Image quality partial order model training and application method, device, equipment and medium

Abstract

The embodiment of the present disclosure discloses an image quality partial order model training and application method, device, equipment and medium. The training method comprises the following steps: controlling an image quality partial order model to execute an image quality judgment training task and an image partial order constraint training task, learning the partial order relationship of the sample images required for training by the image partial order constraint training task, adjusting the parameters of the image quality partial order model according to the image quality judgment training task and the image partial order constraint training task, and obtaining a converged image quality partial order model. The partial order constraint is realized by learning the partial order relationship between images through the image partial order constraint training task, the problem of concentrated model prediction result distribution is solved, the distribution of the image quality output by the model is more balanced, the coverage is wider, and the image quality partial order effect is greatly improved.

Description

Technical Field

[0001] This disclosure relates to the field of image processing technology, and in particular to a method, apparatus, device, and medium for training and applying an image quality partial order model. Background Technology

[0002] In many consumer scenarios that target a broad user base, such as category pages, the image content on category pages can be configured and sorted for display, allowing users to export and use it.

[0003] Currently, image quality is primarily characterized by manually assigning scores based on feedback. However, with the growing scale of the business system and the increasing number and categories of images, it's difficult to maintain regular updates through manual configuration. Furthermore, many images of low quality, with infrequent export usage, are displayed prominently and have high click-through rates, resulting in many high-quality images failing to receive adequate exposure and being missed for export. Therefore, analyzing images on similar category pages to optimize the display of high-quality images becomes crucial. Summary of the Invention

[0004] This disclosure provides a method, apparatus, device, and medium for training and applying an image quality partial order model, in order to optimize the image quality partial order process, thereby achieving a more balanced distribution of image quality partial order and improving its coverage.

[0005] In a first aspect, this disclosure provides a method for training an image quality partial order model, the training method comprising:

[0006] The image quality partial order model is controlled to perform image quality judgment training tasks and image partial order constraint training tasks; wherein, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training.

[0007] Based on the image quality judgment training task and the image partial order constraint training task, the parameters of the image quality partial order model are adjusted to obtain a converged image quality partial order model.

[0008] Secondly, this disclosure also provides a method for applying an image quality partial order model, wherein the image quality partial order model is obtained using any of the image quality partial order model training methods described in this disclosure, and the application method includes:

[0009] The image to be processed is input into the trained image quality partial order model, and the output is the image quality of the image to be processed.

[0010] A partial image ordering operation is performed on the image to be processed based on its image quality.

[0011] Thirdly, this disclosure also provides an image quality partial order model training device, the training device comprising:

[0012] The task control module is used to control the image quality partial order model to perform image quality judgment training tasks and image partial order constraint training tasks; wherein, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training.

[0013] The model training module is used to adjust the parameters of the image quality partial order model according to the image quality judgment training task and the image partial order constraint training task, so as to obtain a converged image quality partial order model.

[0014] Fourthly, this disclosure also provides an image quality partial ordering model application device, wherein the image quality partial ordering model is obtained using any of the image quality partial ordering model training methods described in this disclosure, and the application device includes:

[0015] The image quality determination module is used to input the image to be processed into the trained image quality partial order model and output the image quality of the image to be processed.

[0016] The image partial order processing module is used to perform image partial ordering operations on the image to be processed based on the image quality of the image to be processed.

[0017] Fifthly, this disclosure also provides an electronic device, the electronic device comprising:

[0018] At least one processor; and

[0019] A memory communicatively connected to the at least one processor; wherein,

[0020] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the image quality partial order model training method or the image quality partial order model application method as described in any of the above embodiments.

[0021] Sixthly, this disclosure also provides a computer-readable medium storing computer instructions that, when executed by a processor, implement the image quality partial order model training method or the image quality partial order model application method described in any of the above embodiments.

[0022] The technical solution of this disclosure, after entering the model training phase, controls the image quality partial order model to perform image quality judgment training tasks and image partial order constraint training tasks. Through the image partial order constraint training tasks, the image quality partial order model learns the partial order relationships of the sample images required for training. Based on the image quality judgment training tasks and image partial order constraint training tasks, the parameters of the image quality partial order model are adjusted to obtain a converged image quality partial order model. This disclosure solution, by learning the partial order relationships between images through image partial order constraint training tasks to achieve partial order constraints, solves the problem of concentrated distribution of model prediction results, making the distribution of image quality output by the model more balanced, with a wider coverage, and significantly improving the image quality partial order effect.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0024] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0025] Figure 1 A flowchart of an image quality partial order model training method provided in this disclosure embodiment;

[0026] Figure 2 A flowchart illustrating another image quality partial order model training method provided in this disclosure embodiment;

[0027] Figure 3 A schematic diagram illustrating the applicable method of selecting sample image pairs from a training set, as provided in embodiments of this disclosure;

[0028] Figure 4 A schematic diagram illustrating the principle of performing image quality determination training tasks and image partial order constraint training tasks for the applicable image quality partial order model provided in the embodiments of this disclosure;

[0029] Figure 5 A schematic diagram illustrating the model performance of a single-task implementation of image quality partial order model training provided in the embodiments of this disclosure;

[0030] Figure 6 A schematic diagram illustrating the model performance of a suitable multi-task image quality partial order model training method provided in the embodiments of this disclosure;

[0031] Figure 7 A flowchart of another image quality partial order model training method provided in this disclosure embodiment;

[0032] Figure 8 A schematic diagram illustrating the applicable method of selecting sample image pairs from the training set to construct training data, as provided in embodiments of this disclosure;

[0033] Figure 9 A flowchart illustrating an image quality partial order model application method provided in this embodiment of the disclosure;

[0034] Figure 10 A structural block diagram of an image quality partial order model training device provided in this embodiment of the present disclosure;

[0035] Figure 11 A structural block diagram of an image quality partial order model application device provided in this disclosure embodiment;

[0036] Figure 12 A structural block diagram of an electronic device for implementing the image quality partial order model training method or image quality partial order model application method of the embodiments of this disclosure. Detailed Implementation

[0037] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0038] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0039] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0040] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0041] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0042] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0043] In the following embodiments, each embodiment provides optional features and examples. The various features described in the embodiments can be combined to form multiple optional solutions. Each numbered embodiment should not be regarded as only one technical solution. Furthermore, unless otherwise specified, the embodiments and features in the embodiments of this disclosure can be combined with each other.

[0044] Figure 1 This is a flowchart illustrating an image quality partial order model training method provided in this embodiment. The technical solution of this embodiment is applicable to situations where partial order processing is performed between images to filter high-quality images. This method can be executed by an image quality partial order model training device, which can be implemented by software and / or hardware and is generally integrated into any electronic device with network communication capabilities, including but not limited to: computers, personal digital assistants, etc. Figure 1 As shown, the image quality partial order model training method of this embodiment may include the following steps S110-S120:

[0045] S110, Control the image quality partial order model to perform image quality judgment training task and image partial order constraint training task.

[0046] Among them, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training.

[0047] During the model training phase, when training the image quality partial order model, a training set can be obtained. The training sample images from this set are then input into the image quality partial order model to be trained for image quality prediction. The model is then trained and updated based on the prediction results. Image quality refers to the evaluation of visual perception of an image.

[0048] When it comes to image quality assessment, the training sample images are input into the image quality partial order model for training and updating. However, the partial order information between images is lacking. If the training method of directly fitting the image quality is used for model training, the image quality output by the image quality partial order model will tend to converge towards the median mean. The image skew effect is not obvious, resulting in a relatively concentrated distribution of the image quality predicted by the model, which does not meet the actual image quality distribution requirements.

[0049] Therefore, this application will simultaneously introduce an image partial order constraint training task during the training process of the image quality judgment training task to fit the image quality of the image quality partial order model. The image partial order constraint training task will learn the partial order relationship between images to achieve partial order constraint, so that the distribution of the image quality output by the model is more balanced and the coverage is wider.

[0050] As an optional but not restrictive implementation, the image quality output by the image quality partial order model can be measured and represented by, but is not limited to, the following metrics: image click-to-display ratio, image export-to-display ratio, and image export-to-click ratio.

[0051] The image click-through rate (CTR) is determined by the ratio of the number of times an image is clicked to the number of times it is viewed. The image export through rate (ETR) is determined by the ratio of the number of times an image is exported to the number of times it is viewed. The image export click rate (ECR) is determined by the ratio of the number of times an image is exported to the number of times it is selected by clicking.

[0052] Image quality evaluation primarily relies on manual scoring to construct training datasets, which is heavily influenced by subjective human perception and is difficult to annotate extensively. Therefore, defining image quality skew manually is challenging; only a small number of high-quality head images can be selected from the entire dataset, making it impossible to manually annotate the overall image quality skew or to extract a large number of low-quality images as training negatives. To address this difficulty in manual annotation, posterior metrics related to image consumption, such as image click-through rate, image export rate, and image export click-through rate, are used as metrics to measure model fit. These metrics directly characterize image quality, suggesting that users' image selection process implicitly includes a judgment on image quality skew.

[0053] As an optional but non-restrictive implementation, the image quality partial order model uses the image export click ratio as a metric for judging image quality during model training.

[0054] For image click-to-show ratio, image export-to-show ratio, and image export-to-click ratio, since the image click-to-show ratio (CTR) and image export-to-show ratio (ETR) are greatly affected by the original display order of the image, while the image export-to-click ratio (ECR) is the probability of actually exporting the finished image after the user has already clicked on the image, which has eliminated the influence of the image display order and the user's judgment of image quality is more accurate, the image export-to-click ratio (ECR) is used as the fitting index for model training.

[0055] S120. Based on the image quality judgment training task and the image partial order constraint training task, adjust the parameters of the image quality partial order model to obtain a converged image quality partial order model.

[0056] For the trained image quality partial ordering model, the image quality partial ordering model can perform image quality judgment, especially the image quality judgment operation of the images displayed on the classification page, so as to perform partial ordering of different images according to the output image quality, and select the high-quality images for sorting and display.

[0057] The predicted image quality output by the image quality partial order model can, to some extent, reflect the importance of an image when displayed on the classification page. Thus, based on the predicted image quality, the partial order relationship between an image and other images can be determined, guiding the ranking of images displayed on the classification page.

[0058] According to the technical solution of this disclosure embodiment, after entering the model training stage, the partial order relationship between images is learned through the image partial order constraint training task to realize the partial order constraint. The image quality judgment training task is guided by the partial order, which solves the problem of concentrated distribution of model prediction results. It makes the output image quality of the model as far as possible to be constrained and optimized in the direction with the same partial order as before, so that the distribution of the output image quality of the model is more balanced and the coverage is wider, which greatly improves the image quality partial order effect.

[0059] Figure 2 This is a flowchart illustrating another image quality partial order model training method provided in this embodiment. Based on the above embodiments, this embodiment further optimizes the process of controlling the image quality partial order model to perform image quality judgment training tasks and image partial order constraint training tasks in each embodiment. This embodiment can be combined with various optional solutions in one or more of the above embodiments. Figure 2 As shown, the image quality partial order model training method of this embodiment may include the following steps S210-S240:

[0060] S210. Determine the sample image pairs required for training the image quality partial order model from the training set.

[0061] See Figure 3Based on posterior metrics of business data, such as Image Export Click-Through Rate (ECR) as an image quality metric, the ECR of the massive dataset of acquired images is calculated automatically, eliminating the need for manual scoring and constructing training set data suitable for model training. Simultaneously, to ensure the model learns the partial order relationships between images during training, for each input sample image, a corresponding reference image from the training set is found to form a sample image pair. This allows the partial order relative to the reference image to assist the image quality partial order model in image quality judgment training during training with the training sample images.

[0062] S220. Perform image quality prediction on sample image pairs using an image quality partial order model to execute image quality judgment training tasks and image partial order constraint training tasks.

[0063] Among them, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training.

[0064] S230. Based on the image quality prediction operation, determine the loss information of the image quality judgment training task and the loss information of the image partial order constraint training task.

[0065] Among them, the loss information of the image partial order constraint training task enables the image quality partial order model to learn the partial order relationship between each sample image in the sample image pair.

[0066] By sequentially inputting each sample image in a sample image pair into the image quality partial order model to perform image quality prediction, the predicted image quality of each sample image in the sample image pair can be obtained. Combined with the pre-labeled image quality of each sample image in the sample image pair, the loss information when performing the image quality judgment training task and the loss information when performing the image partial order constraint training task can be calculated.

[0067] As an optional but non-limiting implementation, determining the loss information for the image quality judgment training task and the loss information for the image partial order constraint training task based on the image quality prediction operation may include the following steps A1-A2:

[0068] Step A1: Based on the predicted image quality of each sample image in the sample image pair obtained from the image quality prediction operation and the corresponding pre-labeled image quality, determine the loss function score of the image quality judgment training task.

[0069] After the image quality partial order model performs image quality prediction and obtains the predicted image quality score for each sample image in the sample image pair, the loss can be calculated using, but is not limited to, the root mean square error (RMSE). The loss function score for each sample image in the sample image pair when performing the image quality judgment training task can then be used as the loss function score for the image quality judgment training task. The formula for calculating the RMSE loss is as follows:

[0070]

[0071] Among them, ecr i Pred represents the pre-labeled image quality of the i-th training sample image. i The image quality is represented by the predicted image quality of the i-th training sample image during model training, and n represents the number of sample images participating in the training of the image quality partial order model. The image quality can be measured by the image export click ratio.

[0072] Step A2: Based on the difference in pre-annotated image quality between each sample image in the sample image pair obtained from the image quality prediction operation and the corresponding difference in predicted image quality, determine the loss function score for the image partial order constraint training task.

[0073] After the image quality partial order model performs image quality prediction and obtains the predicted image quality score of each sample image in the sample image pair, it can calculate the difference in predicted image quality between each sample image in the sample image pair and the difference in pre-annotated image quality between each sample image in the sample image pair. Here, the root mean square error (RMSE) can still be used for loss calculation, and the loss between the difference in pre-annotated image quality and the difference in predicted image quality of the sample image pair can be obtained respectively, which serves as the loss function score for the image partial order constraint training task.

[0074] It is understandable that the difference between the loss function score calculation formula for the image partial order constraint training task and the loss function score calculation formula for the aforementioned image quality judgment training task lies in the fact that in the formula, ecr... i With pred i The meaning of ECR ​​has been slightly adjusted. i Change to represent the pre-labeled image quality difference of the i-th sample image pair, pred i Instead, it represents the difference in predicted image quality for the i-th sample image pair during model training.

[0075] See Figure 3 See also Figure 4Taking each sample image in a sample image pair as the baseline image and the reference image as an example, the baseline image A and the reference image B are respectively processed by the image quality partial order model to perform image quality prediction operation and output the corresponding predicted image quality score A and predicted image quality score B. These scores are then fitted with their own pre-labeled image quality scores A and B, and the loss function scores when performing the image quality judgment training task with the baseline image A and the reference image B are calculated. These loss function scores are then used as the loss function scores for the image quality judgment training task.

[0076] See also Figure 4 Simultaneously, the loss function scores of the pre-labeled image quality difference and the corresponding predicted image quality difference between the baseline image A and the reference image B are calculated. The loss information from the image partial order constraint training task is used to further constrain the loss, allowing the image quality partial order model to learn the partial order relationship between sample image pairs. That is, if the pre-labeled image quality score A is greater than the pre-labeled image quality score B, then during model training, it is desired that the predicted image quality score A is greater than the predicted image quality score B when the model outputs the image quality; conversely, if the pre-labeled image quality score A is less than the pre-labeled image quality score B, then during model training, it is desired that the predicted image quality score A is less than the predicted image quality B when the model outputs the image quality. This ensures that the image quality partial order model calculates the output image quality with the same partial order as before, solving the problem of concentrated model prediction distribution.

[0077] S240. Based on the image quality judgment training task and the image partial order constraint training task, adjust the parameters of the image quality partial order model to obtain a converged image quality partial order model.

[0078] Based on this, the image quality partial order model can be trained to output image quality based on the loss information when performing the image quality judgment training task. At the same time, the loss information when performing the image partial order constraint training task is introduced to constrain and optimize the image quality output by the image quality partial order model to have the same partial order direction as the original sample image, so that the distribution of the image quality output by the image quality partial order model is more balanced and the coverage is wider, thus greatly improving the image quality partial order effect.

[0079] As an optional but non-limiting implementation, the parameters of the image quality partial order model are adjusted based on the image quality judgment training task and the image partial order constraint training task, which may include steps B1-B3:

[0080] Step B1: Determine the loss weights for the image quality judgment training task and the image partial order constraint training task.

[0081] Step B2: Combine the loss information from the image quality judgment training task with the loss information from the image partial order constraint training task according to the loss weights of their respective tasks.

[0082] Step B3: Based on the loss information obtained from loss fusion, adjust the parameters of the image quality partial order model until the image quality partial order model converges.

[0083] When controlling the image quality partial order model to perform image quality judgment training tasks and image partial order constraint training tasks, the loss information of the image quality judgment training task and the loss information of the image partial order constraint training task can be obtained respectively during the training task execution process.

[0084] For the image quality judgment training task and the image partial order constraint training task, although different training tasks are executed simultaneously, considering that the image quality partial order model is to complete the image quality judgment, while the image partial order constraint training task is to achieve the correction assistance for the image quality judgment, it is not objective to simply superimpose the loss information of different training tasks. It is necessary to analyze the loss weights of the image quality judgment training task and the image partial order constraint training task to ensure that the loss fusion ratio of the training tasks is more in line with the actual image quality judgment scenario.

[0085] By adopting the above method, the loss of the image quality judgment training task and the loss of the image partial order constraint training task are fused together. This allows the model to learn the partial order information between images while training on the image quality of the images. As a result, the model calculates the output quality score with the same partial order as before, thus solving the problem of the concentrated distribution of the model's predicted output results.

[0086] See Figure 5 This paper illustrates the comparison between the predicted image quality distribution and the actual pre-labeled image quality distribution after the training of the image quality partial order model in a single task has converged. The predicted image quality is more concentrated than the actual pre-labeled image quality, resulting in a certain gap between the predicted and actual values ​​of the head and tail data. The model's predicted distribution is relatively concentrated.

[0087] See Figure 6 This paper illustrates the comparison between the predicted image quality distribution and the actual pre-labeled image quality distribution after the multi-task implementation of the image quality partial order model converges, which incorporates image partial order constraints during training. By increasing the number of sample image pairs to impose partial order constraints during model training, the number of predictions for head and tail data increases, thus narrowing the gap between the predicted image quality distribution of head and tail data and the actual pre-labeled image quality distribution. The predicted distribution is closer to the actual pre-labeled image quality, and the model's predicted distribution is more dispersed.

[0088] According to the technical solution of this disclosure embodiment, after entering the model training stage, the partial order relationship between images is learned through the image partial order constraint training task to realize the partial order constraint, which solves the problem of concentrated distribution of model prediction results. It optimizes the output image quality of the model as much as possible in the direction with the same partial order as before, so that the model learns the partial order information between images while calculating the image quality, so that the model calculates the output quality score with the same partial order as before, so that the distribution of the output image quality of the model is more balanced and the coverage is wider, which greatly improves the image quality partial order effect.

[0089] Figure 7 This is a flowchart illustrating another image quality partial order model training method provided in this disclosure. The technical solution of this embodiment further optimizes the process of determining the sample image pairs required for image quality partial order model training from the training set in each of the above embodiments, based on the above embodiments. This embodiment can be combined with various optional solutions in one or more of the above embodiments. Figure 7 As shown, the image quality partial order model training method of this embodiment may include the following steps S210-S240:

[0090] S710. Using each sample image in the training set as the reference image, select sample images from the training set that satisfy the preset image quality difference relationship with the reference image as the reference image.

[0091] S720. Combine the baseline image with the reference image corresponding to the baseline image to form the sample image pair required for training the image quality partial order model.

[0092] Taking the image export click-through rate (ECR) as an image quality metric as an example, the ECR of each sample image is calculated. To facilitate the learning of partial order relationships during model training, sample image pairs need to be constructed beforehand. Furthermore, each sample image pair should exhibit significant differences in image quality between its constituent images to better demonstrate a clear partial order relationship. Training the model using sample image pairs allows it to learn the partial order information between these pairs while simultaneously calculating image quality, ensuring that the model's output image quality reflects the same partial order relationship as the original sample image pair.

[0093] As an optional but non-limiting implementation, the number of image clicks on the sample images in the training set is greater than a preset threshold, and the image quality distribution of each sample image in the training set is uniform after being sorted by the image export click ratio.

[0094] To ensure the reliability of training sample images, images with more than 50 clicks can be selected for the training set. For the sample images in the training set, the image export click-through rate (ECR) is used as a measure of image quality. The ECR is calculated, and each image is sorted from highest to lowest. If the ECR is found to be evenly distributed and not clustered in a single segment, it can well represent the image quality distribution. Therefore, by selecting images with sufficient posterior data and calculating their ECR, the selection of training set data for model training can be automated.

[0095] As an optional but non-limiting implementation, the preset image quality difference relationship includes the training set standard deviation where the difference between the image export click ratio of the baseline image and the reference image is greater than a preset multiple. The training set standard deviation is determined based on the standard deviation of the image export click ratio of each sample image in the training set.

[0096] See Figure 8 The algorithm iterates through each sample image in the training set as a baseline image and randomly selects another image from the test set as a reference image. For each sample image pair, the difference between the baseline and reference images must be greater than the overall standard deviation of the ECR of the sample images in the training set. A difference of one standard deviation can be considered a threshold for significant differences, thus selecting a reference image for each baseline image. The preset multiplier can be one or more times.

[0097] S730. Perform image quality prediction on sample image pairs using an image quality partial order model to execute image quality judgment training tasks and image partial order constraint training tasks.

[0098] Among them, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training.

[0099] S740. Based on the image quality prediction operation, determine the loss information of the image quality judgment training task and the loss information of the image partial order constraint training task.

[0100] Among them, the loss information of the image partial order constraint training task enables the image quality partial order model to learn the partial order relationship between each sample image in the sample image pair.

[0101] S750. Based on the image quality judgment training task and the image partial order constraint training task, the parameters of the image quality partial order model are adjusted to obtain a converged image quality partial order model.

[0102] According to the technical solution of this disclosure, after entering the model training stage, a suitable sample image pair is selected, and the partial order relationship between the images in the sample image pair is learned through the image partial order constraint training task to achieve partial order constraint. This solves the problem of concentrated distribution of model prediction results, and makes the output image quality of the model as far as possible to be constrained and optimized in the direction with the same partial order as before, so that the distribution of the output image quality of the model is more balanced and the coverage is wider, which greatly improves the image quality partial order effect.

[0103] Figure 9 This is a flowchart illustrating an image quality partial order model application method provided in this embodiment. The technical solution of this embodiment is applicable to situations where partial order processing is performed between images to filter high-quality images. This method can be executed by an image quality partial order model application device, which can be implemented by software and / or hardware and is generally integrated into any electronic device with network communication capabilities, including but not limited to: computers, personal digital assistants, etc. Figure 9 As shown, the image quality partial order model application method of this embodiment may include the following steps S910-S920:

[0104] S910. Input the image to be processed into the trained image quality partial order model and output the image quality of the image to be processed.

[0105] The image quality partial ordering model is obtained using any of the image quality partial ordering model training methods described in the above embodiments. After entering the model application stage, the image to be processed can be input into the trained image quality partial ordering model to output the image quality of the image to be processed.

[0106] S920. Perform an image partial ordering operation on the image to be processed based on the image quality of the image to be processed.

[0107] According to the technical solution of the present disclosure, the image quality partial order model learns the partial order relationship between images through the image partial order constraint training task to achieve partial order constraint, which solves the problem of concentrated distribution of model prediction results. It optimizes the output image quality of the model as much as possible in the direction with the same partial order as the original, so that the distribution of the output image quality of the model is more balanced and the coverage is wider, which greatly improves the image quality partial order effect.

[0108] Figure 10 This is a structural block diagram of an image quality partial order model training device provided in this embodiment. The technical solution of this embodiment can be applied to situations where partial order processing is performed between images to filter high-quality images. This device can be implemented by software and / or hardware and is generally integrated into any electronic device with network communication capabilities, including but not limited to: computers, personal digital assistants, etc. Figure 10As shown, the image quality partial order model training device of this embodiment may include the following: a task control module 1010 and a model training module 1020. Wherein:

[0109] The task control module 1010 is used to control the image quality partial order model to perform image quality judgment training tasks and image partial order constraint training tasks; wherein, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training.

[0110] The model training module 1020 is used to adjust the parameters of the image quality partial order model according to the image quality judgment training task and the image partial order constraint training task, so as to obtain a converged image quality partial order model.

[0111] Based on the above embodiments, optionally, the image quality partial order model is trained using the image export click ratio as a metric for image quality determination.

[0112] Based on the above embodiments, optionally, controlling the image quality partial order model to perform image quality judgment training tasks and image partial order constraint training tasks includes:

[0113] Determine the sample image pairs required for training the image quality partial order model from the training set;

[0114] The image quality partial order model is used to perform image quality prediction on the sample image pairs to execute image quality judgment training tasks and image partial order constraint training tasks.

[0115] The loss information for the image quality judgment training task and the loss information for the image partial order constraint training task are determined based on the image quality prediction operation.

[0116] The loss information of the image partial order constraint training task enables the image quality partial order model to learn the partial order relationship between each sample image in the sample image pair.

[0117] Based on the above embodiments, optionally, the sample image pairs required for training the image quality partial order model are determined from the training set, including:

[0118] Using each sample image in the training set as a reference image, a sample image that satisfies a preset image quality difference relationship with the reference image is selected from the training set as a reference image corresponding to the reference image;

[0119] The reference image and the corresponding reference image are combined to form the sample image pair required for training the image quality partial order model.

[0120] Based on the above embodiments, optionally, the preset image quality difference relationship includes the training set standard deviation where the difference in the image export click ratio between the reference image and the reference image is greater than a preset multiple, and the training set standard deviation is determined based on the standard deviation of the image export click ratio of each sample image in the training set.

[0121] Based on the above embodiments, optionally, the number of image clicks on the sample images in the training set is greater than a preset threshold, and the image quality distribution of each sample image in the training set is uniform after being sorted by the size of the image export click ratio.

[0122] Based on the above embodiments, optionally, determining the loss information for the image quality judgment training task and the loss information for the image partial order constraint training task based on the image quality prediction operation includes:

[0123] Based on the predicted image quality of each sample image in the sample image pair obtained by the image quality prediction operation and their corresponding pre-labeled image quality, the loss function score of the image quality judgment training task is determined.

[0124] Based on the difference in pre-annotated image quality between each sample image in the sample image pair obtained by the image quality prediction operation and the corresponding difference in predicted image quality, the loss function score of the image partial order constraint training task is determined.

[0125] Based on the above embodiments, optionally, the parameters of the image quality partial order model are adjusted according to the image quality judgment training task and the image partial order constraint training task, including:

[0126] Determine the loss weights for the image quality judgment training task and the loss weights for the image partial order constraint training task;

[0127] The loss information of the image quality judgment training task and the loss information of the image partial order constraint training task are fused according to the loss weights of their respective tasks.

[0128] Based on the loss information obtained by loss fusion, the parameters of the image quality partial order model are adjusted until the image quality partial order model converges.

[0129] The image quality partial order model training apparatus provided in this embodiment can execute the image quality partial order model training method provided in any of the above embodiments of this disclosure, and has the corresponding functions and beneficial effects of executing the image quality partial order model training method. For details, please refer to the relevant operations of the image quality partial order model training method in the foregoing embodiments.

[0130] Figure 11This is a structural block diagram of an image quality partial order model application device provided in an embodiment of this disclosure. The technical solution of this embodiment can be applied to situations where partial order processing is performed between images to filter high-quality images. This device can be implemented by software and / or hardware and is generally integrated into any electronic device with network communication capabilities, including but not limited to: computers, personal digital assistants, etc. Figure 11 As shown, the image quality partial order model application device of this embodiment may include the following: an image quality determination module 1110 and an image partial order processing module 1120. Wherein:

[0131] The image quality determination module 1110 is used to input the image to be processed into the trained image quality partial order model and output the image quality of the image to be processed.

[0132] The image quality partial order model is obtained using any of the image quality partial order model training methods described in the above embodiments.

[0133] The image partial order processing module 1120 is used to perform an image partial order operation on the image to be processed based on the image quality of the image to be processed.

[0134] The image quality partial order model application device provided in this embodiment can execute the image quality partial order model application method provided in any of the above embodiments of this disclosure, and has the corresponding functions and beneficial effects of executing the image quality partial order model application method. For details, please refer to the relevant operations of the image quality partial order model application method in the foregoing embodiments.

[0135] The following is for reference. Figure 12 The diagram illustrates a structural schematic of an electronic device 1200 suitable for implementing embodiments of the present disclosure. The terminal devices in the embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 12 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0136] like Figure 12As shown, the electronic device 1200 may include a processing unit (e.g., a central processing unit, a graphics processor, etc.) 1201, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1202 or a program loaded from a storage device 1206 into a random access memory (RAM) 1203. The RAM 1203 also stores various programs and data required for the operation of the electronic device 1200. The processing unit 1201, ROM 1202, and RAM 1203 are interconnected via a bus 1204. An input / output (I / O) interface 1205 is also connected to the bus 1204.

[0137] Typically, the following devices can be connected to I / O interface 1205: input devices 1206 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 1207 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1206 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1209. Communication device 1209 allows electronic device 1200 to communicate wirelessly or wiredly with other devices to exchange data. Although... Figure 12 An electronic device 1200 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0138] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0139] The electronic device provided in this embodiment belongs to the same inventive concept as the image quality partial order model training method or image quality partial order model application method provided in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0140] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the image quality partial order model training method or the image quality partial order model application method shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 1209, or installed from storage device 1206, or installed from ROM 1202. When the computer program is executed by processing device 1201, it performs the functions defined in the image quality partial order model training method or the image quality partial order model application method of the embodiments of this disclosure.

[0141] This disclosure provides a computer storage medium storing a computer program that, when executed by a processor, implements the image quality partial order model training method or the image quality partial order model application method provided in the above embodiments.

[0142] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0143] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0144] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0145] The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: control an image quality partial order model to perform an image quality judgment training task and an image partial order constraint training task; wherein, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training; and adjust the parameters of the image quality partial order model according to the image quality judgment training task and the image partial order constraint training task to obtain a converged image quality partial order model.

[0146] Alternatively, the aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: input the image to be processed into a trained image quality partial ordering model and output the image quality of the image to be processed; and perform an image partial ordering operation on the image to be processed based on the image quality of the image to be processed.

[0147] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0148] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0149] The units described in the embodiments of this disclosure can be implemented in software or in hardware. The name of a unit does not necessarily limit the unit itself; for example, the first acquisition unit can also be described as "a unit that acquires at least two Internet Protocol addresses".

[0150] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0151] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0152] According to one or more embodiments of this disclosure, Example 1 provides a method for training an image quality partial order model, the training method comprising:

[0153] The image quality partial order model is controlled to perform image quality judgment training tasks and image partial order constraint training tasks; wherein, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training.

[0154] Based on the image quality judgment training task and the image partial order constraint training task, the parameters of the image quality partial order model are adjusted to obtain a converged image quality partial order model.

[0155] Example 2: According to the method described in Example 1, the image quality partial order model is trained using the image export click ratio as a metric for image quality determination.

[0156] Example 3, based on the method described in Example 1, controls the image quality partial order model to perform image quality judgment training tasks and image partial order constraint training tasks, including:

[0157] Determine the sample image pairs required for training the image quality partial order model from the training set;

[0158] The image quality partial order model is used to perform image quality prediction on the sample image pairs to execute image quality judgment training tasks and image partial order constraint training tasks.

[0159] The loss information for the image quality judgment training task and the loss information for the image partial order constraint training task are determined based on the image quality prediction operation.

[0160] The loss information of the image partial order constraint training task enables the image quality partial order model to learn the partial order relationship between each sample image in the sample image pair.

[0161] Example 4, according to the method described in Example 3, determines the sample image pairs required for training the image quality partial order model from the training set, including:

[0162] Using each sample image in the training set as a reference image, a sample image that satisfies a preset image quality difference relationship with the reference image is selected from the training set as a reference image corresponding to the reference image;

[0163] The reference image and the corresponding reference image are combined to form the sample image pair required for training the image quality partial order model.

[0164] Example 5: According to the method described in Example 4, the preset image quality difference relationship includes the training set standard deviation where the difference in the image export click ratio between the baseline image and the reference image is greater than a preset multiple. The training set standard deviation is determined based on the standard deviation of the image export click ratio of each sample image in the training set.

[0165] Example 6: According to the method described in Example 4, the number of image clicks on the sample images in the training set is greater than a preset threshold, and the image quality distribution of each sample image in the training set is uniform after being sorted by the size of the image export click ratio.

[0166] Example 7, based on the method described in Example 3, determines the loss information for the image quality judgment training task and the loss information for the image partial order constraint training task based on the image quality prediction operation, including:

[0167] Based on the predicted image quality of each sample image in the sample image pair obtained by the image quality prediction operation and their corresponding pre-labeled image quality, the loss function score of the image quality judgment training task is determined.

[0168] Based on the difference in pre-annotated image quality between each sample image in the sample image pair obtained by the image quality prediction operation and the corresponding difference in predicted image quality, the loss function score of the image partial order constraint training task is determined.

[0169] Example 8, according to the method described in Example 1, adjusts the parameters of the image quality partial order model based on the image quality judgment training task and the image partial order constraint training task, including:

[0170] Determine the loss weights for the image quality judgment training task and the loss weights for the image partial order constraint training task;

[0171] The loss information of the image quality judgment training task and the loss information of the image partial order constraint training task are fused according to the loss weights of their respective tasks.

[0172] Based on the loss information obtained by loss fusion, the parameters of the image quality partial order model are adjusted until the image quality partial order model converges.

[0173] According to one or more embodiments of this disclosure, Example 9 provides a method for applying an image quality partial order model, wherein the image quality partial order model is obtained using any of the image quality partial order model training methods described in Examples 1-8 above, and the application method includes:

[0174] The image to be processed is input into the trained image quality partial order model, and the output is the image quality of the image to be processed.

[0175] A partial image ordering operation is performed on the image to be processed based on its image quality.

[0176] According to one or more embodiments of this disclosure, Example 10 provides an image quality partial order model training apparatus, the training apparatus comprising:

[0177] The task control module is used to control the image quality partial order model to perform image quality judgment training tasks and image partial order constraint training tasks; wherein, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training.

[0178] The model training module is used to adjust the parameters of the image quality partial order model according to the image quality judgment training task and the image partial order constraint training task, so as to obtain a converged image quality partial order model.

[0179] According to one or more embodiments of this disclosure, Example 11 provides an image quality partial order model application device, wherein the image quality partial order model is obtained using any of the image quality partial order model training methods described in Examples 1-8 above, and the application device includes:

[0180] The image quality determination module is used to input the image to be processed into the trained image quality partial order model and output the image quality of the image to be processed.

[0181] The image partial order processing module is used to perform image partial ordering operations on the image to be processed based on the image quality of the image to be processed.

[0182] According to one or more embodiments of this disclosure, Example 12 provides an electronic device, the electronic device comprising:

[0183] At least one processor; and

[0184] A memory communicatively connected to the at least one processor; wherein,

[0185] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the image quality partial order model training method described in any one of Examples 1-8 or the image quality partial order model application method described in Example 9.

[0186] According to one or more embodiments of the present disclosure, Example 13 provides a computer-readable medium storing computer instructions for causing a processor to execute and implement the image quality partial order model training method of any one of Examples 1-8 or the image quality partial order model application method of Example 9.

[0187] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0188] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0189] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for training an image quality partial order model, characterized in that, The training method includes: The image quality partial order model is controlled to perform image quality judgment training tasks and image partial order constraint training tasks; wherein, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training, and the partial order relationship is the partial order relationship between images. Based on the image quality judgment training task and the image partial order constraint training task, the parameters of the image quality partial order model are adjusted to obtain a converged image quality partial order model. The image quality partial order model performs image quality judgment training tasks and image partial order constraint training tasks, including: Determine the sample image pairs required for training the image quality partial order model from the training set; The image quality partial order model is used to perform image quality prediction on the sample image pairs to execute image quality judgment training tasks and image partial order constraint training tasks. The loss information for the image quality judgment training task and the loss information for the image partial order constraint training task are determined based on the image quality prediction operation. The loss information of the image partial order constraint training task enables the image quality partial order model to learn the partial order relationship between each sample image in the sample image pair.

2. The method according to claim 1, characterized in that, The image quality partial order model is trained using the image export click ratio as a metric for judging image quality.

3. The method according to claim 1, characterized in that, The sample image pairs required for training the image quality partial order model are determined from the training set, including: Using each sample image in the training set as a reference image, a sample image that satisfies a preset image quality difference relationship with the reference image is selected from the training set as a reference image corresponding to the reference image; The reference image and the corresponding reference image are combined to form the sample image pair required for training the image quality partial order model.

4. The method according to claim 3, characterized in that, The preset image quality difference relationship includes the training set standard deviation where the difference in the image export click ratio between the baseline image and the reference image is greater than a preset multiple. The training set standard deviation is determined based on the standard deviation of the image export click ratio of each sample image in the training set.

5. The method according to claim 3, characterized in that, The number of image clicks on the sample images in the training set is greater than a preset threshold, and the image quality distribution of each sample image in the training set is uniform after being sorted by the size of the image export click ratio.

6. The method according to claim 1, characterized in that, The loss information for the image quality judgment training task and the loss information for the image partial order constraint training task are determined based on the image quality prediction operation, including: Based on the predicted image quality of each sample image in the sample image pair obtained by the image quality prediction operation and the corresponding pre-labeled image quality, the loss function score of the image quality judgment training task is determined. Based on the difference in pre-annotated image quality between each sample image in the sample image pair obtained by the image quality prediction operation and the corresponding difference in predicted image quality, the loss function score of the image partial order constraint training task is determined.

7. The method according to claim 1, characterized in that, Based on the image quality judgment training task and the image partial order constraint training task, the parameters of the image quality partial order model are adjusted, including: Determine the loss weights for the image quality judgment training task and the loss weights for the image partial order constraint training task; The loss information of the image quality judgment training task and the loss information of the image partial order constraint training task are fused according to the loss weights of their respective tasks. Based on the loss information obtained by loss fusion, the parameters of the image quality partial order model are adjusted until the image quality partial order model converges.

8. A method for applying an image quality partial order model, characterized in that, The image quality partial order model is obtained using the image quality partial order model training method described in any one of claims 1-7 above, and the application method includes: The image to be processed is input into the trained image quality partial order model, and the output is the image quality of the image to be processed. A partial image ordering operation is performed on the image to be processed based on its image quality.

9. An image quality partial order model training device, characterized in that, The training device includes: The task control module is used to control the image quality partial order model to perform image quality judgment training tasks and image partial order constraint training tasks; wherein, the image partial order constraint training task is used to enable the image quality partial order model to learn the partial order relationship of the sample images required for training, and the partial order relationship is the partial order relationship between images. The task control module is specifically used for: The image quality partial order model is trained by determining the sample image pairs required for training from the training set. The image quality partial order model is then used to perform image quality prediction on these sample image pairs to execute image quality judgment training and image partial order constraint training tasks. Based on the image quality prediction operation, the loss information for the image quality judgment training task and the loss information for the image partial order constraint training task are determined. The loss information for the image partial order constraint training task enables the image quality partial order model to learn the partial order relationships between the sample images in the sample image pairs. The model training module is used to adjust the parameters of the image quality partial order model according to the image quality judgment training task and the image partial order constraint training task, so as to obtain a converged image quality partial order model.

10. An application device for an image quality partial order model, characterized in that, The image quality partial order model is obtained using the image quality partial order model training method described in any one of claims 1-7 above, and the application device includes: The image quality determination module is used to input the image to be processed into the trained image quality partial order model and output the image quality of the image to be processed. The image partial order processing module is used to perform image partial ordering operations on the image to be processed based on the image quality of the image to be processed.

11. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the image quality partial order model training method of any one of claims 1-7 or the image quality partial order model application method of claim 8.

12. A computer-readable medium, characterized in that, The computer-readable medium stores computer instructions that cause a processor to execute the image quality partial order model training method of any one of claims 1-7 or the image quality partial order model application method of claim 8.

Images