CN109829969A - A kind of data capture method, device and storage medium - Google Patents
A kind of data capture method, device and storage medium Download PDFInfo
- Publication number
- CN109829969A CN109829969A CN201811615761.2A CN201811615761A CN109829969A CN 109829969 A CN109829969 A CN 109829969A CN 201811615761 A CN201811615761 A CN 201811615761A CN 109829969 A CN109829969 A CN 109829969A
- Authority
- CN
- China
- Prior art keywords
- identified
- information
- label
- image
- model
- 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
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000013481 data capture Methods 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000013136 deep learning model Methods 0.000 claims abstract description 20
- 238000009877 rendering Methods 0.000 claims abstract description 14
- 238000004422 calculation algorithm Methods 0.000 claims description 20
- 238000013507 mapping Methods 0.000 claims description 20
- 238000012545 processing Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 8
- 238000013519 translation Methods 0.000 claims description 4
- 241000282326 Felis catus Species 0.000 description 31
- 238000013527 convolutional neural network Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 230000008901 benefit Effects 0.000 description 5
- 238000013480 data collection Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000004043 dyeing Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 241001465754 Metazoa Species 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000001815 facial effect Effects 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 102000045246 noggin Human genes 0.000 description 1
- 108700007229 noggin Proteins 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The present invention provides a kind of data capture method, device and storage mediums, this method comprises: obtaining object to be identified and the corresponding image information of object to be identified；According to object to be identified and described image information, the corresponding multiple 3D models of object to be identified are generated, multiple 3D models include 3D rendering of multiple objects to be identified under multiple scenes；According to multiple 3D models, first label of at least one of position, size and type that object to be identified is identified in each 3D model is acquired respectively；According to multiple 3D models and the first label, the label information of object to be identified corresponding multiple 2D images and each 2D image is obtained；The label information of multiple 2D images and each 2D image is for training deep learning model.Therefore, the sample data with accurately label information can be got, and eliminates the operation being largely manually labeled, improve the label accuracy of data while simplifying the determining operation of data, reduces artificial investment.
Description
Technical field
The invention belongs to field of image processings, more particularly to a kind of data capture method, device and storage medium.
Background technique
In the prior art, when being trained to deep learning model, need a large amount of quality data to guarantee depth
The effect of learning model, to obtain above data collection, it usually needs carry out data acquisition, and needed later to each data
Sample is manually marked, and during the acquisition of above data collection, it may appear that data category is uneven, does not have label, label
Single equal (such as image resolution ratio) problem of the types of inaccuracy and data, then need the corresponding data of developer's progress clear
Wash work；Meanwhile in the above artificial annotation process, a large amount of manpower and material resources can be expended, however also not can guarantee the matter of mark
Amount and integrality.
Summary of the invention
In view of this, the present invention provides a kind of data capture method, device and storage medium, deposited to solve the prior art
The complicated and of low quality problem of data set acquisition process.
According to the present invention in a first aspect, providing a kind of data capture method, this method may include:
Obtain object to be identified and the corresponding image information of the object to be identified；
According to the object to be identified and described image information, the corresponding multiple 3D models of the object to be identified are generated,
The multiple 3D model includes 3D rendering of multiple objects to be identified under multiple scenes；
According to the multiple 3D model, the position that the object to be identified is identified in each 3D model, size are acquired respectively
With the first label of at least one of type；
According to the multiple 3D model and first label, obtain the corresponding multiple 2D images of the object to be identified and
The label information of each 2D image；The label information of the multiple 2D image and each 2D image is for training deep learning model.
Second aspect according to the present invention provides a kind of data acquisition facility, the apparatus may include:
Object obtains module, for obtaining object to be identified and the corresponding image information of the object to be identified；
3D model generation module, for generating the object to be identified according to the object to be identified and described image information
The corresponding multiple 3D models of body, the multiple 3D model include 3D rendering of multiple objects to be identified under multiple scenes；
Label acquisition module, it is described wait know for according to the multiple 3D model, acquiring the mark in each 3D model respectively
First label of at least one of position, size and the type of other object；
2D data obtains module, for obtaining the object to be identified according to the multiple 3D model and first label
Multiple 2D images of body and the label information of each 2D image；The label information of the multiple 2D image and each 2D image is for training
Deep learning model.
The third aspect according to the present invention provides a kind of computer readable storage medium, the computer-readable storage
Computer program is stored on medium, and data acquisition as described in relation to the first aspect is realized when the computer program is executed by processor
The step of method.
For first technology, the present invention has following advantage:
According to the object to be identified and corresponding image information got, multiple 3D models under multiple scenes are generated,
Building different scenes as much as possible and multiple modelings are carried out to object to be identified, with rich image sample data, and base
In multiple 3D models record generated the object to be identified wherein drawn the first label, for by 3D model projection
To after 2D image, corresponding calculate determines that the label information of 2D image not only makes convenient for the training for deep learning model
The sample data got has accurately label information, and eliminates the operation being largely manually labeled, and improves data
Label accuracy simplify data simultaneously determine operation, reduce artificial investment.
The above description is only an overview of the technical scheme of the present invention, in order to better understand the technical means of the present invention,
And it can be implemented in accordance with the contents of the specification, and in order to allow above and other objects of the present invention, feature and advantage can
It is clearer and more comprehensible, the followings are specific embodiments of the present invention.
Detailed description of the invention
By reading the following detailed description of the preferred embodiment, various other advantages and benefits are common for this field
Technical staff will become clear.The drawings are only for the purpose of illustrating a preferred embodiment, and is not considered as to the present invention
Limitation.And throughout the drawings, the same reference numbers will be used to refer to the same parts.In the accompanying drawings:
Fig. 1 is a kind of step flow chart of data capture method provided in an embodiment of the present invention；
Fig. 2 is a kind of specific steps flow chart of data capture method provided in an embodiment of the present invention；
Fig. 3 is a kind of specific steps flow chart of data capture method provided in an embodiment of the present invention；
Fig. 4 is a kind of specific steps flow chart of data capture method provided in an embodiment of the present invention；
Fig. 5 is a kind of block diagram of data acquisition facility provided in an embodiment of the present invention.
Specific embodiment
The exemplary embodiment that the present invention will be described in more detail below with reference to accompanying drawings.Although showing the present invention in attached drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the present invention without should be by embodiments set forth here
It is limited.It is to be able to thoroughly understand the present invention on the contrary, providing these embodiments, and can be by the scope of the present invention
It is fully disclosed to those skilled in the art.
Fig. 1 is a kind of step flow chart of data capture method provided in an embodiment of the present invention, as shown in Figure 1, this method
May include:
Step 101, object to be identified and the corresponding image information of object to be identified are obtained.
In a particular application, generally in training deep learning model, needing to acquire acquisition has targetedly trained number
According to, that is, according to the corresponding training data of actual application scenarios acquisition.Wherein, there are three types of basic moulds for deep learning model
Type, be respectively multi-layer perception (MLP) (abbreviation: MLP, English: Multi-layer Perceptron), convolutional neural networks (abbreviation:
CNN, English: Convolutional Neural Network) and Recognition with Recurrent Neural Network (abbreviation: RNN, English:
Recurrent Neural Network).By taking CNN as an example, it is widely used in image recognition, target detection, recognition of face
Etc. in many computer vision fields, and before a CNN network model is used in actual image recognition, need to acquire
Training data is trained initial CNN network model, optimizes the parameter in network model, stablizes output to generate to have
CNN network model.Such as when using the progress recognition of face of CNN network, training data collected should be a large amount of facial image
Therefore data when determining the training dataset for training deep learning model, need corresponding according to the deep learning model
Object to be identified carry out, e.g. personage, animal or other objects, such as vehicle, furniture, so as to targetedly
Image information is obtained, and then when the drafting of step progress 3D rendering below, modeling can be extended based on actual conditions, kept away
Exempt from the authenticity and validity of reduction 3D modeling.
In embodiments of the present invention, include in the corresponding image information of object to be identified variform and/or type and/or
Multiple 2D/3D images corresponding to the object to be identified of shape.
In the specific application process, since the feature that deep learning model can learn is abstract characteristics, and
It is to be drawn with approximate scene set to be modeled to real scene in graphics, therefore based on object to be identified
The real scene and actual characteristic of image information determine corresponding multiple 2D/3D images.Illustratively, when object to be identified is
Cat acquires the image data of all types of cats, such as the picture of the cat of 10 kinds of different cultivars, and quantity is 10,000, every size
For 224*224, such as based on above-mentioned 10,000 picture, it can establish the 3D model of cat, to simulate the various forms of cat, lie, sit,
Sleeping, race etc.；Or can also be the 2D image of various true cats, and below in 3D plot step, it is corresponding to generate 3D
Image.For example, including different kinds, pattern, size in corresponding image information in the case where being cat for object to be identified
The 2D/3D image of the cats such as size.
Variform corresponding to the object to be identified that will acquire and/or multiple 2D/3D of type and/or shape figure
Picture constructs the basic basis of multiple 3D models as below step, and the 3D model constructed in step later is accorded with
Actual conditions are closed, are not in unreality content, so that training data has targetedly has workability simultaneously.
Step 102, according to object to be identified and image information, the corresponding multiple 3D models of object to be identified are generated.
Wherein, multiple 3D models include 3D rendering of multiple objects to be identified under multiple scenes.
Illustratively, determine that object to be identified is cat according to step 101, then based on from 10 kinds of different cats, 10,000 every kind
After determining the corresponding model of object (cat) to be identified in picture, using 3D drawing algorithm, e.g. texture-mapping algorithm carries out 3D
Modeling, generates multiple 3D models of the object to be identified under different scenes；The default scene of difference is corresponding for object to be identified
Design, and scene multiple combinations are carried out, to ensure generation 3D model as much as possible, the sample content of abundant data collection,
More it is bonded actual conditions.
In addition, a usual default scene can be understood as, cat lies prone and (sits, is sleeping, running) on desk, cat lies prone and (sits, is sleeping, running)
In bed or cat lies prone and (sits, is sleeping, running) on floor etc., practical when this step generates 3D model, can also preset field to these
Scape is combined, such as desk, chair, bed are provided in a 3D rendering, and cat is in form wherein including running, sitting, crouch, lie prone
Etc., kind, pattern of multiple cats etc., that is to say, that the scene that cat is likely to occur can be designed and be combined, such as designed
Cat on desk, cat on sofa, cat running and the cat litter scenes such as in corners, then carrying out multiple combinations, in a 3D
Combined in model multiple cats on sofa, cat running and cat litter scene in corners, can also be in a 3D model
There is multiple cats, including multiple and different kinds, and then models and generate 3D model, the content in scene abundant as far as possible, simulation
The scene being likely to occur.
Step 103, according to multiple 3D models, position, the size that object to be identified is identified in each 3D model are acquired respectively
With the first label of at least one of type.
It in a particular application, is the addition that object to be identified is carried out according to the scene preset when constructing 3D model, because
This collects and records one in wherein position, size and the type for identifying object to be identified while generating 3D model
Or the first label of more persons, using graph tool, e.g. excel is recorded in detail, using the label as training data,
It can be used in the training to deep learning model.
Illustratively, when being recorded in one 3D model of generation, a three dyeing defect cats coordinate that coordinate is has been placed at first point
Are as follows: 0.23,4.67,3.25；The coordinate of second point are as follows: 0.23,16.78,78.5；And coordinate thirdly is 89.2,7.89,
67.5 position, corresponding first label are above-mentioned 3 points of coordinate and three dyeing defect cats.
Step 104, according to multiple 3D models and the first label, the corresponding multiple 2D images of object to be identified and each 2D are obtained
The label information of image.
Wherein, the label information of multiple 2D images and each 2D image is for training deep learning model.
Under concrete application scene, after generating 3D model data collection through the above steps, 3D model can be carried out
2D projection, such as its corresponding 2D image is obtained by way of unity and coherence in writing projection mapping, as training deep learning model
Target data set, can be image or video format, simultaneously because step 103 is based on determined by step 102
Multiple 3D models are the first labels recorded based on designed scene and object to be identified, and then generate accurate label letter
2D image and the corresponding label information of each 2D image can be used to carry out deep learning model effectively to train, Jin Er by breath
Under the premise of the accuracy for guaranteeing label information, a large amount of human inputs is avoided to carry out the process of image tagged.
It should be noted that data capture method proposed by the present invention, is corresponding to be identified for deep learning model
Object application scenarios, building different scenes as much as possible and carry out multiple modelings to object to be identified, with rich image sample
Notebook data, and the first label in the object to be identified wherein drawn is recorded based on multiple 3D models generated, to be used for
It is corresponding to calculate the label information for determining 2D image, it is ensured that the label information got after by 3D model projection to 2D image
Accurate and effective.
In conclusion data capture method provided by the invention, obtains object to be identified and the corresponding figure of object to be identified
As information；According to object to be identified and described image information, the corresponding multiple 3D models of object to be identified, multiple 3D models are generated
3D rendering including multiple objects to be identified under multiple scenes；According to multiple 3D models, acquires get the bid in each 3D model respectively
Know first label of at least one of position, size and type of object to be identified；According to multiple 3D models and the first label, obtain
Take the label information of object to be identified corresponding multiple 2D images and each 2D image；The label of multiple 2D images and each 2D image letter
Breath is for training deep learning model.Therefore, using 2D image accessed by 3D model projection to 2D as training deep learning
The image data of model to get data source training data abundant and true to nature, and the corresponding projection is calculated
Label of the label information as training data, and avoid manually marking the error that may generate label information, promote label
The accuracy of information, and then simplify the operation that determines of data, the artificial investment of reduction.
Optionally, Fig. 2 is a kind of specific steps flow chart of data capture method provided in an embodiment of the present invention, such as Fig. 2
It is shown, according to object to be identified and image information described in step 102, the corresponding multiple 3D models of object to be identified are generated, it can
To include:
Step 1021, preset scene setting rule, multiple scenes of the corresponding object to be identified of setting are utilized.
In a particular application, different objects to be identified is adapted to different scenes, and preset scene setting rule is used for
Limiting field shadow definer closes the actual natural law, including contents such as position, correspondingly-sized, the ratios occurred to object to be identified
Limitation, the principle of scene setting rule is the spy based on the corresponding image information of object to be identified and object to be identified itself
Property, be arranged under conditions of without prejudice to the natural law correspondence as much as possible preset scene, while also carry out scene extension and
Fine tuning, further to enrich the scene of modeling, to solve the problems, such as that existing training dataset is single, class imbalance.
Illustratively, cat can be bigger than general noggin, cat be not in the sky, shared ratio has cat on all fours in bed
The setting rule such as limit corresponds to cat this object to be identified and sets corresponding multiple scenes, with the true of the 3D model that ensures to generate
Real validity.
Step 1022, according to multiple scenes, drawing processing is carried out to image information, to generate the mesh for drawing 3D model
Logo image.
Wherein, processing of drawing includes at least one of translation, rotation and deformation.
In a particular application, different according to corresponding scene, it needs to handle image information, to guarantee it in corresponding fields
Adaptability and authenticity in scape, therefore, can correspond to scene carry out image information drawing processing, such as to image information into
Row amplification, size of the diminution to be suitble under the scene, translate it, are rotated to meet the scene content, at drawing
Reason, can optimize the generation quality in 3D model.
Step 1023,3D drawing algorithm is called, multiple 3D models of the target image in multiple scenes are drawn.
Wherein, 3D drawing algorithm includes at least one of texture-mapping algorithm and Shadow Mapping algorithm.
Illustratively, the target image that step 1022 determines is added under multiple default scenes and is modeled, generated and correspond to
3D model.
In a particular application, texture-mapping algorithm (Texture Mapping) is a kind of by graphic plotting (mapping) to table
The technology in face can increase the details and the sense of reality of drawn scene significantly, by defining texture object, to generate texture pair
As array, then by selection texture object, to complete the definition of the texture object, later before rendering the scene, for the scenery
Corresponding texture is loaded, finally realizes texture mapping operation.Shadow Mapping algorithm (Shadow Mapping) is to produce in the scene
The technology of organism shade, to increase the sense of reality and spatial impression of scene.The rendering of scene is carried out using the position of light as visual angle,
It is depth buffered due to being provided in a program, so a part of the object or object that are blocked will not be seen, that is,
Tell me less than place, be exactly the shade of object in scene, the depth value that Z-buffering obtains each fragment can be used
(between generally 0 to 1), the depth value of dash area is certainly bigger than the depth value of the shelter of corresponding position, while by depth
Value information is stored into texture, that is, shadow map.Using Texture Mapping and Shadow Mapping to target figure
As progress 3D modeling, such as 3D model of the more cats in the form of different in room is generated, every cat can also be correspondingly arranged
Different size, pattern, kind etc..
Optionally, Fig. 3 is a kind of specific steps flow chart of data capture method provided in an embodiment of the present invention, such as Fig. 3
It is shown, according to multiple 3D models and the first label described in step 104, obtain corresponding multiple 2D images of object to be identified and each
The label information of 2D image may include:
Step 1031,2D projection is carried out to multiple 3D models, to obtain multiple 2D images of preset format.
Wherein, preset format includes picture format or video format.
Illustratively, projection is carried out according to 3D model and determines corresponding 2D image, mapped for example, by using projective textures
3D model projection is gone out 2D format by (Projective Texture Mapping) mode, for training deep learning model, base
2D image is generated in 3D model, increases the authenticity of image itself, and enrich data space.
Step 1032, according to multiple 2D images, label information is determined using the first label is corresponding.
Wherein, label information includes location information, dimension information and type letter of the object to be identified in each 2D image
At least one of breath.
Optionally, the step is as shown in Figure 4, comprising the following steps:
Step 10321, it in the case where the first label is size and/or position for identifying object to be identified, is based on
The projection relation and the first label when 2D projection are carried out to 3D model, obtain the dimension information of object to be identified in the projected
And/or location information；And/or in the case where the first label is the type for identifying object to be identified, using the type as
Type information.
Illustratively, since dimension information and location information are related to the projection relation of 3D model to 2D image, such as using
Selected parameter is different when projective textures map, cause projection angle, Projection Depth etc. different and generate different size and
Location information, and then by carrying out dimension information and/or location information after mapping calculation determines projection to the first label, also
It is specific size and/or position of the object to be identified in 2D image；And the type of object to be identified does not generate in the projected
Variation, therefore type information can be determined directly as.
Step 10322, the dimension information and/or location information and/or type information that will acquire are as label information.
Illustratively, the dimension information and/or location information and/or type information step 10321 got is as mark
Sign information, the corresponding label of accessed training data be it is accurate and perfect, can guarantee to deep learning model
Training quality.
Fig. 5 is a kind of block diagram of data acquisition facility provided in an embodiment of the present invention, as shown in figure 5, the device 500 wraps
It includes:
Object obtains module 510, for obtaining object to be identified and the corresponding image information of object to be identified.
3D model generation module 520, for it is corresponding more to generate object to be identified according to object to be identified and image information
A 3D model, multiple 3D model include 3D rendering of multiple objects to be identified under multiple scenes.
Label acquisition module 530, for acquiring identify object to be identified in each 3D model respectively according to multiple 3D models
At least one of position, size and type the first label.
2D data obtains module 540, for obtaining multiple 2D of object to be identified according to multiple 3D models and the first label
The label information of image and each 2D image；The label information of multiple 2D images and each 2D image is for training deep learning model.
It optionally, include variform and/or type and/or shape in the corresponding image information of object to be identified wait know
Multiple 2D/3D images corresponding to other object.
The 3D model generation module 520 includes:
Scene setting submodule, for utilizing preset scene setting rule, multiple fields of the corresponding object to be identified of setting
Scape.
Image procossing submodule, for carrying out drawing processing to image information according to multiple scenes, to generate for drawing
The target image of 3D model, processing of drawing include at least one of translation, rotation and deformation.
Image Rendering submodule draws multiple 3D moulds of the target image in multiple scenes for calling 3D drawing algorithm
Type.
Wherein, 3D drawing algorithm includes at least one of texture-mapping algorithm and Shadow Mapping algorithm.
Optionally, 2D data obtains module 540, comprising:
Submodule is projected, for carrying out 2D projection to multiple 3D models, to obtain multiple 2D images of preset format, is preset
Format includes picture format or video format；
Label determines submodule, for determining label information using the first label is corresponding according to multiple 2D images.
Optionally, label information includes location information, dimension information and type of the object to be identified in each 2D image
At least one of information, the label determine submodule, comprising:
Size positions determination unit, projection relation and the first label when for based on to the progress 2D projection of 3D model,
Obtain the dimension information and/or location information of object to be identified in the projected；And/or in the first label for for identifying wait know
In the case where the type of other object, using type as type information.
Information determination unit, dimension information and/or location information and/or type information for will acquire are believed as label
Breath.
In addition, the embodiment of the present invention also provides a kind of terminal, including processor, memory, storage is on a memory and can
The computer program run in processing, the computer program realize that above-mentioned data capture method is implemented when being executed by processor
Each process of example, and identical technical effect can be reached, to avoid repeating, which is not described herein again.
The embodiment of the present invention also provides a kind of computer readable storage medium, and meter is stored on computer readable storage medium
Calculation machine program, the computer program realize each process of above-mentioned data capture method embodiment when being executed by processor, and
Identical technical effect can be reached, to avoid repeating, which is not described herein again.Wherein, the computer readable storage medium, can
Think read-only memory (Read-Only Memory, abbreviation ROM), random access memory (Random Access Memory,
Abbreviation RAM), magnetic or disk etc..
For above-mentioned apparatus embodiment, since it is basically similar to the method embodiment, so be described relatively simple,
The relevent part can refer to the partial explaination of embodiments of method.
All the embodiments in this specification are described in a progressive manner, the highlights of each of the examples are with
The difference of other embodiments, the same or similar parts between the embodiments can be referred to each other.
It would have readily occurred to a person skilled in the art that: any combination application of above-mentioned each embodiment is all feasible, therefore
Any combination between above-mentioned each embodiment is all embodiment of the present invention, but this specification exists as space is limited,
This is not just detailed one by one.
Provided herein data capture method not with any certain computer, virtual system or the intrinsic phase of other equipment
It closes.Various general-purpose systems can also be used together with teachings based herein.As described above, construction has present invention side
Structure required by the system of case is obvious.In addition, the present invention is also not directed to any particular programming language.It should be bright
It is white, it can use various programming languages and realize summary of the invention described herein, and retouched above to what language-specific was done
State is in order to disclose the best mode of carrying out the invention.
In the instructions provided here, numerous specific details are set forth.It is to be appreciated, however, that implementation of the invention
Example can be practiced without these specific details.In some instances, well known method, structure is not been shown in detail
And technology, so as not to obscure the understanding of this specification.
Similarly, it should be understood that in order to simplify the present invention and help to understand one or more of the various inventive aspects,
Above in the description of exemplary embodiment of the present invention, each feature of the invention is grouped together into single implementation sometimes
In example, figure or descriptions thereof.However, the disclosed method should not be interpreted as reflecting the following intention: i.e. required to protect
Shield the present invention claims features more more than feature expressly recited in each claim.More precisely, such as right
As claim reflects, inventive aspect is all features less than single embodiment disclosed above.Therefore, it then follows tool
Thus claims of body embodiment are expressly incorporated in the specific embodiment, wherein each claim conduct itself
Separate embodiments of the invention.
Those skilled in the art will understand that can be carried out adaptively to the module in the equipment in embodiment
Change and they are arranged in one or more devices different from this embodiment.It can be the module or list in embodiment
Member or component are combined into a module or unit or component, and furthermore they can be divided into multiple submodule or subelement or
Sub-component.Other than such feature and/or at least some of process or unit exclude each other, it can use any
Combination is to all features disclosed in this specification (including adjoint claim, abstract and attached drawing) and so disclosed
All process or units of what method or apparatus are combined.Unless expressly stated otherwise, this specification is (including adjoint power
Benefit require, abstract and attached drawing) disclosed in each feature can carry out generation with an alternative feature that provides the same, equivalent, or similar purpose
It replaces.
In addition, it will be appreciated by those of skill in the art that although some embodiments described herein include other embodiments
In included certain features rather than other feature, but the combination of the feature of different embodiments mean it is of the invention
Within the scope of and form different embodiments.For example, in detail in the claims, embodiment claimed it is one of any
Can in any combination mode come using.
Various component embodiments of the invention can be implemented in hardware, or to run on one or more processors
Software module realize, or be implemented in a combination thereof.It will be understood by those of skill in the art that can be used in practice
Microprocessor or digital signal processor (DSP) come realize some in data capture method according to an embodiment of the present invention or
The some or all functions of person's whole component.The present invention is also implemented as one for executing method as described herein
Point or whole device or device programs (for example, computer program and computer program product).Such this hair of realization
Bright program can store on a computer-readable medium, or may be in the form of one or more signals.It is such
Signal can be downloaded from an internet website to obtain, and is perhaps provided on the carrier signal or is provided in any other form.
It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and ability
Field technique personnel can be designed alternative embodiment without departing from the scope of the appended claims.In the claims,
Any reference symbol between parentheses should not be configured to limitations on claims.Word "comprising" does not exclude the presence of not
Element or step listed in the claims.Word "a" or "an" located in front of the element does not exclude the presence of multiple such
Element.The present invention can be by means of including the hardware of several different elements and being come by means of properly programmed computer real
It is existing.In the unit claims listing several devices, several in these devices can be through the same hardware branch
To embody.The use of word first, second, and third does not indicate any sequence.These words can be explained and be run after fame
Claim.
Claims (11)
1. a kind of data capture method, which is characterized in that the described method includes:
Obtain object to be identified and the corresponding image information of the object to be identified；
According to the object to be identified and described image information, the corresponding multiple 3D models of the object to be identified are generated, it is described
Multiple 3D models include 3D rendering of multiple objects to be identified under multiple scenes；
According to the multiple 3D model, position, size and the class that the object to be identified is identified in each 3D model are acquired respectively
First label of at least one of type；
According to the multiple 3D model and first label, the corresponding multiple 2D images of object to be identified and each 2D are obtained
The label information of image；The label information of the multiple 2D image and each 2D image is for training deep learning model.
2. the method according to claim 1, wherein including more in the corresponding image information of the object to be identified
Multiple 2D/3D images corresponding to the object to be identified of kind form and/or type and/or shape.
3. according to the method described in claim 2, it is characterized in that, described believe according to the object to be identified and described image
Breath generates the corresponding multiple 3D models of the object to be identified, comprising:
Utilize preset scene setting rule, the multiple scene of the corresponding object to be identified of setting；
According to the multiple scene, drawing processing is carried out to described image information, to generate the mesh for drawing the 3D model
Logo image, the drawing processing include at least one of translation, rotation and deformation；
3D drawing algorithm is called, the multiple 3D model of the target image in the multiple scene is drawn；
Wherein, the 3D drawing algorithm includes at least one of texture-mapping algorithm and Shadow Mapping algorithm.
4. the method according to claim 1, wherein described mark according to the multiple 3D model with described first
Label obtain multiple 2D images of the object to be identified and the label information of each 2D image, comprising:
2D projection is carried out to the multiple 3D model, to obtain the multiple 2D image of preset format, the preset format packet
Include picture format or video format；
According to the multiple 2D image, the label information is determined using first label is corresponding.
5. according to the method described in claim 4, it is characterized in that, the label information includes the object to be identified described
At least one of location information, dimension information and type information in each 2D image, it is described according to the multiple 2D image,
The label information is determined using first label is corresponding, comprising:
In the case where first label is size and/or position for identifying the object to be identified, based on to 3D mould
Type carries out the projection relation and first label when 2D projection, obtains size of the object to be identified after the projection
Information and/or location information；And/or in the case where first label is the type for identifying the object to be identified,
Using the type as the type information；
The dimension information and/or location information and/or type information that will acquire are as the label information.
6. a kind of data acquisition facility, which is characterized in that described device includes:
Object obtains module, for obtaining object to be identified and the corresponding image information of the object to be identified；
3D model generation module, for generating the object pair to be identified according to the object to be identified and described image information
The multiple 3D models answered, the multiple 3D model include 3D rendering of multiple objects to be identified under multiple scenes；
Label acquisition module, for acquiring identify the object to be identified in each 3D model respectively according to the multiple 3D model
First label of at least one of position, size and the type of body；
2D data obtains module, for obtaining the object to be identified according to the multiple 3D model and first label
The label information of multiple 2D images and each 2D image；The label information of the multiple 2D image and each 2D image is for training depth
Learning model.
7. device according to claim 6, which is characterized in that include more in the corresponding image information of the object to be identified
Multiple 2D/3D images corresponding to the object to be identified of kind form and/or type and/or shape.
8. device according to claim 7, which is characterized in that the 3D model generation module, comprising:
Scene setting submodule, for using preset scene setting rule, setting to correspond to the described more of the object to be identified
A scene；
Image procossing submodule is used for for carrying out drawing processing to described image information according to the multiple scene with generating
The target image of the 3D model is drawn, the drawing processing includes at least one of translation, rotation and deformation；
Image Rendering submodule, for calling 3D drawing algorithm, draw the target image in the multiple scene described in
Multiple 3D models；
Wherein, the 3D drawing algorithm includes at least one of texture-mapping algorithm and Shadow Mapping algorithm.
9. device according to claim 6, which is characterized in that the 2D data obtains module, comprising:
Submodule is projected, for carrying out 2D projection to the multiple 3D model, to obtain the multiple 2D image of preset format,
The preset format includes picture format or video format；
Label determines submodule, for determining that the label is believed using first label is corresponding according to the multiple 2D image
Breath.
10. device according to claim 9, which is characterized in that the label information includes the object to be identified in institute
At least one of location information, dimension information and the type information in each 2D image are stated, the label determines submodule, packet
It includes:
Size positions determination unit, projection relation and first label when for based on to the progress 2D projection of 3D model,
Obtain dimension information and/or location information of the object to be identified after the projection；And/or it is in first label
In the case where type for identifying the object to be identified, using the type as the type information；
Information determination unit, the dimension information and/or location information and/or type information for will acquire are as the mark
Sign information.
11. a kind of computer readable storage medium, which is characterized in that store computer journey on the computer readable storage medium
Sequence, the data capture method as described in any one of claims 1 to 5 is realized when the computer program is executed by processor
Step.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811615761.2A CN109829969A (en) | 2018-12-27 | 2018-12-27 | A kind of data capture method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811615761.2A CN109829969A (en) | 2018-12-27 | 2018-12-27 | A kind of data capture method, device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109829969A true CN109829969A (en) | 2019-05-31 |
Family
ID=66860582
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811615761.2A Pending CN109829969A (en) | 2018-12-27 | 2018-12-27 | A kind of data capture method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109829969A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111159506A (en) * | 2019-12-26 | 2020-05-15 | 广州信天翁信息科技有限公司 | Data validity identification method, device and equipment and readable storage medium |
| CN111402401A (en) * | 2020-03-13 | 2020-07-10 | 北京华捷艾米科技有限公司 | A method for acquiring 3D face data, a face recognition method and device |
| CN111401423A (en) * | 2020-03-10 | 2020-07-10 | 北京百度网讯科技有限公司 | Data processing method and apparatus for autonomous vehicle |
| CN111680758A (en) * | 2020-06-15 | 2020-09-18 | 杭州海康威视数字技术股份有限公司 | Image training sample generation method and device |
| CN112269385A (en) * | 2020-10-23 | 2021-01-26 | 北京理工大学 | Cloud unmanned vehicle dynamics control system and method |
| CN112541445A (en) * | 2020-12-16 | 2021-03-23 | 中国联合网络通信集团有限公司 | Facial expression migration method and device, electronic equipment and storage medium |
| CN112651881A (en) * | 2020-12-30 | 2021-04-13 | 北京百度网讯科技有限公司 | Image synthesis method, apparatus, device, storage medium, and program product |
| CN112785681A (en) * | 2019-11-07 | 2021-05-11 | 杭州睿琪软件有限公司 | Method and device for generating 3D image of pet |
| CN113963127A (en) * | 2021-12-22 | 2022-01-21 | 深圳爱莫科技有限公司 | Simulation engine-based model automatic generation method and processing equipment |
| CN115223162A (en) * | 2022-08-22 | 2022-10-21 | 北京海舶无人船科技有限公司 | Floater identification method and system |
| WO2023028777A1 (en) * | 2021-08-30 | 2023-03-09 | 西门子（中国）有限公司 | Method and apparatus for automatically generating labeled image, and computer storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170161590A1 (en) * | 2015-12-07 | 2017-06-08 | Dassault Systemes | Recognition of a 3d modeled object from a 2d image |
| US20170304732A1 (en) * | 2014-11-10 | 2017-10-26 | Lego A/S | System and method for toy recognition |
| CN107578477A (en) * | 2017-09-11 | 2018-01-12 | 南京大学 | A kind of automatic testing method of threedimensional model part |
| CN108961422A (en) * | 2018-06-27 | 2018-12-07 | 百度在线网络技术（北京）有限公司 | The labeling method and device of threedimensional model |
-
2018
- 2018-12-27 CN CN201811615761.2A patent/CN109829969A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170304732A1 (en) * | 2014-11-10 | 2017-10-26 | Lego A/S | System and method for toy recognition |
| US20170161590A1 (en) * | 2015-12-07 | 2017-06-08 | Dassault Systemes | Recognition of a 3d modeled object from a 2d image |
| CN107578477A (en) * | 2017-09-11 | 2018-01-12 | 南京大学 | A kind of automatic testing method of threedimensional model part |
| CN108961422A (en) * | 2018-06-27 | 2018-12-07 | 百度在线网络技术（北京）有限公司 | The labeling method and device of threedimensional model |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112785681B (en) * | 2019-11-07 | 2024-03-08 | 杭州睿琪软件有限公司 | Method and device for generating 3D image of pet |
| CN112785681A (en) * | 2019-11-07 | 2021-05-11 | 杭州睿琪软件有限公司 | Method and device for generating 3D image of pet |
| CN111159506A (en) * | 2019-12-26 | 2020-05-15 | 广州信天翁信息科技有限公司 | Data validity identification method, device and equipment and readable storage medium |
| CN111159506B (en) * | 2019-12-26 | 2023-11-14 | 广州信天翁信息科技有限公司 | Data validity identification method, device, equipment and readable storage medium |
| CN111401423B (en) * | 2020-03-10 | 2023-05-26 | 北京百度网讯科技有限公司 | Data processing method and device for automatic driving vehicle |
| CN111401423A (en) * | 2020-03-10 | 2020-07-10 | 北京百度网讯科技有限公司 | Data processing method and apparatus for autonomous vehicle |
| CN111402401A (en) * | 2020-03-13 | 2020-07-10 | 北京华捷艾米科技有限公司 | A method for acquiring 3D face data, a face recognition method and device |
| CN111402401B (en) * | 2020-03-13 | 2023-08-18 | 北京华捷艾米科技有限公司 | A method for acquiring 3D face data, a face recognition method and device |
| CN111680758A (en) * | 2020-06-15 | 2020-09-18 | 杭州海康威视数字技术股份有限公司 | Image training sample generation method and device |
| CN111680758B (en) * | 2020-06-15 | 2024-03-05 | 杭州海康威视数字技术股份有限公司 | Image training sample generation method and device |
| CN112269385A (en) * | 2020-10-23 | 2021-01-26 | 北京理工大学 | Cloud unmanned vehicle dynamics control system and method |
| CN112269385B (en) * | 2020-10-23 | 2021-09-07 | 北京理工大学 | Cloud unmanned vehicle dynamics control system and method |
| CN112541445B (en) * | 2020-12-16 | 2023-07-18 | 中国联合网络通信集团有限公司 | Transfer method, device, electronic equipment and storage medium of human facial expression |
| CN112541445A (en) * | 2020-12-16 | 2021-03-23 | 中国联合网络通信集团有限公司 | Facial expression migration method and device, electronic equipment and storage medium |
| CN112651881B (en) * | 2020-12-30 | 2023-08-01 | 北京百度网讯科技有限公司 | Image synthesizing method, apparatus, device, storage medium, and program product |
| CN112651881A (en) * | 2020-12-30 | 2021-04-13 | 北京百度网讯科技有限公司 | Image synthesis method, apparatus, device, storage medium, and program product |
| WO2023028777A1 (en) * | 2021-08-30 | 2023-03-09 | 西门子（中国）有限公司 | Method and apparatus for automatically generating labeled image, and computer storage medium |
| CN113963127A (en) * | 2021-12-22 | 2022-01-21 | 深圳爱莫科技有限公司 | Simulation engine-based model automatic generation method and processing equipment |
| CN115223162A (en) * | 2022-08-22 | 2022-10-21 | 北京海舶无人船科技有限公司 | Floater identification method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829969A (en) | A kind of data capture method, device and storage medium | |
| Wu et al. | Single image 3d interpreter network | |
| US11887256B2 (en) | Deferred neural rendering for view extrapolation | |
| CN111783525A (en) | A method for generating target samples of aerial photography images based on style transfer | |
| CN117671138A (en) | Digital twin modeling method and system based on SAM large model and NeRF | |
| CN113822982A (en) | Human body three-dimensional model construction method and device, electronic equipment and storage medium | |
| CN115457188A (en) | A gaze-based 3D rendering display method and system | |
| CN113689578A (en) | A method and device for generating human data set | |
| Ijiri et al. | Digitization of natural objects with micro CT and photographs | |
| JP2006506164A5 (en) | ||
| CN114842127B (en) | Terrain rendering method and device, electronic equipment, medium and product | |
| Ramanarayanan et al. | Feature-based textures | |
| CN115239861A (en) | Face data enhancement method and device, computer equipment and storage medium | |
| CN119206006B (en) | Three-dimensional model data processing method, device, equipment, medium and product | |
| Smit et al. | The online anatomical human: web-based anatomy education. | |
| CN120219640B (en) | A three-dimensional reconstruction method and system of nipple protection area based on super-resolution | |
| JP2024064413A (en) | Image data generating device, image data generating method, and image data generating program | |
| CN116469101A (en) | Data labeling method, device, electronic equipment and storage medium | |
| CN118537472B (en) | An optimized rendering interaction method for the fusion of neural radiation field and three-dimensional system | |
| CN112652056B (en) | 3D information display method and device | |
| CN112669446B (en) | Building scene modeling method and device | |
| CN114120000B (en) | Method, device and system for automatically generating significance detection task annotation data | |
| US8520926B2 (en) | Method for processing image and volume data based on statistical models | |
| CN113496468B (en) | Depth image restoration method, device and storage medium | |
| CN117237524A (en) | Method and apparatus for providing a three-dimensional model of an object |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190531 |
|
| RJ01 | Rejection of invention patent application after publication |