CN108960232A - Model training method, device, electronic equipment and computer readable storage medium - Google Patents
Model training method, device, electronic equipment and computer readable storage medium Download PDFInfo
- Publication number
- CN108960232A CN108960232A CN201810590113.XA CN201810590113A CN108960232A CN 108960232 A CN108960232 A CN 108960232A CN 201810590113 A CN201810590113 A CN 201810590113A CN 108960232 A CN108960232 A CN 108960232A
- Authority
- CN
- China
- Prior art keywords
- deep learning
- learning model
- detection information
- image data
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Abstract
This application involves a kind of model training method, device, electronic equipment and computer readable storage mediums.The above method includes: A: image data being inputted preset deep learning model, obtains the first detection information that the deep learning model detects described image data；B: the second detection information being corrected to first detection information is obtained；C: the deep learning model is trained according to described image data and corresponding second detection information；D: the convergence result of deep learning model after training is obtained；E: when the convergence result is unsatisfactory for the default condition of convergence, iteration executes step A to step D and presets the condition of convergence until the convergence result meets.The above method, training data using the image data of deep learning model mark as training, the training dataset that deep learning model is trained is marked without artificial, the workload of artificial labeled data training set is reduced, saves the cost of trained deep learning model.
Description
Technical field
This application involves field of computer technology is arrived, a kind of model training method, device, electronic equipment are especially related to
And computer readable storage medium.
Background technique
It is more and more normal using machine learning techniques processing big data with the development of big data and machine learning techniques
See.Under normal conditions, the machine learning model that machine learning techniques building can be used handles big data, is using machine
Before learning model handles big data, it usually needs be trained, detect and verify to machine learning model, in machine
Learning model is again handled big data with machine learning model after being verified.
Summary of the invention
The embodiment of the present application provides a kind of model training method, device, electronic equipment and computer readable storage medium, can
To save the cost being trained to model, the efficiency being trained to model is improved.
A kind of model training method, comprising:
A: inputting preset deep learning model for image data, obtains the deep learning model to described image data
The first detection information detected；
B: the second detection information being corrected to first detection information is obtained；
C: the deep learning model is trained according to described image data and corresponding second detection information；
D: the convergence result of deep learning model after training is obtained；
E: when the convergence result is unsatisfactory for the default condition of convergence, iteration executes step A to step D until the convergence
As a result meet the default condition of convergence.
A kind of model training apparatus, comprising:
Detection module obtains the deep learning model pair for image data to be inputted preset deep learning model
The first detection information that described image data are detected；
First obtains module, for obtaining the second detection information being corrected to first detection information；
Training module, for according to described image data and corresponding second detection information to the deep learning model into
Row training；
Second obtain module, for obtain training after deep learning model convergence result；
Iteration module, for when the convergence result is unsatisfactory for the default condition of convergence, by the detection module, described the
One obtains module, the training module and described second obtains the respective function of module iteration execution until the convergence result is full
The default condition of convergence of foot.
A kind of electronic equipment, including memory and processor store computer program, the calculating in the memory
When machine program is executed by the processor, so that the step of processor executes method as described above.
A kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of method as described above is realized when being executed by processor.
Method, apparatus, electronic equipment and computer readable storage medium in the embodiment of the present application, using deep learning model
Image is detected, then the first detection information of image gone out to deep learning model inspection is corrected automatically, is obtained pair
The second detection information that above-mentioned first detection information is corrected；The image data conduct marked using deep learning model
To the training data that deep learning model is trained, the training dataset that deep learning model is trained is marked without artificial
Note, greatly reduces the workload of artificial labeled data training set, saves the cost being trained to deep learning model.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of application for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the flow chart of image processing method in one embodiment；
Fig. 2 is the flow chart of image processing method in another embodiment；
Fig. 3 is the flow chart of image processing method in another embodiment；
Fig. 4 is the flow chart of image processing method in another embodiment；
Fig. 5 is the structural block diagram of image processing apparatus in one embodiment；
Fig. 6 is the structural block diagram of image processing apparatus in another embodiment；
Fig. 7 is the schematic diagram of image processing circuit in one embodiment.
Specific embodiment
It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understood
The application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, and
It is not used in restriction the application.
Electronic equipment can carry out image procossing to image after getting image.For example, being carried out at saturation degree to image
Reason, brightness of image processing, picture contrast processing and image color processing etc..Electronic equipment can also according to image taking when
Quarter, the place of image taking, face etc. is grouped processing to image in image.Image procossing of the above-mentioned electronic equipment to image
It is the camera parameter of color space and electronic equipment based on image to image progress parameter adjustment.Further, electronic equipment
Can also scene Recognition be carried out to image, identify the corresponding scene type of above-mentioned image, and then according to the corresponding scene type of image
Global treatment or Local treatment are carried out to image.Wherein, electronic equipment can carry out scene knowledge to image according to deep learning model
Not.Above-mentioned deep learning model refers to the model constructed according to deep learning algorithm, above-mentioned deep learning model analog human brain
In be used for analytic learning neural network, analyzed further according to data of simulated person's brain mechanism to input.Using depth
Before learning model analyzes the data of input, it usually needs be trained to deep learning model, make deep learning mould
The convergence result of type reaches the default condition of convergence.
Under normal conditions, when being trained to above-mentioned deep learning model, the data after artificial treatment can be inputted
Deep learning model is stated, above-mentioned deep learning model is made to be learnt and be analyzed according to the data after above-mentioned artificial treatment, so that
The convergence result of above-mentioned deep learning model reaches the default condition of convergence.For example, when above-mentioned deep learning model is for image
When the target detection model of scene Recognition, electronic equipment will can manually mark image input target detection model, above-mentioned artificial mark
Note image refers to the image being labeled to the position of target subject in image and classification, and above-mentioned target detection model then can root
Learnt and analyzed according to the artificial mark image of input.It is above-mentioned when the sample of above-mentioned target detection model learning is enough
Target detection model can position to target subject in image and classification carry out automatic identification.Under normal conditions, to depth
Need the artificial treatment data of magnanimity as sample when learning model is trained, so as to cause to deep learning model training
Higher cost.
Fig. 1 is model training method in one embodiment characterized by comprising
Step 102, image data is inputted into preset deep learning model, obtain deep learning model to image data into
The first detection information that row detection obtains.
Above-mentioned deep learning model is the target detection model for image scene identification, and electronic equipment is by above-mentioned picture number
According to input deep learning model after, above-mentioned deep learning model can according to preset algorithm of target detection to above-mentioned image data into
Row target detection identifies the position of the target subject for including in each image and each target subject, and according in image
The position of target subject marks out target subject in image to come.Optionally, above-mentioned deep learning model can be by target in image
Main body is highlighted with rectangle frame, and marks the classification of each target subject in image.
Above-mentioned image data can be inputted preset deep learning model, obtained by electronic equipment after getting image data
Deep learning model is taken to carry out the testing result of target detection, i.e. the first detection information to above-mentioned image data.Above-mentioned picture number
According to the image for as needing to carry out target detection；The classification of above-mentioned target subject is preset classification, it may include: portrait,
Baby, cat, dog, cuisines, blue sky, greenweed, sandy beach etc..It include the target subject for including in above-mentioned first detection information in image
The position of classification and each target subject.
Step 104, the second detection information being corrected to the first detection information is obtained.
After getting the first detection information of image data, above-mentioned first detection information can be corrected to obtain second
Detection information.Above-mentioned be corrected to the first detection information includes: to be identified to target subject out unidentified in image, is right
The target subject of target subject classification identification mistake is corrected, to the mesh of identification mistake in target subject position in image in image
Mark main body is corrected.When the first detection information of image is wrong, above-mentioned first detection information is corrected available
Second detection information, i.e., above-mentioned second detection information are the corresponding target subject classification of image and target subject position letter after correction
Breath.Wherein, it can be corrected by first detection information of the electronic equipment to above-mentioned image data, it can also be by manually to above-mentioned image
First detection information of data is corrected.
Step 106, deep learning model is trained according to image data and corresponding second detection information.
After getting the second detection information being corrected to the first detection information, electronic equipment can be searched above-mentioned
The corresponding image data of second detection information, and above-mentioned image data and corresponding second detection information are inputted into above-mentioned depth again
Spend learning model.The corresponding image data of above-mentioned second detection information is deep learning model to target subject classification in image
Identify the image data of mistake, above-mentioned image data and corresponding second detection information are inputted deep learning model by electronic equipment
Target subject classification information correct in image and image is inputted into deep learning model.
Electronic equipment can be using corresponding second detection information of above-mentioned image data and image data as above-mentioned deep learning
The training set of model, then above-mentioned deep learning model can be according to corresponding second detection information of above-mentioned image data and image data
Above-mentioned deep learning model is trained.Wherein, deep learning model is trained and refers to and will has already passed through target detection
Image data input deep learning model so that deep learning model can be according to the target master detected in above-mentioned image data
The location information of body and target subject is learnt, is optimized, and the accuracy of detection is improved.
Step 108, the convergence result of deep learning model after training is obtained.
After being trained to above-mentioned deep learning model, can obtain training after deep learning model convergence as a result,
Whether the convergence result for detecting deep learning model after above-mentioned training reaches the default condition of convergence.Wherein, to deep learning model
After being trained, deep learning model can increasingly restrain, and above-mentioned convergence result can indicate the precision of above-mentioned deep learning model.
It, i.e., can when the precision of learning model reaches default precision after above-mentioned training when above-mentioned convergence result meets the default condition of convergence
The target detection of image data is carried out using the deep learning model after above-mentioned training.For example, default precision is deep learning mould
Type is 95% to the accuracy that image data identifies, when the precision of deep learning model is not less than above-mentioned 95% after training, then
The convergence result of deep learning model meets the default condition of convergence after training, can be carried out according to deep learning model after above-mentioned training
The target detection of image data.
Step 110, when convergence result is unsatisfactory for the default condition of convergence, 102 are returned to step until convergence result is full
The default condition of convergence of foot.
When the above-mentioned convergence result not up to default condition of convergence, can iteration execute above-mentioned steps 102 to above-mentioned steps
108: obtaining image data and input deep learning model, obtain what above-mentioned deep learning model detected image data
First detection information；The second detection information being corrected to the first detection information of above-mentioned image data, root are obtained again
Above-mentioned deep learning model is trained according to the second detection information of image data and image data, after obtaining above-mentioned training
Deep learning model convergence result.After i.e. deep learning model detects image, deep learning model pair can be obtained
Image, which carries out the image information that target detection obtains, further can also obtain the image information obtained to above-mentioned target detection
Image information after the correction being corrected, deep learning model are trained further according to image information after above-mentioned correction, mention
The precision of high above-mentioned deep learning model.
By taking deep learning model is target detection model as an example, image is inputted into above-mentioned target detection model, then above-mentioned mesh
Mark detection model target subject and can mark out the position of target subject in image in the above-mentioned image of automatic identification, and export to figure
The first detection information obtained as carrying out target detection.Manual verification can be carried out to above-mentioned first detection information, to above-mentioned image
The target subject of middle marking error, the target subject not marked etc. are corrected, and obtain the first detection information to above-mentioned image
The second detection information being corrected.It, can be by above-mentioned image data and corresponding after getting the second detection information
Two detection informations are input to target detection model, according to above-mentioned image data and corresponding second detection information to target detection mould
Type is trained, and promotes the precision of above-mentioned target detection model.The precision for obtaining target detection model after training, in target detection
When the precision of model reaches default precision, above-mentioned target detection model treatment image can be used；When above-mentioned target detection model
When precision not up to presets precision, repeatable above-mentioned steps are trained until above-mentioned target detection model target detection model
Precision reach default precision.
Under normal conditions, when being trained to deep learning model, artificial labeled data training set is needed, further according to people
Work mark data training set deep learning model is trained so that the convergence result of deep learning model meet it is above-mentioned pre-
If the condition of convergence, and deep learning model is trained and usually requires mass data, therefore using the training of artificial labeled data
The method of collection trains the higher cost of deep learning model.
Method in the embodiment of the present application detects image using deep learning model automatically, then to deep learning mould
First detection information of the image that type detects is corrected, and obtains be corrected to above-mentioned first detection information second
Detection information, then above-mentioned image and corresponding second detection information are inputted into deep learning model, deep learning model is carried out
Training；Using the image data of deep learning model mark as the training data being trained to deep learning model, to depth
The training dataset that degree learning model is trained greatly reduces the work of artificial labeled data training set without artificial mark
Amount, saves the cost being trained to deep learning model.
In one embodiment, deep learning model is trained according to image data and corresponding second detection information
It include: that data training set is added in image data and corresponding second detection information；Data training set includes identified target master
The image data of body classification and target subject location information；Deep learning model is trained according to data training set.
When being trained to deep learning model, it will usually using the image data that manually marks as data sample,
Deep learning model is trained according to above-mentioned data sample.Getting corresponding second inspection of image data i.e. image data
After measurement information, data training set can be added in above-mentioned image data and corresponding second detection information, be instructed further according to above-mentioned data
Practice collection to be trained deep learning model, i.e., after deep learning model carries out target detection to image data, obtains picture number
According to the first detection information.If the first detection information of above-mentioned image data is wrong, above-mentioned image data can be corrected, be obtained
Corresponding second detection information of above-mentioned image data is taken, then above-mentioned second detection information includes the target master after image data correction
Body classification and target subject location information.After data training set is added in above-mentioned image data and corresponding second detection information,
Deep learning model can be trained according to above-mentioned data training set, i.e., according to label target principal classes and target subject
The image data of location information is trained deep learning model, improves the convergence result of above-mentioned deep learning model.
Data training set is added in image data and corresponding second detection information by method in the embodiment of the present application, according to
Data training set is trained deep learning model, reduces the workload of artificial statement data sample, improves to depth
The efficiency of learning model training.
In one embodiment, a kind of model training method includes:
Step 202, data training set is obtained.
Step 204, deep learning model is trained according to data training set, so that the precision of deep learning model reaches
To first threshold.
Step 206, image data is inputted into preset deep learning model, obtains the deep learning model to the figure
The first detection information detected as data.
Step 208, the second detection information being corrected to first detection information is obtained.
Step 210, the deep learning model is instructed according to described image data and corresponding second detection information
Practice.
Step 212, the convergence result of deep learning model after training is obtained.
Step 214, when the convergence result is unsatisfactory for the default condition of convergence, 206 are returned to step until the receipts
It holds back result and meets the default condition of convergence.
It is deep if the precision of deep learning model is too low when being detected using deep learning model to image data
The information errors rate that degree learning model obtains image data progress target detection is higher, therefore, is using deep learning model
Before being detected to image data, needs to be trained deep learning model, deep learning model accuracy is made to reach first
Threshold value.The precision of above-mentioned deep learning model refers to that deep learning model carries out the obtained information of target detection to image data
Accuracy, i.e. the convergence result of deep learning model.When the precision of deep learning model is higher, i.e. deep learning model is to image
The accuracy that data carry out the information that target detection obtains is higher.
The method being trained to above-mentioned deep learning model includes: acquisition data training set, according to the training of above-mentioned data
Collection is trained deep learning model.Above-mentioned data training set refers to target subject classification and target subject in identified image
The sets of image data of position.Being trained according to above-mentioned data training set to deep learning model includes: to instruct above-mentioned data
Practice and image data and the corresponding target subject classification of image data and target subject location information concentrated to input deep learning model,
Learn deep learning model according to above-mentioned image data information, improves deep learning model to target subject class in image
It Shi Bie not be with the precision of target subject position identification.After being trained to above-mentioned deep learning model, it can obtain deep after training
Whether the precision for spending learning model, the precision for detecting deep learning model after above-mentioned training reach first threshold, deep after training
When the precision of degree learning model reaches first threshold, image data is inputted into above-mentioned preset deep learning model；After training
When the precision of deep learning model is not up to first threshold, image data sample in data training set is replaced, according to the figure of gains in depth of comprehension
As data sample is trained above-mentioned deep learning model, until the precision of above-mentioned deep learning model reaches first threshold.
When the above-mentioned default condition of convergence is expressed as precision, the above-mentioned above-mentioned default condition of convergence of first threshold first；For example, when above-mentioned pre-
If the condition of convergence is deep learning, model recognition correct rate is 95%, then above-mentioned first threshold can be for the identification of deep learning model just
True rate is 80%.First deep learning model is trained using image pattern in data training set, so that deep learning mould
Type reaches 80% to the recognition correct rate of image data, detects further according to deep learning model after training to image data,
The first detection information for obtaining image data after detecting, is corrected to obtain second to the first detection information of above-mentioned image data
Detection information is again trained above-mentioned deep learning model according to image data and corresponding second detection information, improves
The convergence result of deep learning model.
Method in the embodiment of the present application, before being detected using deep learning model to image data, first to depth
Learning model is trained, and is promoted the precision of deep learning model, that is, is reduced the sample being trained to deep learning model
Quantity, but make deep learning model to image carry out target detection when precision will not be too low, reduce to deep learning model
The cost being trained.
In one embodiment, data training set is to carry out target inspection to sample image according to preset target detection standard
The set of the image information measured.
Before obtaining data training set, need to obtain target detection standard, according to above-mentioned target detection standard to image
Target detection is carried out, identifies target subject classification and target subject location information in image.Above-mentioned target detection standard is mesh
Mark principal classes standard and in the picture label target body position standard.It, can root after getting above-mentioned target detection standard
Target detection is carried out to sample image according to above-mentioned target detection standard, obtains target subject classification and target subject in sample data
Location information, then the set of the target subject classification and target subject location information of above-mentioned sample image and sample image be
Data training set.After getting above-mentioned data training set, deep learning model can be instructed according to above-mentioned data training set
Practice.
In one embodiment, before obtaining the second detection information being corrected to the first detection information, also
Include:
Step 302, the third detection information to image data is obtained.
Step 304, if the first detection information and third detection information of image data be not identical, using image data as to
Image correcting data.
Step 306, obtaining includes: according to figure to be corrected to the second detection information that the first detection information is corrected
As the first detection information of data is corrected to obtain the second detection information.
After getting deep learning model and carrying out the first detection information that target detection obtains to image data, can also it obtain
The third detection information of above-mentioned image data is taken, above-mentioned third detection information includes identifying to target subject classification in image data
With the information of target subject position identification.Wherein, above-mentioned third detection information can be for image data progress manual identified acquisition
Target subject classification and target subject location information.It, can be by same figure after getting the third detection information of image data
The first detection information and third detection information of picture compare, by the first detection information and the different figure of third detection information
As data are as image data to be corrected.Above-mentioned image data to be corrected is target subject classification and/or mesh in image data
Mark the image data that body position may be wrong.It, can be to above-mentioned image to be detected after getting above-mentioned image to be detected data
Data are corrected, and obtain the second detection information of above-mentioned image to be detected data.It wherein, can be to above-mentioned image to be detected data
Manual synchronizing is carried out, electronic equipment can also be used, above-mentioned image to be detected data are corrected.
Under normal conditions, when being trained to deep learning model using artificial mark image pattern, according to artificial mark
The image pattern that the image pattern of note is trained to above-mentioned deep learning model, but manually marks can also have error, use
When image pattern with error is trained deep learning model, training result is poor.
Method in the embodiment of the present application, by carrying out pair the first detection information of image data and third detection information
Than the image data that target subject classification and/or target subject position may be wrong in image data can be filtered out, to above-mentioned
Image data is corrected, and is trained further according to the image data after correction to deep learning model, is avoided image data not
The situation for accurately causing deep learning model training effect difference, improves the efficiency to deep learning model training.
In one embodiment, a kind of model training method includes:
Step 402, image data is inputted into preset deep learning model, obtains the deep learning model to the figure
The first detection information detected as data.
Step 404, the second detection information being corrected to first detection information is obtained.
Step 406, the deep learning model is instructed according to described image data and corresponding second detection information
Practice.
Step 408, the convergence result of deep learning model after training is obtained.
Step 410, when the convergence result is unsatisfactory for the default condition of convergence, 402 are returned to step until the receipts
It holds back result and meets the default condition of convergence.
Step 412, when convergence result meets the default condition of convergence, using deep learning model after training to figure to be detected
As carrying out target detection, the classification of target subject and the location information of target subject in image to be detected are identified.
After being trained to above-mentioned deep learning model, if detecting, the convergence result of deep learning model after training is full
The default condition of convergence of foot, then deep learning model can carry out target detection to image to be detected after above-mentioned training.I.e. to above-mentioned depth
After degree learning model is trained, if detecting, the precision of deep learning model after training reaches default precision, can be according to above-mentioned
Deep learning model after training carries out target detection to image to be detected.Above-mentioned image to be detected, which refers to, to be needed to carry out target inspection
The image of survey, it includes: target master in the above-mentioned image to be detected of identification that deep learning model, which carries out target detection to image to be detected,
The classification of body and the location information of target subject.
Method in the embodiment of the present application can be used when the convergence result of deep learning model meets the default condition of convergence
Deep learning model after training carries out target detection to image to be detected, it can be achieved that the fast target to image detects.
In one embodiment, a kind of model training method, comprising:
A: inputting preset deep learning model for image data, obtains deep learning model and detects to image data
The first obtained detection information；
B: the second detection information being corrected to the first detection information is obtained；
C: deep learning model is trained according to image data and corresponding second detection information；
D: the convergence result of deep learning model after training is obtained；
E: when convergence result is unsatisfactory for the default condition of convergence, iteration executes step A to step D until convergence result meets
The default condition of convergence.
In one embodiment, deep learning model is trained according to image data and corresponding second detection information
It include: that data training set is added in image data and corresponding second detection information；Data training set includes identified target master
The image data of body classification and target subject location information；Deep learning model is trained according to data training set.
In one embodiment, before image data to be inputted to preset deep learning model, further includes: obtain data
Training set；Deep learning model is trained according to data training set, so that the precision of deep learning model reaches the first threshold
Value.
In one embodiment, data training set is to carry out target inspection to sample image according to preset target detection standard
The set of the image information measured.
In one embodiment, before obtaining the second detection information being corrected to the first detection information, also
It include: the third detection information obtained to image data；If the first detection information of image data and third detection information not phase
Together, using image data as image data to be corrected；Obtain the second detection information being corrected to the first detection information
It include: to be corrected to obtain the second detection information according to the first detection information of image data to be corrected.
In one embodiment, the above method further include: when convergence result meets the default condition of convergence, after training
Deep learning model carries out target detection to image to be detected, identifies the classification and target subject of target subject in image to be detected
Location information.
It should be understood that although each step in above-mentioned flow chart is successively shown according to the instruction of arrow, this
A little steps are not that the inevitable sequence according to arrow instruction successively executes.Unless expressly state otherwise herein, these steps
It executes there is no the limitation of stringent sequence, these steps can execute in other order.Moreover, in above-mentioned flow chart at least
A part of step may include that perhaps these sub-steps of multiple stages or stage are not necessarily in same a period of time to multiple sub-steps
Quarter executes completion, but can execute at different times, the execution in these sub-steps or stage be sequentially also not necessarily according to
Secondary progress, but in turn or can replace at least part of the sub-step or stage of other steps or other steps
Ground executes.
Fig. 5 is the structural block diagram of model training apparatus in one embodiment.As shown in figure 5, a kind of model training apparatus, packet
It includes:
Detection module 502 obtains deep learning model to figure for image data to be inputted preset deep learning model
The first detection information detected as data.
First obtains module 504, for obtaining the second detection information being corrected to the first detection information.
Training module 506, for being instructed according to image data and corresponding second detection information to deep learning model
Practice.
Second obtain module 508, for obtain training after deep learning model convergence result.
Iteration module 510, for obtaining mould by detection module, first when convergence result is unsatisfactory for the default condition of convergence
Block, training module and second obtain module iteration and execute respective function until convergence result meets the default condition of convergence.
In one embodiment, training module 506 is according to image data and corresponding second detection information to deep learning
It includes: that data training set is added in image data and corresponding second detection information that model, which is trained,；Data training set includes
The image data of identified target subject classification and target subject location information；According to data training set to deep learning model into
Row training.
In one embodiment, training module 506 be also used to by image data input preset deep learning model it
Before, obtain data training set；Deep learning model is trained according to data training set, so that the precision of deep learning model
Reach first threshold.
In one embodiment, data training set is to carry out target inspection to sample image according to preset target detection standard
The set of the image information measured.
In one embodiment, the first acquisition module 504 is also used to be corrected to obtain to the first detection information in acquisition
The second detection information before, obtain to the third detection information of image data；If the first detection information of image data and the
Three detection informations are not identical, using image data as image data to be corrected；First, which obtains module 504, obtains to the first detection letter
Ceasing the second detection information being corrected includes: to be corrected to obtain according to the first detection information of image data to be corrected
Second detection information.
In one embodiment, a kind of model training apparatus, comprising: detection module 602, first obtains module 604, training
Module 606, second obtains module 608, iteration module 610 and identification module 612.
Identification module 612 is used for when convergence result meets the default condition of convergence, using deep learning model pair after training
Image to be detected carries out target detection, identifies the classification of target subject and the location information of target subject in image to be detected.
The division of modules is only used for for example, in other embodiments in above-mentioned model training apparatus, can be by mould
Type training device is divided into different modules as required, to complete all or part of function of above-mentioned model training apparatus.
Realizing for the modules in model training apparatus provided in the embodiment of the present application can be the shape of computer program
Formula.The computer program can be run in terminal or server.The program module that the computer program is constituted is storable in terminal
Or on the memory of server.When the computer program is executed by processor, method described in the embodiment of the present application is realized
Step.
The embodiment of the present application also provides a kind of computer readable storage mediums.One or more is executable comprising computer
The non-volatile computer readable storage medium storing program for executing of instruction, when computer executable instructions are executed by one or more processors,
So that processor executes the step of model training method in the embodiment of the present application.
The embodiment of the present application also provides a kind of computer program products comprising instruction, when it runs on computers
When, so that the step of computer executes model training method in the embodiment of the present application.
The embodiment of the present application also provides a kind of electronic equipment.It include image processing circuit in above-mentioned electronic equipment, at image
Reason circuit can use hardware and or software component realization, it may include define ISP (Image Signal Processing, figure
As signal processing) the various processing units of pipeline.Fig. 7 is the schematic diagram of image processing circuit in one embodiment.Such as Fig. 7 institute
Show, for purposes of illustration only, only showing the various aspects of image processing techniques relevant to the embodiment of the present application.
As shown in fig. 7, image processing circuit includes the first ISP processor 730, the 2nd ISP processor 740 and control logic
Device 750.First camera 710 includes one or more first lens 712 and the first imaging sensor 714.First image sensing
Device 714 may include colour filter array (such as Bayer filter), and the first imaging sensor 714 can be obtained with the first imaging sensor
The luminous intensity and wavelength information that 714 each imaging pixel captures, and one group for being handled by the first ISP processor 730 is provided
Image data.Second camera 720 includes one or more second lens 722 and the second imaging sensor 724.Second image passes
Sensor 724 may include colour filter array (such as Bayer filter), and the second imaging sensor 724 can be obtained with the second image sensing
The luminous intensity and wavelength information that each imaging pixel of device 724 captures, and can be handled by the 2nd ISP processor 740 one is provided
Group image data.
First image transmitting of the first camera 710 acquisition is handled to the first ISP processor 730, the first ISP processing
It, can be by statistical data (brightness of such as image, the contrast value of image, the face of image of the first image after device 730 handles the first image
Color etc.) it is sent to control logic device 750, control logic device 750 can determine the control ginseng of the first camera 710 according to statistical data
Number, so that the first camera 710 can carry out the operation such as auto-focusing, automatic exposure according to control parameter.First image is by the
One ISP processor 730 can store after being handled into video memory 760, and the first ISP processor 730 can also read figure
As the image that stores in memory 760 is with to handling.In addition, the first image can after ISP processor 730 is handled
It is sent directly to display 770 to be shown, display 770 can also read the image in video memory 760 to be shown
Show.
Wherein, the first ISP processor 730 handles image data pixel by pixel in various formats.For example, each image slices
Element can have the bit depth of 7,10,12 or 14 bits, and the first ISP processor 730 can carry out one or more figures to image data
Statistical information as processing operation, collection about image data.Wherein, image processing operations can be by identical or different bit depth
Precision carries out.
Video memory 760 can be independent dedicated in a part, storage equipment or electronic equipment of memory device
Memory, and may include DMA (Direct Memory Access, direct direct memory access (DMA)) feature.
When receiving from the first 714 interface of imaging sensor, the first ISP processor 730 can carry out one or more
Image processing operations, such as time-domain filtering.Image data that treated can be transmitted to video memory 760, to be shown it
It is preceding to carry out other processing.First ISP processor 730 receives processing data from video memory 760, and carries out to processing data
Image real time transfer in RGB and YCbCr color space.Treated that image data may be output to is aobvious for first ISP processor 730
Show device 770, so that user watches and/or by graphics engine or GPU (Graphics Processing Unit, graphics processor)
It is further processed.In addition, the output of the first ISP processor 730 also can be transmitted to video memory 760, and display 770 can be from
Video memory 760 reads image data.In one embodiment, video memory 760 can be configured to realize one or more
A frame buffer.
The statistical data that first ISP processor 730 determines can be transmitted to control logic device 750.For example, statistical data can wrap
Include automatic exposure, automatic white balance, automatic focusing, flicker detection, black level compensation, 712 shadow correction of the first lens etc. first
714 statistical information of imaging sensor.Control logic device 750 may include the processor for executing one or more routines (such as firmware)
And/or microcontroller, one or more routines can statistical data based on the received, determine the control parameter of the first camera 710
And the first ISP processor 730 control parameter.For example, the control parameter of the first camera 710 may include gain, spectrum assignment
The time of integration, stabilization parameter, flash of light control parameter, 712 control parameter of the first lens (such as focus or zoom focal length) or
The combination etc. of these parameters.ISP control parameter may include for automatic white balance and color adjustment (for example, in RGB process phase
Between) 712 shadow correction parameter of gain level and color correction matrix and the first lens.
Similarly, the second image transmitting that second camera 720 acquires is handled to the 2nd ISP processor 740, and second
After ISP processor 740 handles the first image, can by the statistical data of the second image (brightness of such as image, image contrast value,
The color etc. of image) it is sent to control logic device 750, control logic device 750 can determine second camera 720 according to statistical data
Control parameter, so that second camera 720 can carry out auto-focusing, the operation such as automatic exposure according to control parameter.Second figure
As that can store after the 2nd ISP processor 740 is handled into video memory 760, the 2nd ISP processor 740 can also
To read the image stored in video memory 760 with to handling.In addition, the second image is carried out by ISP processor 740
It can be sent directly to display 770 after processing and shown that display 770 can also read the image in video memory 760
To be shown.Second camera 720 and the 2nd ISP processor 740 also may be implemented such as the first camera 710 and the first ISP
Treatment process described in processor 730.Model training side in the embodiment of the present application can be realized with image processing techniques in Fig. 7
Method.
Any reference to memory, storage, database or other media used in this application may include non-volatile
And/or volatile memory.Suitable nonvolatile memory may include read-only memory (ROM), programming ROM (PROM),
Electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include arbitrary access
Memory (RAM), it is used as external cache.By way of illustration and not limitation, RAM is available in many forms, such as
It is static RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDR SDRAM), enhanced
SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM).
Above embodiments only express the several embodiments of the application, and the description thereof is more specific and detailed, but can not
Therefore it is interpreted as the limitation to the application the scope of the patents.It should be pointed out that for those of ordinary skill in the art,
Without departing from the concept of this application, various modifications and improvements can be made, these belong to the protection model of the application
It encloses.Therefore, the scope of protection shall be subject to the appended claims for the application patent.
Claims (10)
1. a kind of model training method characterized by comprising
A: inputting preset deep learning model for image data, obtains the deep learning model and carries out to described image data
Detect the first obtained detection information；
B: the second detection information being corrected to first detection information is obtained；
C: the deep learning model is trained according to described image data and corresponding second detection information；
D: the convergence result of deep learning model after training is obtained；
E: when the convergence result is unsatisfactory for the default condition of convergence, iteration executes step A to step D until the convergence result
Meet the default condition of convergence.
2. the method according to claim 1, wherein it is described according to described image data and it is corresponding second detection
Information is trained the deep learning model
Data training set is added in described image data and corresponding second detection information；The data training set includes identified
The image data of target subject classification and target subject location information；
The deep learning model is trained according to the data training set.
3. the method according to claim 1, wherein image data is inputted preset deep learning mould described
Before type, further includes:
Obtain data training set；
The deep learning model is trained according to the data training set, so that the precision of the deep learning model reaches
To first threshold.
4. according to the method described in claim 3, it is characterized by:
The data training set is to carry out the image that target detection obtains to sample image according to preset target detection standard to believe
The set of breath.
5. method according to claim 1 to 4, which is characterized in that obtain described to first detection
Before the second detection information that information is corrected, further includes:
Obtain the third detection information to described image data；
If first detection information and the third detection information of described image data be not identical, described image data are made
For image data to be corrected；
The acquisition includes: according to the figure to be corrected to the second detection information that first detection information is corrected
As the first detection information of data is corrected to obtain second detection information.
6. method according to claim 1 to 4, which is characterized in that further include:
When the convergence result meets the default condition of convergence, mesh is carried out to image to be detected using deep learning model after training
Mark detection, identifies the location information of the classification of target subject and target subject in described image to be detected.
7. a kind of model training apparatus characterized by comprising
Detection module obtains the deep learning model to described for image data to be inputted preset deep learning model
The first detection information that image data is detected；
First obtains module, for obtaining the second detection information being corrected to first detection information；
Training module, for being instructed according to described image data and corresponding second detection information to the deep learning model
Practice；
Second obtain module, for obtain training after deep learning model convergence result；
Iteration module, for being obtained by the detection module, described first when the convergence result is unsatisfactory for the default condition of convergence
Modulus block, the training module and described second obtain module iteration and execute respective function until the convergence result meets in advance
If the condition of convergence.
8. model training apparatus according to claim 7, which is characterized in that the training module is according to described image data
And corresponding second detection information is trained the deep learning model and includes:
Data training set is added in described image data and corresponding second detection information；The data training set includes identified
The image data of target subject classification and target subject location information；
The deep learning model is trained according to the data training set.
9. a kind of electronic equipment, including memory and processor, computer program, the computer are stored in the memory
When program is executed by the processor, so that the processor executes the step such as method described in any one of claims 1 to 6
Suddenly.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
It realizes when being executed by processor such as the step of method described in any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810590113.XA CN108960232A (en) | 2018-06-08 | 2018-06-08 | Model training method, device, electronic equipment and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810590113.XA CN108960232A (en) | 2018-06-08 | 2018-06-08 | Model training method, device, electronic equipment and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108960232A true CN108960232A (en) | 2018-12-07 |
Family
ID=64488091
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810590113.XA Pending CN108960232A (en) | 2018-06-08 | 2018-06-08 | Model training method, device, electronic equipment and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108960232A (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109753910A (en) * | 2018-12-27 | 2019-05-14 | 北京字节跳动网络技术有限公司 | Crucial point extracting method, the training method of model, device, medium and equipment |
| CN109815991A (en) * | 2018-12-29 | 2019-05-28 | 北京城市网邻信息技术有限公司 | Training method, device, electronic equipment and the storage medium of machine learning model |
| CN109886928A (en) * | 2019-01-24 | 2019-06-14 | 平安科技（深圳）有限公司 | A kind of target cell labeling method, device, storage medium and terminal device |
| CN109948613A (en) * | 2019-03-22 | 2019-06-28 | 国网重庆市电力公司电力科学研究院 | A kind of Infrared image recognition and device of arrester |
| CN110175507A (en) * | 2019-04-09 | 2019-08-27 | 文远知行有限公司 | Model evaluation method, apparatus, computer equipment and storage medium |
| CN110399803A (en) * | 2019-07-01 | 2019-11-01 | 北京邮电大学 | A kind of vehicle checking method and device |
| CN110427985A (en) * | 2019-07-16 | 2019-11-08 | 北京京航安机场工程有限公司 | The method for realizing machine vision training using character |
| CN110705596A (en) * | 2019-09-04 | 2020-01-17 | 北京三快在线科技有限公司 | White screen detection method and device, electronic equipment and storage medium |
| CN111126493A (en) * | 2019-12-25 | 2020-05-08 | 东软睿驰汽车技术（沈阳）有限公司 | Deep learning model training method and device, electronic equipment and storage medium |
| CN111178442A (en) * | 2019-12-31 | 2020-05-19 | 北京容联易通信息技术有限公司 | Service implementation method for improving algorithm precision |
| CN111210024A (en) * | 2020-01-14 | 2020-05-29 | 深圳供电局有限公司 | Model training method and device, computer equipment and storage medium |
| CN111325278A (en) * | 2020-02-26 | 2020-06-23 | 重庆金山医疗技术研究院有限公司 | Image processing method, device and storage medium |
| CN111339905A (en) * | 2020-02-22 | 2020-06-26 | 郑州铁路职业技术学院 | CIM well lid state visual detection system based on deep learning and multi-view angle |
| CN111401102A (en) * | 2019-01-02 | 2020-07-10 | 中国移动通信有限公司研究院 | Deep learning model training method and device, electronic equipment and storage medium |
| CN111401563A (en) * | 2018-12-28 | 2020-07-10 | 杭州海康威视数字技术股份有限公司 | Machine learning model updating method and device |
| CN111695199A (en) * | 2020-06-09 | 2020-09-22 | 北京百度网讯科技有限公司 | Automatic driving test method, device, equipment and storage medium |
| CN112560541A (en) * | 2019-09-10 | 2021-03-26 | 富士通株式会社 | Training device and method for target detection model and electronic equipment |
| CN112614573A (en) * | 2021-01-27 | 2021-04-06 | 北京小白世纪网络科技有限公司 | Deep learning model training method and device based on pathological image labeling tool |
| CN112801933A (en) * | 2019-11-14 | 2021-05-14 | 纬创资通股份有限公司 | Object detection method, electronic device and object detection system |
| WO2021098779A1 (en) * | 2019-11-20 | 2021-05-27 | Oppo广东移动通信有限公司 | Target detection method, apparatus and device, and computer-readable storage medium |
| CN113159212A (en) * | 2021-04-30 | 2021-07-23 | 上海云从企业发展有限公司 | OCR recognition model training method, device and computer readable storage medium |
| CN113240125A (en) * | 2021-01-13 | 2021-08-10 | 深延科技(北京)有限公司 | Model training method and device, labeling method and device, equipment and storage medium |
| CN113469205A (en) * | 2020-03-31 | 2021-10-01 | 阿里巴巴集团控股有限公司 | Data processing method and system, network model and training method thereof, and electronic device |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103914704A (en) * | 2014-03-04 | 2014-07-09 | 西安电子科技大学 | Polarimetric SAR image classification method based on semi-supervised SVM and mean shift |
| CN103955718A (en) * | 2014-05-15 | 2014-07-30 | 厦门美图之家科技有限公司 | Image subject recognition method |
| CN104318242A (en) * | 2014-10-08 | 2015-01-28 | 中国人民解放军空军工程大学 | High-efficiency SVM active half-supervision learning algorithm |
| CN105956524A (en) * | 2016-04-22 | 2016-09-21 | 北京智芯原动科技有限公司 | Method and device for identifying traffic signs |
| US20170061625A1 (en) * | 2015-08-26 | 2017-03-02 | Digitalglobe, Inc. | Synthesizing training data for broad area geospatial object detection |
| CN106529565A (en) * | 2016-09-23 | 2017-03-22 | 北京市商汤科技开发有限公司 | Target identification model training and target identification method and device, and computing equipment |
| CN106530305A (en) * | 2016-09-23 | 2017-03-22 | 北京市商汤科技开发有限公司 | Semantic segmentation model training and image segmentation method and device, and calculating equipment |
| CN106707293A (en) * | 2016-12-01 | 2017-05-24 | 百度在线网络技术（北京）有限公司 | Obstacle recognition method and device for vehicles |
| CN106803071A (en) * | 2016-12-29 | 2017-06-06 | 浙江大华技术股份有限公司 | Object detecting method and device in a kind of image |
-
2018
- 2018-06-08 CN CN201810590113.XA patent/CN108960232A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103914704A (en) * | 2014-03-04 | 2014-07-09 | 西安电子科技大学 | Polarimetric SAR image classification method based on semi-supervised SVM and mean shift |
| CN103955718A (en) * | 2014-05-15 | 2014-07-30 | 厦门美图之家科技有限公司 | Image subject recognition method |
| CN104318242A (en) * | 2014-10-08 | 2015-01-28 | 中国人民解放军空军工程大学 | High-efficiency SVM active half-supervision learning algorithm |
| US20170061625A1 (en) * | 2015-08-26 | 2017-03-02 | Digitalglobe, Inc. | Synthesizing training data for broad area geospatial object detection |
| CN105956524A (en) * | 2016-04-22 | 2016-09-21 | 北京智芯原动科技有限公司 | Method and device for identifying traffic signs |
| CN106529565A (en) * | 2016-09-23 | 2017-03-22 | 北京市商汤科技开发有限公司 | Target identification model training and target identification method and device, and computing equipment |
| CN106530305A (en) * | 2016-09-23 | 2017-03-22 | 北京市商汤科技开发有限公司 | Semantic segmentation model training and image segmentation method and device, and calculating equipment |
| CN106707293A (en) * | 2016-12-01 | 2017-05-24 | 百度在线网络技术（北京）有限公司 | Obstacle recognition method and device for vehicles |
| CN106803071A (en) * | 2016-12-29 | 2017-06-06 | 浙江大华技术股份有限公司 | Object detecting method and device in a kind of image |
Non-Patent Citations (1)
| Title |
|---|
| 余永权，唐平: "《单片机模糊逻辑开发软件》", 30 November 1996 * |
Cited By (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109753910B (en) * | 2018-12-27 | 2020-02-21 | 北京字节跳动网络技术有限公司 | Key point extraction method, model training method, device, medium and equipment |
| CN109753910A (en) * | 2018-12-27 | 2019-05-14 | 北京字节跳动网络技术有限公司 | Crucial point extracting method, the training method of model, device, medium and equipment |
| CN111401563A (en) * | 2018-12-28 | 2020-07-10 | 杭州海康威视数字技术股份有限公司 | Machine learning model updating method and device |
| CN111401563B (en) * | 2018-12-28 | 2023-11-03 | 杭州海康威视数字技术股份有限公司 | Machine learning model updating method and device |
| CN109815991A (en) * | 2018-12-29 | 2019-05-28 | 北京城市网邻信息技术有限公司 | Training method, device, electronic equipment and the storage medium of machine learning model |
| CN109815991B (en) * | 2018-12-29 | 2021-02-19 | 北京城市网邻信息技术有限公司 | Training method and device of machine learning model, electronic equipment and storage medium |
| CN111401102A (en) * | 2019-01-02 | 2020-07-10 | 中国移动通信有限公司研究院 | Deep learning model training method and device, electronic equipment and storage medium |
| CN111401102B (en) * | 2019-01-02 | 2023-11-14 | 中国移动通信有限公司研究院 | Deep learning model training method and device, electronic equipment and storage medium |
| CN109886928B (en) * | 2019-01-24 | 2023-07-14 | 平安科技（深圳）有限公司 | Target cell marking method, device, storage medium and terminal equipment |
| CN109886928A (en) * | 2019-01-24 | 2019-06-14 | 平安科技（深圳）有限公司 | A kind of target cell labeling method, device, storage medium and terminal device |
| CN109948613A (en) * | 2019-03-22 | 2019-06-28 | 国网重庆市电力公司电力科学研究院 | A kind of Infrared image recognition and device of arrester |
| CN110175507B (en) * | 2019-04-09 | 2023-07-04 | 文远知行有限公司 | Model evaluation method, device, computer equipment and storage medium |
| CN110175507A (en) * | 2019-04-09 | 2019-08-27 | 文远知行有限公司 | Model evaluation method, apparatus, computer equipment and storage medium |
| CN110399803A (en) * | 2019-07-01 | 2019-11-01 | 北京邮电大学 | A kind of vehicle checking method and device |
| CN110427985A (en) * | 2019-07-16 | 2019-11-08 | 北京京航安机场工程有限公司 | The method for realizing machine vision training using character |
| CN110705596A (en) * | 2019-09-04 | 2020-01-17 | 北京三快在线科技有限公司 | White screen detection method and device, electronic equipment and storage medium |
| CN112560541A (en) * | 2019-09-10 | 2021-03-26 | 富士通株式会社 | Training device and method for target detection model and electronic equipment |
| CN112801933A (en) * | 2019-11-14 | 2021-05-14 | 纬创资通股份有限公司 | Object detection method, electronic device and object detection system |
| WO2021098779A1 (en) * | 2019-11-20 | 2021-05-27 | Oppo广东移动通信有限公司 | Target detection method, apparatus and device, and computer-readable storage medium |
| CN111126493A (en) * | 2019-12-25 | 2020-05-08 | 东软睿驰汽车技术（沈阳）有限公司 | Deep learning model training method and device, electronic equipment and storage medium |
| CN111126493B (en) * | 2019-12-25 | 2023-08-01 | 东软睿驰汽车技术（沈阳）有限公司 | Training method and device for deep learning model, electronic equipment and storage medium |
| CN111178442A (en) * | 2019-12-31 | 2020-05-19 | 北京容联易通信息技术有限公司 | Service implementation method for improving algorithm precision |
| CN111178442B (en) * | 2019-12-31 | 2023-05-12 | 北京容联易通信息技术有限公司 | Service realization method for improving algorithm precision |
| CN111210024A (en) * | 2020-01-14 | 2020-05-29 | 深圳供电局有限公司 | Model training method and device, computer equipment and storage medium |
| CN111210024B (en) * | 2020-01-14 | 2023-09-15 | 深圳供电局有限公司 | Model training method, device, computer equipment and storage medium |
| CN111339905B (en) * | 2020-02-22 | 2022-07-08 | 郑州铁路职业技术学院 | CIM well lid state visual detection system based on deep learning and multiple visual angles |
| CN111339905A (en) * | 2020-02-22 | 2020-06-26 | 郑州铁路职业技术学院 | CIM well lid state visual detection system based on deep learning and multi-view angle |
| CN111325278A (en) * | 2020-02-26 | 2020-06-23 | 重庆金山医疗技术研究院有限公司 | Image processing method, device and storage medium |
| CN111325278B (en) * | 2020-02-26 | 2023-08-29 | 重庆金山医疗技术研究院有限公司 | Image processing method, device and storage medium |
| CN113469205A (en) * | 2020-03-31 | 2021-10-01 | 阿里巴巴集团控股有限公司 | Data processing method and system, network model and training method thereof, and electronic device |
| CN113469205B (en) * | 2020-03-31 | 2023-01-17 | 阿里巴巴集团控股有限公司 | Data processing method and system, network model and training method thereof, and electronic device |
| CN111695199B (en) * | 2020-06-09 | 2023-09-05 | 北京百度网讯科技有限公司 | Automatic driving test method, device, equipment and storage medium |
| CN111695199A (en) * | 2020-06-09 | 2020-09-22 | 北京百度网讯科技有限公司 | Automatic driving test method, device, equipment and storage medium |
| CN113240125A (en) * | 2021-01-13 | 2021-08-10 | 深延科技(北京)有限公司 | Model training method and device, labeling method and device, equipment and storage medium |
| CN112614573A (en) * | 2021-01-27 | 2021-04-06 | 北京小白世纪网络科技有限公司 | Deep learning model training method and device based on pathological image labeling tool |
| CN113159212A (en) * | 2021-04-30 | 2021-07-23 | 上海云从企业发展有限公司 | OCR recognition model training method, device and computer readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108960232A (en) | Model training method, device, electronic equipment and computer readable storage medium | |
| CN108921161B (en) | Model training method and device, electronic equipment and computer readable storage medium | |
| US11138478B2 (en) | Method and apparatus for training, classification model, mobile terminal, and readable storage medium | |
| CN108764372B (en) | Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set | |
| CN108764208B (en) | Image processing method and device, storage medium and electronic equipment | |
| CN108805103B (en) | Image processing method and device, electronic equipment and computer readable storage medium | |
| CN108304821B (en) | Image recognition method and device, image acquisition method and device, computer device and non-volatile computer-readable storage medium | |
| CN108764321B (en) | Image-recognizing method and device, electronic equipment, storage medium | |
| CN109002843A (en) | Image processing method and device, electronic equipment, computer readable storage medium | |
| CN109712192A (en) | Camera module scaling method, device, electronic equipment and computer readable storage medium | |
| CN108804658B (en) | Image processing method and device, storage medium and electronic equipment | |
| CN107948617B (en) | Image processing method, image processing device, computer-readable storage medium and computer equipment | |
| CN110956679B (en) | Image processing method and device, electronic equipment and computer readable storage medium | |
| CN108734214A (en) | Image-recognizing method and device, electronic equipment, storage medium | |
| CN108063891A (en) | Image processing method, device, computer readable storage medium and computer equipment | |
| CN107743200A (en) | Method, apparatus, computer-readable recording medium and the electronic equipment taken pictures | |
| CN109360254A (en) | Image processing method and device, electronic equipment, computer readable storage medium | |
| CN109712177A (en) | Image processing method, device, electronic equipment and computer readable storage medium | |
| CN110490196A (en) | Subject detection method and apparatus, electronic equipment, computer readable storage medium | |
| CN108763580A (en) | Image processing method and device, electronic equipment, computer storage media | |
| CN107948618A (en) | Image processing method, device, computer-readable recording medium and computer equipment | |
| CN107920205A (en) | Image processing method, device, storage medium and electronic equipment | |
| CN109068060A (en) | Image processing method and device, terminal device, computer readable storage medium | |
| US8655102B2 (en) | Method and system for identifying tokens in an image | |
| CN108737797A (en) | White balancing treatment method, device and electronic equipment |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181207 |