CN114638999A - Training method of image quality evaluation model, data uploading processing method and device - Google Patents
Training method of image quality evaluation model, data uploading processing method and device Download PDFInfo
- Publication number
- CN114638999A CN114638999A CN202210287406.7A CN202210287406A CN114638999A CN 114638999 A CN114638999 A CN 114638999A CN 202210287406 A CN202210287406 A CN 202210287406A CN 114638999 A CN114638999 A CN 114638999A
- Authority
- CN
- China
- Prior art keywords
- image
- image quality
- model
- sample set
- 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
- 238000013441 quality evaluation Methods 0.000 title claims abstract description 155
- 238000012549 training Methods 0.000 title claims abstract description 93
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000003672 processing method Methods 0.000 title claims abstract description 9
- 239000000178 monomer Substances 0.000 claims abstract description 118
- 238000011156 evaluation Methods 0.000 claims abstract description 60
- 238000002372 labelling Methods 0.000 claims abstract description 27
- 238000012545 processing Methods 0.000 claims abstract description 16
- 102000054766 genetic haplotypes Human genes 0.000 claims description 28
- 238000004590 computer program Methods 0.000 claims description 19
- 239000013598 vector Substances 0.000 claims description 14
- 238000001303 quality assessment method Methods 0.000 claims description 13
- 238000013145 classification model Methods 0.000 claims description 9
- 238000013210 evaluation model Methods 0.000 claims description 5
- 238000013138 pruning Methods 0.000 claims description 5
- 238000011157 data evaluation Methods 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 description 10
- 238000005516 engineering process Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 238000013140 knowledge distillation Methods 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 3
- 238000012550 audit Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000001934 delay Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011022 operating instruction Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The specification provides a training method of an image quality evaluation model, a data uploading processing method and a data uploading processing device. The training method comprises the following steps: performing ensemble learning on at least two trained monomer models for evaluating image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of image quality; evaluating the image quality of a first image sample set based on the ensemble learning model, and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set; training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
Description
Technical Field
The document belongs to the technical field of artificial intelligence, and particularly relates to a training method of an image quality evaluation model, and a data uploading processing method and device.
Background
In a business scenario involving uploading of image data by a user, background personnel are usually required to check whether the image data meets business requirements. And if the image data can not be checked, the background personnel return the image data and then upload the image data again by the user. It can be seen that under the condition of low manual efficiency, the time cost of both parties is increased by image review.
Therefore, how to improve the efficiency of examining and verifying image data is a technical problem to be solved by the present application.
Disclosure of Invention
Embodiments of the present disclosure provide a training method for an image quality assessment model, a data uploading processing method, and an apparatus, which can improve the efficiency of examining and verifying image data uploaded by a user.
In order to achieve the above object, the embodiments of the present specification are implemented as follows:
in a first aspect, a training method for an image quality assessment model is provided, including:
performing ensemble learning on at least two trained monomer models for evaluating image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of image quality;
evaluating the image quality of a first image sample set based on the ensemble learning model, and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set;
training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
In a second aspect, a data uploading processing method is provided, including:
receiving uploaded data of a user side, wherein the uploaded data comprise image data;
inputting the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by carrying out ensemble learning on at least two trained monomer models for evaluating the image quality, and different monomer models have different evaluation deviation of the image quality;
and executing corresponding operation on the uploaded data based on the image quality evaluation result of the image data, wherein the operation which can be executed on the uploaded data comprises accepting the uploaded data and returning the uploaded data.
In a third aspect, a training apparatus for an image quality assessment model is provided, including:
the ensemble learning module is used for performing ensemble learning on at least two trained monomer models for evaluating the image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of the image quality;
the sample labeling module is used for evaluating the image quality of a first image sample set based on the ensemble learning model and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set;
and the model training module is used for training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
In a fourth aspect, an electronic device is provided comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being executed by the processor to:
performing ensemble learning on at least two trained monomer models for evaluating image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of image quality;
evaluating the image quality of a first image sample set based on the ensemble learning model, and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set;
training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
In a fifth aspect, a computer-readable storage medium is provided, having stored thereon a computer program which, when executed by a processor, performs the steps of:
performing ensemble learning on at least two trained monomer models for evaluating image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of image quality;
evaluating the image quality of a first image sample set based on the ensemble learning model, and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set;
training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
A sixth aspect provides a data uploading processing apparatus, including:
the data receiving module is used for receiving uploaded data of a user side, wherein the uploaded data comprises image data;
the data evaluation module is used for inputting the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by carrying out ensemble learning on at least two trained monomer models used for evaluating the image quality, and different monomer models have different evaluation deviation of the image quality;
and the data processing module executes corresponding operation on the uploaded data based on the image quality evaluation result of the image data, wherein the operation executable on the uploaded data comprises accepting the uploaded data and returning the uploaded data.
In a seventh aspect, an electronic device is provided that includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being executed by the processor to:
receiving uploaded data of a user side, wherein the uploaded data comprise image data;
inputting the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by carrying out ensemble learning on at least two trained monomer models for evaluating the image quality, and different monomer models have different evaluation deviation of the image quality;
and executing corresponding operation on the uploaded data based on the image quality evaluation result of the image data, wherein the operation which can be executed on the uploaded data comprises accepting the uploaded data and returning the uploaded data.
In an eighth aspect, a computer-readable storage medium is provided, having stored thereon a computer program which, when executed by a processor, performs the steps of:
receiving uploading data of a user end, wherein the uploading data comprises image data;
inputting the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by carrying out ensemble learning on at least two trained haplotype models for evaluating the image quality, and different haplotype models have different evaluation deviation of the image quality;
and executing corresponding operation on the uploaded data based on the image quality evaluation result of the image data, wherein the operation which can be executed on the uploaded data comprises accepting the uploaded data and returning the uploaded data.
The method of the embodiment of the specification adopts an ensemble learning technology to train monomer models with different evaluation biases aiming at image quality, and fuses the monomer models into an ensemble learning model. The ensemble learning model gives consideration to the evaluation bias of each monomer model, so that the ensemble learning model has very comprehensive image quality evaluation capability. And then, a knowledge distillation technology is adopted, the evaluation bias of each monomer model considered by the integrated learning model is migrated and learned to a target image quality evaluation model with a more simplified structure, and the target image quality evaluation model is deployed in an audit scene of image data uploaded by a user in a light weight mode, so that the user can automatically decide whether to accept the image data uploaded by the user or return the image data uploaded by the user. The whole process does not depend on manual operation of background personnel, so that the auditing efficiency of the image data is greatly improved, and the time cost of both the user and the background personnel is reduced.
Drawings
In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only some embodiments described in the embodiments of the present specification, and for a person of ordinary skill in the relevant art, other drawings can be obtained according to the drawings without inventive labor.
Fig. 1 is a schematic flowchart of a training method of an image quality assessment model provided in an embodiment of the present specification.
Fig. 2 is a flowchart illustrating a data upload processing method according to an embodiment of the present disclosure.
Fig. 3 is a schematic structural diagram of a training device for an image quality assessment model provided in an embodiment of the present disclosure.
Fig. 4 is a schematic structural diagram of a data upload processing apparatus according to an embodiment of the present disclosure.
Fig. 5 is a schematic structural diagram of an electronic device provided in an embodiment of this specification.
Detailed Description
In order to make those skilled in the art better understand the technical solutions in the present specification, the technical solutions in the embodiments of the present specification will be clearly and completely described below with reference to the drawings in the embodiments of the present specification, and it is obvious that the described embodiments are only a part of the embodiments of the present specification, and not all of the embodiments. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments in the present specification without any creative efforts shall fall within the protection scope of the present specification.
As mentioned above, in a business scenario involving uploading image data by a user, background personnel are usually required to check whether the image data meets business requirements. And if the image data can not be checked, the background personnel return the image data and then upload the image data again by the user. Taking image data of certificates as an example, after backstage personnel take the image data sent by the user, the backstage personnel need to carefully check whether the certificate information in the image data is clear, so that the checking efficiency is not too high, and the time cost investment of the two parties is large. Especially if the first upload is not approved, the method delays more time and brings more working pressure to background personnel. Therefore, the document aims to provide a technical scheme which can automatically check the image data uploaded by the user and decide the processing result based on the artificial intelligence technology, and compared with the traditional manual checking mode, the technical scheme has higher checking efficiency.
Fig. 1 is a flowchart of a training method of an image quality assessment model according to an embodiment of the present disclosure, where the method shown in fig. 1 may be performed by the following corresponding apparatus, and specifically includes the following steps:
s102, performing ensemble learning on at least two trained monomer models for evaluating image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of image quality.
In the ensemble learning, the weakly supervised haplotype models are combined to obtain a more comprehensive strongly supervised model, namely an ensemble learning model. The potential idea of ensemble learning is that even if one monomer model gets a wrong prediction, other monomer models can correct the error back. Therefore, in general, the ensemble learning model is equivalent to introducing a weighted sum calculator at the output end of each monomer model, so as to calculate a final result by integrating according to the image quality evaluation result output by each monomer model. In the present specification embodiment, the final result output by the ensemble learning model may be regarded as a reliable image quality evaluation result.
It should be noted that ensemble learning belongs to the prior art, and since the implementation manner is not unique, detailed description is not provided herein.
In addition, the individual monomer models that make up the ensemble learning model may also be constructed in different ways.
For example, embodiments of the present disclosure may select different types of model algorithms to construct different monomer models.
Or, feature vectors of different image index dimensions can be selected to construct different monomer models. For example, feature vectors of image gray scale, image brightness and color depth are selected to construct one monomer model, and feature vectors of image resolution, image element ratio and image contrast are selected to construct another monomer model.
Or, in order to reduce the modeling difficulty, an initial model may be constructed in advance, and different monomer models may be derived based on the initial model by performing model trunk replacement, model pruning and the like on the initial model. I.e., different haplotype models used to assess image quality, at least one of which is based on another, model stem replacement and/or model pruning.
In the embodiment of the present specification, the haplotype has an ability to score the image quality, and specifically, a regression model capable of analyzing the influence strength on the image quality (dependent variable) based on the image characteristics (independent variable) may be used.
By way of exemplary introduction, a haplotype employing a regression model may be supervised trained using the labeled second image sample set. Wherein the second image sample set comprises two images with different image qualities, and the image quality difference between the two images belonging to the same image sample is marked. By training the monomer model in the labeling mode, the monomer model can learn to obtain the probability distribution of image quality difference, and therefore the method has the capability of scoring the image quality.
Specifically, in the supervised training process of any monomer model, after the image samples in the second image sample set are input to the monomer model, the image quality scores of the images in the sample image can be obtained. And the image quality scores of the two images under the same sample image are the training results of the sample image given by the haplotype. The difference between the image quality scores of the two images indicated by the training result has a certain error from the difference of the image quality of the labels. The supervised training is to calculate the error between the training result and the label based on a preset loss function, and adjust the parameters in the monomer model with the aim of reducing the error.
On the basis of the training principle, the labeling difficulty of the image quality difference can be simplified, namely, the second image sample set only needs to label which image quality is good or which image quality is bad in two images under the same image sample, and then supervised training can be achieved. For example, if a sample image in the second image sample set includes two images a and b, the second image sample set only needs to indicate the image quality of the sample image a > b. Correspondingly, the loss function used by the haplotype in the supervised training only needs to calculate the error according to whether the image quality score difference of the two images in the training result is matched with the image quality of the two images marked, or the image quality is bad, so that the training gradient is determined according to the error, and the parameter adjustment is carried out on the haplotype.
Here, the embodiments of the present specification may quantify the image quality corresponding to an image in terms of the degree of sharpness of elements in the image. Namely, the image quality difference between two images belonging to the same image sample in the second sample image set is marked according to the image element definition difference between the two images. The quantization mode has the advantage that the definition of the image elements of one image has a relatively clear quality standard, so that the method is very suitable for automatically realizing the marking of the evaluation standard by an artificial intelligence technology.
That is, the embodiments of the present specification may train a model capable of recognizing the sharpness of an image, that is, a sharpness evaluation model, based on a third image sample set labeled with respect to the sharpness of an image element. Correspondingly, in the embodiment of the present specification, only two images belonging to the same image sample in a second sample image set need to be input into the definition classification model, so that the image element definitions of the two different images can be obtained, and the image element definition difference between the two different images can be directly determined based on the image element definitions of the two different images, so that the image samples belonging to the two different images are labeled according to the obtained image element definition difference.
Wherein, if only need concentrate to the second image sample and belong to two image marks under the same image sample which image quality is good, which image quality is bad, then the definition evaluation model can adopt the classification model, and that is the definition evaluation model only need the definition rank of discernment image can realize: and completing the labeling of which image has good quality and which image has bad quality according to the high-low resolution of the definition levels of the two images under the same image sample.
And S104, evaluating the image quality of the first image sample set based on the ensemble learning model, and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set.
It should be understood that the ensemble learning model is obtained by fusing at least two monomer models, the model structure is complex, the number of channels is large, and the ensemble learning model is not suitable for lightweight deployment on a terminal in many application scenarios.
Here, in the embodiments of the present specification, labeling the first image sample set by the ensemble learning model is to perform supervised training on the target image quality assessment model by using the first image sample set, so as to migrate the knowledge of the ensemble learning model to the target image quality assessment model with a relatively simplified structure for commissioning.
S106, training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is smaller than that of the integrated learning model.
In the embodiments of the present description, the image quality of the image data uploaded by the user is evaluated by using the target image quality evaluation model, so as to decide whether to accept the image data of the user. For this requirement, it is only necessary to determine whether the image quality of the image data meets the acceptance standard. Therefore, the target image quality evaluation model can adopt a classification model, that is, two accepted or not accepted categories of the image data uploaded by the user can be distinguished according to the quality of the image (more categories are also applicable).
For a target image quality evaluation model adopting a classification model, image samples in a first image sample set used for training are single image samples, and each image sample in the first image sample set can be labeled with a matched image quality classification label through an image quality evaluation result of the integrated learning model on the image samples in the first image sample set.
Assuming that the image quality is classified into three grades of "excellent", "normal", and "poor", the image quality classification is correspondingly classified into three grades of "excellent", "normal", and "poor". Wherein, the image data uploaded by the user can be accepted if it is determined as "excellent" or "normal", and can be returned if it is determined as "poor".
For the training of the target image quality evaluation model using the classification model, after the sample image (but the image) in the first sample image labeled with the image quality classification label is input into the haplotype model, the classification result for the image quality of the sample image provided by the haplotype model can be obtained, and the classification result is the training result of the target image quality. Certain errors exist between the image quality of the training result and the image quality marked by the image quality classification label. The supervised training is to calculate the error between the training result and the image quality classification label based on a preset loss function, and adjust the parameters in the monomer model with the aim of reducing the error. It can be seen that the image quality classification label marked on one image can be regarded as an expected value of model training, and the model is subjected to iterative training under the supervision of the image quality classification label, and gradually converges on the prediction of the image quality classification label, so that optimization is realized.
The fact that the number of channels of the target image quality evaluation model is less than that of the channels of the ensemble learning model means that: the number of calculation logic channels and the dimension of input/output feature vectors of the target image quality evaluation model are less than those of the ensemble learning model. In the embodiments of the present specification, the target image quality evaluation model may be a single classifier structure (a non-composite model usually has one classifier), and the target image quality evaluation model is greatly simplified in model structure compared to the ensemble learning model obtained by fusing multiple single models.
In addition, in order to reduce the implementation difficulty of the scheme, the target image quality evaluation model is obtained based on the basic construction of the monomer model. That is, in the embodiments of the present specification, feature vectors meeting a preset weight criterion may be selected from each monomer model, and a model may be reconstructed to serve as a target image quality evaluation model. For example, a haplotype a constructed by feature vectors based on the selection of the image resolution, the image element proportion and the image contrast is trained, and the feature vector with the image element proportion has a higher weight (reaching a preset weight standard), which means that the factor with the image element proportion is an important factor influencing the image quality through the supervised training of the haplotype a, so that the image element proportion can be selected as one feature vector of the target image quality evaluation model. In practical application, the number of feature vectors selectable from one monomer model is not unique, and the number of the monomer models in the ensemble learning model is not limited to one, so that a plurality of feature vectors can be finally obtained from each monomer model through the modeling mode, and a final target image quality evaluation model can be directly constructed.
Based on the above, it can be seen that the method in the embodiment of the present specification firstly adopts an ensemble learning technique to train monomer models having different evaluation biases for image quality, and fuses the monomer models into an ensemble learning model. The ensemble learning model gives consideration to the evaluation bias of each monomer model, so that the ensemble learning model has very comprehensive image quality evaluation capability. And then, a knowledge distillation technology is adopted, the evaluation deviation of each monomer model considered by the integrated learning model is migrated and learned to a target image quality evaluation model with a more simplified structure, and therefore light weight deployment is facilitated. For example, the method is applied to an auditing scene of image data uploaded by a user to replace a manual mode.
The method of the embodiments of the present disclosure is described in detail below with reference to practical application scenarios.
The method of the embodiments of the present description mainly comprises the following stages:
firstly, a monomer model training stage:
this stage begins with the construction of different haplotype models for assessing image quality.
In particular, as previously described, different types of model algorithms may be employed to construct different monomer models. Or, feature vectors of different image index dimensions can be selected to construct different monomer models. Or, in order to reduce the modeling difficulty, an initial model may be constructed in advance, and different monomer models may be derived based on the initial model by performing model trunk replacement, model pruning and the like on the initial model.
It should be appreciated that where there is a distinction between haplotype models, in the case of any of the algorithms, feature vectors, and model limbs, there will be a distinction in the evaluation bias that will occur after training.
Under the condition that sample images for training are limited, a bootstrap-based mode can be adopted in the stage to configure a training set required by each haplotype model, namely a first image sample set.
bootstrap, also known as bootstrap, is a sampling method with a fallback in order to get the distribution of statistics and confidence intervals. Wherein, the bootstrap comprises the following specific steps: a certain number of samples are extracted from original samples by adopting a resampling method (with a back sampling method); calculating a statistic T to be obtained according to the extracted sample; repeating the above N times (generally more than 1000) to obtain N statistics T; from these N statistics, the confidence interval of the statistics can be calculated.
Using the bootstrap method, a replaced random sampling may be taken of the sample image set adapted to train the regression model as described above to obtain a second image sample set of each monomer model. At least some of the sample images in the second image sample set of different haplotype models are different, thereby forming training data with differences between the haplotype models based on the limited sample images.
And then, performing supervised training on each monomer model through a second image sample set of each monomer model to obtain the monomer models with different evaluation biases.
II, integrating the learning stage:
in this stage, the monomer models are fused into a learning model, i.e. an integrated learning model.
Specifically, the output end of each monomer model can be introduced into a weighted sum calculator, so as to synthesize a final result according to the image quality evaluation result output by each monomer model.
For simplicity, assume that there are three monomer models A, B, C, where the algorithm for monomer model A is algorithmmAThe algorithm for the haplotype B is algorithmmBThe algorithm for the monomer model C is algorithmmCThen the algorithm of the ensemble learning model may be:
KA×algorithmA+KB×algorithmB+KC×algorithmCwherein, K isA、KB、KCEach represents a weighting coefficient of the monomer model A, B, C.
It can be seen that the ensemble learning model is obtained by combining three types of the single model A, B, C, and is relatively complex in model structure.
Thirdly, a transfer learning stage:
the stage is to distill the knowledge of the integrated learning model to an image quality evaluation model with a relatively simplified structure, such as a target image quality evaluation model with a single classifier structure.
The specific process is as described above, and the ensemble learning model is used to label the sample images of the second image sample set. And then training the target image quality evaluation model according to the labeled second image sample set, so that the parameters of the image quality evaluation model are continuously adjusted in the multi-iteration training process, the image quality evaluation result output by the target image quality evaluation model gradually converges to the image quality evaluation result output by the ensemble learning model, and finally the target image quality evaluation model can replace the ensemble learning model.
Fourthly, an active learning stage:
and in the stage, a fine-tuning active learning mode is adopted to further fine-tune and optimize a target image quality evaluation model finished by the three-stage training.
First, the image quality classification labels are re-labeled on the first sample image of the low-score training result in the first image sample set.
Wherein, the sample image of the low-score training result may refer to, but is not limited to:
and the sample images with the training results inconsistent with the image quality classification labels, namely the sample images with the inaccurate evaluation of the target image quality evaluation model. The reason for this is that the labeling of the sample image of this type is usually wrong, and therefore, the image quality classification label can be corrected. And/or the presence of a gas in the gas,
and the sample image with the confidence value of the training result not reaching the effective threshold value is the sample image with ambiguous evaluation result of the target image quality evaluation model. For example, the confidence values of "excellent", "normal" and "poor" of a certain sample image are close to each other, the "excellent" confidence value is 33%, the "normal" confidence value is 37%, and the "poor" confidence value is 40%, which indicates that the target image quality evaluation model cannot grade the sample image quality to provide effective evaluation. The reason why the type evaluation result is ambiguous is generally that an error occurs in labeling of the sample image of the type, and therefore, the image quality classification label can be corrected.
After the second image sample set is re-labeled, the target image quality evaluation model can be retrained based on the re-labeled image quality classification labels of the first image sample set and the first image sample set, so as to finely adjust parameters in the target image quality evaluation model.
Fig. 2 is a flowchart of a data upload processing method according to an embodiment of the present disclosure, where the method shown in fig. 2 may be executed by a corresponding apparatus, and specifically includes the following steps:
step S202, receiving the uploading data of the user terminal, wherein the uploading data includes the image data.
It should be understood that, in the embodiments of the present disclosure, image quality evaluation needs to be performed on the image data uploaded by the user side, so as to determine whether the image data passes the audit according to the image quality evaluation result. The data uploading mode of the user side is not unique, and is not described in detail herein for example.
Step S204, inputting image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by performing ensemble learning on at least two trained monomer models for evaluating the image quality, and different monomer models have different evaluation bias of the image quality.
It should be understood that the model used for image quality evaluation in this step, i.e., the target image quality evaluation model trained based on the method shown in fig. 1. In practical applications, the uploaded data of the user side is not limited to only include image data, and may also include data of other data types and that need to be checked, and accordingly, the image quality evaluation may check the image data in the uploaded data, which is not described herein again.
Furthermore, it should be noted that this step may be automatically executed based on a computer program, that is, the application process of the target image quality assessment model is compiled into an automatically executed script in a code language, and the automated image quality assessment is realized by running the script instead of manually.
Step S206, based on the image quality evaluation result of the image data, corresponding operation is performed on the uploaded data, wherein the operation performed on the uploaded data includes accepting the uploaded data and returning the uploaded data.
Specifically, returning the uploaded image data may be to return the image data provided by the user and inform that the image data is not approved. The user end may retry submitting the uploaded image data after the image data is returned, or only provide the image data that is not approved, which is not limited in this document.
Based on the above, it can be known that the method in the embodiments of the present specification adopts an ensemble learning technique to train monomer models having different evaluation biases for image quality, and fuses the monomer models into an ensemble learning model. The ensemble learning model gives consideration to the evaluation bias of each monomer model, so that the ensemble learning model has very comprehensive image quality evaluation capability. And then, a knowledge distillation technology is adopted, the evaluation bias of each monomer model considered by the integrated learning model is migrated and learned to a target image quality evaluation model with a more simplified structure, and the target image quality evaluation model is deployed in an audit scene of image data uploaded by a user in a light weight mode, so that the user can automatically decide whether to accept the image data uploaded by the user or return the image data uploaded by the user. The whole process does not depend on manual operation of background personnel, so that the auditing efficiency of the image data is greatly improved, and the time cost of both the user and the background personnel is reduced.
In addition, for the method shown in fig. 1, the embodiment of the present specification further provides a training device for an image quality evaluation model. Fig. 3 is a schematic structural diagram of a training device 300 according to an embodiment of the present disclosure, including:
the ensemble learning module 310 is used for performing ensemble learning on at least two trained monomer models for evaluating the image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of the image quality;
and the sample labeling module 320 is configured to evaluate the image quality of a first image sample set based on the ensemble learning model, and label the sample image of the first image sample set according to the obtained image quality evaluation result of the first image sample set.
And the model training module 330 is configured to train a target image quality evaluation model based on the labeled first image sample set, where the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
The apparatus according to the embodiment of the present disclosure first trains monomer models having different evaluation biases for image quality by using ensemble learning techniques, and fuses the monomer models into an ensemble learning model. The ensemble learning model gives consideration to the evaluation bias of each monomer model, so that the ensemble learning model has very comprehensive image quality evaluation capability. And then, a knowledge distillation technology is adopted, the evaluation deviation of each monomer model considered by the integrated learning model is migrated and learned to a target image quality evaluation model with a more simplified structure, and therefore light weight deployment is facilitated. For example, the method is applied to an auditing scene of image data uploaded by a user to replace a manual mode.
Optionally, at least one of the at least two haplotype models for evaluating image quality is a regression model, and the image samples in the second image sample set used for training include images with two different image qualities, and the second image sample set is labeled with the image quality difference between two images belonging to the same image sample.
Optionally, the image quality difference between two images belonging to the same image sample in the second sample image set is determined based on the image element sharpness difference between the two images. The difference in sharpness of image elements between two images belonging to the same image sample in the second sample image set is determined based on: inputting two images belonging to the same image sample in the second sample image set into a definition classification model to obtain the definition of image elements of the two different images, and determining the definition difference of the image elements between the two different images based on the definition of the image elements of the two different images; wherein the sharpness evaluation model is trained based on a third set of image samples that have been labeled for sharpness of image elements.
Optionally, the target image quality evaluation model is a classification model, and the image samples in the first image sample set are single image samples; specifically, the sample labeling module 320 of the embodiment of the present invention evaluates the image samples of the first image sample set based on the ensemble learning model to obtain an image quality evaluation result of each image sample in the first image sample set;
and labeling each image sample in the first image sample set with a matched image quality classification label according to the image quality evaluation result of each image sample in the first image sample set.
Optionally, the model training module 330 of the embodiment of the present invention is further configured to: re-labeling the image quality classification labels on the first sample images of the low-score training results in the first image sample set, wherein the sample images of the low-score training results refer to sample images with training results inconsistent with the image quality classification labels, and/or sample images with the confidence values of the training results not reaching effective threshold values; and training the target image quality evaluation model based on the first image sample set and the re-labeled image quality classification label of the first image sample set.
Optionally, the target image quality evaluation model is a single classifier model.
Optionally, at least one of the at least two haplotype models used for evaluating image quality is obtained after model trunk replacement and/or model pruning based on the other.
Optionally, the model training module 330 of the embodiment of the present invention is further configured to: and respectively selecting the characteristic vectors reaching the preset weight standard from the two monomer models at least used for evaluating the image quality to construct a target image quality evaluation model.
Obviously, the training device of the present specification can be used as the execution subject of the method shown in fig. 1, and thus can implement the steps and functions of the method implemented in fig. 1. Since the principle is the same, the detailed description is omitted here.
In addition, for the method shown in fig. 2, an embodiment of the present disclosure further provides a data upload processing apparatus. Fig. 4 is a schematic structural diagram of a data upload processing apparatus 400 according to an embodiment of the present disclosure, including:
the data receiving module 410 receives the uploaded data of the client, wherein the uploaded data includes image data.
And the data evaluation module 420 is configured to input the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, where the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled for image quality based on an ensemble learning model, the ensemble learning model is obtained by ensemble learning at least two trained haplotype models for evaluating image quality, and different haplotype models have different evaluation biases for image quality.
The data processing module 430 performs corresponding operations on the uploaded data based on the image quality evaluation result of the image data, wherein the operations that can be performed on the uploaded data include accepting the uploaded data and returning the uploaded data.
Based on the above, it can be known that the apparatus in the embodiments of the present specification uses an ensemble learning technique to train monomer models having different evaluation biases for image quality, and fuses the monomer models into one ensemble learning model. The ensemble learning model gives consideration to the evaluation bias of each monomer model, so that the ensemble learning model has very comprehensive image quality evaluation capability. And then, migrating and learning the evaluation deviation of each monomer model taken into consideration by the integrated learning model to a target image quality evaluation model with a more simplified structure by adopting a knowledge distillation technology, so that the target image quality evaluation model is deployed in a checking scene of image data uploaded by a user in a light weight manner, and automatically deciding whether to accept the image data uploaded by the user or to return the image data uploaded by the user. The whole process does not depend on manual operation of background personnel, so that the auditing efficiency of the image data is greatly improved, and the time cost of both the user and the background personnel is reduced.
Obviously, the data upload processing device in the embodiment of the present specification can be used as the execution main body of the method shown in fig. 2, and thus can implement the steps and functions of the method implemented in fig. 2. Since the principle is the same, the detailed description is omitted here.
Fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present specification. Referring to fig. 5, at a hardware level, the electronic device includes a processor, and optionally further includes an internal bus, a network interface, and a memory. The Memory may include a Memory, such as a Random-Access Memory (RAM), and may further include a non-volatile Memory, such as at least 1 disk Memory. Of course, the electronic device may also include hardware required for other services.
The processor, the network interface, and the memory may be connected to each other via an internal bus, which may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one double-headed arrow is shown in FIG. 5, but this does not indicate only one bus or one type of bus.
And the memory is used for storing programs. In particular, the program may include program code comprising computer operating instructions. The memory may include both memory and non-volatile storage and provides instructions and data to the processor.
The processor reads a corresponding computer program from the nonvolatile memory into the memory and runs the computer program, and the training device of the image quality evaluation model is formed on a logic level. The processor is used for executing the program stored in the memory and is specifically used for executing the following operations:
and performing ensemble learning on at least two trained monomer models for evaluating the image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of the image quality.
And evaluating the image quality of a first image sample set based on the ensemble learning model, and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set.
Training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
Or the processor reads the corresponding computer program from the nonvolatile memory to the memory and then runs the computer program, and the data uploading processing device is formed on the logic level. The processor is used for executing the program stored in the memory and is specifically used for executing the following operations:
receiving the uploading data of the user terminal, wherein the uploading data comprises the image data.
Inputting the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by carrying out ensemble learning on at least two trained monomer models used for evaluating the image quality, and different monomer models have different evaluation bias of the image quality.
And executing corresponding operation on the uploaded data based on the image quality evaluation result of the image data, wherein the operation which can be executed on the uploaded data comprises accepting the uploaded data and returning the uploaded data.
The embodiments disclosed in fig. 1 or fig. 2 of the present specification may be applied to or implemented by a processor. The processor may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in a processor or instructions in the form of software. The Processor may be a general-purpose Processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but also Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components. The various methods, steps and logic blocks disclosed in the embodiments of the present specification may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of a method disclosed in connection with the embodiments of the present specification may be embodied directly in a hardware decoding processor, or in a combination of hardware and software modules in the decoding processor. The software module may be located in ram, flash memory, rom, prom, or eprom, registers, etc. storage media as is well known in the art. The storage medium is located in a memory, and a processor reads information in the memory and completes the steps of the method in combination with hardware of the processor.
It should be understood that the electronic device of the embodiments of the present specification can implement the functions of the above-described forensic method shown in fig. 1 or fig. 2. Since the principle is the same, the detailed description is omitted here.
Of course, besides the software implementation, the electronic device in the present specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may also be hardware or logic devices.
Furthermore, the present specification embodiments also propose a computer-readable storage medium storing one or more programs, the one or more programs including instructions.
Wherein the instructions, when executed by a portable electronic device comprising a plurality of application programs, enable the portable electronic device to perform the method of the embodiment shown in fig. 1, and are specifically configured to perform the following steps:
and performing ensemble learning on at least two trained monomer models for evaluating the image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of the image quality.
And evaluating the image quality of a first image sample set based on the ensemble learning model, and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set.
Training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
Alternatively, the above instructions, when executed by a portable electronic device comprising a plurality of application programs, can cause the portable electronic device to perform the method of the embodiment shown in fig. 2, and specifically to perform the following steps:
receiving the uploading data of the user terminal, wherein the uploading data comprises the image data.
Inputting the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by carrying out ensemble learning on at least two trained monomer models used for evaluating the image quality, and different monomer models have different evaluation bias of the image quality.
And executing corresponding operation on the uploaded data based on the image quality evaluation result of the image data, wherein the operation which can be executed on the uploaded data comprises accepting the uploaded data and returning the uploaded data.
As will be appreciated by one of ordinary skill in the art, the embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The foregoing description has been directed to specific embodiments of this disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
The above description is only an example of the present specification, and is not intended to limit the present specification. Various modifications and changes may occur to those skilled in the art to which it pertains. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification. Moreover, all other embodiments obtained by persons of ordinary skill in the art without making any inventive step shall fall within the scope of protection of this document.
Claims (17)
1. A training method of an image quality assessment model comprises the following steps:
performing ensemble learning on at least two trained monomer models for evaluating image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of image quality;
evaluating the image quality of a first image sample set based on the ensemble learning model, and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set;
training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
2. The method of claim 1, wherein the first and second light sources are selected from the group consisting of,
in the at least two monomer models for evaluating the image quality, different monomer models are trained based on different second image sample sets, wherein the image sample of each second image sample set comprises two images with different image qualities, and an image quality difference between the two images belonging to the same image sample is marked.
3. The method of claim 2, wherein the first and second light sources are selected from the group consisting of,
at least one of the at least two haplotype models used to assess image quality is a regression model.
4. The method of claim 2, wherein the first and second light sources are selected from the group consisting of,
the image quality difference between two images belonging to the same image sample in each second sample image set is determined based on the image element definition difference between the two images.
5. The method of claim 4, wherein the first and second light sources are selected from the group consisting of,
the difference of the definition of the image element between the two images belonging to the same image sample in each second sample image set is determined based on the following mode:
inputting two images belonging to the same image sample in the second sample image set into a definition classification model to obtain the definition of image elements of the two different images, and determining the definition difference of the image elements between the two different images based on the definition of the image elements of the two different images; wherein the sharpness evaluation model is trained based on a third set of image samples that have been labeled for sharpness of image elements.
6. The method of claim 1, wherein the first and second light sources are selected from the group consisting of,
the target image quality evaluation model is a classification model, and the image samples in the first image sample set are single image samples;
evaluating the image quality of a first image sample set based on the ensemble learning model, and labeling the first image sample set according to the image quality evaluation result of the first image sample set, including:
evaluating image samples of a first image sample set based on the ensemble learning model to obtain an image quality evaluation result of each image sample in the first image sample set;
and labeling each image sample in the first image sample set with a matched image quality classification label according to the image quality evaluation result of each image sample in the first image sample set.
7. The method of claim 6, wherein the first and second light sources are selected from the group consisting of,
after training a target image quality assessment model based on the labeled first image sample set, the method further comprises:
re-labeling the image quality classification labels on the first sample images of the low-score training results in the first image sample set, wherein the sample images of the low-score training results refer to sample images with training results inconsistent with the image quality classification labels, and/or sample images with the confidence values of the training results not reaching effective threshold values;
and training the target image quality evaluation model based on the first image sample set and the re-labeled image quality classification label of the first image sample set.
8. The method of claim 6, wherein the first and second light sources are selected from the group consisting of,
the target image quality evaluation model is a single classifier model.
9. The method according to any one of claims 1 to 8,
at least one of the at least two monomer models used for evaluating the image quality is obtained after model trunk replacement and/or model pruning is carried out on the basis of the other monomer model.
10. The method according to any one of claims 1 to 8,
before training a target image quality assessment model based on the labeled first image sample set, the method further includes:
and respectively selecting the characteristic vectors reaching the preset weight standard from the two monomer models at least used for evaluating the image quality to construct a target image quality evaluation model.
11. A data uploading processing method comprises the following steps:
receiving uploaded data of a user side, wherein the uploaded data comprise image data;
inputting the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by carrying out ensemble learning on at least two trained haplotype models for evaluating the image quality, and different haplotype models have different evaluation deviation of the image quality;
and executing corresponding operation on the uploaded data based on the image quality evaluation result of the image data, wherein the operation which can be executed on the uploaded data comprises accepting the uploaded data and returning the uploaded data.
12. A training apparatus of an image quality evaluation model, comprising:
the ensemble learning module is used for performing ensemble learning on at least two trained monomer models for evaluating the image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of the image quality;
the sample labeling module is used for evaluating the image quality of a first image sample set based on the ensemble learning model and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set;
and the model training module is used for training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
13. An electronic device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being executed by the processor to:
performing ensemble learning on at least two trained monomer models for evaluating image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of image quality;
evaluating the image quality of a first image sample set based on the ensemble learning model, and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set;
training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
14. A computer-readable storage medium having a computer program stored thereon, which when executed by a processor, performs the steps of:
performing ensemble learning on at least two trained monomer models for evaluating image quality to obtain an ensemble learning model considering evaluation bias of each monomer model, wherein different monomer models have different evaluation bias of image quality;
evaluating the image quality of a first image sample set based on the ensemble learning model, and labeling the sample images of the first image sample set according to the obtained image quality evaluation result of the first image sample set;
training a target image quality evaluation model based on the labeled first image sample set, wherein the number of channels of the target image quality evaluation model is less than that of the integrated learning model.
15. A data uploading processing device comprises:
the data receiving module is used for receiving uploaded data of a user side, wherein the uploaded data comprises image data;
the data evaluation module is used for inputting the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by carrying out ensemble learning on at least two trained monomer models used for evaluating the image quality, and different monomer models have different evaluation deviation of the image quality;
and the data processing module executes corresponding operation on the uploaded data based on the image quality evaluation result of the image data, wherein the operation executable on the uploaded data comprises accepting the uploaded data and returning the uploaded data.
16. An electronic device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program executed by the processor:
receiving uploading data of a user end, wherein the uploading data comprises image data;
inputting the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by carrying out ensemble learning on at least two trained monomer models for evaluating the image quality, and different monomer models have different evaluation deviation of the image quality;
and executing corresponding operation on the uploaded data based on the image quality evaluation result of the image data, wherein the operation which can be executed on the uploaded data comprises accepting the uploaded data and returning the uploaded data.
17. A computer-readable storage medium having a computer program stored thereon, which when executed by a processor, performs the steps of:
receiving uploaded data of a user side, wherein the uploaded data comprise image data;
inputting the image data into a target image quality evaluation model to obtain an image quality evaluation result of the target image data, wherein the target image quality evaluation model is obtained by training based on a first image sample set, the first image sample set is labeled aiming at the image quality based on an ensemble learning model, the ensemble learning model is obtained by carrying out ensemble learning on at least two trained haplotype models for evaluating the image quality, and different haplotype models have different evaluation deviation of the image quality;
and executing corresponding operation on the uploaded data based on the image quality evaluation result of the image data, wherein the operation which can be executed on the uploaded data comprises accepting the uploaded data and returning the uploaded data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210287406.7A CN114638999A (en) | 2022-03-23 | 2022-03-23 | Training method of image quality evaluation model, data uploading processing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210287406.7A CN114638999A (en) | 2022-03-23 | 2022-03-23 | Training method of image quality evaluation model, data uploading processing method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114638999A true CN114638999A (en) | 2022-06-17 |
Family
ID=81950588
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210287406.7A Pending CN114638999A (en) | 2022-03-23 | 2022-03-23 | Training method of image quality evaluation model, data uploading processing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114638999A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114880637A (en) * | 2022-07-07 | 2022-08-09 | 深圳竹云科技股份有限公司 | Account risk verification method and device, computer equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104318562A (en) * | 2014-10-22 | 2015-01-28 | 百度在线网络技术（北京）有限公司 | Method and device for confirming quality of internet images |
| CN112672143A (en) * | 2020-12-21 | 2021-04-16 | 北京金山云网络技术有限公司 | Video quality evaluation method and device and server |
| CN112926496A (en) * | 2021-03-19 | 2021-06-08 | 京东方科技集团股份有限公司 | Neural network for predicting image definition, training method and prediction method |
| CN113822254A (en) * | 2021-11-24 | 2021-12-21 | 腾讯科技（深圳）有限公司 | Model training method and related device |
-
2022
- 2022-03-23 CN CN202210287406.7A patent/CN114638999A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104318562A (en) * | 2014-10-22 | 2015-01-28 | 百度在线网络技术（北京）有限公司 | Method and device for confirming quality of internet images |
| CN112672143A (en) * | 2020-12-21 | 2021-04-16 | 北京金山云网络技术有限公司 | Video quality evaluation method and device and server |
| CN112926496A (en) * | 2021-03-19 | 2021-06-08 | 京东方科技集团股份有限公司 | Neural network for predicting image definition, training method and prediction method |
| CN113822254A (en) * | 2021-11-24 | 2021-12-21 | 腾讯科技（深圳）有限公司 | Model training method and related device |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114880637A (en) * | 2022-07-07 | 2022-08-09 | 深圳竹云科技股份有限公司 | Account risk verification method and device, computer equipment and storage medium |
| CN114880637B (en) * | 2022-07-07 | 2022-09-20 | 深圳竹云科技股份有限公司 | Account risk verification method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108090902B (en) | Non-reference image quality objective evaluation method based on multi-scale generation countermeasure network | |
| CN108399428B (en) | Triple loss function design method based on trace ratio criterion | |
| CN109740657B (en) | Training method and device of neural network model for image data classification | |
| CN109242149B (en) | Student score early warning method and system based on education data mining | |
| Breinlich et al. | Machine learning in international trade research-evaluating the impact of trade agreements | |
| CN111160406A (en) | Training method of image classification model, and image classification method and device | |
| CN112910690A (en) | Network traffic prediction method, device and equipment based on neural network model | |
| WO2021056914A1 (en) | Automatic modeling method and apparatus for object detection model | |
| US20230066499A1 (en) | Method for establishing defect detection model and electronic apparatus | |
| CN112668809B (en) | Method for establishing autism children rehabilitation effect prediction model | |
| CN111340233B (en) | Training method and device of machine learning model, and sample processing method and device | |
| CN111709371A (en) | Artificial intelligence based classification method, device, server and storage medium | |
| CN111797320A (en) | Data processing method, device, equipment and storage medium | |
| CN111815169A (en) | Business approval parameter configuration method and device | |
| CN114638999A (en) | Training method of image quality evaluation model, data uploading processing method and device | |
| CN112672143A (en) | Video quality evaluation method and device and server | |
| CN111444094B (en) | Test data generation method and system | |
| CN111310918B (en) | Data processing method, device, computer equipment and storage medium | |
| CN112365007A (en) | Model parameter determination method, device, equipment and storage medium | |
| CN109255389B (en) | Equipment evaluation method, device, equipment and readable storage medium | |
| CN111639230A (en) | Similar video screening method, device, equipment and storage medium | |
| TW202117664A (en) | Optical inspection secondary image classification method which can effectively improve the accuracy of image recognition and classification | |
| CN114494756A (en) | Improved clustering algorithm based on Shape-GIoU | |
| CN117078670A (en) | Production control system of cloud photo frame | |
| CN113096070A (en) | Image segmentation method based on MA-Unet |
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 |