Video quality assessment model training method, quality assessment method, device and medium

Abstract

The disclosed embodiments provide a video quality assessment model training method, a video quality assessment method, an apparatus, an electronic device, and a storage medium. The method comprises: obtaining a training data set including a plurality of video quality training samples; the video quality training samples include corresponding video samples and quality text samples; extracting features from the video samples by a video encoder to obtain video text features; extracting features from the quality text samples by a text encoder to obtain quality text features; and fine-tuning a multimodal large language model to be trained based on the video text features and the quality text features to obtain a target video quality assessment model. The method fine-tunes a multimodal large language model to be trained by using video text features and quality text features, so that the video quality assessment model can more accurately identify and assess video quality, thereby being suitable for video quality assessment application scenarios.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a video quality assessment model training method, a video quality assessment method, a device, an electronic device, a computer program product, and a storage medium. Background Art

[0002] With the rapid development of video streaming, social media, and video conferencing, users' expectations for the quality of video content are increasing. Traditionally, video quality assessment relies mainly on visual models that analyze pixel-level features in video frames, calculate image quality indicators such as peak signal-to-noise ratio and structural similarity index, or use deep learning techniques such as convolutional neural networks (CNNs) to generate image quality scores. Although this method can reveal video quality issues to a certain extent, its output is usually limited to a score, which cannot deeply explain the specific causes of quality problems or provide improvement suggestions.

[0003] In addition, traditional visual models are unable to cope with complex and multi-dimensional video content and are unable to fully capture all factors that affect video quality. This limitation makes it difficult for existing evaluation methods to provide comprehensive and detailed quality analysis when faced with complex video content, limiting their effectiveness and practicality in practical applications. Summary of the invention

[0004] The embodiments of the present disclosure provide a video quality assessment model training method, a video quality assessment method, an apparatus, an electronic device, a computer program product, and a storage medium.

[0005] According to a first aspect of an embodiment of the present disclosure, a video quality assessment model training method is provided, comprising: obtaining a training data set including a plurality of video quality training samples; the video quality training samples include corresponding video samples and quality text samples; the quality text samples are texts for assessing the video quality of the video samples; performing feature extraction on the video samples by a video encoder to obtain video text features; performing feature extraction on the quality text samples by a text encoder to obtain quality text features; and fine-tuning a multimodal large language model to be trained based on the video text features and the quality text features to obtain a target video quality assessment model.

[0006] In some exemplary embodiments of the present disclosure, the quality text sample includes: a quality description text and at least one quality question and answer text; the quality description text is a text describing the video quality of the video sample; the quality question and answer text is a question and answer text evaluating the video quality of the video sample; each of the quality question and answer texts includes a corresponding quality question text and a quality answer text.

[0007] In some exemplary embodiments of the present disclosure, the method of fine-tuning the multimodal large language model to be trained according to the video text features and the quality text features to obtain a target video quality assessment model includes: fine-tuning the multimodal large language model to be trained according to the video text features and the quality description text features to obtain a first video quality assessment model; the quality description text features are obtained by extracting features of the quality description text through a first text encoder; fine-tuning the first video quality assessment model according to the video text features and the quality question and answer pair text features to obtain the target video quality assessment model; the quality question and answer pair text features are obtained by extracting features of the quality question and answer pair text through a second text encoder.

[0008] In some exemplary embodiments of the present disclosure, the first video quality assessment model is fine-tuned and trained according to the video text features and the quality question and answer pair text features to obtain the target video quality assessment model, including: the first video quality assessment model is fine-tuned and trained according to the video text features, the quality question and answer pair text features and the first quality description text features to obtain the target video quality assessment model; wherein the first quality description text features belong to the quality description text features.

[0009] In some exemplary embodiments of the present disclosure, the acquiring of a training data set comprising a plurality of video quality training samples comprises: performing video quality scoring on the video samples by means of at least one video quality scoring model to obtain at least one quality scoring text; performing video content feature extraction on the video samples by means of a video content description model to obtain a content description text; and integrating the quality scoring text and the content description text to obtain the quality text sample.

[0010] In some exemplary embodiments of the present disclosure, the video quality scoring of the video sample by using at least one video quality scoring model to obtain at least one quality scoring text includes: scoring the video quality of the video sample by using the at least one video quality scoring model to obtain at least one quality assessment score; the quality assessment score corresponds to the video quality scoring model; the video quality scoring model corresponds to at least one quality assessment dimension; mapping the at least one quality assessment score by using a score-text mapping relationship to obtain the at least one quality scoring text; the quality scoring text corresponds to the quality assessment dimension; the score-text mapping relationship stores a mapping relationship between a preset score interval and a quality scoring text.

[0011] In some exemplary embodiments of the present disclosure, the integrating the quality scoring text and the content description text to obtain the quality text sample includes: splicing the quality scoring text and the content description text to obtain the quality text sample.

[0012] In some exemplary embodiments of the present disclosure, the integrating the quality scoring text and the content description text to obtain the quality text sample includes: integrating the quality scoring text and the content description text through a first large language model to obtain the quality text sample.

[0013] In some exemplary embodiments of the present disclosure, the integrating the quality scoring text and the content description text to obtain the quality text sample includes: integrating the quality scoring text and the content description text to obtain the quality description text; and performing question-answer pair extraction on the quality description text through a second largest language model to obtain the at least one quality question-answer pair text.

[0014] According to a second aspect of an embodiment of the present disclosure, a video quality assessment method is provided, comprising: obtaining quality assessment prompt words and a video to be assessed; inputting the quality assessment prompt words and the video to be assessed into a video quality assessment model; the video quality assessment model is obtained by training using any one of the video quality assessment model training methods described above; and the video quality assessment model generates a video quality assessment result for the video to be assessed based on the quality assessment prompt words and the video to be assessed.

[0015] According to a third aspect of an embodiment of the present disclosure, a video quality assessment model training device is provided, comprising: a sample acquisition module, configured to acquire a training data set including multiple video quality training samples; the video quality training samples include corresponding video samples and quality text samples; the quality text samples are texts for assessing the video quality of the video samples; a video text feature module, configured to perform feature extraction on the video samples through a video encoder to obtain video text features; a quality text feature module, configured to perform feature extraction on the quality text samples through a text encoder to obtain quality text features; and a model training module, configured to perform fine-tuning training on a multimodal large language model to be trained according to the video text features and the quality text features to obtain a target video quality assessment model.

[0016] According to a fourth aspect of an embodiment of the present disclosure, a video quality assessment device is provided, comprising: an assessment request acquisition module, configured to obtain quality assessment prompt words and a video to be assessed; an assessment request input module, configured to input the quality assessment prompt words and the video to be assessed into a video quality assessment model; the video quality assessment model is obtained by training using any one of the video quality assessment model training methods; and a quality assessment module, configured so that the video quality assessment model generates a video quality assessment result for the video to be assessed based on the quality assessment prompt words and the video to be assessed.

[0017] According to the fifth aspect of an embodiment of the present disclosure, an electronic device is provided, characterized in that it includes: a processor; a memory for storing executable instructions of the processor; wherein the processor is configured to execute the executable instructions to implement any one of the video quality assessment model training methods or any one of the video quality assessment methods.

[0018] According to the sixth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute any one of the video quality assessment model training methods or any one of the video quality assessment methods.

[0019] According to a seventh aspect of an embodiment of the present disclosure, a computer program product is provided, comprising a computer program, wherein a processor executes any one of the video quality assessment model training methods or any one of the video quality assessment methods.

[0020] The video quality assessment model training method provided in the embodiment of the present disclosure fine-tunes the multimodal large language model to be trained based on multiple video quality training samples including video samples and quality text samples to obtain a target video quality assessment model. The method fine-tunes the multimodal large language model to be trained through video text features and quality text features, so that the video quality assessment model can more accurately identify and assess video quality, thereby being suitable for video quality assessment application scenarios.

[0021] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0023] Figure 1 A schematic diagram showing an exemplary system architecture to which the method according to an embodiment of the present disclosure can be applied is shown.

[0024] Figure 2 The figure is a flowchart of a video quality assessment model training method according to an exemplary embodiment.

[0025] Figure 3 The present invention is a schematic diagram of the implementation process of a video quality assessment model training method according to an example.

[0026] Figure 4 is a flow chart of a video quality assessment model fine-tuning training process according to an example.

[0027] Figure 5 is a flowchart of a method for obtaining a training data set according to an example.

[0028] Figure 6 The figure is a flow chart of a quality scoring text generation process according to an example.

[0029] Figure 7 The figure is a flowchart of a video quality assessment method according to an exemplary embodiment.

[0030] Figure 8 The present invention is a schematic diagram of an implementation process of a video quality assessment method according to an example.

[0031] Fig. 9 It is a block diagram of a video quality assessment model training device according to an exemplary embodiment.

[0032] Fig.10 The invention is a block diagram of a video quality assessment device according to an exemplary embodiment.

[0033] Fig.11 It is a schematic diagram showing the structure of an electronic device suitable for implementing the exemplary embodiments of the present disclosure according to an exemplary embodiment. DETAILED DESCRIPTION

[0034] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be comprehensive and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted.

[0035] The features, structures or characteristics described in the present disclosure may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the present disclosure.

[0036] The accompanying drawings are only schematic diagrams of the present disclosure, and the same reference numerals in the drawings represent the same or similar parts, so their repeated description will be omitted. Some block diagrams shown in the accompanying drawings do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in at least one hardware module or integrated circuit, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0037] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and steps, nor must they be executed in the order described. For example, some steps can be decomposed, and some steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0038] In this specification, the terms "a", "an", "the", "said" and "at least one" are used to indicate the presence of at least one element / component / etc.; the terms "comprising", "including" and "having" are used to express an open-ended inclusion and mean that additional elements / components / etc. may exist in addition to the listed elements / components / etc.; the terms "first", "second" and "third" etc. are used only as labels and are not intended to limit the quantity of their objects.

[0039] Figure 1 A schematic diagram showing an exemplary system architecture to which the method according to an embodiment of the present disclosure can be applied is shown.

[0040] like Figure 1 As shown, the system architecture may include a server 101, a network 102, a terminal device 103, a terminal device 104, and a terminal device 105. The network 102 is used to provide a medium for a communication link between the terminal device 103, the terminal device 104, or the terminal device 105 and the server 101. The network 102 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0041] The server 101 may be a server that provides various services, such as a background management server that provides support for devices operated by users using the terminal device 103, the terminal device 104, or the terminal device 105. The background management server may analyze and process the received request and other data, and feed back the processing results to the terminal device 103, the terminal device 104, or the terminal device 105.

[0042] Terminal device 103, terminal device 104 and terminal device 105 can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a wearable smart device, a virtual reality device, an augmented reality device, etc., but are not limited thereto.

[0043] It should be understood that Figure 1 The number of terminal devices 103, terminal devices 104, terminal devices 105, networks 102 and servers 101 are merely illustrative. Server 101 may be a physical server, a server cluster consisting of multiple servers, or a cloud server. Depending on actual needs, any number of terminal devices, networks and servers may be provided.

[0044] Below, each step of the method in the exemplary embodiment of the present disclosure will be described in more detail with reference to the accompanying drawings and embodiments.

[0045] Figure 2 The figure is a flowchart of a video quality assessment model training method according to an exemplary embodiment. Figure 3 The present invention is a schematic diagram of the implementation process of a video quality assessment model training method according to an example. Figure 2 The method provided in the embodiment can be executed by any electronic device, such as the above Figure 1 The terminal device in Figure 1 Servers in, or Figure 1 The terminal device and the server in the embodiment are jointly executed, but the present disclosure is not limited to this.

[0046] In step S210, a training data set including a plurality of video quality training samples is obtained; the video quality training samples include corresponding video samples and quality text samples; the quality text samples are texts for evaluating the video quality of the video samples.

[0047] In the related technology, the Multimodal Large Language Model (MLLM) has gradually emerged in recent years. The Multimodal Large Language Model (MLLM) is an artificial intelligence system that can process and integrate multiple forms of data (such as text, images, audio and video, etc.), and improves the accuracy and richness of understanding and generation tasks by fusing information from different modalities. Through image data, the Multimodal Large Language Model can accurately identify the visual elements in the video and their potential content meanings, and then use the text information module to generate text description content related to the video. However, existing Multimodal Large Language Models are usually limited to understanding video content, and do not yet have the ability to evaluate video quality and output relevant text evaluation results.

[0048] In the disclosed embodiments, a video quality assessment model training method is provided to address the above-mentioned problems. By fine-tuning the multimodal large language model, it is possible to use multimodal fusion of images, videos and texts to not only accurately assess the overall quality of the video, but also perform fine-grained attribution analysis on image quality issues.

[0049] In the disclosed embodiment, in order to fine-tune the multimodal large language model, a training data set related to video quality assessment is first obtained. The training data set includes multiple video quality training samples. Each video quality training sample includes at least one set of video samples and quality text samples. The video samples and the quality text samples are in a corresponding relationship. The quality text sample is a text for evaluating the video quality of the video sample.

[0050] In an exemplary embodiment, in order to enable the trained video quality assessment model to make accurate video quality assessment for videos of different video qualities, video samples of high, medium and low video qualities can be selected from the training data set to form the video quality training samples.

[0051] In an exemplary embodiment, according to the selected video samples with different video qualities of high, medium and low, video quality is evaluated through various video quality evaluation models and a large language model, and corresponding quality text samples are generated.

[0052] In an exemplary embodiment, the quality text sample further includes: a quality description text and at least one quality question-answer pair text. The quality description text and the quality question-answer pair text are both quality texts corresponding to the video sample.

[0053] The quality description text is a text that comprehensively describes the video quality of the video sample.

[0054] For example, "On a scale of 1 to 5, I rate the video quality of this video a 4. The analysis is as follows: Clarity: The video is clear overall, but there is slight blur in fast-motion scenes, resulting in a slight loss of details. Color accuracy: The overall color performance of the video is accurate, but the restoration of some colors under complex lighting conditions is slightly deviated, resulting in a slight decrease in the realism of the color. Noise: The overall picture of the video is clean, but there is slight noise in low-light scenes, which slightly affects the purity of the dark area."

[0055] The quality question-answer pair text is a question-answer pair text for evaluating the video quality of the video sample. Each quality question-answer pair text includes a quality question text and a quality answer text in a corresponding relationship. Several quality question-answer pair texts constitute a quality question-answer pair group corresponding to the video sample.

[0056] For example, a quality question-answer pair group may include the following quality question-answer pair texts:

[0057] Q1: Based on the overall quality of the following video, how would you rate its quality using a rating scale of 1 to 5?

[0058] A1: On a scale of 1 to 5, I rate the video quality of this video a 4.

[0059] Q2: Please analyze the video in terms of clarity level and give a brief explanation.

[0060] A2: Clarity. The video is clear overall, but there is slight blur in fast-motion scenes, resulting in a slight loss of details.

[0061] Q3: Please analyze the video in terms of color accuracy level and give a brief explanation.

[0062] A3: Color Accuracy: The overall color performance of the video is accurate, but the restoration of some colors is slightly deviated under complex lighting conditions, resulting in a slight decrease in the realism of the colors.

[0063] Q4: Please analyze the video in terms of noise level and give a brief explanation.

[0064] A4: Noise: The overall video is clean, but there is slight noise in low-light scenes, which slightly affects the purity of dark areas.

[0065] In the disclosed embodiment, the quality description text and quality question-answer text of the quality text sample can, on the one hand, provide the model with text samples for overall description of quality assessment, and on the other hand, provide the model with text samples for evaluating different quality assessment dimensions of the video, so that the model can be trained on different quality assessment dimensions more quickly and accurately.

[0066] In step S220, a video encoder is used to extract features from the video sample to obtain video text features.

[0067] In the present disclosure, Figure 3 As shown in the figure, a series of video frames in the video sample are input into the video encoder. The video encoder processes each frame through a deep learning model, such as a convolutional neural network (CNN), and extracts key visual features. These features include not only the features of static images, but also the dynamic change information between frames, such as motion trajectories, scene transitions, etc. In this way, the video encoder can capture important visual information in the video and convert it into high-dimensional feature vectors, which can comprehensively describe the video content and provide a basis for subsequent video understanding and analysis.

[0068] In view of the multimodal characteristics of the multimodal large language model, a video projector is further provided after the video encoder. The video projector further processes the input high-dimensional feature vector to adapt to the fusion with the text features, and finally forms the video text features for the subsequent fine-tuning training process of the multimodal large language model.

[0069] In step S230, feature extraction is performed on the quality text sample by a text encoder to obtain quality text features.

[0070] In the present disclosure, Figure 3 As shown in the figure, the quality text samples are input into the text encoder. The text encoder processes these texts through deep learning models, such as Transformer, and extracts key semantic features. These features include not only vocabulary and grammatical information, but also the contextual relationships and emotional colors in the text. In this way, the text encoder can capture important semantic information in the text and convert it into high-dimensional feature vectors, ultimately forming quality text features. These vectors can comprehensively describe the text content and are used in the subsequent fine-tuning training process of the multimodal large language model.

[0071] In step S240, the multimodal large language model to be trained is fine-tuned according to the video text features and the quality text features to obtain a target video quality assessment model.

[0072] As mentioned earlier, although the existing multimodal large language model has the ability to process multimodal data such as video and text, it is limited to understanding the video content and does not yet have the ability to evaluate the video quality and output relevant text evaluation results.

[0073] In the present disclosure, Figure 3As shown, according to the aforementioned processing of each video quality training sample in the training data set, the multimodal large language model to be trained is fine-tuned based on the video text features and quality text features to obtain the target video quality assessment model. The so-called fine-tuning training refers to further training using a data set for a specific task on the basis of the pre-trained model to optimize the performance of the model on the task. In this process, video text features and quality text features are integrated into the multimodal large language model, and the model parameters are adjusted through the back propagation algorithm to make it better adapt to the video quality assessment task. Through this fine-tuning training, the video quality assessment model can more accurately identify and evaluate video quality, and is thus suitable for video quality assessment application scenarios.

[0074] It should be pointed out that the video quality assessment model trained by the above method can not only output assessment results for video quality, but also further attribute fine-grained quality issues and put forward suggestions for improving video quality.

[0075] The video quality assessment model training method provided by the disclosed embodiment fine-tunes the multimodal large language model to be trained based on multiple video quality training samples including video samples and quality text samples to obtain a target video quality assessment model. The method fine-tunes the multimodal large language model to be trained through video text features and quality text features, so that the video quality assessment model can more accurately identify and assess video quality, thereby being suitable for video quality assessment application scenarios. The video quality assessment model trained by the above method can not only output assessment results for video quality, but can also further attribute fine-grained quality issues and make suggestions for improving video quality.

[0076] Figure 4 FIG. 1 is a flow chart of a video quality assessment model fine-tuning training process according to an example. Figure 4 As shown, in the embodiment of the present disclosure, the fine-tuning training process of the aforementioned step S240 may include the following steps.

[0077] In step S410, the multimodal large language model to be trained is fine-tuned according to the video text features and the quality description text features to obtain a first video quality assessment model; the quality description text features are obtained by extracting features from the quality description text using a first text encoder.

[0078] As mentioned above, the quality text sample further includes: a quality description text and at least one quality question-answer text. The quality description text and the quality question-answer text are both quality texts corresponding to the video sample. Through the quality description text and the quality question-answer text of the quality text sample, on the one hand, a text sample describing the overall quality assessment can be provided to the model, and on the other hand, a text sample evaluating different quality assessment dimensions of the video can be provided to the model, so that the model can be trained more quickly and accurately on different quality assessment dimensions.

[0079] In the disclosed embodiment, the fine-tuning training process is further divided into two training stages according to the characteristics of the quality description text and the quality question-answer text. In the first stage, the multimodal large language model to be trained is fine-tuned according to the video text features and the quality description text features, thereby obtaining a first video quality assessment model that focuses on the overall description of video quality assessment.

[0080] The quality description text feature is obtained by extracting features from the quality description text through the first text encoder. The feature extraction process is similar to the aforementioned step S230, except that the object of feature extraction is the quality description text in the quality text sample, so it is not described again. The first text encoder can be a part of the text encoder in the aforementioned step S230, or the text encoder can be used directly.

[0081] During the training process, on the one hand, the video encoder processes the video samples and extracts high-dimensional visual features that can represent the video content and structure; on the other hand, the first text encoder extracts features of the quality description text to obtain holistic description text features that can reflect the video quality assessment. By integrating these features and continuously optimizing the model parameters using the back propagation algorithm, the multimodal large language model to be trained can learn the complex relationship between video quality and its description, thereby obtaining the first video quality assessment model that focuses on the holistic description of video quality assessment.

[0082] In step S420, the first video quality assessment model is fine-tuned and trained according to the video text features and the quality question and answer text features to obtain the target video quality assessment model; the quality question and answer text features are obtained by extracting features from the quality question and answer text using a second text encoder.

[0083] In the disclosed embodiment, in the second stage, the first video quality assessment model obtained in the first stage is fine-tuned and trained based on the video text features and the quality question and answer text features, so as to obtain a target video quality assessment model.

[0084] The quality question-answer text feature is obtained by extracting features from the quality question-answer text through the second text encoder. The feature extraction process is similar to the aforementioned step S230, except that the object of feature extraction is the quality question-answer text in the quality text sample, so it will not be described again. The second text encoder can be a part of the text encoder in the aforementioned step S230, or the text encoder can be used directly.

[0085] In the training process, since the quality question and answer pair text is a question and answer pair text for evaluating the video quality of the video sample. Each quality question and answer pair text includes a quality question text and a quality answer text in a corresponding relationship. As mentioned above, each quality question and answer pair text is a question and answer pair for evaluating the video quality from different quality assessment dimensions, and its corresponding quality question text and quality answer text are more concise and accurate than the quality description text. Therefore, fine-tuning the first video quality assessment model based on the quality question and answer pair text can train different quality assessment dimensions more quickly and accurately, thereby obtaining a target video quality assessment model with higher assessment quality.

[0086] In an exemplary embodiment, step S420 may further include: fine-tuning and training the first video quality assessment model according to the video text features, the quality question-answer text features and the first quality description text features to obtain the target video quality assessment model; wherein the first quality description text features belong to the quality description text features.

[0087] In the disclosed embodiment, a first quality description text feature is further introduced on the basis of the aforementioned step S420 to participate in the process of fine-tuning the first video quality assessment model to obtain a target video quality assessment model. The first quality description text feature is a portion of the quality description text features selected from the aforementioned quality description text features. The first quality description text feature corresponds to a portion of the quality description text features with relatively high text description quality in the quality description text features. By introducing the partially preferred first quality description text features into the training process of the first video quality assessment model, the target video quality assessment model finally obtained by training can better learn the partially preferred text descriptions, thereby further optimizing the quality of the text description of the video quality assessment output by the video quality assessment model.

[0088] The video quality assessment model training method provided by the embodiment of the present disclosure is divided into two stages to fine-tune the multimodal large language model to be trained according to the quality description text and the quality question-answer text in the quality text sample. Through the training based on the quality description text, the video quality assessment model obtains the overall description capability for video quality assessment. Through the training based on the quality question-answer text, the video quality assessment model is trained separately for different quality assessment dimensions, so that different quality assessment dimensions can be trained more quickly and accurately, and a target video quality assessment model with higher assessment quality can be obtained.

[0089] Figure 5 FIG. 1 is a flow chart of a method for obtaining a training data set according to an example. Figure 5 As shown, in the embodiment of the present disclosure, the training data set acquisition process of the aforementioned step S210 may include the following steps.

[0090] In step S510, the video quality of the video sample is scored using at least one video quality scoring model to obtain at least one quality scoring text.

[0091] As mentioned above, for the video quality assessment model training method provided by the present disclosure, how to obtain a large number of high-quality video quality training samples, especially quality text samples, is the key to determining the training effect of the video quality assessment model. The present disclosure provides a method for obtaining corresponding quality text samples based on video samples of high, medium and low video qualities.

[0092] In the disclosed embodiments, the obtained video samples are scored for video quality by using several video quality scoring models to obtain several corresponding quality scoring texts. In the related art, there are various model tools for evaluating and scoring video quality from different dimensions. These existing model tools are usually trained using deep learning technologies such as convolutional neural networks (CNN), such as VMAF, PSNR, SSIM, etc. On the one hand, these model tools can only evaluate the video quality for a certain video quality assessment dimension, and on the other hand, the evaluation results they output are often specific score values, lacking a text description of video quality problems, and not having the ability to attribute quality problems and provide improvement suggestions.

[0093] Figure 6 FIG. 1 is a flow chart of a quality scoring text generation process according to an example. Figure 6 As shown, the quality scoring text generation process may include the following steps.

[0094] In step S610, the video quality of the video sample is scored by using the at least one video quality scoring model to obtain at least one quality assessment score; the quality assessment score corresponds to the video quality scoring model; the video quality scoring model corresponds to at least one quality assessment dimension.

[0095] In the disclosed embodiment, a video quality scoring tool library is constructed in advance using these existing video quality scoring models. Video quality is scored for video samples using the existing video quality scoring capabilities of these video quality scoring models to obtain at least one quality assessment score. Among them, each quality assessment score corresponds to each video quality scoring model. Each video quality scoring model corresponds to at least one quality assessment dimension.

[0096] In an exemplary embodiment, the video quality scoring tool library includes at least: scoring model 1, scoring model 2, scoring model 3, and scoring model 4. The scoring model 1 is responsible for scoring the overall video quality of the video; the scoring model 2 is responsible for scoring the clarity dimension of the video; the scoring model 3 is responsible for scoring the color accuracy dimension of the video; and the scoring model 4 is responsible for scoring the noise dimension of the video. The video quality of the video sample is scored by each video quality scoring model to obtain an overall score, a clarity score, a color score, and a noise score, respectively. Each quality assessment score corresponds to each video quality scoring model. On a scale of 1 to 5, assume that the quality assessment scores are: an overall score of 4 points, a clarity score of 3.5 points, a color score of 3.8 points, and a noise score of 4.2 points.

[0097] In step S620, the at least one quality assessment score is mapped through a score text mapping relationship to obtain the at least one quality scoring text; the quality scoring text corresponds to the quality assessment dimension; the score text mapping relationship stores a mapping relationship between a preset score interval and a quality scoring text.

[0098] In the disclosed embodiment, since the quality assessment score obtained by each video quality scoring model is only a score value and does not have the text logic of natural language, in order to more accurately extract the text features, the score text mapping relationship is used to convert it into a corresponding quality scoring text in this step. Among them, the score text mapping relationship stores the mapping relationship between the preset score interval and the quality scoring text. According to the score text mapping relationship, the above-mentioned each quality assessment score obtained is mapped to a corresponding text description word, that is, a quality scoring text. Since the above-mentioned each video quality scoring model corresponds to at least one quality assessment dimension. Therefore, each quality scoring text here also corresponds to a quality assessment dimension respectively.

[0099] In an exemplary embodiment, the score text mapping relationship divides the score into five score intervals, and each of them corresponds to a different quality scoring text. Based on the standard of 1 to 5 points, the score text mapping relationship corresponds the [0, 1] interval to the quality scoring text "Bad"; the [1, 2] interval to the quality scoring text "Poor"; the [2, 3] interval to the quality scoring text "Fair"; the [3, 4] interval to the quality scoring text "Good"; and the [4, 5] interval to the quality scoring text "Excellent".

[0100] In an exemplary embodiment, according to the above score text mapping relationship, the quality assessment scores corresponding to different quality assessment dimensions can be mapped to corresponding quality scoring texts, that is, the overall is Excellent, the clarity is Good, the color is Good, and the noise is Excellent.

[0101] In the embodiments of the present disclosure, the above method can be used to score the video quality of the video sample using various existing video quality scoring models to obtain quality score texts corresponding to different video quality assessment dimensions. It should be noted that any video quality scoring tool can be built into the video quality scoring tool library as a tool for generating quality text samples in the present disclosure, and the present disclosure does not limit the specific model tool used.

[0102] In step S520, video content features are extracted from the video sample using a video content description model to obtain a content description text.

[0103] In the disclosed embodiment, the video content feature extraction is performed on the video sample through the video content description model to obtain a content description text. The video content description model can adopt a multimodal large language model, a convolutional neural network (CNN), a recurrent neural network (RNN) or a long short-term memory network (LSTM) and other models. The video content description model is used to extract the video content features and obtain the corresponding content description text. The content description text is a text that describes the video content of the video sample. The content description text can be text content related to the video quality, or it can be text content unrelated to the video quality.

[0104] In step S530, the quality scoring text and the content description text are integrated to obtain the quality text sample.

[0105] In the embodiment of the present disclosure, the quality score texts obtained in the aforementioned step S510 and the content description texts obtained in the aforementioned step S520 are integrated to obtain a quality text sample. Through the integration process, the quality text sample includes both the description text information of the video content and the quality score texts describing different quality assessment dimensions of the video.

[0106] In an exemplary embodiment, the integration method of the quality score text and the content description text may include: splicing the quality score text and the content description text to obtain a quality text sample. This method directly performs simple text splicing on the quality score text and the content description text to quickly obtain a quality text sample that incorporates video quality assessment information.

[0107] In an exemplary embodiment, the method of integrating the quality score text and the content description text may include: integrating the quality score text and the content description text through a first language model to obtain the quality text sample. This method is different from the above-mentioned direct splicing method. Instead, the quality score text and the content description text are fused through the first language model to obtain a quality text sample that incorporates video quality assessment information. Compared with the direct splicing method, the text quality obtained by this method is better, but it requires a certain amount of computing cost.

[0108] In an exemplary embodiment, the above-mentioned direct concatenation method and the large language model integration method can also be combined to fuse the result of direct concatenation with the result of model integration to obtain the quality text sample.

[0109] In an exemplary embodiment, the step S530 may further include:

[0110] The quality scoring text and the content description text are integrated to obtain the quality description text.

[0111] The quality description text is subjected to question-answer pair extraction through the second largest language model to obtain the at least one quality question-answer pair text.

[0112] In an embodiment of the present disclosure, any of the above-mentioned methods are used to integrate the quality scoring text and the content description text, and the integrated fused text is used as the aforementioned quality description text. On this basis, the quality description text is subjected to question-answer pair extraction through the second largest language model, thereby obtaining a number of quality question-answer pair texts corresponding to the video sample. In the related art, there are a variety of large language models that have the ability to generate a number of corresponding question-answer pairs based on text content. For example, vicuna, chatGPT, etc. The present disclosure utilizes the relevant capabilities in the existing model to generate a number of corresponding quality question-answer pair texts based on the quality description text for use in the aforementioned related model training.

[0113] The video quality assessment model training method provided by the embodiment of the present disclosure, on the one hand, utilizes various existing video quality scoring models to score the video quality of video samples, and obtains quality score texts corresponding to different video quality assessment dimensions. On the other hand, the content description text is obtained through the video content description model. By integrating the quality score text and the content description text, a quality text sample containing quality assessment information is obtained. In the above manner, a large number of quality text samples corresponding to video samples can be quickly generated, thereby providing sufficient training data sets for fine-tuning the multimodal large language model to be trained. In addition, based on the integrated quality description text, a number of quality question-answer pair texts corresponding to the video sample are generated, thereby providing corresponding training data for the aforementioned staged training.

[0114] Figure 7 The figure is a flowchart of a video quality assessment method according to an exemplary embodiment. Figure 8 The present invention is a schematic diagram of an implementation process of a video quality assessment method according to an example. Figure 7 , 8 The method provided in the embodiment can be executed by any electronic device, such as the above Figure 1 The terminal device in Figure 1 Servers in, or Figure 1 The terminal device and the server in the embodiment are jointly executed, but the present disclosure is not limited to this.

[0115] In step S710, quality assessment prompt words and a video to be assessed are obtained.

[0116] In the disclosed embodiment, a video quality assessment request of a user is obtained. The video quality assessment request includes at least: a quality assessment prompt and a video to be assessed. The quality assessment prompt is a description text used to describe the video quality assessment requirement. Depending on the needs of the user, the quality assessment prompt may be a video quality evaluation prompt, a video quality attribution prompt, or a video quality improvement suggestion prompt. The video to be assessed is the target video content to be assessed.

[0117] In an exemplary embodiment, the quality assessment prompt word is a video quality evaluation prompt word, and an example is as follows:

[0118] “Based on the overall quality of the following video frames, rate their visual quality on a scale of 1 to 5 (5 being the highest quality and 1 being the lowest quality). Please analyze the clarity, brightness, color accuracy, and noise levels in the video, then give an overall rating and briefly explain the main factors.”

[0119] This prompt guides the model to focus on key visual quality factors (such as clarity and color accuracy) and give a comprehensive score, thereby obtaining a picture quality score that meets the requirements.

[0120] In an exemplary embodiment, the quality assessment prompt word is a video quality attribution prompt word, and an example is as follows:

[0121] “Please analyze the video for quality defects, including blurriness, noise, brightness or color distortion, etc. For each defect detected, please describe in detail its cause, such as compression, transmission loss, camera shake, etc. Also indicate the impact these defects may have on the user experience.”

[0122] This prompt breaks down the problem, encourages the model to find different defects in the video, analyze possible technical causes, and explain the impact of the defects on the user experience, so as to provide a deeper quality attribution analysis.

[0123] In an exemplary embodiment, the quality assessment prompt word is a video quality improvement suggestion prompt word, and an example is as follows:

[0124] “Based on the quality defects detected, please provide specific suggestions for improving the quality of the video. For example, for blur issues, you can recommend using higher resolution or stable equipment when shooting; for noise issues, you can recommend using a noise reduction filter in post-production. Please list the improvement methods for each problem.”

[0125] This prompt requires the model to give detailed improvement suggestions based on specific defects, ensuring that the model's answers can help video producers or editors in actual operations and provide practical guidance information.

[0126] In step S720, the quality assessment prompt words and the video to be assessed are input into a video quality assessment model; the video quality assessment model is obtained by training using any of the aforementioned video quality assessment model training methods.

[0127] In step S730, the video quality assessment model generates a video quality assessment result for the video to be assessed according to the quality assessment prompt words and the video to be assessed.

[0128] In the disclosed embodiment, the quality assessment prompt words in the video quality assessment request and the video to be assessed are input into the video quality assessment model trained by any of the aforementioned video quality assessment model training methods. The video quality assessment model generates a detailed video quality assessment result after comprehensive analysis based on the prompt words and the video content. This process makes full use of the association between the video features and the text description learned by the model during training, ensuring that the assessment results are both accurate and constructive, and providing users with reliable video quality assessment feedback.

[0129] The video quality assessment method provided by the embodiment of the present disclosure can evaluate the quality of the video to be evaluated based on the video quality assessment model obtained by the above training according to the quality assessment prompt words input by the user, and output the corresponding video quality assessment result. According to different quality assessment prompt words, the video quality assessment model can not only output the assessment result of the video quality, but also further attribute the fine-grained quality problems and put forward suggestions for improving the video quality.

[0130] The following are embodiments of the device disclosed herein, which can be used to execute the method embodiments disclosed herein. For details not disclosed in the device embodiments disclosed herein, please refer to the method embodiments disclosed herein.

[0131] Fig. 9 is a block diagram of a video quality assessment model training device according to an exemplary embodiment. Fig. 9 The device 900 may include: a sample acquisition module 910, a video text feature module 920, a quality text feature module 930 and a model training module 940.

[0132] The sample acquisition module 910 is configured to acquire a training data set including a plurality of video quality training samples; the video quality training samples include corresponding video samples and quality text samples; the quality text samples are texts for evaluating the video quality of the video samples.

[0133] The video text feature module 920 is configured to extract features from the video sample through a video encoder to obtain video text features.

[0134] The quality text feature module 930 is configured to extract features from the quality text sample through a text encoder to obtain quality text features.

[0135] The model training module 940 is configured to fine-tune the multimodal large language model to be trained according to the video text features and quality text features to obtain a target video quality assessment model.

[0136] In some exemplary embodiments of the present disclosure, the quality text sample includes: a quality description text and at least one quality question and answer text; the quality description text is a text describing the video quality of the video sample; the quality question and answer text is a question and answer text evaluating the video quality of the video sample; each of the quality question and answer texts includes a corresponding quality question text and a quality answer text.

[0137] In some exemplary embodiments of the present disclosure, the model training module 940 is further configured to fine-tune the multimodal large language model to be trained according to the video text features and the quality description text features to obtain a first video quality assessment model; the quality description text features are obtained by extracting features from the quality description text through a first text encoder; the first video quality assessment model is fine-tuned according to the video text features and the quality question and answer pair text features to obtain the target video quality assessment model; the quality question and answer pair text features are obtained by extracting features from the quality question and answer pair text through a second text encoder.

[0138] In some exemplary embodiments of the present disclosure, the model training module 940 is also configured to fine-tune the first video quality assessment model based on the video text features, quality question and answer text features and the first quality description text features to obtain the target video quality assessment model; wherein the first quality description text features belong to the quality description text features.

[0139] In some exemplary embodiments of the present disclosure, the sample acquisition module 910 is further configured to perform video quality scoring on the video sample through at least one video quality scoring model to obtain at least one quality scoring text; perform video content feature extraction on the video sample through a video content description model to obtain a content description text; and integrate the quality scoring text and the content description text to obtain the quality text sample.

[0140] In some exemplary embodiments of the present disclosure, the sample acquisition module 910 is further configured to perform video quality scoring on the video sample through the at least one video quality scoring model to obtain at least one quality assessment score; the quality assessment score corresponds to the video quality scoring model; the video quality scoring model corresponds to at least one quality assessment dimension; the at least one quality assessment score is mapped through a score text mapping relationship to obtain the at least one quality scoring text; the quality scoring text corresponds to the quality assessment dimension; the score text mapping relationship stores a mapping relationship between a preset score interval and a quality scoring text.

[0141] In some exemplary embodiments of the present disclosure, the sample acquisition module 910 is further configured to concatenate the quality scoring text and the content description text to obtain the quality text sample.

[0142] In some exemplary embodiments of the present disclosure, the sample acquisition module 910 is further configured to integrate the quality scoring text and the content description text through a first large language model to obtain the quality text sample.

[0143] In some exemplary embodiments of the present disclosure, the sample acquisition module 910 is further configured to integrate the quality scoring text and the content description text to obtain the quality description text; and to extract question-answer pairs from the quality description text through a second largest language model to obtain the at least one quality question-answer pair text.

[0144] Fig.10 is a block diagram of a video quality assessment device according to an exemplary embodiment. Fig.10 , the device 1000 may include: an evaluation request acquisition module 1010 , an evaluation request input module 1020 and a quality evaluation module 1030 .

[0145] The evaluation request acquisition module 1010 is configured to acquire quality evaluation prompt words and a video to be evaluated.

[0146] The evaluation request input module 1020 is configured to input the quality evaluation prompt words and the video to be evaluated into a video quality evaluation model; the video quality evaluation model is obtained by training using any of the video quality evaluation model training methods.

[0147] The quality assessment module 1030 is configured to generate a video quality assessment result of the video to be assessed based on the quality assessment prompt words and the video to be assessed by the video quality assessment model.

[0148] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0149] Refer to the following Fig.11 1100 according to this embodiment of the present disclosure is described. Fig.11 The electronic device 1100 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0150] like Fig.11 As shown, the electronic device 1100 is in the form of a general computing device. The components of the electronic device 1100 may include, but are not limited to: the at least one processing unit 1110, the at least one storage unit 1120, a bus 1130 connecting different system components (including the storage unit 1120 and the processing unit 1110), and a display unit 1140.

[0151] The storage unit stores program codes, which can be executed by the processing unit 1110, so that the processing unit 1110 executes the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification. For example, the processing unit 1110 can execute the following steps: Figure 2The steps shown.

[0152] For example, electronic devices can achieve Figure 2 or Figure 7 The steps shown.

[0153] The storage unit 1120 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 1121 and / or a cache memory unit 1122 , and may further include a read-only memory unit (ROM) 1123 .

[0154] The storage unit 1120 may also include a program / utility 1124 having a set (at least one) of program modules 1125, such program modules 1125 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0155] Bus 1130 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0156] The electronic device 1100 may also communicate with one or more external devices 1170 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 1100, and / or communicate with any device that enables the electronic device 1100 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1150. Furthermore, the electronic device 1100 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 1160. As shown, the network adapter 1160 communicates with other modules of the electronic device 1100 via a bus 1130. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1100, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0157] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiment of the present disclosure.

[0158] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory including instructions, and the above instructions can be executed by a processor of the device to complete the above method. Optionally, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0159] In an exemplary embodiment, a computer program product is also provided, including a computer program, and when the computer program is executed by a processor, the method in the above embodiment is implemented.

[0160] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0161] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A video quality assessment model training method, characterized in that: include: Acquire a training data set including a plurality of video quality training samples; the video quality training samples include corresponding video samples and quality text samples; the quality text samples are texts for evaluating the video quality of the video samples; Extracting features from the video sample using a video encoder to obtain video text features; Extracting features from the quality text sample using a text encoder to obtain quality text features; According to the video text features and quality text features, the multimodal large language model to be trained is fine-tuned to obtain a target video quality assessment model.

2. The method according to claim 1, characterized in that The quality text sample includes: a quality description text and at least one quality question-answer pair text; The quality description text is a text describing the video quality of the video sample; The quality question-answer text pair is a question-answer text pair for evaluating the video quality of the video sample; each of the quality question-answer text pairs includes a quality question text and a quality answer text in a corresponding relationship.

3. The method according to claim 2, characterized in that The method of fine-tuning the multimodal large language model to be trained according to the video text features and the quality text features to obtain a target video quality assessment model includes: According to the video text features and the quality description text features, the multimodal large language model to be trained is fine-tuned to obtain a first video quality assessment model; the quality description text features are obtained by extracting features from the quality description text through a first text encoder; The first video quality assessment model is fine-tuned and trained according to the video text features and the quality question and answer text features to obtain the target video quality assessment model; the quality question and answer text features are obtained by extracting features from the quality question and answer text using a second text encoder.

4. The method according to claim 3, characterized in that The step of fine-tuning the first video quality assessment model according to the video text features and the quality question-answer pair text features to obtain the target video quality assessment model includes: The first video quality assessment model is fine-tuned and trained according to the video text features, the quality question-answer text features and the first quality description text features to obtain the target video quality assessment model; wherein the first quality description text features belong to the quality description text features.

5. The method according to claim 1, characterized in that: The step of obtaining a training data set including a plurality of video quality training samples comprises: Performing video quality scoring on the video sample by using at least one video quality scoring model to obtain at least one quality scoring text; Extracting video content features from the video sample using a video content description model to obtain a content description text; The quality scoring text and the content description text are integrated to obtain the quality text sample.

6. The method according to claim 5, characterized in that The step of scoring the video quality of the video sample by using at least one video quality scoring model to obtain at least one quality scoring text includes: Scoring the video quality of the video sample by using the at least one video quality scoring model to obtain at least one quality assessment score; the quality assessment score corresponds to the video quality scoring model; the video quality scoring model corresponds to at least one quality assessment dimension; The at least one quality assessment score is mapped through a score text mapping relationship to obtain the at least one quality scoring text; the quality scoring text corresponds to the quality assessment dimension; the score text mapping relationship stores a mapping relationship between a preset score interval and a quality scoring text.

7. The method according to claim 5, characterized in that The step of integrating the quality scoring text and the content description text to obtain the quality text sample includes: The quality scoring text and the content description text are concatenated to obtain the quality text sample.

8. The method according to claim 5, characterized in that The step of integrating the quality scoring text and the content description text to obtain the quality text sample includes: The quality scoring text and the content description text are integrated through a first language model to obtain the quality text sample.

9. The method according to claim 5, characterized in that The step of integrating the quality scoring text and the content description text to obtain the quality text sample includes: Integrate the quality scoring text and the content description text to obtain the quality description text; The quality description text is subjected to question-answer pair extraction through the second largest language model to obtain the at least one quality question-answer pair text.

10. A video quality assessment method, characterized in that: include: Obtain quality assessment prompt words and videos to be assessed; Inputting the quality assessment prompt words and the video to be assessed into a video quality assessment model; the video quality assessment model is obtained by training the video quality assessment model training method according to any one of claims 1 to 9; The video quality assessment model generates a video quality assessment result of the video to be assessed according to the quality assessment prompt words and the video to be assessed.

11. A video quality assessment model training device, characterized in that: include: The sample acquisition module is configured to acquire a training data set including a plurality of video quality training samples; the video quality training samples include corresponding video samples and quality text samples; the quality text samples are texts for evaluating the video quality of the video samples; A video text feature module is configured to extract features from the video sample through a video encoder to obtain video text features; A quality text feature module is configured to extract features from the quality text sample through a text encoder to obtain quality text features; The model training module is configured to fine-tune the multimodal large language model to be trained according to the video text features and quality text features to obtain a target video quality assessment model.

12. A video quality assessment device, characterized in that: include: An evaluation request acquisition module is configured to acquire quality evaluation prompt words and a video to be evaluated; An evaluation request input module, configured to input the quality evaluation prompt word and the video to be evaluated into a video quality evaluation model; the video quality evaluation model is obtained by training the video quality evaluation model training method according to any one of claims 1 to 9; The quality assessment module is configured so that the video quality assessment model generates a video quality assessment result for the video to be assessed according to the quality assessment prompt words and the video to be assessed.

13. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the executable instructions to implement the video quality assessment model training method as described in any one of claims 1 to 9 or the video quality assessment method as described in claim 10.

14. A computer-readable storage medium, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to execute the video quality assessment model training method as described in any one of claims 1 to 9 or the video quality assessment method as described in claim 10.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the video quality assessment model training method according to any one of claims 1 to 9 or the video quality assessment method according to claim 10 is implemented.

Images