Video quality evaluation method and system, electronic device and storage medium

Abstract

The embodiment of the application discloses a video quality evaluation method and system, electronic equipment and storage medium. The technical scheme provided by the embodiment of the application comprises the following steps: obtaining a target video, then determining the correlation features between adjacent two video frames in the target video based on a space-time attention mechanism, and fusing the correlation features to obtain the global features of the target video. Then, the global features of the target video are input into a pre-constructed video quality evaluation model, and the video quality evaluation result of the target video is output. The video quality evaluation model pre-determines the global features of each training video through the space-time attention mechanism and feature fusion, and performs model training based on the global features of each training video. The above technical means can make the video quality evaluation model learn the video features sufficiently, realize more comprehensive video quality evaluation, and improve the video quality evaluation precision of the model.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a video quality assessment method, system, electronic device, and storage medium. Background Technology

[0002] Currently, in video quality assessment scenarios, deep neural networks are typically used to build video quality assessment models to improve the assessment results. These models are trained based on video frame samples, and once trained, they can output corresponding video quality assessment results for the input video frames.

[0003] However, traditional video quality assessment models typically only consider the image features of individual video frames when evaluating video quality, lacking the ability to learn the feature relationships between video frames. This results in the video quality assessment model failing to fully learn video features, thus limiting its evaluation performance. Summary of the Invention

[0004] This application provides a video quality assessment method, system, electronic device, and storage medium that can combine the spatial and temporal domains of video frames for feature learning, thereby improving the accuracy of video quality assessment and solving the technical problem of low assessment accuracy in traditional video quality assessment models.

[0005] In a first aspect, embodiments of this application provide a video quality assessment method, comprising:

[0006] The target video is acquired, and the correlation features between two adjacent video frames in the target video are determined based on the spatiotemporal attention mechanism. The global features of the target video are obtained by fusing the various correlation features.

[0007] The global features of the target video are input into a pre-built video quality assessment model, which outputs the video quality assessment result of the target video. The video quality assessment model pre-determines the global features of each training video through a spatiotemporal attention mechanism and feature fusion, and trains the model based on the global features of each training video.

[0008] In a second aspect, embodiments of this application provide a video quality assessment system, comprising:

[0009] The fusion module is used to acquire the target video, determine the correlation features between two adjacent video frames in the target video based on the spatiotemporal attention mechanism, and fuse the various correlation features to obtain the global features of the target video.

[0010] The evaluation module is used to input the global features of the target video into a pre-built video quality evaluation model and output the video quality evaluation result of the target video. The video quality evaluation model pre-determines the global features of each training video through a spatiotemporal attention mechanism and feature fusion, and trains the model based on the global features of each training video.

[0011] In a third aspect, embodiments of this application provide a video quality assessment device, comprising:

[0012] Memory and one or more processors;

[0013] The memory is configured to store one or more programs;

[0014] When the one or more programs are executed by the one or more processors, the one or more processors implement the video quality assessment method as described in the first aspect.

[0015] In a fourth aspect, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions configured, when executed by a computer processor, to perform the video quality assessment method as described in the first aspect.

[0016] In a fifth aspect, embodiments of this application provide a computer program product containing instructions that, when executed on a computer or processor, cause the computer or processor to perform the video quality assessment method as described in the first aspect.

[0017] This application embodiment acquires a target video, then determines the correlation features between two adjacent video frames based on a spatiotemporal attention mechanism, and fuses these correlation features to obtain the global features of the target video. Subsequently, the global features of the target video are input into a pre-constructed video quality assessment model, which outputs the video quality assessment result of the target video. Specifically, this video quality assessment model pre-determines the global features of each training video through a spatiotemporal attention mechanism and feature fusion, and trains the model based on these global features. By employing the above techniques, and combining the spatiotemporal attention mechanism to acquire temporal and spatial features between video frames for model training, the video quality assessment model can fully learn video features, achieving a more comprehensive video quality assessment and improving the model's video quality assessment accuracy. Subsequent video quality assessment using the temporal and spatial features of the target video yields a more accurate video quality assessment result. Attached Figure Description

[0018] Figure 1 This is a flowchart of a video quality assessment method provided in an embodiment of this application;

[0019] Figure 2 This is a flowchart illustrating the training process of the video quality assessment model in this application embodiment;

[0020] Figure 3 This is a schematic diagram of video frame processing in an embodiment of this application;

[0021] Figure 4 This is a schematic diagram illustrating the generation of relevance features in an embodiment of this application;

[0022] Figure 5 This is a flowchart illustrating the generation of relevance features in an embodiment of this application;

[0023] Figure 6 This is a schematic diagram of the structure of a video quality assessment system provided in an embodiment of this application;

[0024] Figure 7 This is a schematic diagram of the structure of a video quality assessment device provided in an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0026] This application provides a video quality assessment method that aims to acquire the spatial features of video frames and the temporal features between video frames through a spatiotemporal attention mechanism. Then, a video quality assessment model is trained based on the temporal and spatial features, enabling the video quality assessment model to learn more comprehensive video features and have the ability to mine the feature relationships between video frames, thereby achieving a more accurate and comprehensive video quality assessment.

[0027] Traditional video quality assessment models typically extract features and train based on individual video frames, lacking the ability to learn the relationships between features between video frames. Consequently, these models struggle to uncover the connections between video frames, resulting in biased assessments and relatively low accuracy. Therefore, this application provides a video quality assessment method that addresses the low accuracy of traditional video quality assessment models.

[0028] Example:

[0029] Figure 1 A flowchart of a video quality assessment method provided in an embodiment of this application is given. The video quality assessment method provided in this embodiment can be executed by a video quality assessment device, which can be implemented by software and / or hardware. The video quality assessment device can consist of two or more physical entities, or it can consist of a single physical entity. Generally, the video quality assessment device can be a video processing server, computer, server host, or other processing equipment.

[0030] The following description uses the video quality assessment equipment as an example to illustrate the video quality assessment method. (Refer to...) Figure 1 The video quality assessment method specifically includes:

[0031] S110. Obtain the target video, determine the correlation features between two adjacent video frames in the target video based on the spatiotemporal attention mechanism, and fuse the various correlation features to obtain the global features of the target video.

[0032] S120. Input the global features of the target video into the pre-built video quality assessment model and output the video quality assessment result of the target video. The video quality assessment model pre-determines the global features of each training video through spatiotemporal attention mechanism and feature fusion, and trains the model based on the global features of each training video.

[0033] In this embodiment of the application, when performing video quality assessment, for a given video, to more comprehensively evaluate video quality, video features are collected from video frames, and the correlation features between two adjacent video frames are obtained based on a spatiotemporal attention mechanism. Then, feature fusion is used to obtain the global features of the entire video. This yields the spatial domain features (video features of individual video frames) and temporal domain features (correlation features between video frames) of the entire video. A more accurate and comprehensive video quality assessment is then performed based on these global features.

[0034] For the video input to the video quality assessment model, this video is defined as the target video. By extracting the global features of the target video and inputting them into the video quality assessment model, the video quality assessment model can accurately calculate the corresponding video quality assessment result based on these global features.

[0035] Prior to this, a pre-built video quality assessment model is used to accurately assess video quality based on the global features of the input model and output the corresponding video quality assessment results. Furthermore, since the video quality assessment model needs to perform video quality assessment based on global features, during the training of the video quality assessment model, it is necessary to enable the model to discover the global features of the training samples and then train the model based on these global features.

[0036] Specifically, refer to Figure 2 The training process for the video quality assessment model includes:

[0037] S1001. Obtain multiple training videos and determine the global features of each training video based on spatiotemporal attention mechanism and feature fusion.

[0038] S1002. After passing the global features through the fully connected layer of the video quality assessment model, the video quality assessment result of the training video is obtained. Based on the video quality assessment result of the training video, the loss function of the video quality assessment model is calculated until the loss function reaches the set value, thus completing the training of the video quality assessment model.

[0039] Based on the training samples constructed from the training videos, the global features of each training video are extracted before model training. This embodiment employs a spatiotemporal attention mechanism combined with feature fusion to collect the global features of each training video.

[0040] Specifically, such as Figure 3 As shown, for each video frame of the training video, a feature extraction network is used to extract features, resulting in video features representing the spatial characteristics of each frame. Then, an inter-frame feature attention module is used to extract the correlation features between adjacent video frames. These correlation features are then fused into global features. The global features are then processed through a fully connected layer to obtain the corresponding video quality assessment results.

[0041] For example, when using a training video for model computation, for each video frame I input to the feature extraction network... t First, a feature extraction network is used to obtain the video feature vector V of the frame. t .

[0042]

[0043] Where θ represents the feature extraction network.

[0044] Optionally, multiple feature extraction networks can be used, and each network shares its network parameters. That is, each feature extraction network has the same network parameters, ensuring that the extracted video features meet the same standards, facilitating subsequent feature processing. Furthermore, using multiple feature extraction networks can improve feature extraction efficiency, shorten feature extraction time, and thus improve video quality assessment efficiency.

[0045] Furthermore, for a training video containing T video frames, the feature vectors [V0, V1, ..., V] of the T frames can be obtained. T These feature vectors are input together into the inter-frame feature attention module. Through inter-frame correlation feature acquisition and feature fusion, a global feature f representing the overall video quality of the training video can be obtained. G .

[0046] Among them, reference Figure 4 For two adjacent video frames, the feature vector of their video features is represented as V. i V i+1 The inter-frame feature attention module is used to mine V. i and V i+1 When obtaining correlation features, first analyze V. i and V i+1 Perform linear transformations on each part to obtain the eigenvector W. k and W q Then, the transformed feature vectors are multiplied pixel by pixel to calculate the correlation score (softmax) between the two. The correlation score (softmax) is then multiplied by the feature vector V. i The eigenvector W obtained by another linear transformation V Pixel multiplication yields the correlation feature f. i Among them, the correlation feature f i The calculation formula is:

[0047]

[0048] in, This is a pixel multiplication operation.

[0049] In practical applications, the inter-frame feature attention module can use a variety of different computation methods to collect its global features. The embodiments of this application do not impose fixed restrictions on the specific global feature collection methods, and will not be described in detail here.

[0050] Optionally, during actual computation, when calculating correlation features based on the correlation scores of two adjacent video frames, if the correlation scores of two adjacent video frames are lower than a certain set threshold, the correlation features of these video frames can be ignored, and they will not be included in the global feature calculation. It is understood that for video frames with lower correlation scores, their temporal feature relationships have a smaller impact on the global features. To reduce computational load and improve feature fusion efficiency, the calculation of these correlation features can be ignored, thus optimizing the efficiency and effectiveness of video quality assessment.

[0051] Based on the correlation features between two adjacent video frames determined above, a feature fusion method is used to fuse all the correlation features of a training video together to obtain the global features of the training video. Feature fusion can be performed by feature concatenation. This application does not impose fixed restrictions on the specific fusion method, and will not elaborate further here.

[0052] Optionally, to reduce the complexity of feature fusion while ensuring the network's representational ability, embodiments of this application obtain global features f by continuously performing momentum fine-tuning on the relevant features. G The global feature calculation formula is expressed as follows:

[0053] The formula for calculating global features is:

[0054]

[0055] Among them, f i f represents the correlation feature between two adjacent video frames in the i-th group. Gi f represents the global feature obtained by fusing the correlation features between two adjacent video frames in the first i groups. Gi-1 The global feature is obtained by fusing the correlation features between two adjacent video frames in the first i-1 groups. λ represents the set fine-tuning amount, which is a hyperparameter. In this embodiment, λ is set to 0.1.

[0056] Based on the above global feature calculation formula, the features can be fine-tuned continuously according to the correlation features of each group of adjacent video frames, and finally the global features f of the training video can be obtained. G .

[0057] Then, the global feature f G The video quality assessment model uses two fully connected layers to obtain the final video quality score, which is the video quality assessment result. The video quality score is represented as:

[0058] score = W2ReLU(W1f) G +b1)+b2

[0059] Where W1 and W2 are the weight matrices of the two fully connected layers, and b1 and b2 are the offsets of the fully connected layers.

[0060] Then, based on the calculated video quality assessment results of the training videos, a loss function is calculated. The model parameters are continuously adjusted according to the value of the loss function. The adjusted video quality assessment model is then used to repeatedly assess the video quality of the training videos, calculating the loss function based on the assessment results, and adjusting the model parameters again based on the loss function's value. This process is repeated until the loss function converges to a set value, confirming that the model possesses high-accuracy video quality assessment capabilities. At this point, the training of the video quality assessment model is complete. Optionally, the batch size during training is set to 16, the Adam optimizer is used, and the initial learning rate is set to 10. -4 .

[0061] Pearson linear correlation coefficient (PLCC) and Spearman rank correlation coefficient (SRCC) were used as evaluation metrics. A differentiable PLCC function was used as the loss function.

[0062]

[0063] The batch size during training was 16, the AdamW optimizer was used, and the initial learning rate was set to 10. -4 .

[0064] The loss function is constructed based on the Pearson correlation coefficient between the video quality assessment results of the training videos and the quality scores of the labeled videos.

[0065] The loss function formula is expressed as:

[0066]

[0067] Among them, s pred The video quality score, i.e., the video quality assessment result, is S. gt The labeled video quality score represents the training video, and PLCC is the Pearson linear correlation coefficient function. By using the Pearson correlation coefficient as an evaluation metric, more reasonable video quality assessment results can be obtained through training.

[0068] Furthermore, based on the trained video quality assessment model, the first step in video quality assessment is to extract global features from the target video. Referring to the global feature extraction method used in the training video, inter-frame correlation features are obtained using the inter-frame feature attention module, and then the global features of the target video are obtained by fusing these correlation features.

[0069] Among them, reference Figure 5 The process of generating relevance features includes:

[0070] S1101. Extract the video features of each video frame in the target video;

[0071] S1102. Based on the video features of two adjacent video frames, perform linear transformation and pixel multiplication to obtain the correlation features between the two adjacent video frames.

[0072] By inputting each video frame in the target video into multiple pre-constructed feature extraction networks, video features are extracted based on the feature extraction networks, and the network parameters of each feature extraction network are the same.

[0073] Furthermore, video features are processed based on the inter-frame feature attention module to obtain the correlation features between two adjacent video frames in the target video.

[0074] For two adjacent video frames, the video features of the current video frame are transformed linearly to obtain the first transformation feature and the second transformation feature. The video features of the next video frame are transformed linearly to obtain the third transformation feature. The first transformation feature and the third transformation feature are multiplied pixel by pixel to obtain the correlation score. The correlation score is multiplied pixel by pixel by the second transformation feature to obtain the correlation feature between the two adjacent video frames.

[0075] Referring to the aforementioned relevance feature extraction method for training videos, based on the video feature vectors of each video frame in the target video, an inter-frame feature attention module is used to mine the feature relationships between frames and obtain inter-frame relevance features. Specifically, by performing linear transformations on the feature vectors of two adjacent video frames, first and third transformed features are obtained. Then, pixel-wise multiplication of the transformed feature vectors is used to calculate the relevance score between them. Finally, the relevance score is multiplied pixel-wise with the second transformed feature obtained by performing another linear transformation on the video features of the current video frame to obtain the relevance feature.

[0076] Further based on this correlation feature, and referring to the feature fusion method of the training videos mentioned above, correlation feature fusion is performed to obtain the global features of the target video. This global feature is input into the video quality assessment model, and through model computation via the fully connected layers, the corresponding video quality score, i.e., the video quality assessment result of the target video, is obtained. Throughout the video quality assessment process, a spatiotemporal attention mechanism is used to fully mine the video quality features in the target video. This simultaneously considers video quality features in both the spatial and temporal domains, allowing the model to focus on relatively important quality information in both domains. This more effectively and comprehensively captures the video quality features of the target video for model computation, resulting in a more accurate video quality assessment result and effectively improving the accuracy of video quality assessment.

[0077] As described above, by acquiring the target video and then determining the correlation features between adjacent video frames based on a spatiotemporal attention mechanism, the global features of the target video are obtained by fusing these correlation features. Subsequently, the global features of the target video are input into a pre-constructed video quality assessment model, which outputs the video quality assessment result of the target video. This video quality assessment model pre-determines the global features of each training video through a spatiotemporal attention mechanism and feature fusion, and trains the model based on these global features. By employing the above techniques and combining the spatiotemporal attention mechanism to acquire temporal and spatial features between video frames for model training, the video quality assessment model can fully learn video features, achieving a more comprehensive video quality assessment and improving the model's video quality assessment accuracy. Subsequent video quality assessment using the temporal and spatial features of the target video yields a more accurate video quality assessment result.

[0078] Based on the above embodiments, Figure 6 A schematic diagram of the structure of a video quality assessment system provided in this application. (Reference) Figure 6 The video quality assessment system provided in this embodiment specifically includes: a fusion module 21 and an assessment module 22.

[0079] The fusion module 21 is used to acquire the target video, determine the correlation features between two adjacent video frames in the target video based on the spatiotemporal attention mechanism, and fuse the various correlation features to obtain the global features of the target video.

[0080] The evaluation module 22 is used to input the global features of the target video into the pre-built video quality evaluation model and output the video quality evaluation result of the target video. The video quality evaluation model pre-determines the global features of each training video through a spatiotemporal attention mechanism and feature fusion, and trains the model based on the global features of each training video.

[0081] Specifically, the correlation features between two adjacent video frames in the target video are determined based on a spatiotemporal attention mechanism, including:

[0082] Extract video features from each frame of the target video;

[0083] The correlation features between two adjacent video frames are obtained by performing linear transformation and pixel multiplication based on the video features of two adjacent video frames.

[0084] Specifically, the correlation features between two adjacent video frames are obtained by performing linear transformation and pixel multiplication based on the video features of two adjacent video frames, including:

[0085] The video features of the current video frame are transformed linearly to obtain the first and second transformed features, and the video features of the next video frame are transformed linearly to obtain the third transformed feature.

[0086] The correlation score is obtained by multiplying the first and third transformation features by pixels. The correlation score is then multiplied by the second transformation feature by pixels to obtain the correlation feature between two adjacent video frames.

[0087] Specifically, the video features of each frame in the target video are extracted, including:

[0088] Each video frame in the target video is input into multiple pre-constructed feature extraction networks, and video features are extracted based on the feature extraction networks. The network parameters of each feature extraction network are the same.

[0089] Specifically, the formula for calculating global features is:

[0090]

[0091] Among them, f i f represents the correlation feature between two adjacent video frames in the i-th group. Gi f represents the global feature obtained by fusing the correlation features between two adjacent video frames in the first i groups. Gi-1 The global feature is obtained by fusing the correlation features between two adjacent video frames in the first i-1 groups, and λ represents the set fine-tuning amount.

[0092] Specifically, the training process for the video quality assessment model includes:

[0093] Multiple training videos are acquired, and global features of each training video are determined based on spatiotemporal attention mechanism and feature fusion.

[0094] The global features are passed through the fully connected layer of the video quality assessment model to obtain the video quality assessment results of the training video. The loss function of the video quality assessment model is calculated based on the video quality assessment results of the training video until the loss function reaches the set value, thus completing the training of the video quality assessment model.

[0095] Specifically, the loss function is constructed based on the Pearson correlation coefficient between the video quality assessment results of the training videos and the quality scores of the labeled videos.

[0096] As described above, by acquiring the target video and then determining the correlation features between adjacent video frames based on a spatiotemporal attention mechanism, the global features of the target video are obtained by fusing these correlation features. Subsequently, the global features of the target video are input into a pre-constructed video quality assessment model, which outputs the video quality assessment result of the target video. This video quality assessment model pre-determines the global features of each training video through a spatiotemporal attention mechanism and feature fusion, and trains the model based on these global features. By employing the above techniques and combining the spatiotemporal attention mechanism to acquire temporal and spatial features between video frames for model training, the video quality assessment model can fully learn video features, achieving a more comprehensive video quality assessment and improving the model's video quality assessment accuracy. Subsequent video quality assessment using the temporal and spatial features of the target video yields a more accurate video quality assessment result.

[0097] The video quality assessment system provided in this application embodiment can be configured to execute the video quality assessment method provided in the above embodiment, and has corresponding functions and beneficial effects.

[0098] Based on the above practical examples, this application also provides a video quality assessment device, referring to... Figure 7 The video quality assessment device includes a processor 31, a memory 32, a communication module 33, an input device 34, and an output device 35. The memory 32, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, and modules, such as program instructions / modules corresponding to the video quality assessment method described in any embodiment of this application (e.g., a fusion module and an assessment module in a video quality assessment system). The communication module 33 is configured to perform data transmission. The processor 31 executes various functional applications and data processing of the device by running the software programs, instructions, and modules stored in the memory, thereby implementing the aforementioned video quality assessment method. The input device 34 can be configured to receive input digital or character information and generate key signal inputs related to user settings and function control of the device. The output device 35 may include a display screen or other display device. The video quality assessment device provided above can be configured to execute the video quality assessment method provided in the above embodiments, possessing corresponding functions and beneficial effects.

[0099] Based on the above embodiments, this application also provides a computer-readable storage medium storing computer-executable instructions. These computer-executable instructions, when executed by a computer processor, are configured to perform a video quality assessment method. The storage medium can be any type of memory device or storage device. Of course, the computer-readable storage medium provided in this application is not limited to the video quality assessment method described above; it can also perform related operations within the video quality assessment method provided in any embodiment of this application.

[0100] Based on the above embodiments, this application also provides a computer program product. The technical solution of this application, in essence or in other words, the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer program product is stored in a storage medium and includes several instructions to cause a computer device, mobile terminal, or processor therein to execute all or part of the steps of the video quality assessment method described in the various embodiments of this application.

Claims

1. A method of video quality assessment, characterized by, The method comprises: obtaining a target video, extracting video features of each frame of the target video, based on two adjacent frames of the target video, performing linear transformation on the video features of the current frame to obtain first transformed features and second transformed features, performing linear transformation on the video features of the next frame to obtain third transformed features, performing pixel multiplication on the first transformed features and the third transformed features to obtain a correlation score, performing pixel multiplication on the correlation score and the second transformed features to obtain a correlation feature between the two adjacent frames of the target video, and fusing each correlation feature to obtain a global feature of the target video; inputting the global feature of the target video into a pre-constructed video quality evaluation model, and outputting a video quality evaluation result of the target video, wherein the video quality evaluation model is pre-determined by a spatio-temporal attention mechanism and feature fusion to determine a global feature of each training video, and the video quality evaluation model is trained based on the global feature of each training video.

2. The video quality assessment method of claim 1, wherein, The extraction of the video features of each frame of the target video comprises: inputting each frame of the target video into a plurality of pre-constructed feature extraction networks, extracting video features based on the feature extraction networks, and the network parameters of each feature extraction network are the same.

3. The method of claim 1, wherein, The global feature calculation formula is: wherein, represents the correlation feature between the i-th group of two adjacent video frames, represents the global feature obtained by feature fusion of the correlation features between the first i groups of two adjacent video frames, represents the global feature obtained by feature fusion of the correlation features between the first i-1 groups of two adjacent video frames, represents the set fine-tuning momentum.

4. The method of claim 1, wherein, The training process of the video quality evaluation model comprises: obtaining a plurality of training videos, determining a global feature of each training video based on a spatio-temporal attention mechanism and feature fusion; obtaining a video quality evaluation result of the training video by passing the global feature through a fully connected layer of the video quality evaluation model, calculating a loss function of the video quality evaluation model based on the video quality evaluation result of the training video, and completing the training of the video quality evaluation model until the loss function reaches a set value.

5. The method of video quality assessment according to claim 4, characterized in that, The loss function is constructed based on the Pearson correlation coefficient between the video quality evaluation result of the training video and a label video quality score.

6. A video quality assessment system characterized by, The method comprises: a fusion module configured to obtain a target video, extract video features of each frame of the target video, perform linear transformation on the video features of the current frame based on two adjacent frames of the target video to obtain first transformed features and second transformed features, perform linear transformation on the video features of the next frame to obtain third transformed features, perform pixel multiplication on the first transformed features and the third transformed features to obtain a correlation score, perform pixel multiplication on the correlation score and the second transformed features to obtain a correlation feature between the two adjacent frames of the target video, and fuse each correlation feature to obtain a global feature of the target video; an evaluation module configured to input the global feature of the target video into a pre-constructed video quality evaluation model, and output a video quality evaluation result of the target video, wherein the video quality evaluation model is pre-determined by a spatio-temporal attention mechanism and feature fusion to determine a global feature of each training video, and the video quality evaluation model is trained based on the global feature of each training video.

7. A video quality assessment device, characterized by comprising: The method comprises: a memory and one or more processors; the memory is configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the video quality assessment method as claimed in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions which, when executed by a computer processor, are configured to perform the video quality assessment method as claimed in any one of claims 1-5.

9. A computer program product, characterised in that, The computer program product comprises instructions which, when executed on a computer or processor, cause the computer or processor to perform the video quality assessment method as claimed in any one of claims 1-5.

Images