CN109829932A - A kind of collecting method and device of automatic foreground extraction - Google Patents
A kind of collecting method and device of automatic foreground extraction Download PDFInfo
- Publication number
- CN109829932A CN109829932A CN201910049969.0A CN201910049969A CN109829932A CN 109829932 A CN109829932 A CN 109829932A CN 201910049969 A CN201910049969 A CN 201910049969A CN 109829932 A CN109829932 A CN 109829932A
- Authority
- CN
- China
- Prior art keywords
- sku
- module
- information
- foreground
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 60
- 238000000034 method Methods 0.000 title claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 51
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 30
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 30
- 238000004321 preservation Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims description 35
- 230000002708 enhancing effect Effects 0.000 claims description 18
- 238000009499 grossing Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 238000012805 post-processing Methods 0.000 claims description 5
- 230000002194 synthesizing effect Effects 0.000 claims description 3
- 238000013434 data augmentation Methods 0.000 abstract description 3
- 230000008901 benefit Effects 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 235000013399 edible fruits Nutrition 0.000 description 4
- 229910052500 inorganic mineral Inorganic materials 0.000 description 4
- 239000011707 mineral Substances 0.000 description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 4
- 235000013361 beverage Nutrition 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Abstract
The invention proposes the collecting methods and device of a kind of automatic foreground extraction, comprising: carries out foreground extraction to video image to be processed, obtains foreground image；It is used as a prospect sku information preservation into sku database in pairs the foreground image and corresponding mask, is used for subsequent training；The sku information of list to be trained is searched according to training mission and reads required prospect sku information from the sku database, and pulls sample as negative sample from sku database；Background information is chosen from basic database, then synthesizes the prospect sku information and background information；The data after synthesis are trained using the negative sample.The present invention does not need human assistance foreground extraction, accomplishes to be fully automated；Various complex backgrounds, data augmentation Huge value can be merged.
Description
Technical field
The present invention relates to technical field at image data, in particular to the collecting method of a kind of automatic foreground extraction and
Device.
Background technique
Existing dataset acquisition includes two kinds of usual tasks, first is that Detection task, second is that semantic segmentation task.Wherein,
The most common mask method is artificial mark, time-consuming and laborious, and accuracy rate is low.Existing solution is auxiliary using artificial intelligence
The notation methods helped, such as the notation methods based on tracing algorithm, but this mode has the following deficiencies:
1, precondition is needed to correspond to the detection model of classification；
2, background information can be added while automatic marking, cause relatively simple background, thus to the multiplicity of data
Property has an impact.
Summary of the invention
The purpose of the present invention aims to solve at least one of described technological deficiency.
For this purpose, it is an object of the invention to propose the collecting method and device of a kind of automatic foreground extraction.
To achieve the goals above, the embodiment of the present invention provides a kind of collecting method of automatic foreground extraction, packet
Include following steps:
Step S1 carries out foreground extraction to video image to be processed, obtains foreground image；
Step S2 is used as a prospect sku information preservation to sku number the foreground image and corresponding mask in pairs
According in library, used for subsequent training；
Step S3 searches the sku information of list to be trained according to training mission and needed for reading in the sku database
Prospect sku information, and pull sample as negative sample from sku database；
Step S4 chooses background information from basic database, then carries out the prospect sku information and background information
Synthesis；
Step S5 is trained the data after synthesis using the negative sample.
Further, in the step S1, using foreground extraction algorithm, prospect is carried out to video image to be processed and is mentioned
It takes.
Further, further include following steps between the step S1 and step S2: the foreground image is optimized
Processing, including smoothing processing and fill up processing.
Further, in the step S3, described to pull sample from sku database and walked as negative sample, including as follows
It is rapid: when the training mission is non-closed trained, to be calculated from existing sku information and belong to same major class with the sku information
But the sku information of not same group, and the sku information with the sku information dissmilarity, pull from sku database and meet
The sample of above-mentioned condition is as negative sample.
Further, further include following steps between the step S4 and step S5:
Data enhancing processing is carried out to the data after step S4 synthesis, then treated that generated data is sent by data enhancing
Enter step S5 to be trained.
The embodiment of the present invention also provides a kind of data acquisition device of automatic foreground extraction, comprising: data acquisition module, preceding
Scape extraction module, sku information searching module, front and back scape synthesis module, training module, wherein the data acquisition module is used for
To video image to be processed；The foreground extracting module is used to carry out foreground extraction to the video image, obtains foreground picture
Picture, and it is used as a prospect sku information preservation into sku database in pairs the foreground image and corresponding mask, it supplies
Subsequent training uses；The sku information searching module is used to search the sku information of list to be trained according to training mission and from institute
It states and reads required prospect sku information in sku database, and pull sample as negative sample from sku database；The front and back
Then scape synthesis module carries out the prospect sku information and background information for choosing background information from basic database
Synthesis；The training module is for being trained the data after synthesis using the negative sample.
Further, the foreground extracting module is used to use foreground extraction algorithm, before carrying out to video image to be processed
Scape extracts.
Further, the invention also includes: mask post-processing module, the mask post-processing module are mentioned with the prospect respectively
Modulus block is connected, for optimizing processing to the foreground image, including smoothing processing with fill up processing, then will be at optimization
Foreground image and corresponding mask after reason are used as a prospect sku information preservation into sku database in pairs, for subsequent instruction
Practice and uses.
Further, when the training mission is non-closed trained, the sku information searching module is believed from existing sku
The sku information for belonging to same major class but not same group with the sku information is calculated in breath, and dissimilar with the sku information
Sku information, the sample for meeting above-mentioned condition is pulled from sku database as negative sample.
Further, the invention also includes: data to enhance processing module, the data enhancing processing module and the front and back scape
Synthesis module is connected with the training module, for carrying out at data enhancing to the data after front and back scape synthesis module synthesis
Reason, then by data enhancing, treated that generated data is sent to the training module, by the training module using described negative
Sample is trained the data after synthesis.
The collecting method and device of automatic foreground extraction according to an embodiment of the present invention can be accurately partitioned into wait mark
The object of note, while after eliminating background information, can be more convenient generated for subsequent training data, training data enhancing
Etc. schemes.The fast automatic video foreground extracting method of the present invention, does not need the auxiliary manually demarcated, can do it is automatic, quick,
Accurate foreground extraction, the prospect extracted are needed to use and are somebody's turn to do in subsequent training in this way as the basic data collection of target object
, can be with significantly more efficient expanding data when object, and support that randomly selecting target data is trained task.
The collecting method and device of the automatic foreground extraction of the embodiment of the present invention, have the advantages that
1, human assistance foreground extraction is not needed, accomplishes to be fully automated；
2, a series of subsequent processings have been done to Mask, has solved the problems, such as that Mask is coarse and empty；
3, without being directed to target classification training pattern in advance, it is only necessary to optimize foreground extraction algorithm；
4, various complex backgrounds, data augmentation Huge value can be merged.
The additional aspect of the present invention and advantage will be set forth in part in the description, and will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect of the invention and advantage will become from the description of the embodiment in conjunction with the following figures
Obviously and it is readily appreciated that, in which:
Fig. 1 is the flow chart according to the collecting method of the automatic foreground extraction of the embodiment of the present invention；
Fig. 2 a to Fig. 2 c is the schematic diagram according to the collecting method of the automatic foreground extraction of the embodiment of the present invention；
Fig. 3 is the structure chart according to the data acquisition device of the automatic foreground extraction of one embodiment of the invention；
Fig. 4 is the structure chart according to the data acquisition device of the automatic foreground extraction of another embodiment of the present invention.
Specific embodiment
The embodiment of the present invention is described below in detail, the example of embodiment is shown in the accompanying drawings, wherein identical from beginning to end
Or similar label indicates same or similar element or element with the same or similar functions.It is retouched below with reference to attached drawing
The embodiment stated is exemplary, it is intended to is used to explain the present invention, and is not considered as limiting the invention.
The embodiment of the present invention and the collecting method and device of a kind of automatic foreground extraction are provided, may be implemented to calculate
Target detection, semantic segmentation and data enhance algorithm in machine visual field.
As shown in Figure 1, the collecting method of the automatic foreground extraction of the embodiment of the present invention, includes the following steps:
Step S1 carries out foreground extraction to video image to be processed, obtains foreground image.
In this step, using foreground extraction algorithm, foreground extraction is carried out to video image to be processed, such as Fig. 2 a to figure
Shown in 2c.
Preferably, the foreground extraction algorithm in step S1 can be real using conspicuousness detection deep neural network model algorithm
It is existing.
It should be noted that above-mentioned is only for exemplary purposes, to be not intended to be limiting of the invention.The present invention can be with
Other kinds of foreground extraction algorithm is taken, may be implemented, details are not described herein.
In an embodiment of the present invention, further include following steps between step S1 and step S2: foreground image is carried out
Optimization processing, including smoothing processing and fill up processing.
Preferably, this step can be realized to the smooth of foreground image and be filled out using subsequent algorithms such as condition random field CRF
Benefit processing.It should be noted that this step realizes that the algorithm optimized to foreground image is not limited to the example above, it can also be used
The algorithm of his type is only able to satisfy above-mentioned optimization requirement, and details are not described herein.
Step S2 is used as a prospect sku information preservation to sku number foreground image and corresponding mask mask in pairs
According in library, used for subsequent training.
Step S3, according to training mission search list to be trained sku information and from sku database read needed for before
Scape sku information, and sample is pulled as negative sample from sku database.
Specifically, searching the sku information of list to be trained from sku management system according to training mission first.Wherein, to
The sku information of training list includes: the information such as the title, brand, original image of kinds of goods.
Then, then from the sku database in step S2 read required prospect sku information.
In addition, pulling sample as negative sample in slave sku database in this step, include the following steps: when training
When task is non-closed training (needing to identify the target in non-known class), calculates from existing sku information and believe with the sku
Breath belongs to the sku information of same major class but not same group, and the sku information with the sku information dissmilarity, from sku number
According to pulling the sample for meeting above-mentioned condition in library as negative sample.
For example, sku information corresponds to mineral water kinds of goods, then belong to same major class but not same group with the sku information
Sku information can be fruit drink kinds of goods, the fruit drink and mineral water belong to beverage (belonging to same major class), but not
Drink (not same group) of the same race.
It should be noted that the sku information with the sku information dissmilarity refers among the above, it is right with current sku information institute
Answer product dissimilar sku information in shape.
Step S4 chooses background information from basic database, then closes prospect sku information and background information
At.
In one embodiment of the invention, background information is chosen from basic database, such as can be shelf, room
Interior, outdoor, refrigerator etc..Then according to prospect sku information and background information, a picture is merged into according to queueing discipline, is realized
The random combine of prospect foreground and background information completes synthetic operation.
Preferably, further include following steps between step S4 and step S5: the data after step S4 synthesis are counted
It is handled according to enhancing, data enhancing treated generated data is then sent into step S5 and is trained.For example, to the number after synthesis
According to sheared, rotated, the operation such as brightness adjustment.
Step S5 is trained the data after synthesis using negative sample.
As shown in figure 3, the data acquisition device of the automatic foreground extraction of the embodiment of the present invention, comprising: data acquisition module
1, foreground extracting module 2, sku information searching module 3, front and back scape synthesis module 4, training module 5.
Data acquisition module 1 is used for video image to be processed.
Foreground extracting module 2 is used to carry out foreground extraction to video image, obtains foreground image, and by foreground image and
Corresponding mask is used as a prospect sku information preservation into sku database in pairs, uses for subsequent training.
In one embodiment of the invention, foreground extracting module 2 is used to use foreground extraction algorithm, to view to be processed
Frequency image carries out foreground extraction.
Preferably, the foreground extraction algorithm in foreground extracting module 2 can detect deep neural network mould using conspicuousness
Type algorithm is realized.
It should be noted that above-mentioned is only for exemplary purposes, to be not intended to be limiting of the invention.The present invention can be with
Other kinds of foreground extraction algorithm is taken, may be implemented, details are not described herein.
In addition, as shown in figure 4, the data acquisition device of the automatic foreground extraction of the embodiment of the present invention, further includes: after mask
Processing module 6, mask post-processing module 6 are connected with foreground extracting module 2 respectively, for optimizing processing to foreground image,
Including smoothing processing and filling up processing, then by the foreground image after optimization processing and before corresponding mask is used as one in pairs
Scape sku information preservation is used into sku database for subsequent training.
Preferably, mask post-processing module 6 can be realized using subsequent algorithms such as condition random field CRF to foreground image
It is smooth and fill up processing.It should be noted that realizing that the algorithm optimized to foreground image is not limited to the example above, can also adopt
With other kinds of algorithm, it is only able to satisfy above-mentioned optimization requirement, details are not described herein.
Sku information searching module 3 is used to search the sku information of list to be trained according to training mission and from sku database
Prospect sku information needed for middle reading, and sample is pulled as negative sample from sku database.
Specifically, sku information searching module 3 searches list to be trained from sku management system first according to training mission
Sku information.Wherein, the sku information of list to be trained includes: the information such as the title, brand, original image of kinds of goods.Then, then from
Required prospect sku information is read in sku database.
When training mission is non-closed training (needing to identify the target in non-known class), sku information searching module 3
Calculated from existing sku information and belong to the sku information of same major class but not same group with the sku information, and with this
The sku information of sku information dissmilarity, pulls the sample for meeting above-mentioned condition as negative sample from sku database.
For example, sku information corresponds to mineral water kinds of goods, then belong to same major class but not same group with the sku information
Sku information can be fruit drink kinds of goods, the fruit drink and mineral water belong to beverage (belonging to same major class), but not
Drink (not same group) of the same race.
It should be noted that the sku information with the sku information dissmilarity refers among the above, it is right with current sku information institute
Answer product dissimilar sku information in shape.
Front and back scape synthesis module 4 from basic database for choosing background information, then by prospect sku information and background
Information is synthesized.
In one embodiment of the invention, front and back scape synthesis module 4 chooses background information from basic database, such as
It can be shelf, indoor and outdoor, refrigerator etc..Then it according to prospect sku information and background information, is merged into according to queueing discipline
One picture realizes the random combine of prospect foreground and background information, completes synthetic operation.
The data acquisition device of the automatic foreground extraction of the embodiment of the present invention further include: data enhance processing module 7, data
Enhancing processing module 7 is connect with front and back scape synthesis module 4 and training module 5, for the number after synthesizing to front and back scape synthesis module 4
According to data enhancing processing is carried out, for example, being sheared, being rotated to the data after synthesis, the operation such as brightness adjustment.Then by data
Treated that generated data is sent to training module 5 for enhancing, is instructed using negative sample to the data after synthesizing by training module 5
Practice.Specifically, training module 5 is for being trained the data after synthesis using negative sample.
The collecting method and device of automatic foreground extraction according to an embodiment of the present invention can be accurately partitioned into wait mark
The object of note, while after eliminating background information, can be more convenient generated for subsequent training data, training data enhancing
Etc. schemes.The fast automatic video foreground extracting method of the present invention, does not need the auxiliary manually demarcated, can do it is automatic, quick,
Accurate foreground extraction, the prospect extracted are needed to use and are somebody's turn to do in subsequent training in this way as the basic data collection of target object
, can be with significantly more efficient expanding data when object, and support that randomly selecting target data is trained task.
The collecting method and device of the automatic foreground extraction of the embodiment of the present invention, have the advantages that
1, human assistance foreground extraction is not needed, accomplishes to be fully automated；
2, a series of subsequent processings have been done to Mask, has solved the problems, such as that Mask is coarse and empty；
3, without being directed to target classification training pattern in advance, it is only necessary to optimize foreground extraction algorithm；
4, various complex backgrounds, data augmentation Huge value can be merged.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
Centainly refer to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be any
One or more embodiment or examples in can be combined in any suitable manner.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art are not departing from the principle of the present invention and objective
In the case where can make changes, modifications, alterations, and variations to the above described embodiments within the scope of the invention.The scope of the present invention
It is extremely equally limited by appended claims.
Claims (10)
1. a kind of collecting method of automatic foreground extraction, which comprises the steps of:
Step S1 carries out foreground extraction to video image to be processed, obtains foreground image；
Step S2 is used as a prospect sku information preservation to sku database the foreground image and corresponding mask in pairs
In, it is used for subsequent training；
Step S3, according to training mission search list to be trained sku information and from the sku database read needed for before
Scape sku information, and sample is pulled as negative sample from sku database；
Step S4 chooses background information from basic database, then closes the prospect sku information and background information
At；
Step S5 is trained the data after synthesis using the negative sample.
2. the collecting method of automatic foreground extraction as described in claim 1, which is characterized in that in the step S1,
Using foreground extraction algorithm, foreground extraction is carried out to video image to be processed.
3. the collecting method of automatic foreground extraction as described in claim 1, which is characterized in that in the step S1 and step
Further include following steps between rapid S2: processing optimized to the foreground image, including smoothing processing with fill up processing.
4. the collecting method of automatic foreground extraction as described in claim 1, which is characterized in that in the step S3,
The sample that pulls from sku database includes the following steps: that when the training mission be non-closed training as negative sample
When, the sku information for belonging to same major class but not same group with the sku information, Yi Jiyu are calculated from existing sku information
The sku information of the sku information dissmilarity, pulls the sample for meeting above-mentioned condition as negative sample from sku database.
5. the collecting method of automatic foreground extraction as described in claim 1, which is characterized in that in the step S4 and step
Further include following steps between rapid S5:
Data enhancing processing is carried out to the data after step S4 synthesis, data enhancing treated generated data is then sent into step
Rapid S5 is trained.
6. a kind of data acquisition device of automatic foreground extraction characterized by comprising data acquisition module, foreground extraction mould
Block, sku information searching module, front and back scape synthesis module, training module, wherein
The data acquisition module is used for video image to be processed；
The foreground extracting module is used to carry out foreground extraction to the video image, obtains foreground image, and by the prospect
Image and corresponding mask are used as a prospect sku information preservation into sku database in pairs, use for subsequent training；
The sku information searching module is used to search the sku information of list to be trained according to training mission and from the sku data
Required prospect sku information is read in library, and pulls sample as negative sample from sku database；
The front and back scape synthesis module for choosing background information from basic database, then by the prospect sku information and
Background information is synthesized；
The training module is for being trained the data after synthesis using the negative sample.
7. the data acquisition device of automatic foreground extraction as claimed in claim 6, which is characterized in that the foreground extracting module
For using foreground extraction algorithm, foreground extraction is carried out to video image to be processed.
8. the data acquisition device of automatic foreground extraction as claimed in claim 6, which is characterized in that further include: after mask
Module is managed, the mask post-processing module is connected with the foreground extracting module respectively, excellent for carrying out to the foreground image
Change processing, including smoothing processing and fill up processing, then by after optimization processing foreground image and corresponding mask make in pairs
It is a prospect sku information preservation into sku database, is used for subsequent training.
9. the data acquisition device of automatic foreground extraction as claimed in claim 6, which is characterized in that when the training mission is
When non-closed trained, the sku information searching module is calculated from existing sku information belongs to same major class with the sku information
But the sku information of not same group, and the sku information with the sku information dissmilarity, pull from sku database and meet
The sample of above-mentioned condition is as negative sample.
10. the data acquisition device of automatic foreground extraction as claimed in claim 6, which is characterized in that further include: data enhancing
Processing module, the data enhancing processing module is connect with the front and back scape synthesis module and the training module, for institute
Data after stating the synthesis of front and back scape synthesis module carry out data enhancing processing, then send out data enhancing treated generated data
It send to the training module, the data after synthesizing is trained using the negative sample by the training module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910049969.0A CN109829932A (en) | 2019-01-18 | 2019-01-18 | A kind of collecting method and device of automatic foreground extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910049969.0A CN109829932A (en) | 2019-01-18 | 2019-01-18 | A kind of collecting method and device of automatic foreground extraction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109829932A true CN109829932A (en) | 2019-05-31 |
Family
ID=66860992
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910049969.0A Pending CN109829932A (en) | 2019-01-18 | 2019-01-18 | A kind of collecting method and device of automatic foreground extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109829932A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110288614A (en) * | 2019-06-24 | 2019-09-27 | 睿魔智能科技(杭州)有限公司 | Image processing method, device, equipment and storage medium |
| CN110503146A (en) * | 2019-08-21 | 2019-11-26 | 杭州比智科技有限公司 | Data enhancement methods, calculate equipment and computer storage medium at device |
| CN110689477A (en) * | 2019-09-07 | 2020-01-14 | 创新奇智(重庆)科技有限公司 | Universal flaw image simulation method |
| CN111062861A (en) * | 2019-12-13 | 2020-04-24 | 广州市玄武无线科技股份有限公司 | Method and device for generating display image samples |
| CN112508132A (en) * | 2021-01-29 | 2021-03-16 | 广州市玄武无线科技股份有限公司 | Training method and device for identifying SKU |
| CN114333038A (en) * | 2022-03-03 | 2022-04-12 | 百度在线网络技术（北京）有限公司 | Training method of object recognition model, object recognition method, device and equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102789578A (en) * | 2012-07-17 | 2012-11-21 | 北京市遥感信息研究所 | Infrared remote sensing image change detection method based on multi-source target characteristic support |
| CN104915972A (en) * | 2014-03-13 | 2015-09-16 | 欧姆龙株式会社 | Image processing apparatus, image processing method and program |
| CN107679557A (en) * | 2017-09-19 | 2018-02-09 | 平安科技（深圳）有限公司 | Driving model training method, driver's recognition methods, device, equipment and medium |
| CN107808373A (en) * | 2017-11-15 | 2018-03-16 | 北京奇虎科技有限公司 | Sample image synthetic method, device and computing device based on posture |
| CN108986058A (en) * | 2018-06-22 | 2018-12-11 | 华东师范大学 | The image interfusion method of lightness Consistency Learning |
-
2019
- 2019-01-18 CN CN201910049969.0A patent/CN109829932A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102789578A (en) * | 2012-07-17 | 2012-11-21 | 北京市遥感信息研究所 | Infrared remote sensing image change detection method based on multi-source target characteristic support |
| CN104915972A (en) * | 2014-03-13 | 2015-09-16 | 欧姆龙株式会社 | Image processing apparatus, image processing method and program |
| CN107679557A (en) * | 2017-09-19 | 2018-02-09 | 平安科技（深圳）有限公司 | Driving model training method, driver's recognition methods, device, equipment and medium |
| CN107808373A (en) * | 2017-11-15 | 2018-03-16 | 北京奇虎科技有限公司 | Sample image synthetic method, device and computing device based on posture |
| CN108986058A (en) * | 2018-06-22 | 2018-12-11 | 华东师范大学 | The image interfusion method of lightness Consistency Learning |
Non-Patent Citations (2)
| Title |
|---|
| 康伟德等: "嵌入式系统中实时行人检测方法", 《计算机系统应用》 * |
| 赵雪, 吉林大学出版社 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110288614A (en) * | 2019-06-24 | 2019-09-27 | 睿魔智能科技(杭州)有限公司 | Image processing method, device, equipment and storage medium |
| CN110288614B (en) * | 2019-06-24 | 2022-01-11 | 睿魔智能科技(杭州)有限公司 | Image processing method, device, equipment and storage medium |
| CN110503146A (en) * | 2019-08-21 | 2019-11-26 | 杭州比智科技有限公司 | Data enhancement methods, calculate equipment and computer storage medium at device |
| CN110503146B (en) * | 2019-08-21 | 2021-12-14 | 杭州比智科技有限公司 | Data enhancement method and device, computing equipment and computer storage medium |
| CN110689477A (en) * | 2019-09-07 | 2020-01-14 | 创新奇智(重庆)科技有限公司 | Universal flaw image simulation method |
| CN110689477B (en) * | 2019-09-07 | 2023-01-10 | 创新奇智(重庆)科技有限公司 | Universal flaw image simulation method |
| CN111062861A (en) * | 2019-12-13 | 2020-04-24 | 广州市玄武无线科技股份有限公司 | Method and device for generating display image samples |
| CN112508132A (en) * | 2021-01-29 | 2021-03-16 | 广州市玄武无线科技股份有限公司 | Training method and device for identifying SKU |
| CN114333038A (en) * | 2022-03-03 | 2022-04-12 | 百度在线网络技术（北京）有限公司 | Training method of object recognition model, object recognition method, device and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829932A (en) | A kind of collecting method and device of automatic foreground extraction | |
| CN110956185B (en) | Method for detecting image salient object | |
| CN107833213B (en) | Weak supervision object detection method based on false-true value self-adaptive method | |
| CN110188635A (en) | A kind of plant pest recognition methods based on attention mechanism and multi-level convolution feature | |
| CN101604325B (en) | Method for classifying sports video based on key frame of main scene lens | |
| CN110472467A (en) | The detection method for transport hub critical object based on YOLO v3 | |
| CN109754009B (en) | Article identification method, article identification device, vending system and storage medium | |
| Korschens et al. | Elpephants: A fine-grained dataset for elephant re-identification | |
| CN102831239B (en) | A kind of method and system building image data base | |
| CN107730553B (en) | Weak supervision object detection method based on false-true value search method | |
| CN103336835B (en) | Image retrieval method based on weight color-sift characteristic dictionary | |
| CN109034247B (en) | Tracking algorithm-based higher-purity face recognition sample extraction method | |
| CN109753853A (en) | One kind being completed at the same time pedestrian detection and pedestrian knows method for distinguishing again | |
| CN110827312A (en) | Learning method based on cooperative visual attention neural network | |
| CN103399870A (en) | Visual word bag feature weighting method and system based on classification drive | |
| CN114140665A (en) | Dense small target detection method based on improved YOLOv5 | |
| Puri | COCO dataset stuff segmentation challenge | |
| CN111462162A (en) | Foreground segmentation algorithm for specific class of pictures | |
| CN109271927A (en) | A kind of collaboration that space base is multi-platform monitoring method | |
| Arun et al. | Effective and efficient multi-crop pest detection based on deep learning object detection models | |
| CN105404682B (en) | A kind of book retrieval method based on digital image content | |
| Shuai et al. | An improved YOLOv5-based method for multi-species tea shoot detection and picking point location in complex backgrounds | |
| CN106920255B (en) | Moving object extraction method and device for image sequence | |
| CN112200862A (en) | Training method of target detection model, target detection method and device | |
| CN114743045B (en) | Small sample target detection method based on double-branch area suggestion network |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200509 Address after: 201900 569n, building 1, No. 668, SHANGDA Road, Baoshan District, Shanghai Applicant after: Innovation wisdom (Shanghai) Technology Co.,Ltd. Address before: 315000 Lane 41, Tiantong South Road, Yinzhou District, Ningbo City, Zhejiang Province Applicant before: ALNNOVATION (NINGBO) TECHNOLOGY Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190531 |
|
| RJ01 | Rejection of invention patent application after publication |