangle of testing result and the annotation results；
According to the position of the position of the central point of the testing result and the central point of the annotation results, determination is described most Between the height of small boundary rectangle, and the vertical range of the central point of the central point and annotation results of the testing result Difference as the first difference, and determines the width of the minimum circumscribed rectangle, with the central point of the testing result and described Difference between the horizontal distance of the central point of annotation results, as the second difference；
According to the height of the minimum circumscribed rectangle and width, first difference and second difference, institute is determined State whole departure degree of the testing result relative to the annotation results.
Wherein, described according to the height and width of the minimum circumscribed rectangle, first difference and described second poor Value, determines whole departure degree of the testing result relative to the annotation results, comprising:
According to first difference and the height of the minimum circumscribed rectangle ratio and second difference with it is described The sum of ratio of width of minimum circumscribed rectangle, it is negative relative to the whole departure degree of the annotation results with the testing result Relevant relationship determines whole departure degree of the testing result relative to the annotation results.
Optionally, detection integrity degree of the determination testing result to the target, comprising:
By detecting whether the testing result includes completely the target, determine the testing result to the target Detect integrity degree.
Optionally, described by detecting whether the testing result includes completely the target, determine the testing result To the detection integrity degree of the target, comprising:
Whether the starting content and/or end content for determining the testing result are preset content；Wherein, described default interior Hold is that cannot function as the starting content of content entirety and/or the content element of end content in the target fields；
If it is preset content, it is determined that the testing result does not completely include the target, and then determines the inspection Surveying result is the first integrity degree to the detection integrity degree of the target；
If not preset content, it is determined that the testing result completely includes the target, and then determines the detection It as a result is the second integrity degree to the detection integrity degree of the target；Wherein, first integrity degree is less than second integrity degree.
Optionally, described by detecting whether the testing result includes completely the target, determine the testing result To the detection integrity degree of the target, comprising:
The testing result is aligned with the annotation results, determines the difference of the testing result Yu the annotation results Part；
Judge the difference section whether include the target content；
If the difference section includes the content of the target, it is determined that the testing result is not completely comprising described Target, and then determine that the testing result is third integrity degree to the detection integrity degree of the target；
If the difference section does not include the content of the target, it is determined that the testing result completely includes the mesh Mark, and then determine that the testing result is the 4th integrity degree to the detection integrity degree of the target；
Wherein, the third integrity degree is less than the 4th integrity degree.
Optionally, described by detecting whether the testing result includes completely the target, determine the testing result To the detection integrity degree of the target, comprising:
Determine the content of the target；
By detecting the overlapping cases of the outer boundary of the testing result and the content of the target, the detection knot is determined Detection integrity degree of the fruit to the target.
Wherein, the overlapping cases of the content of the outer boundary by detecting the testing result and the target, determine Detection integrity degree of the testing result to the target, comprising:
It is overlapping to determine whether the outer boundary of the testing result and the content of the target have；
If be not overlapped, it is determined that the testing result is the 5th integrity degree to the detection integrity degree of the target；
If there is overlapping, the then position according to lap in the outer boundary of the testing result and lap picture The total pixel quantity of the outer boundary of vegetarian refreshments quantity and the testing result, be calculated the outer boundary of the testing result with it is described The overlapping degree of the content of target, as the testing result to the detection integrity degree of the target.
Wherein, the position of the outer boundary according to lap in the testing result and lap pixel The total pixel quantity of the outer boundary of quantity and the testing result, be calculated the testing result outer boundary and the target Content overlapping degree, comprising:
According to lap in the position of the outer boundary of the testing result, determine that overlapping degree calculates the factor；Wherein, weight Folded part identified overlapping degree when on the longer outer boundary of the testing result calculates the factor, is less than lap in institute Identified overlapping degree calculates the factor when stating on the shorter outer boundary of testing result；
According to the total pixel quantity of the outer boundary of lap pixel quantity and the testing result, it is calculated described In the outer boundary pixel of testing result, the not pixel ratio Chong Die with the content of the target；
In the outer boundary pixel that the overlapping degree is calculated to the factor and the testing result, not with the content of the target The pixel ratio of overlapping is multiplied, and the overlapping degree of the outer boundary of the testing result and the content of the target is calculated.
Optionally, the whole departure degree and the inspection according to the testing result relative to the annotation results Result is surveyed to the detection integrity degree of the target, determines the evaluation result to the testing result, comprising:
By the testing result relative to the whole departure degree of the annotation results and the testing result to the mesh Target detects integrity degree, inputs preset testing result evaluation and test model, determines the evaluation result to the testing result；
Wherein, the testing result evaluation and test model is at least through according to testing result relative to the whole departure of annotation results Degree and testing result carry out evaluation and test training to testing result and obtain to the detection integrity degree of target.
Optionally, described corresponding to the testing result in region locating for goal-selling and corresponding to region locating for the target Annotation results, for corresponding to region locating for preset multiple targets each testing result and correspond to the multiple target Each annotation results in locating region；
After determining to the evaluation result of each testing result, the method also includes:
According to the evaluation result to each testing result, determine to obtaining the target detection of each testing result The evaluation result of algorithm.
A kind of evaluating apparatus of object detection results, comprising:
Data capture unit, for obtaining the testing result corresponding to region locating for goal-selling respectively and corresponding to institute State the annotation results in region locating for target；
Parameter determination unit, for determining whole departure degree of the testing result relative to the annotation results, and/ Or determine the testing result to the detection integrity degree of the target；
Evaluation process unit, for the whole departure degree according to the testing result relative to the annotation results, and/ Or the testing result determines the evaluation result to the testing result to the detection integrity degree of the target.
Optionally, the parameter determination unit determines whole departure journey of the testing result relative to the annotation results When spending, it is specifically used for:
According to central point and the outside of the position and the annotation results of the central point of the testing result and outer boundary The position on boundary determines whole departure degree of the testing result relative to the annotation results.
Optionally, the parameter determination unit, comprising:
First parameter determination unit, for according to the outer of the position of the outer boundary of the testing result and the annotation results The position on boundary determines the height and width of the minimum circumscribed rectangle comprising the testing result and the annotation results；
Second parameter determination unit, in the position and the annotation results according to the central point of the testing result The position of heart point, determines the height of the minimum circumscribed rectangle, with the central point of the testing result and the annotation results Difference between the vertical range of central point as the first difference, and determines the width of the minimum circumscribed rectangle, and described Difference between the central point of testing result and the horizontal distance of the central point of the annotation results, as the second difference；
Calculation processing unit, for height and width, first difference according to the minimum circumscribed rectangle, Yi Jisuo The second difference is stated, determines whole departure degree of the testing result relative to the annotation results.
Wherein, height and width, first difference of the calculation processing unit according to the minimum circumscribed rectangle, with And second difference is specifically used for when determining whole departure degree of the testing result relative to the annotation results:
According to first difference and the height of the minimum circumscribed rectangle ratio and second difference with it is described The sum of ratio of width of minimum circumscribed rectangle, it is negative relative to the whole departure degree of the annotation results with the testing result Relevant relationship determines whole departure degree of the testing result relative to the annotation results.
Optionally, when the parameter determination unit determines detection integrity degree of the testing result to the target, specifically For:
By detecting whether the testing result includes completely the target, determine the testing result to the target Detect integrity degree.
Optionally, the parameter determination unit is determined by detecting whether the testing result completely includes the target When the testing result is to the detection integrity degree of the target, it is specifically used for:
Whether the starting content and/or end content for determining the testing result are preset content；Wherein, described default interior Hold is that cannot function as the starting content of content entirety and/or the content element of end content in the target fields；
If it is preset content, it is determined that the testing result does not completely include the target, and then determines the inspection Surveying result is the first integrity degree to the detection integrity degree of the target；
If not preset content, it is determined that the testing result completely includes the target, and then determines the detection It as a result is the second integrity degree to the detection integrity degree of the target；Wherein, first integrity degree is less than second integrity degree.
Optionally, the parameter determination unit is determined by detecting whether the testing result completely includes the target When the testing result is to the detection integrity degree of the target, it is specifically used for:
The testing result is aligned with the annotation results, determines the difference of the testing result Yu the annotation results Part；
Judge the difference section whether include the target content；
If the difference section includes the content of the target, it is determined that the testing result is not completely comprising described Target, and then determine that the testing result is third integrity degree to the detection integrity degree of the target；
If the difference section does not include the content of the target, it is determined that the testing result completely includes the mesh Mark, and then determine that the testing result is the 4th integrity degree to the detection integrity degree of the target；
Wherein, the third integrity degree is less than the 4th integrity degree.
Optionally, the parameter determination unit is determined by detecting whether the testing result completely includes the target When the testing result is to the detection integrity degree of the target, it is specifically used for:
Determine the content of the target；
By detecting the overlapping cases of the outer boundary of the testing result and the content of the target, the detection knot is determined Detection integrity degree of the fruit to the target.
Wherein, the overlapping cases of the content of the outer boundary by detecting the testing result and the target, determine Detection integrity degree of the testing result to the target, comprising:
It is overlapping to determine whether the outer boundary of the testing result and the content of the target have；
If be not overlapped, it is determined that the testing result is the 5th integrity degree to the detection integrity degree of the target；
If there is overlapping, the then position according to lap in the outer boundary of the testing result and lap picture The total pixel quantity of the outer boundary of vegetarian refreshments quantity and the testing result, be calculated the outer boundary of the testing result with it is described The overlapping degree of the content of target, as the testing result to the detection integrity degree of the target.
Wherein, the position of the outer boundary according to lap in the testing result and lap pixel The total pixel quantity of the outer boundary of quantity and the testing result, be calculated the testing result outer boundary and the target Content overlapping degree, comprising:
According to lap in the position of the outer boundary of the testing result, determine that overlapping degree calculates the factor；Wherein, weight Folded part identified overlapping degree when on the longer outer boundary of the testing result calculates the factor, is less than lap in institute Identified overlapping degree calculates the factor when stating on the shorter outer boundary of testing result；
According to the total pixel quantity of the outer boundary of lap pixel quantity and the testing result, it is calculated described In the outer boundary pixel of testing result, the not pixel ratio Chong Die with the content of the target；
In the outer boundary pixel that the overlapping degree is calculated to the factor and the testing result, not with the content of the target The pixel ratio of overlapping is multiplied, and the overlapping degree of the outer boundary of the testing result and the content of the target is calculated.
Optionally, whole departure journey of the evaluation process unit according to the testing result relative to the annotation results Degree and the testing result when determining the evaluation result to the testing result, specifically use the detection integrity degree of the target In:
By the testing result relative to the whole departure degree of the annotation results and the testing result to the mesh Target detects integrity degree, inputs preset testing result evaluation and test model, determines the evaluation result to the testing result；
Wherein, the testing result evaluation and test model is at least through according to testing result relative to the whole departure of annotation results Degree and testing result carry out evaluation and test training to testing result and obtain to the detection integrity degree of target.
Optionally, when the testing result corresponding to region locating for goal-selling and corresponding to area locating for the target The annotation results in domain for each testing result corresponding to region locating for preset multiple targets and correspond to the multiple mesh When marking each annotation results in locating region, described device further include:
Algorithm evaluates and tests unit, for the evaluation result according to the evaluation process unit to each testing result, really The fixed evaluation result to the algorithm of target detection for obtaining each testing result.
A kind of evaluation and test equipment of object detection results, comprising:
Memory and processor；
Wherein, the memory is connected to the processor, for storing program；
The processor, for realizing following functions by the program in the operation memory:
It obtains respectively and corresponds to the testing result in region locating for goal-selling and corresponding to region locating for the target Annotation results；It determines whole departure degree of the testing result relative to the annotation results, and/or determines the detection knot Detection integrity degree of the fruit to the target；Whole departure degree according to the testing result relative to the annotation results, and/ Or the testing result determines the evaluation result to the testing result to the detection integrity degree of the target.
A kind of storage medium is stored with computer program on the storage medium, and the computer program is held by processor When row, each step of the evaluating method of above-mentioned object detection results is realized.
The evaluating method for the object detection results that the application proposes, can be by determining testing result relative to annotation results Whole departure degree and/or testing result to the detection integrity degree of target, determine the evaluation result to testing result.The party Method can be realized the evaluation and test of the automation to object detection results, so that it is determined that the superiority and inferiority of testing result.
Further, technical scheme can combine the evaluation and test of multiple dimensions to testing result, realize from Multiple dimensions evaluate and test testing result, so that more objective to the evaluation and test of testing result, more comprehensively.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this The embodiment of application for those of ordinary skill in the art without creative efforts, can also basis The attached drawing of offer obtains other attached drawings.
Fig. 1 is a kind of flow diagram of the evaluating method of object detection results provided by the embodiments of the present application；
Fig. 2 is the schematic diagram provided by the embodiments of the present application for being detected and being marked to goal-selling；
Fig. 3 is processing of the determining testing result provided by the embodiments of the present application relative to the whole departure degree of annotation results Flow diagram；
The annotation results and testing result that Fig. 4 is the same target of correspondence provided by the embodiments of the present application are under the same coordinate system Expression schematic diagram；
Fig. 5 is a kind of testing result schematic diagram of corresponding goal-selling provided by the embodiments of the present application；
Fig. 6 is that a kind of determining testing result provided by the embodiments of the present application shows the process flow of the detection integrity degree of target It is intended to；
Fig. 7 is the signal that the two difference section is determined after annotation results provided by the embodiments of the present application are aligned with testing result Figure；
Fig. 8 is another process flow for detecting integrity degree for determining testing result to target provided by the embodiments of the present application Schematic diagram；
Fig. 9 is another testing result schematic diagram of corresponding goal-selling provided by the embodiments of the present application；
Figure 10 is that the figure image of cutting provided by the embodiments of the present application to testing result shown in Fig. 9 carries out binary conversion treatment Schematic diagram；
Figure 11 is the PR curve synoptic diagram of corresponding algorithm of target detection provided by the embodiments of the present application；
Figure 12 is a kind of structural schematic diagram of the evaluating apparatus of object detection results provided by the embodiments of the present application；
Figure 13 is a kind of structural schematic diagram of the evaluation and test equipment of object detection results provided by the embodiments of the present application.
Specific embodiment
The embodiment of the present application technical solution is suitable for the application scenarios evaluated and tested to object detection results, using the application Embodiment technical solution may be implemented the evaluation and test to object detection results, i.e., make judge to the superiority and inferiority of object detection results.
Illustratively, the embodiment of the present application technical solution can be applied to the hardware devices such as hardware processor, or be packaged into soft Part program is run, when hardware processor execute the embodiment of the present application technical solution treatment process or above-mentioned software program quilt When operation, the automatic judgment to object detection results may be implemented.The embodiment of the present application is only to the specific of technical scheme Treatment process carries out exemplary introduction, is not defined to the specific execution form of technical scheme, any form of The technology way of realization that technical scheme treatment process can be executed, can be used by the embodiment of the present application.
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete Site preparation description, it is clear that described embodiments are only a part of embodiments of the present application, instead of all the embodiments.It is based on Embodiment in the application, it is obtained by those of ordinary skill in the art without making creative efforts every other Embodiment shall fall in the protection scope of this application.
The embodiment of the present application proposes a kind of evaluating method of object detection results, shown in Figure 1, this method comprises:
S101, the testing result for corresponding to region locating for goal-selling is obtained respectively and corresponds to area locating for the target The annotation results in domain.
Wherein, above-mentioned goal-selling refers to the preset target image content in image, includes but is not limited to be Picture and text, character, figure, image in image etc..Such as mathematical operation formula " the 51 ÷ 34=shown in Fig. 2 in image 51 ÷, 17 ÷ 2 ", can be used as goal-selling.
Region locating for above-mentioned goal-selling refers to above-mentioned goal-selling region locating in the image where it.It can To understand, picture material inherently occupies certain image-region in the picture, and region locating for picture material may be greater than figure As content and comprising the image-region of picture material, or the picture position region exactly the same with picture material position. When being detected to picture material, as long as detecting region locating for picture material from image, also it has been detected by image Hold, therefore, when detecting to the object content in image, usually detects region locating for object content, obtained detection knot Fruit is an area detection result.
Based on above-mentioned theory, carried out when being detected to above-mentioned goal-selling, and to region locating for goal-selling Detection, obtained testing result is an area detection result, such as the region that the solid box in Fig. 2 is included, i.e., to default mesh The testing result that mark " 51 ÷ 34=51 ÷, 17 ÷ 2 " is detected.
Wherein, the testing result in region locating for above-mentioned goal-selling can be through algorithm of target detection from shown in Fig. 2 The testing result that goal-selling " 51 ÷ 34=51 ÷, 17 ÷ 2 " is obtained is detected in image.
Meanwhile in order to which the testing result objectively to region locating for above-mentioned goal-selling is evaluated and tested, the embodiment of the present application While obtaining the testing result in region locating for above-mentioned goal-selling, the mark for region locating for the goal-selling is also obtained As a result.
The annotation results in region locating for above-mentioned goal-selling refer to and are accurately marked region locating for goal-selling As a result.As shown in the dotted line frame in Fig. 2, the accurate mark in region locating for as above-mentioned goal-selling " 51 ÷ 34=51 ÷, 17 ÷ 2 " Note is as a result, the region that the dotted line frame is included, as region locating for the goal-selling.
Wherein, the annotation results in region locating for above-mentioned goal-selling, can be manually to region locating for goal-selling into The result of rower note.Since the annotation results are the accurate annotation results in the region locating for above-mentioned goal-selling, the mark Note result can be used as evaluation and test for the evaluating standard of the testing result in region locating for above-mentioned goal-selling.
S102, it determines whole departure degree of the testing result relative to the annotation results, and/or determines the inspection Result is surveyed to the detection integrity degree of the target.
Whole departure degree of the above-mentioned testing result relative to annotation results, refers to testing result whole relative to mark As a result departure degree is whole with the difference of annotation results in a variety of dimensions such as size, position comprising testing result Body Diversity measure can be used to indicate that testing result and the difference of annotation results on the whole.
It is appreciated that being substantially the inspection to region locating for goal-selling due to above-mentioned testing result and annotation results Result and annotation results are surveyed, therefore the two is the expression to the same region.As shown in Fig. 2, corresponding goal-selling " 51 ÷ 34 The testing result and annotation results in region locating for 17 ÷ 2 " of=51 ÷ be essentially all in the picture frame have selected an image district Domain.Unlike, above-mentioned annotation results are the accurate mark to region locating for goal-selling, and above-mentioned testing result is to default The detection in region locating for target marks.Theoretically, above-mentioned testing result is got over relative to the whole departure degree of above-mentioned annotation results Small, then testing result and annotation results are more identical, and above-mentioned testing result is better to the detection effect of above-mentioned goal-selling；On the contrary, Above-mentioned testing result is bigger relative to the whole departure degree of above-mentioned annotation results, then the difference of testing result and annotation results is got over Greatly, above-mentioned testing result is poorer to the detection effect of above-mentioned goal-selling.Therefore, by determining that testing result is tied relative to mark The whole departure degree of fruit can make judge to the superiority and inferiority of testing result.
Illustratively, whole departure degree of the above-mentioned testing result relative to annotation results, can be with numerical value, graded category Forms such as (basic, normal, high etc.) indicate, and position coordinates that can according to testing result with annotation results in the same coordinate system And it determines.
Above-mentioned testing result refers to testing result to object content comprising journey the detection integrity degree of the target Degree.Illustratively, if testing result completely includes all the elements of target, testing result is to the detection integrity degree of target 100%, if testing result only includes the half content of target, testing result is 50% to the detection integrity degree of target.
It is appreciated that testing result is higher to the detection integrity degree of target, then illustrate inspection of the testing result to object content Survey is more complete, better to the detection effect of target；On the contrary, testing result is lower to the detection integrity degree of target, then illustrate to examine It is more imperfect to the detection of object content to survey result, it is poorer to the detection effect of target.Therefore, pass through the above-mentioned detection knot of determination Fruit can make judge to the superiority and inferiority of testing result to the detection integrity degree of the target.
Illustratively, above-mentioned testing result, can be with shapes such as numerical value or graded categories to the detection integrity degree of the target Formula indicates, can compare determination with object content by will test the content that result is included.
S103, it is tied according to the testing result relative to the whole departure degree of the annotation results and/or the detection Fruit determines the evaluation result to the testing result to the detection integrity degree of the target.
The above-mentioned evaluation result to the testing result, refers to the detection effect to testing result about above-mentioned goal-selling Superiority and inferiority qualitative or quantitative evaluation, include but is not limited to can be testing result superiority and inferiority grade determine, or to detection tie The scoring etc. of fruit.
By above-mentioned introduction it was determined that by determining whole departure degree of the testing result relative to annotation results, or Person determines that testing result to the detection integrity degree of target, can make judge to the superiority and inferiority of testing result.
Therefore, the embodiment of the present application, can after obtaining above-mentioned testing result relative to the whole departure degree of annotation results Relative to the whole departure degree of annotation results, to determine the evaluation result to testing result according to testing result.Theoretically, it examines Survey result is bigger relative to the whole departure degree of annotation results, then the whole difference of the testing result and annotation results is bigger, The testing result is poorer；On the contrary, testing result is smaller relative to the whole departure degree of annotation results, then the testing result and mark The whole difference for infusing result is smaller, and the testing result is better.
Alternatively, after obtaining above-mentioned testing result to the detection integrity degree of the target, it can be according to testing result to institute The detection integrity degree of target is stated, determines the evaluation result to testing result.Theoretically, testing result is complete to the detection of the target Whole degree is higher, then the testing result is better；On the contrary, testing result is lower to the detection integrity degree of the target, then the detection knot Fruit is poorer.
Further, in order to improve the objectivity of the evaluation and test to testing result and comprehensive, the embodiment of the present application is also proposed It can detection to the target of whole departure degree and above-mentioned testing result by above-mentioned testing result relative to annotation results Integrity degree combines, and is provided commonly for determining the evaluation result to above-mentioned testing result.
Theoretically, above-mentioned testing result is relative to the whole departure degree of annotation results is smaller and above-mentioned testing result Higher to the detection integrity degree of the target, then above-mentioned testing result is better；On the contrary, above-mentioned testing result is relative to annotation results Whole departure degree it is bigger and above-mentioned testing result is lower to the detection integrity degree of the target, then above-mentioned testing result It is poorer.
It is appreciated that by above-mentioned testing result relative to the whole departure degree of annotation results and above-mentioned testing result to institute The detection integrity degree for stating target combines to evaluate and test to testing result, actually realizes from multiple dimensions and ties to detection Fruit is evaluated and tested, the one-sidedness that can be evaluated and tested to avoid single dimension so as to the evaluation and test of testing result more comprehensively, it is more objective,
Illustratively, when above-mentioned testing result relative to the whole departure degree of annotation results, above-mentioned testing result to institute (wherein, above-mentioned detection knot when stating the detection integrity degree of target, and being indicated with score value the evaluation result of above-mentioned testing result Fruit is bigger relative to the whole departure degree of annotation results, and score value is lower；Its opposite score value is higher), the embodiment of the present application is Above-mentioned testing result is complete to the detection of the target relative to the whole departure degree of annotation results and above-mentioned testing result Degree, is respectively set weighted value a and b.Determine above-mentioned testing result relative to annotation results respectively when passing through execution step S102 After the score A of whole departure degree and above-mentioned testing result are to the score B of the detection integrity degree of the target, the application is real It applies example and A and B is weighted summation process, that is, calculate the value of aA+bB, obtained calculated result (assuming that aA+bB=C) i.e. conduct The evaluation result score of above-mentioned testing result.It is appreciated that the size of final score C is influenced by two kinds of scores of A and B, it is single The variation of one type score can affect for final score, but the variation of single kind score can not determine final obtain completely again Divide size.
By above-mentioned introduction as it can be seen that the embodiment of the present application propose object detection results evaluating method, can be by true Testing result is determined relative to the whole departure degree and/or testing result of annotation results to the detection integrity degree of target, is determined pair The evaluation result of testing result.This method can be realized the evaluation and test of the automation to object detection results, so that it is determined that testing result Superiority and inferiority.
Further, the embodiment of the present application technical solution can combine the evaluation and test of multiple dimensions to testing result, Realization evaluates and tests testing result from multiple dimensions, so that more objective to the evaluation and test of testing result, more comprehensively.
It as a kind of optional implementation, is disclosed in another embodiment of the application, the above-mentioned determination detection As a result it relative to the whole departure degree of the annotation results, specifically includes:
According to central point and the outside of the position and the annotation results of the central point of the testing result and outer boundary The position on boundary determines whole departure degree of the testing result relative to the annotation results.
Since above-mentioned testing result and annotation results are that region locating for above-mentioned goal-selling is detected and marked respectively Note as a result, therefore, the geometry of testing result and annotation results is identical as the geometry in region locating for goal-selling.Then The central point of the geometry of the central point, that is, testing result and annotation results of testing result and annotation results；Testing result and mark Infuse the outer boundary of outer boundary, that is, testing result of result and the geometry of annotation results.
For example, with reference to shown in Fig. 2, the region as locating for goal-selling " 51 ÷ 34=51 ÷, 17 ÷ 2 " is rectangular area, because The geometry of this corresponding testing result and annotation results is rectangle frame.The central point of testing result and annotation results, i.e., The geometric center point of the rectangle frame of testing result and annotation results；The outer boundary of testing result and annotation results, i.e. testing result With each side of the rectangle frame of annotation results；The position of the central point and outer boundary of testing result and annotation results, i.e. detection knot The position coordinates of the geometric center point of the rectangle frame of fruit and annotation results and each side in a coordinate system.
In order to determine whole departure degree of the testing result relative to annotation results, the embodiment of the present application make testing result and Annotation results are in the same coordinate system, in the position of central point and outer boundary according to testing result and annotation results The position of the heart point and outer boundary determines whole departure degree of the testing result relative to annotation results.
Illustratively, shown in Figure 3, it is disclosed in another embodiment of the application, it is described according to the testing result The central point of the position and the annotation results of central point and outer boundary and the position of outer boundary, determine the testing result Whole departure degree relative to the annotation results, comprising:
S301, the position according to the outer boundary of the position and annotation results of the outer boundary of the testing result, determine The height and width of minimum circumscribed rectangle comprising the testing result and the annotation results；
Shown in Figure 4, the above-mentioned minimum circumscribed rectangle comprising the testing result and the annotation results refers to simultaneously Minimum rectangle frame including testing result and annotation results, the rectangle frame are each vertex by testing result and annotation results In, the rectangle that maximum abscissa, the smallest abscissa, maximum ordinate and the smallest ordinate are determined, and should The side of rectangle frame is parallel with reference axis.
Illustratively, after determining the minimum circumscribed rectangle comprising testing result and annotation results, according to testing result and The position of each outer boundary of annotation results, can determine the position on each side of the minimum circumscribed rectangle, and then can determine The height and width of the minimum circumscribed rectangle.
As another optional implementation, the minimum circumscribed rectangle comprising annotation results a and testing result b is being determined Height and when width, can first determine relative positional relationship when annotation results a and testing result b are in the same coordinate system, Shown in Figure 4, the top left corner apex coordinate of annotation results a is rect a (x, y) at this time, and the width of annotation results a is rect It a (width), is highly rect a (height)；The top left corner apex coordinate of testing result b is rect b (x, y), testing result The width of b is rect b (width), is highly rect tb (height).Since the coboundary of annotation results a is higher than detection knot The ordinate of the coboundary of the minimum circumscribed rectangle can be then set as the vertical seat of the coboundary of annotation results a by the coboundary of fruit b Mark, i.e. Y (up_line)=rect a (y)；Since the lower boundary of testing result b is lower than the lower boundary of annotation results a, then can incite somebody to action The ordinate of the lower boundary of the minimum circumscribed rectangle is set as the ordinate of the lower boundary of testing result b, i.e. Y (down_line) =rect b (y)+rect b (height), thus height Y=Y (down_line)-Y (up_ of entire minimum circumscribed rectangle line)。
Similarly, the above-mentioned external square of minimum can be found out according to annotation results a and the relative position in the horizontal direction testing result b Abscissa X (left_line)=rect a (x) of the left border of the shape and abscissa X (right_line) of right side boundary= Rect b (x)+rect b (width), thus width XX=X (left_line)-X (right_ of entire minimum circumscribed rectangle line)。
S302, the position according to the central point of the position and annotation results of the central point of the testing result, determine The height of the minimum circumscribed rectangle, the vertical range with the central point of the central point and annotation results of the testing result Between difference, as the first difference, and determine the width of the minimum circumscribed rectangle, the central point with the testing result Difference between the horizontal distance of the central point of the annotation results, as the second difference；
It is shown in Figure 4, in the same coordinate system, the shape and each outer boundary of above-mentioned annotation results a and testing result b After position coordinates determine, the position coordinates of the central point of annotation results a and testing result b can also be determined.Above-mentioned minimum is outer Meet the difference between the height of rectangle and the central point of annotation results a and the vertical range of the central point of testing result b, i.e. Y- (vertical range of the central point of the central point and testing result b of annotation results a).According to the phase of annotation results a and testing result b To positional relationship, it is assumed that the distance of the central point of annotation results a to the coboundary of above-mentioned minimum circumscribed rectangle is Ya1, detection knot The central point of fruit b to above-mentioned minimum circumscribed rectangle lower boundary distance be Yb1, then Ya1+Yb1 is the above-mentioned external square of minimum Difference between the height of shape, and the vertical range of the central point of the central point and testing result b of annotation results a, i.e. Ya1+Yb1 =Y- (vertical range of the central point of the central point and testing result b of annotation results a).
Similarly, the central point of the central point and testing result b of the width of above-mentioned minimum circumscribed rectangle and annotation results a Difference between horizontal distance is X- (horizontal distance of the central point of the central point and testing result b of annotation results a).According to mark Infuse the relative positional relationship of result a and testing result b, it is assumed that the central point of annotation results a a to left side for above-mentioned minimum circumscribed rectangle The distance on boundary is Xa1, and the distance of the right margin of the central point of testing result b to above-mentioned minimum circumscribed rectangle is Xb1, then Xa1+ Xb1 is the width of above-mentioned minimum circumscribed rectangle, with the central point of the central point and testing result b of annotation results a it is horizontal away from Difference between, i.e. Xa1+Xb1=X- (horizontal distance of the central point of the central point and testing result b of annotation results a).
As another optional implementation, the calculating of above-mentioned first difference and the second difference can also be by direct The vertical range and horizontal distance for calculating the central point of annotation results a and testing result b, then calculate above-mentioned minimum circumscribed rectangle Height and the difference of above-mentioned vertical range obtain the first difference, calculate above-mentioned minimum circumscribed rectangle width and it is above-mentioned it is horizontal away from From difference obtain the second difference.
Although calculation is different, numerically, the first above-mentioned difference is equal to above-mentioned Ya1+Yb1, above-mentioned Second difference is equal to above-mentioned Xa1+Xb1.
S303, height and width, first difference and second difference according to the minimum circumscribed rectangle, Determine whole departure degree of the testing result relative to the annotation results.
Illustratively, the embodiment of the present application is set, according to above-mentioned first difference Ya1+Yb1 and above-mentioned minimum circumscribed rectangle The ratio of height YAnd the ratio of above-mentioned second difference Xa1+Xb1 and the width X of above-mentioned minimum circumscribed rectangleThe sum of, the relationship with above-mentioned testing result b relative to the whole departure degree negative correlation of annotation results a, in determination State whole departure degree of the testing result b relative to annotation results a.
Specifically, formula is calculated as follows in the embodiment of the present application, determine above-mentioned testing result b relative to annotation results a Whole departure degree d:
The value range of the d acquired according to above-mentioned calculation formula is 0~1.D value is smaller, closer to 0, then testing result b Whole departure degree relative to annotation results a is bigger；D value is bigger, and closer to 1, then testing result b is relative to annotation results The whole departure degree of a is smaller.
It is appreciated that the testing result obtained according to the above-mentioned calculating process of the embodiment of the present application is relative to the whole of annotation results Body departure degree, can be directly as score of the testing result in the whole departure degree dimension relative to annotation results, should Score is higher, i.e., the above-mentioned d value being calculated is bigger, then illustrates that testing result is got over relative to the whole departure degree of annotation results Small, i.e., testing result is more accurate；The score is lower, i.e., the above-mentioned d value being calculated is smaller, then illustrates testing result relative to mark The whole departure degree for infusing result is bigger, i.e., testing result is more inaccurate.
It as a kind of optional implementation, is disclosed in another embodiment of the application, the determination detection knot Detection integrity degree of the fruit to the target, comprising:
By detecting whether the testing result includes completely the target, determine the testing result to the target Detect integrity degree.
Illustratively, if above-mentioned testing result completely includes all the elements of the target, the detection can be determined As a result it completely detected the target, the testing result should be higher to the detection integrity degree of the target at this time；And such as The above-mentioned testing result of fruit does not completely include all the elements of the target, then can determine that the testing result does not detect completely The target out, the testing result should be lower to the detection integrity degree of the target at this time.Therefore, by detecting above-mentioned detection As a result whether complete packet contains the target, is determined for the testing result to the detection integrity degree of the target.
As a kind of optional implementation, another embodiment of the application discloses described by detecting the testing result Whether include completely the target, determine the testing result to the detection integrity degree of the target, comprising:
Whether the starting content and/or end content for determining the testing result are preset content；
If it is preset content, it is determined that the testing result does not completely include the target, and then determines the inspection Surveying result is the first integrity degree to the detection integrity degree of the target；
If not preset content, it is determined that the testing result completely includes the target, and then determines the detection It as a result is the second integrity degree to the detection integrity degree of the target；Wherein, first integrity degree is less than second integrity degree.
Wherein, the preset content be in the target fields cannot function as the starting content of content entirety with/ Or the content element of end content.
Specifically, for the content element of above-mentioned preset content fields, when it is used for component content entirety, Some content elements are the starting content and/or end content that cannot function as content entirety.Such as when number and mathematical operation When symbol forms mathematical operation formula shown in Fig. 2, add, subtract, the oeprators such as multiplication and division cannot function as mathematical operation formula Starting content and end content；For another example for content of text, starting content cannot be comma, fullstop, exclamation Number equal punctuation marks.
The embodiment of the present application utilizes rule when above-mentioned content element constitution content entirety, to judge that testing result has not Have completely comprising target.It is appreciated that if the starting content and/or end content of testing result are its fields, no That testing result is included can be then determined as the starting content of content entirety and/or the content element of end content Beginning content and/or end content are not content whole really starting content and/or end content, and then can determine that its is right The detection of entire content is incomplete.On the contrary, if the starting content and/or end content of testing result are not belonging to it Field, it cannot function as the starting content of content entirety and/or the content element of end content, it may be considered that testing result is complete Whole includes target.
Such as shown in Fig. 5, when above-mentioned goal-selling is mathematical operation formula " 51 ÷ 34=51 ÷, 17 ÷ 2 ", if inspection Surveying result is region shown in rectangle frame shown in fig. 5, then can determine that the starting content of testing result is " ÷ ".And due to " ÷ " is the starting content that cannot function as mathematical operation formula, thus may determine that testing result does not completely include target " 51 ÷ 34=51 ÷, 17 ÷ 2 ".
Based on above-mentioned theory, after getting the testing result of corresponding above-mentioned goal-selling, the embodiment of the present application plays it Beginning content and/or end content are identified judge whether its starting content and/or end content are preset content, that is, are judged Whether be the starting content and/or end content that cannot function as content entirety in the target fields content element.
If the starting content and/or end content of testing result are above-mentioned preset content, the detection knot can be determined Fruit does not completely include above-mentioned goal-selling, and then can determine that testing result is the to the detection integrity degree of above-mentioned goal-selling One integrity degree, for example, 0.
If the starting content and/or end content of testing result are not above-mentioned preset contents, the detection can be determined As a result completely include above-mentioned goal-selling, and then can determine that the testing result is the to the detection integrity degree of above-mentioned preset content Two integrity degrees, for example, 1.
It is shown in Figure 6 as another optional implementation, it is also disclosed in another embodiment of the application, it is described By detecting whether the testing result includes completely the target, determine that the testing result is complete to the detection of the target Degree, comprising:
S601, the testing result is aligned with the annotation results, determines the testing result and the annotation results Difference section.
Illustratively, shown in Figure 7, it is assumed that the annotation results to goal-selling " 51 ÷ 34=51 ÷, 17 ÷ 2 " are void Region shown in wire frame, the testing result to the goal-selling are region shown in solid box.Then, the embodiment of the present application is by above-mentioned inspection Result and annotation results alignment are surveyed, it illustratively, can be corresponding by will test the same coordinate position of result and annotation results It is overlapped, realizes the alignment of testing result and annotation results.
For example, will test the coordinate pair on four vertex of four vertex of the rectangle frame of result and the rectangle frame of annotation results It should be labeled into the same coordinate system, then the same coordinate point in two rectangle frames can be aligned overlapping, so that reaching will examine Survey the purpose of result and annotation results alignment.
After it will test result and annotation results alignment, the pixel grey scale of the two corresponding part is compared, determines two The same section and difference section of person.Illustratively, the embodiment of the present application determines testing result and mark knot by setting threshold value The same section and difference section of fruit.For the testing result and annotation results after alignment, if the pixel grey scale of corresponding part Difference be less than the threshold value, it may be considered that the part is identical, if the difference of the pixel grey scale of corresponding part is greater than the threshold value, It may be considered that the part is different, after successively determining the testing result part different with annotation results, that is, detection knot can determine The difference section of fruit and annotation results.Such as dash area shown in Fig. 7, as testing result and dotted line frame shown in solid box Shown in annotation results difference section.
S602, judge the difference section whether include the target content.
After determining the difference section of testing result and annotation results, the embodiment of the present application further judges in difference section Whether include above-mentioned target content.
The implementation of property as an example, when the content for determining that content that testing result detects only has above-mentioned target When, it can determine whether by the way that whether the above-mentioned difference section of detection includes the pixel that gray value is 0 comprising in target Hold.
Shown in Figure 7, for difference section shown in dash area, detection is 0 with the presence or absence of gray value in the part Pixel.For character position, since it shows as black trace, the pixel gray value of character position is 0, if In above-mentioned difference section there are the pixel that gray value is 0, then it can determine that there are characters in the difference section, that is, exist The content of target；If above-mentioned difference section can be determined and not deposited in the difference section there is no the pixel that gray value is 0 In character, the content of target is also just not present.
If the difference section includes the content of the target, thens follow the steps S603, determines that the testing result does not have Have completely comprising the target, and then determines that the testing result is third integrity degree to the detection integrity degree of the target.
If the difference section does not include the content of the target, thens follow the steps S604, determines the testing result Completely include the target, and then determines that the testing result is the 4th integrity degree to the detection integrity degree of the target.
Wherein, the third integrity degree is less than the 4th integrity degree.
It is appreciated that if above-mentioned difference section there are the contents of target, that is to say, that there are also partial target contents to be in The difference section of testing result and annotation results can then determine that above-mentioned testing result, at this time can be with not completely comprising target Determine that above-mentioned testing result is third integrity degree to the detection integrity degree of the target, for example, 0.
If the content of target is not present in above-mentioned difference section, it may be considered that above-mentioned testing result completely includes mesh Mark can determine that above-mentioned testing result is the 4th integrity degree to the detection integrity degree of the target at this time, for example, 1.
As another optional implementation, another embodiment of the application also discloses above-mentioned by detecting the inspection Surveying result, whether complete packet contains the target, determines that the testing result realizes the another of detection integrity degree of the target Mode, shown in Figure 8, which specifically includes:
S801, the content for determining the target.
The content of above-mentioned target, refers to the particular content that target is included, and the content of target as shown in Figure 9 is mathematics The number and mathematical operator number that operational formula " 51 ÷ 34=51 ÷, 17 ÷ 2 " is included.
Illustratively, by carrying out content recognition to above-mentioned target, the content of above-mentioned target can be determined.
Then, the embodiment of the present application is by detection in the outer boundary of the testing result of above-mentioned target and above-mentioned target The overlapping cases of appearance determine the testing result to the detection integrity degree of the target.
It is appreciated that as shown in figure 9, if testing result and target in have Chong Die, can determine testing result It completely cannot include all the elements of target, and then can determine that testing result does not detect target completely；If detection knot Fruit and the content of target be not be overlapped, it may be considered that testing result completely includes all the elements of target, and then can consider Testing result completely detects target.Therefore, pass through the weight of the content of the outer boundary and above-mentioned target of the above-mentioned testing result of detection Folded situation, can determine testing result to the detection integrity degree of target.
Illustratively, above by the overlapping cases of the content for the outer boundary and above-mentioned target for detecting above-mentioned testing result, Determine that the testing result includes: to the concrete processing procedure of the detection integrity degree of the target
S802, that whether the outer boundary for determining the testing result has with the content of the target is overlapping.
Illustratively, pass through the position of the content of the position coordinates and above-mentioned target of the outer boundary of the above-mentioned testing result of detection Whether coordinate has overlapping, and it is overlapping can to determine whether the outer boundary of testing result has with the content of target.
As another optional implementation, when the content for determining that content that testing result is included only has target, or Person is when the outer boundary for determining testing result can not overlap with the content in addition to the content of above-mentioned target, by inspection The figure image of cutting for surveying results area carries out binary conversion treatment, and the content of the outer boundary and above-mentioned target that can determine testing result is It is no to have overlapping.
For example, the figure image of cutting in testing result region shown in Fig. 9 is carried out binary conversion treatment, for the outer of testing result The not no part Chong Die with object content in boundary, the pixel gray value of binaryzation are 255, and the outer boundary of testing result with The part of object content overlapping, the pixel gray value of binaryzation are 0.Then, will test results area cuts figure image Binarization result carries out image expression, and the pixel for being 255 to its outer boundary gray value is indicated with white, to its outer boundary gray scale The pixel that value is 0 indicates with black, then available expression result as shown in Figure 10.Knot is detected as shown in Fig. 9 The right side outer boundary of fruit has occurred be overlapped with object content " 2 ", therefore cuts figure image in testing result region shown in Fig. 10 Binarization result in, Right Border there are three sections of blank parts, the blank parts be testing result right side outer boundary with The lap of object content " 2 ".
If be not overlapped, thens follow the steps S803, determines the testing result and be to the detection integrity degree of the target 5th integrity degree.
It is appreciated that if the outer boundary of testing result and the content of target be not be overlapped, it may be considered that testing result The content intact of target is included in inside testing result region, i.e., completely includes object content, at this time it is considered that detection knot Fruit is the 5th integrity degree to the detection integrity degree of target, such as can be set to 1.
If there is overlapping, the then position according to lap in the outer boundary of the testing result and lap picture The total pixel quantity of the outer boundary of vegetarian refreshments quantity and the testing result, be calculated the outer boundary of the testing result with it is described The overlapping degree of the content of target, as the testing result to the detection integrity degree of the target.
It is appreciated that if the outer boundary of testing result and target in have Chong Die, can determine testing result not It can completely include the content of target.In this case, if the lap of the content of the outer boundary and target of testing result It is more, then illustrate that the content for the target that testing result includes is fewer, can determine that testing result gets over the detection integrity degree of target It is low；If the lap of the content of the outer boundary and target of testing result is fewer, illustrate the target that testing result includes Content is more, can determine that testing result is higher to the detection integrity degree of target.
That is, in the case where having overlapping in the outer boundary and target for determining testing result, testing result The overlapping degree of the content of outer boundary and target, can directly reaction detection result to the detection integrity degree of target.And detect knot The overlapping degree of the content of the outer boundary and target of fruit, by lap in the position of the outer boundary of testing result, Yi Jichong The factors such as the pixel quantity of folded part influence.Therefore, the embodiment of the present application is in the outer boundary and mesh for determining above-mentioned testing result In the case where having overlapping in target, further according to lap in the position and overlapping portion of the outer boundary of testing result The total pixel quantity of outer boundary of point pixel quantity and testing result, be calculated testing result outer boundary and target it is interior The overlapping degree of appearance, as testing result to the detection integrity degree of target.
Illustratively, determined in S802 through the above steps have in the outer boundary and target of testing result it is overlapping after, The treatment process of the overlapping degree of the content of the outer boundary and target of above-mentioned determining testing result specifically includes:
S804, according to lap in the position of the outer boundary of the testing result, determine that overlapping degree calculates the factor；Its In, lap identified overlapping degree when on the longer outer boundary of the testing result calculates the factor, is less than overlapping portion Identified overlapping degree when on the shorter outer boundary of the testing result is divided to calculate the factor.
Above-mentioned overlapping degree calculates the factor, be can reflect the object content that testing result is included account for target it is all in The factor of the scale of appearance.The shorter outer boundary weight shown in Figure 9 for being appreciated that the content when target and testing result When folding, such as overlapping with right side outer boundary, testing result is affected only to the detection integrity degree of single character, is detected at this time As a result the object content ratio for being included is relatively large；And if the content of target and the longer outer boundary weight of testing result It is folded, such as overlap with top outer boundary, then influence whether testing result to the detection integrity degree of characters all in target, this When the testing result object content ratio that is included it is relatively small.
For above situation, the embodiment of the present application, in the position of the outer boundary of testing result, determines weight according to lap Folded degree calculates the factor.
Illustratively, firstly, the embodiment of the present application first carries out really lap in the position of the outer boundary of testing result It is fixed, such as shown in Fig. 9, determine that lap is in the lateral outer boundary of testing result or in longitudinal outer boundary.
When lateral outer boundary of the lap in testing result, which is all neglected Slightly, weight is obtained than upper single transverse and longitudinal outer boundary all pixels point quantity X+Y using the pixel quantity Y of single longitudinal outer boundary Folded degree calculates the factorSimilarly, when longitudinal outer boundary when lap in testing result, by this, there are laps Longitudinal outer boundary all ignore, using the pixel quantity X of single lateral outer boundary, pictures more all than upper single transverse and longitudinal outer boundary Vegetarian refreshments quantity X+Y obtains overlapping degree and calculates the factor
It is appreciated that according to the above embodiments of the present application introduce determined by overlapping degree calculate the size of the factor, and again The outer boundary length of the testing result where folded part is inversely proportional, longer outer boundary of the lap in the testing result Identified overlapping degree calculates the factor when upper, is determined when on the shorter outer boundary of the testing result less than lap Overlapping degree calculate the factor.
S805, the total pixel quantity of outer boundary according to lap pixel quantity and the testing result, calculate Into the outer boundary pixel of the testing result, the not pixel ratio Chong Die with the content of the target.
Illustratively, calculation formula is utilizedAbove-mentioned detection knot can be calculated In the outer boundary pixel of fruit, pixel ratio K not Chong Die with the content of target.
S806, the overlapping degree is calculated in the outer boundary pixel of the factor and the testing result, not with the target Content overlapping pixel ratio be multiplied, the outer boundary of the testing result and the overlapping journey of the content of the target is calculated Degree.
The case where lateral outer boundary of testing result is located at for lap, the outer boundary of testing result and target it is interior The overlapping degree of appearance is
The case where longitudinal outer boundary of testing result is located at for lap, the outer boundary of testing result and target it is interior The overlapping degree of appearance is
The overlapping degree of the content of the outer boundary and target for the testing result being calculated according to the method described above, can be used as Detection integrity degree of the testing result to target.
For example, with reference to shown in Fig. 9, it is assumed that the outer boundary of testing result shares 600 pixels, one of laterally outside Boundary X includes pixel 220, and a longitudinal direction outer boundary Y includes pixel 80, and assumes the outer boundary and mesh of testing result It marks content and is overlapped 30 pixels.Then when 30 pixels are located at the lateral outer boundary of testing result, testing result at this time Outer boundary and the overlapping degree of content of target beWhen 30 pixels are located at inspection When surveying longitudinal outer boundary of result, the overlapping degree of the content of the outer boundary and target of testing result is at this time
It is appreciated that the embodiment of the present application fully takes into account the different outer boundary positions that above-mentioned lap is located at testing result Be testing result to the detection integrity degree bring Different Effects of target when setting, by be arranged above-mentioned overlapping degree calculate because Son is enabled to when the longer outer boundary of testing result is Chong Die with object content, testing result to the detection integrity degree of target, Lower than when the shorter outer boundary of testing result is Chong Die with object content, detection integrity degree of the testing result to target.Namely It says, the above-mentioned overlapping degree of the embodiment of the present application setting calculates the factor, can objectively embody the different outer boundary of testing result It is that testing result influences the detection integrity degree bring of target when overlapping with object content.
It should be noted that above-mentioned each embodiment describes inspection of a variety of different determining testing results to target respectively The specific implementation for surveying integrity degree can be from above-mentioned various realization sides when the embodiment of the present application technical solution is embodied In formula, the mode of choosing any one kind of them determines that testing result is complete to the detection of target.Further, above-mentioned a variety of different determining inspections Surveying result can also be in conjunction with execution to the implementation of the detection integrity degree of target, it can will be in above-mentioned various implementations It is arbitrarily a variety of to be used in combination, it is provided commonly for determining testing result to the detection integrity degree of target.
As a kind of optional implementation, disclosed in another embodiment of the application, when according to above-mentioned testing result Relative to the whole departure degree of annotation results and above-mentioned testing result to the detection integrity degree of target, determine to above-mentioned detection When evaluation result as a result, specifically include:
Detection to target of whole departure degree and above-mentioned testing result by above-mentioned testing result relative to annotation results Integrity degree inputs preset testing result evaluation and test model, determines the evaluation result to above-mentioned testing result.
Wherein, above-mentioned testing result evaluation and test model is at least through according to testing result relative to the whole departure of annotation results Degree and testing result carry out evaluation and test training to testing result and obtain to the detection integrity degree of target.
Specifically, the embodiment of the present application, which passes through training testing result, evaluates and tests model, the evaluation and test to testing result is realized.The inspection It is complete to the detection of target relative to the whole departure degree and testing result of annotation results with testing result to survey result evaluation and test model Whole degree is input, realizes the testing result evaluation and test of automation.Wherein, above-mentioned testing result is to the detection integrity degree of target, can be with It is that any one or more testing result disclosed in the various embodiments described above determines the detection integrity degree method of determination of target Result.
As a kind of optional implementation, above-mentioned testing result evaluation and test model can be set to disaggregated model, be used for The detection integrity degree of target ties evaluation and test relative to the whole departure degree of annotation results and testing result according to testing result Fruit is classified, and determines evaluation result classification, for example, evaluation result it is excellent, it is good, can, difference etc..
At this point, including: that will test result relative to annotation results to the training process of above-mentioned testing result evaluation and test model Whole departure degree sample and testing result input testing result evaluation and test model to the detection integrity degree sample of target simultaneously, make The model carries out classification learning to evaluation result, until the model can accurately classify to evaluation result.The mould at this time Type can be used to according to any testing result relative to the whole departure degree of annotation results and the testing result to target Integrity degree is detected, is classified to the testing result, so that it is determined that the evaluation result of the testing result.
As another optional implementation, above-mentioned testing result evaluation and test model can be set to scoring and calculate mould Type, the model can be complete to the detection of target relative to the whole departure degree and testing result of annotation results to testing result Weight is respectively set in degree, is then based on whole departure degree of the weight to testing result relative to annotation results, and detection As a result summation process is weighted to the detection integrity degree of target, obtained score value is as the evaluation result to testing result.
In the training process to above-mentioned scoring computation model, will be marked the testing result sample that really scores relative to The whole departure degree and testing result sample of annotation results input the scoring to the detection integrity degree of target simultaneously and calculate mould Type makes the scoring of model sample according to testing result, determines whole departure degree of the testing result relative to annotation results Weight and testing result to the weight of the detection integrity degree of target, enable the scoring computation model according to testing result The weight and testing result of whole departure degree relative to annotation results to target detection integrity degree weight, accurately The scoring of testing result is calculated in ground.After completing above-mentioned training, which can determine according to by training Testing result relative to annotation results whole departure degree weight and testing result to the detection integrity degree of target Weight, it is complete to the detection of target with it relative to the whole departure degree of annotation results according to it for arbitrary testing result Whole degree is calculated its scoring, that is, determines the evaluation result to testing result.
It as a kind of optional implementation, is also disclosed in another embodiment of the application, when above-mentioned corresponding to default The testing result in region locating for target and annotation results corresponding to region locating for the target, for corresponding to preset multiple Each testing result in region locating for target and each annotation results corresponding to region locating for the multiple target, and on When stating each testing result and being the testing result that identical algorithm of target detection detects different target respectively, needle To above-mentioned each testing result, the evaluating method of the object detection results by implementing the above embodiments of the present application introduction can be with The evaluation result to above-mentioned each testing result is determined respectively.
After determining to the evaluation result of each testing result, the evaluation and test side for the object detection results that the application proposes Method further include:
According to the evaluation result to each testing result, determine to obtaining the target detection of each testing result The evaluation result of algorithm.
It illustratively, is that identical algorithm of target detection carries out different targets due to above-mentioned each testing result Detect obtained testing result, therefore, after obtaining to the evaluation result of each testing result, in conjunction with each evaluation result pair The target annotation results answered are concluded the evaluation result of above-mentioned each testing result, are analyzed, be may further determine that this The evaluation result of algorithm of target detection.For example, by each testing result obtained by above-mentioned algorithm of target detection, and it is corresponding Target annotation results, determine the algorithm of target detection to the recall ratio and precision ratio of target detection, and then can draw out as figure The PR curve that the algorithm of target detection is corresponded to shown in 11 determines the evaluation result to the algorithm of target detection by the curve.
Corresponding with the evaluating method of above-mentioned object detection results, the embodiment of the present application also discloses a kind of target detection As a result evaluating apparatus, shown in Figure 12, which includes:
Data capture unit 100, for obtaining testing result and correspondence corresponding to region locating for goal-selling respectively The annotation results in the region locating for the target；
Parameter determination unit 110, for determining whole departure degree of the testing result relative to the annotation results, And/or determine the testing result to the detection integrity degree of the target；
Evaluation process unit 120, for the whole departure degree according to the testing result relative to the annotation results, And/or the testing result determines the evaluation result to the testing result to the detection integrity degree of the target.
The evaluating apparatus for the object detection results that the embodiment of the present application proposes can be determined by parameter determination unit 110 Testing result relative to annotation results whole departure degree and/or testing result to the detection integrity degree of target, then by commenting Survey the inspection relative to the whole departure degree and/or testing result of annotation results to target according to testing result of processing unit 120 Integrity degree is surveyed, determines the evaluation result to testing result.The device can be realized the evaluation and test of the automation to object detection results, from And determine the superiority and inferiority of testing result.
Further, technical scheme can combine the evaluation and test of multiple dimensions to testing result, realize from Multiple dimensions evaluate and test testing result, so that more objective to the evaluation and test of testing result, more comprehensively.
Optionally, it is proposed in another embodiment of the application, the parameter determination unit 110 determines the detection knot When whole departure degree of the fruit relative to the annotation results, it is specifically used for:
According to central point and the outside of the position and the annotation results of the central point of the testing result and outer boundary The position on boundary determines whole departure degree of the testing result relative to the annotation results.
As a kind of optional implementation, the parameter determination unit 110, comprising:
First parameter determination unit, for according to the outer of the position of the outer boundary of the testing result and the annotation results The position on boundary determines the height and width of the minimum circumscribed rectangle comprising the testing result and the annotation results；
Second parameter determination unit, in the position and the annotation results according to the central point of the testing result The position of heart point, determines the height of the minimum circumscribed rectangle, with the central point of the testing result and the annotation results Difference between the vertical range of central point as the first difference, and determines the width of the minimum circumscribed rectangle, and described Difference between the central point of testing result and the horizontal distance of the central point of the annotation results, as the second difference；
Calculation processing unit, for height and width, first difference according to the minimum circumscribed rectangle, Yi Jisuo The second difference is stated, determines whole departure degree of the testing result relative to the annotation results.
As a kind of optional implementation, the calculation processing unit is according to the height and width of the minimum circumscribed rectangle Degree, first difference and second difference, determine whole departure of the testing result relative to the annotation results When degree, it is specifically used for:
According to first difference and the height of the minimum circumscribed rectangle ratio and second difference with it is described The sum of ratio of width of minimum circumscribed rectangle, it is negative relative to the whole departure degree of the annotation results with the testing result Relevant relationship determines whole departure degree of the testing result relative to the annotation results.
Optionally, it is proposed in another embodiment of the application, the parameter determination unit 110 determines the detection knot When fruit is to the detection integrity degree of the target, it is specifically used for:
By detecting whether the testing result includes completely the target, determine the testing result to the target Detect integrity degree.
As a kind of optional implementation, whether the parameter determination unit 110 is complete by detecting the testing result It is whole to be specifically used for when determining detection integrity degree of the testing result to the target comprising the target:
Whether the starting content and/or end content for determining the testing result are preset content；Wherein, described default interior Hold is that cannot function as the starting content of content entirety and/or the content element of end content in the target fields；
If it is preset content, it is determined that the testing result does not completely include the target, and then determines the inspection Surveying result is the first integrity degree to the detection integrity degree of the target；
If not preset content, it is determined that the testing result completely includes the target, and then determines the detection It as a result is the second integrity degree to the detection integrity degree of the target；Wherein, first integrity degree is less than second integrity degree.
As a kind of optional implementation, whether the parameter determination unit 110 is complete by detecting the testing result It is whole to be specifically used for when determining detection integrity degree of the testing result to the target comprising the target:
The testing result is aligned with the annotation results, determines the difference of the testing result Yu the annotation results Part；
Judge the difference section whether include the target content；
If the difference section includes the content of the target, it is determined that the testing result is not completely comprising described Target, and then determine that the testing result is third integrity degree to the detection integrity degree of the target；
If the difference section does not include the content of the target, it is determined that the testing result completely includes the mesh Mark, and then determine that the testing result is the 4th integrity degree to the detection integrity degree of the target；
Wherein, the third integrity degree is less than the 4th integrity degree.
As a kind of optional implementation, whether the parameter determination unit 110 is complete by detecting the testing result It is whole to be specifically used for when determining detection integrity degree of the testing result to the target comprising the target:
Determine the content of the target；
By detecting the overlapping cases of the outer boundary of the testing result and the content of the target, the detection knot is determined Detection integrity degree of the fruit to the target.
As a kind of optional implementation, the outer boundary by the detection testing result is interior with the target The overlapping cases of appearance determine the testing result to the detection integrity degree of the target, comprising:
It is overlapping to determine whether the outer boundary of the testing result and the content of the target have；
If be not overlapped, it is determined that the testing result is the 5th integrity degree to the detection integrity degree of the target；
If there is overlapping, the then position according to lap in the outer boundary of the testing result and lap picture The total pixel quantity of the outer boundary of vegetarian refreshments quantity and the testing result, be calculated the outer boundary of the testing result with it is described The overlapping degree of the content of target, as the testing result to the detection integrity degree of the target.
As a kind of optional implementation, the position of the outer boundary according to lap in the testing result, And the total pixel quantity of outer boundary of lap pixel quantity and the testing result, the testing result is calculated Outer boundary and the target content overlapping degree, comprising:
According to lap in the position of the outer boundary of the testing result, determine that overlapping degree calculates the factor；Wherein, weight Folded part identified overlapping degree when on the longer outer boundary of the testing result calculates the factor, is less than lap in institute Identified overlapping degree calculates the factor when stating on the shorter outer boundary of testing result；
According to the total pixel quantity of the outer boundary of lap pixel quantity and the testing result, it is calculated described In the outer boundary pixel of testing result, the not pixel ratio Chong Die with the content of the target；
In the outer boundary pixel that the overlapping degree is calculated to the factor and the testing result, not with the content of the target The pixel ratio of overlapping is multiplied, and the overlapping degree of the outer boundary of the testing result and the content of the target is calculated.
Optionally, it is proposed in another embodiment of the application, the evaluation process unit 120 is tied according to the detection Fruit, to the detection integrity degree of the target, determines relative to the whole departure degree of the annotation results and the testing result When to the evaluation result of the testing result, it is specifically used for:
By the testing result relative to the whole departure degree of the annotation results and the testing result to the mesh Target detects integrity degree, inputs preset testing result evaluation and test model, determines the evaluation result to the testing result；
Wherein, the testing result evaluation and test model is at least through according to testing result relative to the whole departure of annotation results Degree and testing result carry out evaluation and test training to testing result and obtain to the detection integrity degree of target.
Optionally, it is proposed in another embodiment of the application, when the inspection corresponding to region locating for goal-selling Result and the annotation results corresponding to region locating for the target are surveyed, for corresponding to each of region locating for preset multiple targets A testing result and corresponding to region locating for the multiple target each annotation results when, described device further include:
Algorithm evaluates and tests unit, for the evaluation result according to the evaluation process unit to each testing result, really The fixed evaluation result to the algorithm of target detection for obtaining each testing result.
Another embodiment of the application also discloses a kind of evaluation and test equipment of object detection results, and shown in Figure 13, this sets It is standby to include:
Memory 200 and processor 210；
Wherein, the memory 200 is connect with the processor 210, for storing program；
The processor 210 realizes following functions for the program by storing in the operation memory 200:
It obtains respectively and corresponds to the testing result in region locating for goal-selling and corresponding to region locating for the target Annotation results；It determines whole departure degree of the testing result relative to the annotation results, and/or determines the detection knot Detection integrity degree of the fruit to the target；Whole departure degree according to the testing result relative to the annotation results, and/ Or the testing result determines the evaluation result to the testing result to the detection integrity degree of the target.
Specifically, the evaluation and test equipment of above-mentioned object detection results can also include: bus, communication interface 220, input equipment 230 and output equipment 240.
Processor 210, memory 200, communication interface 220, input equipment 230 and output equipment 240 are mutual by bus Connection.Wherein:
Bus may include an access, transmit information between computer system all parts.
Processor 210 can be general processor, such as general central processor (CPU), microprocessor etc., be also possible to Application-specific integrated circuit (application-specific integrated circuit, ASIC), or one or more use In the integrated circuit that control the present invention program program executes.It can also be digital signal processor (DSP), specific integrated circuit (ASIC), ready-made programmable gate array (FPGA) either other programmable logic device, discrete gate or transistor logic, Discrete hardware components.
Processor 210 may include primary processor, may also include baseband chip, modem etc..
The program for executing technical solution of the present invention is preserved in memory 200, can also preserve operating system and other Key business.Specifically, program may include program code, and program code includes computer operation instruction.More specifically, it stores Device 200 may include read-only memory (read-only memory, ROM), the other types that can store static information and instruction Static storage device, random access memory (random access memory, RAM), can store information and instruction its The dynamic memory of his type, magnetic disk storage, flash etc..
Input equipment 230 may include receive user input data and information device, such as keyboard, mouse, camera, Scanner, light pen, speech input device, touch screen, pedometer or gravity sensor etc..
Output equipment 240 may include allowing output information to the device, such as display screen, printer, loudspeaker etc. of user.
Communication interface 220 may include using the device of any transceiver one kind, so as to logical with other equipment or communication network Letter, such as Ethernet, wireless access network (RAN), WLAN (WLAN) etc..
Processor 2102 executes the program stored in memory 200, and calls other equipment, can be used for realizing this Shen Please object detection results provided by embodiment evaluating method each step.
Another embodiment of the application additionally provides a kind of storage medium, is stored with computer program on the storage medium, should When computer program is executed by processor, each of the evaluating method for the object detection results that any of the above-described embodiment provides is realized Step.
For the various method embodiments described above, for simple description, therefore, it is stated as a series of action combinations, but Be those skilled in the art should understand that, the application is not limited by the described action sequence because according to the application, certain A little steps can be performed in other orders or simultaneously.Secondly, those skilled in the art should also know that, it is retouched in specification The embodiment stated belongs to preferred embodiment, necessary to related actions and modules not necessarily the application.
It should be noted that all the embodiments in this specification are described in a progressive manner, each embodiment weight Point explanation is the difference from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For device class embodiment, since it is basically similar to the method embodiment, so being described relatively simple, related place ginseng See the part explanation of embodiment of the method.
Step in each embodiment method of the application can be sequentially adjusted, merged and deleted according to actual needs.
Each embodiment kind device of the application and module in terminal and submodule can be merged according to actual needs, be drawn Divide and deletes.
In several embodiments provided herein, it should be understood that disclosed terminal, device and method, Ke Yitong Other modes are crossed to realize.For example, terminal embodiment described above is only schematical, for example, module or submodule Division, only a kind of logical function partition, there may be another division manner in actual implementation, for example, multiple submodule or Module may be combined or can be integrated into another module, or some features can be ignored or not executed.Another point is shown The mutual coupling, direct-coupling or communication connection shown or discussed can be through some interfaces, between device or module Coupling or communication connection are connect, can be electrical property, mechanical or other forms.
Module or submodule may or may not be physically separated as illustrated by the separation member, as mould The component of block or submodule may or may not be physical module or submodule, it can and it is in one place, or It may be distributed on multiple network modules or submodule.Some or all of mould therein can be selected according to the actual needs Block or submodule achieve the purpose of the solution of this embodiment.
In addition, each functional module or submodule in each embodiment of the application can integrate in a processing module In, it is also possible to modules or submodule physically exists alone, it can also be integrated with two or more modules or submodule In a module.Above-mentioned integrated module or submodule both can take the form of hardware realization, can also use software function Energy module or the form of submodule are realized.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered Think beyond scope of the present application.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor The combination of capable software unit or the two is implemented.Software unit can be placed in random access memory (RAM), memory, read-only deposit Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology In any other form of storage medium well known in field.
Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged Except there is also other identical elements in the process, method, article or apparatus that includes the element.
The foregoing description of the disclosed embodiments makes professional and technical personnel in the field can be realized or use the application. Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein General Principle can be realized in other embodiments without departing from the spirit or scope of the application.Therefore, the application It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one The widest scope of cause.

Claims (15)

1. a kind of evaluating method of object detection results characterized by comprising
The testing result for corresponding to region locating for goal-selling and the mark corresponding to region locating for the target are obtained respectively As a result；
It determines whole departure degree of the testing result relative to the annotation results, and/or determines the testing result pair The detection integrity degree of the target；
According to the testing result relative to the whole departure degree of the annotation results and/or the testing result to described The detection integrity degree of target determines the evaluation result to the testing result.
2. the method according to claim 1, wherein the determination testing result is tied relative to the mark The whole departure degree of fruit, comprising:
According to the central point of the position and the annotation results of the central point of the testing result and outer boundary and outer boundary Position determines whole departure degree of the testing result relative to the annotation results.
3. according to the method described in claim 2, it is characterized in that, the central point and outer boundary according to the testing result Position and the annotation results central point and outer boundary position, determine the testing result relative to the mark As a result whole departure degree, comprising:
According to the position of the position of the outer boundary of the testing result and the outer boundary of the annotation results, determine to include the inspection Survey the height and width of the minimum circumscribed rectangle of result and the annotation results；
According to the position of the position of the central point of the testing result and the central point of the annotation results, determine described minimum outer Difference between the height for connecing rectangle, and the vertical range of the central point of the central point and annotation results of the testing result Value as the first difference, and determines the width of the minimum circumscribed rectangle, central point and the mark with the testing result The difference between the horizontal distance of the central point of result is infused, as the second difference；
According to the height of the minimum circumscribed rectangle and width, first difference and second difference, the inspection is determined Survey whole departure degree of the result relative to the annotation results.
4. according to the method described in claim 3, it is characterized in that, the height and width according to the minimum circumscribed rectangle Degree, first difference and second difference, determine whole departure of the testing result relative to the annotation results Degree, comprising:
Ratio and second difference and the minimum according to first difference and the height of the minimum circumscribed rectangle The sum of ratio of width of boundary rectangle, it is negatively correlated relative to the whole departure degree of the annotation results with the testing result Relationship, determine whole departure degree of the testing result relative to the annotation results.
5. the method according to claim 1, wherein detection of the determination testing result to the target Integrity degree, comprising:
By detecting whether the testing result includes completely the target, detection of the testing result to the target is determined Integrity degree.
6. according to the method described in claim 5, it is characterized in that, it is described by detect the testing result whether completely include The target determines the testing result to the detection integrity degree of the target, comprising:
Whether the starting content and/or end content for determining the testing result are preset content；Wherein, the preset content is In the target fields, the starting content of content entirety and/or the content element of end content cannot function as；
If it is preset content, it is determined that the testing result does not completely include the target, and then determines the detection knot Fruit is the first integrity degree to the detection integrity degree of the target；
If not preset content, it is determined that the testing result completely includes the target, and then determines the testing result Detection integrity degree to the target is the second integrity degree；Wherein, first integrity degree is less than second integrity degree.
7. according to the method described in claim 5, it is characterized in that, it is described by detect the testing result whether completely include The target determines the testing result to the detection integrity degree of the target, comprising:
The testing result is aligned with the annotation results, determines the difference portion of the testing result Yu the annotation results Point；
Judge the difference section whether include the target content；
If the difference section includes the content of the target, it is determined that the testing result does not completely include the mesh Mark, and then determine that the testing result is third integrity degree to the detection integrity degree of the target；
If the difference section does not include the content of the target, it is determined that the testing result completely includes the target, And then determine that the testing result is the 4th integrity degree to the detection integrity degree of the target；
Wherein, the third integrity degree is less than the 4th integrity degree.
8. according to the method described in claim 5, it is characterized in that, it is described by detect the testing result whether completely include The target determines the testing result to the detection integrity degree of the target, comprising:
Determine the content of the target；
By detecting the overlapping cases of the outer boundary of the testing result and the content of the target, the testing result pair is determined The detection integrity degree of the target.
9. according to the method described in claim 8, it is characterized in that, the outer boundary and institute by detecting the testing result The overlapping cases for stating the content of target determine the testing result to the detection integrity degree of the target, comprising:
It is overlapping to determine whether the outer boundary of the testing result and the content of the target have；
If be not overlapped, it is determined that the testing result is the 5th integrity degree to the detection integrity degree of the target；
If there is overlapping, the then position according to lap in the outer boundary of the testing result and lap pixel The total pixel quantity of the outer boundary of quantity and the testing result, be calculated the testing result outer boundary and the target Content overlapping degree, as the testing result to the detection integrity degree of the target.
10. according to the method described in claim 9, it is characterized in that, it is described according to lap in the outer of the testing result The total pixel quantity of outer boundary of the position and lap pixel quantity and the testing result on boundary, is calculated The overlapping degree of the content of the outer boundary of the testing result and the target, comprising:
According to lap in the position of the outer boundary of the testing result, determine that overlapping degree calculates the factor；Wherein, overlapping portion Divide identified overlapping degree when on the longer outer boundary of the testing result to calculate the factor, is less than lap in the inspection Identified overlapping degree calculates the factor when surveying on the shorter outer boundary of result；
According to the total pixel quantity of the outer boundary of lap pixel quantity and the testing result, the detection is calculated As a result in outer boundary pixel, the not pixel ratio Chong Die with the content of the target；
It is not Chong Die with the content of the target in the outer boundary pixel that the overlapping degree is calculated to the factor and the testing result Pixel ratio be multiplied, the overlapping degree of the outer boundary of the testing result and the content of the target is calculated.
11. according to claim 1 to method described in any claim in 10, which is characterized in that described according to the detection As a result relative to the whole departure degree of the annotation results and the testing result to the detection integrity degree of the target, really The fixed evaluation result to the testing result, comprising:
By the testing result relative to the whole departure degree of the annotation results and the testing result to the target Integrity degree is detected, preset testing result evaluation and test model is inputted, determines the evaluation result to the testing result；
Wherein, the testing result evaluation and test model is at least through according to testing result relative to the whole departure degree of annotation results With testing result to the detection integrity degree of target, evaluation and test training is carried out to testing result and is obtained.
12. according to claim 1 to method described in any claim in 10, which is characterized in that described to correspond to default mesh The testing result in locating region and the annotation results corresponding to region locating for the target are marked, for corresponding to preset multiple mesh Mark each testing result in locating region and each annotation results corresponding to region locating for the multiple target；
After determining to the evaluation result of each testing result, the method also includes:
According to the evaluation result to each testing result, determine to obtaining the algorithm of target detection of each testing result Evaluation result.
13. a kind of evaluating apparatus of object detection results characterized by comprising
Data capture unit, for obtaining the testing result corresponding to region locating for goal-selling respectively and corresponding to the mesh Mark the annotation results in locating region；
Parameter determination unit, for determining whole departure degree of the testing result relative to the annotation results, and/or really Detection integrity degree of the fixed testing result to the target；
Evaluation process unit, for according to the testing result relative to the annotation results whole departure degree and/or institute Testing result is stated to the detection integrity degree of the target, determines the evaluation result to the testing result.
14. a kind of evaluation and test equipment of object detection results characterized by comprising
Memory and processor；
Wherein, the memory is connected to the processor, for storing program；
The processor, for realizing following functions by the program in the operation memory:
The testing result for corresponding to region locating for goal-selling and the mark corresponding to region locating for the target are obtained respectively As a result；It determines whole departure degree of the testing result relative to the annotation results, and/or determines the testing result pair The detection integrity degree of the target；Whole departure degree and/or institute according to the testing result relative to the annotation results Testing result is stated to the detection integrity degree of the target, determines the evaluation result to the testing result.
15. a kind of storage medium, which is characterized in that be stored with computer program, the computer program on the storage medium When being executed by processor, the evaluating method of the object detection results as described in any claim in claim 1 to 12 is realized Each step.
CN201910328909.2A 2019-04-23 2019-04-23 Evaluation method, device, equipment and storage medium for target detection result Active CN110070533B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910328909.2A CN110070533B (en) 2019-04-23 2019-04-23 Evaluation method, device, equipment and storage medium for target detection result

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910328909.2A CN110070533B (en) 2019-04-23 2019-04-23 Evaluation method, device, equipment and storage medium for target detection result

Publications (2)

Publication Number Publication Date
CN110070533A true CN110070533A (en) 2019-07-30
CN110070533B CN110070533B (en) 2023-05-30

Family

ID=67368542

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910328909.2A Active CN110070533B (en) 2019-04-23 2019-04-23 Evaluation method, device, equipment and storage medium for target detection result

Country Status (1)

Country Link
CN (1) CN110070533B (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110598799A (en) * 2019-09-20 2019-12-20 科大讯飞股份有限公司 Target detection result evaluation method, device, equipment and storage medium
CN111160219A (en) * 2019-12-26 2020-05-15 深圳云天励飞技术有限公司 Object integrity evaluation method and device, electronic equipment and storage medium
CN111986254A (en) * 2020-08-21 2020-11-24 四川大学华西医院 Target area contour analysis method and device, storage medium and electronic equipment
CN112200217A (en) * 2020-09-09 2021-01-08 天津津航技术物理研究所 Identification algorithm evaluation method and system based on infrared image big data
CN112540352A (en) * 2019-09-20 2021-03-23 初速度（苏州）科技有限公司 Method and device for evaluating target detection algorithm based on unmanned vehicle
CN113435535A (en) * 2021-07-13 2021-09-24 北京达佳互联信息技术有限公司 Training method of image recognition model, image recognition method and device

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108764235A (en) * 2018-05-23 2018-11-06 中国民用航空总局第二研究所 Neural network model, object detection method, equipment and medium
WO2019000653A1 (en) * 2017-06-30 2019-01-03 清华大学深圳研究生院 Image target identification method and apparatus

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019000653A1 (en) * 2017-06-30 2019-01-03 清华大学深圳研究生院 Image target identification method and apparatus
CN108764235A (en) * 2018-05-23 2018-11-06 中国民用航空总局第二研究所 Neural network model, object detection method, equipment and medium

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
李荣龙等: "字符候选集筛选算法评测方法研究", 《信息技术》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110598799A (en) * 2019-09-20 2019-12-20 科大讯飞股份有限公司 Target detection result evaluation method, device, equipment and storage medium
CN112540352A (en) * 2019-09-20 2021-03-23 初速度（苏州）科技有限公司 Method and device for evaluating target detection algorithm based on unmanned vehicle
CN112540352B (en) * 2019-09-20 2022-09-27 魔门塔(苏州)科技有限公司 Method and device for evaluating target detection algorithm based on unmanned vehicle
CN110598799B (en) * 2019-09-20 2023-09-26 科大讯飞股份有限公司 Evaluation method, device, equipment and storage medium for target detection result
CN111160219A (en) * 2019-12-26 2020-05-15 深圳云天励飞技术有限公司 Object integrity evaluation method and device, electronic equipment and storage medium
CN111986254A (en) * 2020-08-21 2020-11-24 四川大学华西医院 Target area contour analysis method and device, storage medium and electronic equipment
CN112200217A (en) * 2020-09-09 2021-01-08 天津津航技术物理研究所 Identification algorithm evaluation method and system based on infrared image big data
CN112200217B (en) * 2020-09-09 2023-06-09 天津津航技术物理研究所 Identification algorithm evaluation method and system based on infrared image big data
CN113435535A (en) * 2021-07-13 2021-09-24 北京达佳互联信息技术有限公司 Training method of image recognition model, image recognition method and device

Also Published As

Publication number Publication date
CN110070533B (en) 2023-05-30

Similar Documents

Publication Publication Date Title
CN110070533A (en) A kind of evaluating method of object detection results, device, equipment and storage medium
US20190266434A1 (en) Method and device for extracting information from pie chart
CN107886074A (en) A kind of method for detecting human face and face detection system
CN102436587B (en) Multi-instance study is used to train the method and system of landmark detector
CN112801146B (en) Target detection method and system
CN105574550A (en) Vehicle identification method and device
CN107273002A (en) Handwriting input answer method, terminal and computer-readable recording medium
CN105068993B (en) A method of assessment text difficulty
CN107025174A (en) For the method for the user interface abnormality test of equipment, device and readable storage media
CN109241857A (en) A kind of recognition methods and system of document information
CN107273870A (en) The pedestrian position detection method of integrating context information under a kind of monitoring scene
CN110136153A (en) A kind of image processing method, equipment and storage medium
CN108205685A (en) Video classification methods, visual classification device and electronic equipment
CN109345553A (en) A kind of palm and its critical point detection method, apparatus and terminal device
WO2022247823A1 (en) Image detection method, and device and storage medium
CN107256379A (en) Information collecting method, mobile terminal and storage medium based on image recognition
US20190163499A1 (en) Simulating human interface interaction resilient to user interface changes
CN100361136C (en) Evaluation method and system for face detection system
CN107506162A (en) Coordinate mapping method, computer-readable recording medium and projecting apparatus
CN102081742B (en) Method for automatically evaluating writing ability
CN109086336A (en) Paper date storage method, device and electronic equipment
CN113111817A (en) Semantic segmentation face integrity measurement method, system, equipment and storage medium
CN117290694B (en) Question-answering system evaluation method, device, computing equipment and storage medium
JP5520505B2 (en) Scoring apparatus, scoring method, and program
JP2010231529A (en) Determination device and program

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant