Image data processing method, and system, electronic device and storage medium

Abstract

Disclosed in the present disclosure are an image data processing method, and a system, an electronic device and a storage medium. The method comprises: monitoring image data to be processed; acquiring text data corresponding to the image data, wherein the text data is used for recording at least one type of target task; acquiring task description information of the target task, wherein the task description information is used for describing semantic information of the target task and semantic information of an associated task, and an association relationship exists between the semantic information of the target task and the semantic information of the associated task; and inputting the task description information and the image data into a vision model, and using the task description information to guide the vision model to execute the target task on the image data, and obtaining a task execution result.

Description

[0001]Image Data Processing Method, System, Electronic Device, and Storage Medium Cross-Reference This disclosure claims priority to Chinese patent application No. 202410195033.X, filed with the China Patent Office on February 21, 2024, entitled “Image Data Processing Method, System, Electronic Device, and Storage Medium,” the entire contents of which are incorporated herein by reference. Technical Field This disclosure relates to large model technology and image processing, and more specifically, to an image data processing method, system, electronic device, and storage medium. Background: Currently, building open-world, unified, and interactive visual models is a trend in general artificial intelligence. However, the time-consuming and labor-intensive task of labeling training data during visual model construction limits the types of tasks that can be handled by existing visual models. Furthermore, traditional task-specific visual models are trained using task-specific data, resulting in significant technical limitations in the model's ability to process image data. Currently, no effective solution has been proposed to address these issues. SUMMARY OF THE INVENTION Embodiments of the present disclosure provide a method, system, electronic device, and storage medium for processing image data to at least address the technical issue of limited model processing of image data. According to one aspect of an embodiment of the present disclosure, a method for processing image data is provided. The method may include: monitoring image data to be processed; obtaining text data corresponding to the image data, wherein the text data describes at least one type of target task; obtaining task description information for the target task, wherein the task description information describes semantic information of the target task and semantic information of associated tasks, wherein the semantic information of the target task and the semantic information of the associated tasks are associated with each other; inputting the task description information and image data into a visual model, and using the task description information to guide the visual model to perform the target task on the image data, thereby obtaining a task execution result. According to another aspect of an embodiment of the present disclosure, another method for generating a model is also provided.The method may include: querying a corpus for semantic information samples of associated tasks based on semantic information samples and association relationships of task types, wherein the corpus includes at least semantic information samples of different task types and association relationships between the semantic information samples of different task types, wherein the task types are recorded in text data samples, and there are association relationships between the semantic information samples of the task types and the semantic information samples of the associated tasks; combining the semantic information samples of the task types and the semantic information samples of the associated tasks to form a task description information sample that matches the text data sample, wherein the task description information sample is used to describe the semantic information sample of the task types and the semantic information sample of the associated tasks; and training a visual model using image data samples corresponding to the task description information samples and the text data samples, wherein the visual model is used to perform the task types on input image data. According to another aspect of an embodiment of the present disclosure, another method for processing image data is also provided. The method may include: displaying image data to be processed on the operation interface in response to an input operation on the operation interface; displaying text data corresponding to the image data on the operation interface, wherein the text data describes at least one type of target task; and displaying a task execution result corresponding to the image data on the operation interface in response to a task execution operation on the operation interface, wherein the task execution result is obtained by guiding a visual model to perform the target task on the image data using task description information of the target task, wherein the task description information describes semantic information of the target task and semantic information of associated tasks, wherein the semantic information of the target task and the semantic information of associated tasks are associated with each other. According to another aspect of an embodiment of the present disclosure, a system for processing image data is also provided. The system may include: a data input terminal configured to monitor image data to be processed; a task execution terminal configured to obtain text data corresponding to the image data, wherein the text data describes at least one type of target task; obtaining task description information for the target task, wherein the task description information describes semantic information of the target task and semantic information of associated tasks, with the semantic information of the target task and the semantic information of the associated tasks being associated with each other; inputting the task description information and the image data into a visual model, and using the task description information to guide the visual model to perform the target task on the image data to obtain a task execution result; and a result output terminal configured to output the task execution result. According to another aspect of an embodiment of the present disclosure, an electronic device is also provided. The electronic device may include a memory and a processor; the memory is configured to store computer-executable instructions, and the processor is configured to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the above-mentioned method of the embodiment of the present disclosure is implemented.According to another aspect of an embodiment of the present disclosure, a processor is provided. The processor is configured to run a program, wherein when the program is executed, the processor performs any of the aforementioned methods. According to another aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored program, wherein when the program is executed, the processor controls the device containing the storage medium to perform any of the aforementioned methods. According to another aspect of an embodiment of the present disclosure, a computer program product is provided. The computer program product may include computer instructions that, when executed by the processor, implement any of the aforementioned methods. In an embodiment of the present disclosure, image data to be processed is monitored; text data corresponding to the image data is acquired, wherein the text data describes at least one type of target task; task description information of the target task is acquired, wherein the task description information describes semantic information of the target task and semantic information of associated tasks, wherein the semantic information of the target task and the semantic information of the associated tasks are associated with each other; the task description information and the image data are input into a visual model, and the task description information is used to guide the visual model to perform the target task on the image data, thereby obtaining a task execution result. That is, in embodiments of the present disclosure, by acquiring text data corresponding to image data, task description information corresponding to the input image data can be generated. This task description information can be used to guide the visual model to better understand the learning process of different tasks, thereby better performing the target task on the image data. This provides a unified, fine-grained, and hierarchical understanding of the target for multi-task learning, thereby achieving the technical effect of reducing the limitations of the model in processing image data and resolving the technical problem of the large limitations of the model in processing image data. It should be noted that the general description above and the detailed description that follow are merely illustrative and illustrative of the present disclosure and do not constitute limitations of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS The drawings described herein are provided to provide a further understanding of the present disclosure and constitute a part of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are provided to explain the present disclosure and do not constitute undue limitations of the present disclosure.In the accompanying drawings: Figure 1 is a schematic diagram of an application scenario of a method for processing image data according to an embodiment of the present disclosure; Figure 2 is a flow chart of a method for processing image data according to an embodiment of the present disclosure; Figure 3 is a flow chart of a method for deploying a model according to an embodiment of the present disclosure; Figure 4 is a flow chart of another method for processing image data according to an embodiment of the present disclosure; Figure 5 is a schematic diagram of an image data processing system according to an embodiment of the present disclosure; Figure 6 is a schematic diagram of a general open visual understanding framework according to an embodiment of the present disclosure; Figure 7 is a hardware structure block diagram of a computer terminal (or mobile device) for a method for processing image data according to an embodiment of the present disclosure; Figure 8 is a schematic diagram of an image data processing device according to an embodiment of the present disclosure; Figure 9 is a schematic diagram of a model generation device according to an embodiment of the present disclosure; Figure 10 is a schematic diagram of another image data processing device according to an embodiment of the present disclosure; Figure 11 is a structural block diagram of a computer terminal according to an embodiment of the present disclosure; Figure 12 is a block diagram of an electronic device for a method for processing image data according to an embodiment of the present disclosure. DETAILED DESCRIPTION To help those skilled in the art better understand the present disclosure, the following will provide a clear and complete description of the technical solutions in the embodiments of the present disclosure, in conjunction with the accompanying drawings. It should be noted that the described embodiments represent only a portion of the embodiments of the present disclosure, and are not exhaustive. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present disclosure without inventive effort should fall within the scope of protection of the present disclosure. It should be noted that the terms "first," "second," and so on, in the specification and claims of the present disclosure, and in the accompanying drawings, are used to distinguish similar objects, and are not necessarily intended to describe a specific order or sequential sequence. It should be understood that such terms are interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than that illustrated or described herein. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or components is not necessarily limited to the steps or components expressly listed, but may include other steps or components not expressly listed or inherent to such process, method, product, or apparatus. The technical solutions provided in this disclosure are mainly implemented using large-scale model technology. The large model here refers to a deep learning model with large-scale model parameters, which can typically include hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than ten trillion model parameters.Large models, also known as foundational models, are pre-trained on large amounts of unlabeled corpora, producing pre-trained models with over 100 million parameters. These models are adaptable to a wide range of downstream tasks and exhibit good generalization capabilities. Examples include large language models (LLMs) and multi-modal pre-training models. It should be noted that in practical applications, large models can be fine-tuned using a small number of samples, allowing them to be applied to different tasks. For example, large models can be widely applied in fields such as natural language processing (NLP), computer vision, and speech processing. Specifically, they can be applied to computer vision tasks such as visual question answering (VQA), image captioning (IC), and image generation. They can also be widely used in natural language processing tasks such as text-based sentiment classification, text summarization, and machine translation. Therefore, the main application scenarios of large models include, but are not limited to, digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design. In the embodiments of this disclosure, data processing using visual models in the context of image processing is used as an example for explanation. First, some terms or terms used in describing the embodiments of this disclosure are subject to the following interpretation: A specialized / expert visual model can be a model specifically proposed and trained to solve a specific type of objective, such as a semantic segmentation model, an instance segmentation model, or an object detection model. A general visual model can be a unified large-scale visual model that integrates multiple objectives. A general large-scale visual model can be selected to simultaneously solve multiple objectives. Open-world understanding can refer to a model's ability to solve objects of categories not seen during training. Visual concepts can refer to a person's cognition and understanding of visual features such as the appearance, shape, color, size, and spatial relationships of objects. According to embodiments of this disclosure, a method for processing image data is provided. It should be noted that the steps shown in the flowcharts of the accompanying figures can be executed in a computer system, such as a set of computer-executable instructions. Although the flowcharts illustrate a logical order, in some cases, the steps shown or described may be executed in a different order.Considering the large number of model parameters in large models and the limited computing resources of mobile terminals, the image data processing method provided in the embodiments of the present disclosure can be applied to the application scenario shown in Figure 1, but is not limited thereto. Figure 1 is a schematic diagram of an application scenario of the image data processing method according to the embodiments of the present disclosure. In the application scenario shown in Figure 1, the large model is deployed on a server 10. Server 10 can be connected to one or more client devices 20 via a local area network, a wide area network, the Internet, or other types of data networks. Client devices 20 herein may include, but are not limited to, smartphones, tablet computers, laptops, PDAs, personal computers, smart home devices, and in-vehicle devices. Client devices 20 can interact with users through an operation interface to invoke the large model, thereby implementing the image data processing method provided in the embodiments of the present disclosure. In an embodiment of the present disclosure, a system consisting of a client device and a server can perform the following steps: the client device executes image data to be processed; the server can perform the following tasks: Step S102: monitoring the image data to be processed; Step S104: obtaining text data corresponding to the image data, wherein the text data is used to record at least one type of target task; Step S106: obtaining task description information for the target task, wherein the task description information is used to describe the semantic information of the target task and the semantic information of associated tasks, and there is an association between the semantic information of the target task and the semantic information of the associated tasks; Step S108: inputting the task description information and the image data into a visual model, and using the task description information to guide the visual model to execute the target task on the image data, thereby obtaining a task execution result. It should be noted that, if the operating resources of the client device can meet the deployment and operation requirements of the large model, the present disclosure can be implemented in the client device. In this operating environment, the present disclosure provides an image data processing method as shown in Figure 2. Figure 2 is a flowchart of an image data processing method according to an embodiment of the present disclosure. As shown in FIG2 , the method may include the following steps: Step S202: Monitoring image data to be processed. In the technical solution provided in step S202 of the present disclosure, the image data to be processed may be monitored. The image data to be processed may be image data required to perform a target task, such as image data to be subjected to semantic segmentation, instance segmentation, or panoptic segmentation. These examples are provided herein for illustrative purposes only and are not intended to limit the image data. Step S204: Acquiring text data corresponding to the image data, wherein the text data is used to record at least one type of target task.In the technical solution provided in step S204 of the present disclosure, text data corresponding to the image data can be obtained. The text data may include at least one type of target task, which may be recorded in the form of task text. The target task may be a visual understanding task or visual task to be performed on the image data, such as semantic segmentation, instance segmentation, object localization, or object monitoring. It should be noted that this is merely an example and does not specifically limit the type of target task. Optionally, upon detecting image data to be processed, text data corresponding to the image data is obtained to determine at least one target task to be performed on the image data. The at least one target task to be performed on the image data can be determined by determining the task name. In step S206, task description information for the target task is obtained. The task description information describes the semantic information of the target task and the semantic information of associated tasks, with the semantic information of the target task and the semantic information of the associated tasks being associated with each other. In the technical solution provided in step S206 of the present disclosure, a target task is determined based on text data, and task description information for the target task is obtained. The task description information can be used to describe the semantic information of the target task and the semantic information of associated tasks. It can be information for visual positioning description and can be in a fixed format or in the form of strong directives. Optionally, the semantic information can be used to represent visual concepts (also referred to as concepts) of visual perception elements, reflecting the cognition and understanding of visual features. For example, it can be used to represent the categories of objects in image data. Associations can include hierarchical relationships, hierarchical relationships, partial relationships, negative relationships, positive relationships, and other relationships between semantic information. These can be represented through a knowledge graph. This is merely an example and does not impose any specific restrictions on the types of associations. An associated task can be a task that has an association with the target task. Optionally, there can be a semantic or conceptual association between the semantic information of the associated task and the semantic information of the target task. For example, if both the semantic segmentation task and the object detection task focus on cats and dogs in an image, there is an association between the two tasks. Associations can be represented in the corpus through common vocabulary or data elements. For example, if the semantic information of the target task is to monitor animals, and since there is a hierarchical relationship between animals, cats, and dogs in the corpus, the semantic information of the associated task can be determined to be monitoring cats and dogs. It should be noted that this is merely an example and does not impose any specific restrictions on the method for determining the semantic information of the associated task.In step S208, the task description information and image data are input into the visual model, and the task description information is used to guide the visual model to perform the target task on the image data, thereby obtaining a task execution result. In the technical solution provided in step S208 of the present disclosure, the task description information and image data can be input into the visual model, and the visual model performs the target task on the image data based on the task description information to obtain a task execution result. The visual model can be a large visual model trained based on large-scale image data, and can include a specialized / expert model, such as a general visual model. The task execution result can be the result obtained after the visual model performs the target task on the image data, such as a semantic segmentation result of the image data, an instance segmentation result of the image data, or an object detection result of the image data. It should be noted that this is merely an example and does not impose any specific restrictions on the type of task execution result. Because the visual model's ability to understand task types is very limited, far less than the vocabulary humans use to describe the real world, there is a technical problem of significant limitations in the visual model's ability to process image data. To address the above issues, in this embodiment, text data is adjusted to generate multi-granularity task description information. This generated task description information is used to guide the visual model in performing the target task on the image data. Semantic information and rich visual data elements provide important semantic support for the visual model, thereby improving the model's generalization and multi-task processing capabilities. This, in turn, reduces the model's limitations in processing image data, resolving the technical issue of the model's significant limitations in processing image data. Alternatively, the multi-granularity task description information can include semantic information describing the target task and semantic information related to the task. For example, if the text data is "Identify mammals in an image," the task description information can be "Identify cats, dogs, and chickens among mammals." This demonstrates that the task description information further refines the hierarchy of the text data. In this embodiment, by utilizing multi-granularity task description information, the visual model accurately understands the target task to be performed and performs the target task on the image data, thereby improving the visual model's processing efficiency. Optionally, the task description information and the image data are input into the visual model, and the task description information can be used to guide the visual model to perform the target task on the image data. Since the task description information can enable the visual model to more accurately determine what the target task to be processed is, accurate task execution results can be obtained.For example, suppose a visual model needs to be used to identify animals in an image. The image data could be a photo of a cat and a dog, and the text data could be "Identify animals in an image." The task description could then be multi-granular information such as "Identify cats, dogs, sheep, and chickens in an image." The task description and image data can be input into the visual model, and the task description can be used to guide the visual model in performing the target task on the image data. "Guidance" means that the visual model will perform the target task on the image data based on the content indicated by the task description, i.e., identifying cats, dogs, sheep, and chickens in the image. In other words, the output of the visual model is influenced by the task description. The visual model will analyze the features in the image and identify the cat and dog, ultimately obtaining the task execution result, i.e., identifying the cat and dog in the image. Through the above steps, the task description completes the process of guiding the visual model to perform the target task on the image data. It should be noted that this is merely an example and does not impose any specific limitation on the manner in which the task description guides the visual model. Through steps S202 to S208 above, the image data to be processed is monitored; text data corresponding to the image data is obtained, wherein the text data is used to record at least one type of target task; task description information for the target task is obtained, wherein the task description information describes the semantic information of the target task and the semantic information of associated tasks, and there is a correlation between the semantic information of the target task and the semantic information of the associated tasks; the task description information and the image data are input into a visual model, and the task description information is used to guide the visual model to perform the target task on the image data, thereby obtaining a task execution result. That is, in the embodiment of the present disclosure, obtaining text data corresponding to the image data can generate task description information corresponding to the input image data. The task description information is used to guide the visual model to better understand the learning process of different tasks, thereby better performing the target task on the image data. This provides a unified, fine-grained, and hierarchical target understanding for multi-task learning, thereby achieving the technical effect of reducing the limitations of the model in processing image data and solving the technical problem of the large limitations of the model in processing image data. The above-mentioned method of this embodiment is further described below.As an optional implementation, step S206, obtaining task description information for the target task, includes: identifying at least one type of target task from the text data corresponding to the image data; querying a corpus for semantic information of associated tasks based on the semantic information and associations of the target task, wherein the corpus includes at least semantic information of different types of tasks and associations between the semantic information of different types of tasks; and combining the semantic information of the target task with the semantic information of associated tasks to form task description information that matches the text data. In this embodiment, at least one type of target task is identified from the text data. Based on the semantic information and associations of the target task, semantic information of the associated tasks is querying the corpus for semantic information of the associated tasks. The semantic information of the target task with the semantic information of the associated tasks can be combined to obtain task description information that matches the text data. The corpus can include at least semantic information of different types of tasks and associations between the semantic information of different types of tasks, and can be a multi-granular concept-relation corpus, such as a visual concept-relation corpus. The type of corpus is not specifically limited herein. For example, the image data to be processed contains an image of a cat and a dog. The text data corresponding to the image data is obtained as follows: "Perform semantic segmentation and object monitoring on image data, segmentation targets are animals, and monitoring targets are animals." The target task is identified as semantic segmentation and object monitoring. Based on the association between the semantic information of the target task and the semantic information of different types of tasks, the semantic information of the associated tasks is determined in the corpus to be "segment cats and dogs, monitor cats and dogs." The semantic information of the target task can be combined with the semantic information of the associated tasks to obtain task description information that matches the text data: "Perform semantic segmentation and object monitoring on image data, segmentation targets are cats and dogs, and monitoring targets are cats and dogs." As an optional implementation, querying the corpus for semantic information of associated tasks based on the semantic information and association relationships of the target task includes: if a type task matching the target task's type is identified in the corpus, using the semantic information of the matching type task as the semantic information of the target task; determining the association relationship corresponding to the semantic information of the target task in the corpus; and querying the corpus based on the association relationship based on the semantic information of the target task to obtain semantic information of associated tasks associated with the target task. In this embodiment, a type task matching the target task's type is identified in the corpus. If a task type matching the target task's type is identified, the semantic information of the matching type task can be used as the semantic information of the target task.In the corpus, associations corresponding to the semantic information of the target task can be determined. Based on the semantic information of the target task, the corpus is queried according to the associations to obtain semantic information of associated tasks associated with the target task. In other words, before determining the semantic information of the associated tasks, it is necessary to determine whether there are tasks of a type matching the target task in the corpus. Semantic information of the associated tasks can then be further determined based on the semantic information of the matching tasks. For example, assuming the text data is "Identify dogs in image data," the semantic information of the target task can be determined to be "dog." In the corpus, dogs may be hierarchically structured, for example, their parent categories may include animals and mammals. If "dog" is found in the corpus, the semantic information of the associated tasks associated with "dog" can be determined to be "cat," "bird," "mammal," and so on, based on the association relationships. This is merely an example and does not impose any specific restrictions on the types of text data and semantic information. As an optional implementation, the method may further include: obtaining text features of the text data and image features of the image data; concatenating the text features and image features to obtain concatenated features; determining a target feature that matches the concatenated features from features of semantic information of different types of tasks; and determining the type of task corresponding to the target feature in the corpus as a type of task that matches the type of the target task. In this embodiment, concatenating the text features and image features achieves visual-linguistic alignment. By concatenating the image features with the text features, This can help the visual model better understand the relationship between different visual tasks, thereby improving the accuracy of the visual model in multi-task processing. For example, text data is tokenized and encoded using a text encoder to obtain accurate text features. Image features of the image data are determined. The text features and image features are concatenated to obtain concatenated features. From the semantic information features of different types of tasks, a target task that matches the concatenated features is determined. The type of task corresponding to the target feature in the corpus can be determined as a task that matches the type of the target task. As an optional implementation, obtaining image features of the image data includes: invoking an encoder in the visual model to extract image features from the image data. In this embodiment, the encoder in the visual model can be invoked to extract image features from the image data. The encoder can be an image encoder. oFor example, image data to be segmented, image data to be monitored, and image data to be located can be obtained, and an encoder can be invoked to encode the aforementioned image data to obtain image features corresponding to the image data. As an optional implementation, semantic information of the target task and semantic information of associated tasks can be combined to form task description information that matches the text data. This includes invoking a generation template that matches the visual model to fuse the semantic information of the target task and the semantic information of the associated tasks to obtain the task description information. The generation template is used to indicate the format that the task description information meets. In this embodiment, the generation template that matches the visual model can be invoked to fuse the semantic information of the target task and the semantic information of the associated tasks to obtain the task description information. The generation template can be used to indicate the format that the task description information must meet. This can be a pre-set format or a unified instruction condition. For example, it can be [Task Name], [Concept Name], or semantic information, such as [Person: Human] and [Cat: Animal] for image segmentation. It should be noted that this is merely an example and does not impose any specific restrictions on the content of the generation template. Since visual models can handle different types of tasks, when using a visual model to process image data, the visual model must first be informed of the task type to be processed. This can be done through the task description information. Therefore, in this embodiment, a generation template that matches the visual model is called, and the semantic information of the target task and the semantic information of related tasks are integrated through the generation template to obtain the task description information. The resulting task description information is a description that combines multi-granular concepts from the corpus. For example, assume that the beginning of the generation template is used to place semantic information for the target task, such as image classification and object detection. Let's assume the semantic information for the target task is "locate animals in an image," and the semantic information for the associated task is "cat, dog, bird." The semantic information of the target and associated tasks can be combined to produce the task description "locate cats, dogs, and birds in an image." This approach enriches and refines the text data, allowing the visual model to better understand the task content when performing the target task on image data, thereby improving the accuracy of the model's data processing.As an optional implementation, step S208, inputting the task description information and image data into the visual model, and using the task description information to guide the visual model to perform the target task on the image data to obtain the task execution result, includes: inputting the task description information and image features of the image data into a decoder in the visual model; and using the task description information to guide the decoder to perform a decoding operation corresponding to the target task on the image features to obtain the task execution result. In this embodiment, the task description and image features of the image data can be input into the decoder in the visual model, and the task description information can be used to guide the decoder to perform a decoding operation corresponding to the target task on the image features to obtain the task execution result. The decoding operation may include scene decoding and object decoding. This is for illustrative purposes only and does not specifically limit the type of decoding operation. As an optional implementation, the method may also include updating model parameters of the visual model using the task execution result. In this embodiment, task execution results can be used as training data to update the model parameters of the visual model, thereby further training the visual model. In this embodiment, a hierarchical, multi-granular corpus is constructed. This corpus can be used to generate task description information corresponding to the input image data, guiding the visual model to better understand the learning process of different tasks, thereby better performing tasks on image data. This provides a unified, fine-grained, and hierarchical understanding of the objectives of multi-task learning, thereby achieving the technical effect of reducing the limitations of the model's image data processing and resolving the technical problem of the model's large limitations in image data processing. This embodiment also provides a model generation method. Figure 3 is a flow chart of a model generation method according to an embodiment of the present disclosure. As shown in Figure 3, the method may include the following steps. In step S302, based on the semantic information samples and association relationships of the task type, the corpus is searched for semantic information samples of the associated task. The corpus includes at least semantic information samples of different task types and association relationships between the semantic information samples of the different task types. The task type is recorded in the text data sample, and association relationships exist between the semantic information samples of the task type and the semantic information samples of the associated task. In the technical solution provided in step S302 of the present disclosure, the semantic information samples can be used to represent the category of objects in the image sample data, and can be visual concepts (hereinafter referred to as concepts) of visual perception elements, which can be used to reflect the cognition and understanding of visual features.The corpus includes at least semantic information samples of different task types and the relationships between semantic information samples of different task types. For example, these may include hierarchical relationships, hierarchical relationship diagrams, hierarchical relationships, correlation relationships, knowledge graphs, and other content between semantic information. It should be noted that this is merely an example and does not impose any specific limitations on the content contained in the corpus. For example, there may be a relationship between semantic information samples of a task type and semantic information samples of associated tasks. For example, if the semantic information sample of the task type is "locating animals in image data," the semantic information sample of the associated task may be "locating dogs, cats, and pigs in image data." In step S304, the semantic information samples of the task type and the semantic information samples of the associated task are combined to form a task description information sample that matches the text data sample. The task description information sample is used to describe the semantic information samples of the task type and the semantic information samples of the associated task. In the technical solution provided in step S304 of the present disclosure, the semantic information samples of the type task and the semantic information samples of the associated tasks are spliced, merged, or otherwise processed to obtain a task description information sample that matches the text data sample. The resulting task description information sample can be a multi-granularity information sample. It should be noted that this is merely an example and does not impose any specific limitation on the method for generating the task description information sample. In step S306, a visual model is trained using the task description information samples and the image data samples corresponding to the text data samples. The visual model is used to perform the type task on the input image data. In the technical solution provided in step S306 of the present disclosure, the visual model can be trained using the task description information samples and the image data samples. The trained visual model can be used to perform the target task on the input image data. As an optional implementation, the method may further include: sampling semantic information from text data samples corresponding to different types of tasks; determining the semantic information of the sampled text data samples as the semantic information of the corresponding type of task; and constructing a corpus based on the semantic information of the different types of tasks. In this embodiment, text data samples corresponding to different types of tasks are obtained. Semantic information of the text data samples can be sampled online for identification to determine semantic information for different types of tasks. Semantic information can be used to characterize the type of data element, such as positive categories or negative categories. oA corpus can be constructed based on semantic information. Optionally, the text data samples can be from different datasets and can be open vocabularies (also referred to as raw vocabularies). Different datasets can include open vocabularies for segmentation and / or monitoring. The different types of tasks can include target tasks, including segmentation, monitoring, and positioning tasks. For example, semantic information can be obtained for text data samples corresponding to different types of tasks. For example, a text data sample can be "window, wall, cat, table, sofa, person, potted plant, plant." Online discriminative sampling is performed on the semantic information of the raw vocabularies to determine the negative and positive categories in the data elements, thereby generating a corpus. Data elements in the negative category can be used as negative samples, and data elements in the positive category can be used as positive samples. Optionally, to address cognitive differences between visual concepts in datasets, a concept-relation corpus based on multimodal datasets is introduced. This concept-relation corpus unifies the semantic information of data elements and enriches visual concepts, providing important semantic support for multi-task learning, thereby improving the model's generalization and multi-task performance. For example, a concept-relationship corpus containing approximately 18 kilobytes (KB) of semantic information and a hierarchical relationship graph of this semantic information can be constructed. The constructed corpus can be used for unified and differentiated multi-task open vocabulary learning. In multi-task learning model training using multiple datasets, data elements in different datasets, such as visual perception elements, can have significant semantic differences, leading to inconsistent definitions of data element categories, which in turn affects model training effectiveness. To address this issue, this embodiment introduces a hierarchical, multi-granular concept-relationship corpus. By introducing this concept-relationship corpus, the type definitions of data elements in different datasets are unified, thereby guiding model training and achieving the technical effect of reducing the model's limitations in processing image data. This resolves the technical issue of the model's significant limitations in processing image data. As an optional implementation, constructing a corpus based on semantic information of different types of tasks includes: constructing a knowledge graph from the semantic information of different types of tasks and the relationships between the semantic information of different types of tasks, wherein the corpus includes the knowledge graph. For example, multimodal datasets corresponding to different types of tasks can be collected, including multiple data formats such as images, videos, and text descriptions. For example, a dataset for animal monitoring can be a dataset containing multimodal data such as animal images, animal names, and animal descriptions.Semantic information for different types of tasks, as well as the relationships between these semantic information, can be sampled from the dataset. This extracted semantic information and relationships can then be integrated into a knowledge graph. For example, an animal knowledge graph can be constructed, which includes information such as the names, feature descriptions, and genus-species relationships of different animals. This knowledge graph clearly displays the hierarchical relationships between different animals, such as those between mammals and birds, or between felines and canines, thereby generating a relation-concept corpus for a multimodal visual dataset. In this embodiment, a knowledge graph containing a rich set of visual concepts and their hierarchical relationships is constructed. Through a hierarchical, multi-granular concept-relationship corpus, visual perception elements across multiple datasets are unified, alleviating the problem of annotation discrepancies between datasets. As an optional implementation, the method may further include: determining, in the knowledge graph, multiple pieces of semantic information whose similarity between hierarchical relationships on the knowledge graph exceeds a first similarity threshold; and / or determining, in the knowledge graph, multiple pieces of semantic information whose semantic similarity is less than a second similarity threshold; and using the multiple pieces of semantic information as negative samples for alignment training of the visual model. In this embodiment, multiple pieces of semantic information whose similarity between hierarchical relationships on the knowledge graph exceeds a first similarity threshold, and / or multiple pieces of semantic information whose semantic similarity is less than a second similarity threshold, may be determined in the knowledge graph, and the multiple pieces of semantic information may be used as negative samples for alignment training of the visual model. The first and second similarity thresholds may be pre-set similarity thresholds. The multiple pieces of semantic information whose semantic similarity exceeds the first similarity threshold may be hierarchically similar semantic information. The multiple pieces of semantic information whose semantic similarity is less than the second similarity threshold may be semantically significant categories and may serve as negative samples. Optionally, when constructing the open vocabulary visual-linguistic alignment loss, online discriminative sampling (ODS) is proposed. By introducing hierarchically similar and potentially confusing categories and / or semantically distinct categories as negative samples from the constructed concept-relation corpus, this method maintains training balance and generalization. The concepts of positive and negative samples encompass various elements in the alignment training process, improving the model's discriminability and robustness.Optionally, online discriminative sampling introduces a selective negative sample strategy. This strategy dynamically selects semantic information with a similar conceptual hierarchy to the positive sample (i.e., the similarity between the hierarchical relationships on the knowledge graph exceeds a first similarity threshold) and / or with significant semantic differences (i.e., the semantic similarity is less than a second similarity threshold) as negative samples in training. Because these negative samples are more similar to the positive samples, they are more likely to confuse the model. Therefore, by introducing negative samples with similar hierarchical relationships, the model's ability to distinguish between positive and negative samples can be enhanced, thereby improving the model's discriminability and robustness. For example, when constructing the vision-language alignment loss function, an online discriminative sampling method can be introduced to identify multiple pieces of semantic information with hierarchical relationships on the knowledge graph exceeding a first similarity threshold and / or multiple pieces of semantic information within the knowledge graph with semantic similarity less than a second similarity threshold. These multiple pieces of semantic information are used as negative samples, thereby achieving the goal of dynamically selecting appropriate negative samples. By utilizing highly discriminative negative samples, training balance can be maintained and the model's generalization ability can be enhanced. As an optional implementation, step S306 combines the semantic information samples of the type task and the semantic information samples of the associated tasks to form a generated task description information sample that matches the text data sample. This includes: invoking a generation template that matches the visual model to fuse the semantic information samples of the type task and the semantic information samples of the associated tasks to obtain a task description information sample that matches the text data sample. The generation template indicates the format that the task description information sample meets. In this embodiment, the generation template can be a pre-set fixed-format task description template or a unified template instruction that can be used to indicate the task to be processed by the visual model. The task description information sample can be a mandatory instruction similar to a visual positioning description that can be used to indicate the type of task to be processed by the model. Due to the inherent task domain gaps in vision tasks, existing frameworks often experience significant performance degradation when training multiple tasks simultaneously. To address the inherent task gaps in multi-task joint training, this embodiment introduces the concept of task-conditional joint training. This uses a concept-relation corpus to develop unified instruction-conditional training. Using a fixed-format task description (i.e., a generation template) as an instruction, semantic information samples corresponding to text sample data and semantic information samples of associated tasks are fused to generate fixed-format task description information samples. This task description information sample guides the vision model in processing image sample data, thereby improving the model's efficiency in processing the task description information samples and alleviating the task processing gap. The unified instruction condition can be a task description, which can be used to instruct the model on how to specifically handle the task.Optionally, this embodiment proposes the generation of strong directive instructions that combine visual-linguistic descriptions. By prepending the task name (i.e., the semantic information sample corresponding to the text sample data) to the description and incorporating multi-granularity concepts (i.e., the semantic information of the associated task), the gaps between different visual tasks are effectively bridged. As an optional implementation, step S308 utilizes the task description information samples and image data samples to train a visual model. This includes: determining the task description information samples and image data samples as target training samples corresponding to the associated task; and training the visual model using the target training samples and initial training samples corresponding to the associated task. The initial training samples include initial task description information samples and initial image data samples. The initial task description information samples are used to describe the semantic information sample of the associated task and the semantic information sample of the type task. The text data samples corresponding to the initial image data samples are used to record the associated task. The visual model is configured to perform the associated task on the input image data. In this embodiment, the task description information samples and image data samples can be determined as target training samples corresponding to the associated task. A visual model is trained using target training samples and initial training samples corresponding to associated tasks. The target training samples may include available visual elements and semantic information. The initial training samples may include initial task description information samples and initial image data samples. The initial task description information samples may be configured to include semantic information samples describing the associated task and semantic information samples of the target task. The text data samples corresponding to the initial image data samples may be configured to indicate the associated task to be performed on the initial image data samples. The visual model is configured to perform the associated task on the input image data. Optionally, initial training samples corresponding to the associated task are obtained. The initial training samples may include initial task description information samples and initial image data samples. A generation template matching the visual model may be invoked to fuse the semantic information samples corresponding to the text sample data with the semantic information samples of the associated task to obtain a task description information sample. The task description information sample and the image data sample are determined as target training samples corresponding to the associated task. The visual model is trained using the target training samples and the initial training samples corresponding to the associated task. Optionally, the cross-task collaborative learning of visual elements and semantics proposed in this embodiment implements a dynamic collaborative learning mechanism. Under this mechanism, a visual model can be trained using target training samples and initial training samples corresponding to associated tasks. This allows the visual model to share and learn visual elements and semantic information across different tasks, enhancing the model's performance in multi-task learning scenarios. The visual elements can refer to images, and the semantic information can be the categories of objects in the images. These categories can be unified into the corpus.As an optional implementation, the method may further include: obtaining text feature samples from text data samples and image feature samples from image data samples; concatenating the text feature samples and image feature samples to obtain concatenated feature samples; and performing contrastive learning training on the visual model using features of semantic information samples from different types of tasks in a corpus and the concatenated feature samples. In this embodiment, during model training of the visual model, text feature samples from text data samples and image feature samples from image data samples may be obtained. The text feature samples and image feature samples may be concatenated to obtain concatenated feature samples. Simultaneously, features of semantic information of data elements may be obtained from the corpus. Contrastive learning training of the visual model may be performed using features of semantic information samples from different types of tasks and the concatenated feature samples. oIn an embodiment of the present disclosure, a corpus is constructed, wherein the corpus includes at least semantic information samples of different types of tasks and associations between the semantic information samples of different types of tasks. Within the corpus, semantic information samples of associated tasks are retrieved based on the semantic information samples and associations of the type of tasks. The type of tasks are recorded in text data samples, and there is an association between the semantic information samples of the type of tasks and the semantic information samples of the associated tasks. The semantic information samples of the type of tasks and the semantic information samples of the associated tasks are combined to form task description information samples that match the text data samples. The task description information samples are used to describe the semantic information samples of the type of tasks and the semantic information samples of the associated tasks. A visual model is trained using the image data samples corresponding to the task description information samples and the text data samples. The visual model is used to perform the type of tasks on input image data, thereby achieving the technical effect of reducing the limitations of the model in processing image data and resolving the technical problem of the large limitations of the model in processing image data. The present disclosure also provides another method for processing image data. FIG4 is a flowchart of another method for processing image data according to an embodiment of the present disclosure. As shown in FIG4 , the method may include the following steps. Step S402: In response to an input operation on the operation interface, the image data to be processed is displayed on the operation interface. In the technical solution provided in step S402 of the present disclosure, the user can trigger the input operation on the operation interface by clicking a control, manually operating the interface, or operating the mouse. The visual model responds to the input operation on the operation interface and can display the image data to be processed on the operation interface. The operation interface can be an interface of a device such as a mobile terminal or a computer. This is merely an example and does not limit the type of operation interface. Step S404: Text data corresponding to the image data is displayed on the operation interface. The text data is used to record at least one type of target task. In the technical solution provided in step S404 of the present disclosure, text data corresponding to the image data is obtained. The image data and the text data corresponding to the image data can be displayed on the operation interface. The text data can be used to record at least one type of target task. Step S406: In response to the task execution operation performed on the operation interface, the task execution result corresponding to the image data is displayed on the operation interface, where the task execution result is obtained by guiding the visual model to perform the target task on the image data using the task description information of the target task. The task description information is used to describe the semantic information of the target task and the semantic information of the associated tasks. There is an association relationship between the semantic information of the target task and the semantic information of the associated tasks.In the technical solution provided in step S406 of the present disclosure, task description information is generated based on text data, and the task description information and image data are input into a visual model. The task description information is used to guide the visual model to perform a target task on the image data to obtain a task execution result. In response to a task execution operation on an operation interface, the task execution result corresponding to the image data can be displayed on the operation interface. In an embodiment of the present disclosure, in response to an input operation on the operation interface, image data to be processed is displayed on the operation interface; text data corresponding to the image data is displayed on the operation interface, wherein the text data describes at least one type of target task; and in response to the task execution operation on the operation interface, the task execution result corresponding to the image data is displayed on the operation interface. The task execution result is obtained by guiding the visual model to perform the target task on the image data using the task description information of the target task. The task description information describes the semantic information of the target task and the semantic information of associated tasks, and there is an association between the semantic information of the target task and the semantic information of the associated tasks. This achieves the technical effect of reducing the limitations of the model in processing image data and solves the technical problem of the large limitations of the model in processing image data. According to an embodiment of the present disclosure, an embodiment of an image data processing system is also provided. It should be noted that the image data processing system of this embodiment can be used to execute the image data processing method of the embodiment of the present disclosure. Figure 5 is a schematic diagram of an image data processing system according to an embodiment of the present disclosure. As shown in Figure 5, image data processing system 500 may include a data input terminal 502, a task execution terminal 504, and a result output terminal 506. Data input terminal 502 is configured to monitor image data to be processed. In this embodiment, the data input terminal can monitor image data to be processed. The data input terminal can be a mobile terminal, a computer, or other device. It should be noted that this is for illustrative purposes only and does not impose any specific limitation on the type of data input terminal. The task execution terminal 504 is configured to obtain text data corresponding to the image data, wherein the text data is used to record at least one type of target task; obtain task description information of the target task, wherein the task description information is used to describe the semantic information of the target task and the semantic information of the associated tasks, and there is an association relationship between the semantic information of the target task and the semantic information of the associated tasks; input the task description information and the image data into the visual model, and use the task description information to guide the visual model to execute the target task on the image data to obtain a task execution result.In this embodiment, the task execution terminal can obtain text data corresponding to the image data and process the text data using the semantic information of the data elements recorded in the corpus and the associations between data elements in different data sets to generate task description information. Based on the task description information, the semantic information of the target task and the semantic information of related tasks can be determined. The task description information and the image data can be input into a visual model, which can be guided to execute the target model on the image data based on the task description information to obtain a task execution result. The task execution terminal can be a terminal that includes the visual model. The result output terminal 506 is configured to output the task execution result. In this embodiment, the task execution result can be output using the result output terminal. The result output terminal can be the same device terminal as the data input terminal or a different device terminal. In this embodiment, image data to be processed is monitored via a data input terminal 502. Text data corresponding to the image data is obtained via a task execution terminal 504, where the text data describes at least one type of target task. Task description information for the target task is obtained, where the task description information describes the semantic information of the target task and the semantic information of associated tasks, with the semantic information of the target task and the semantic information of the associated tasks being associated with each other. The task description information and image data are input into a visual model, and the task description information is used to guide the visual model to perform the target task on the image data, obtaining a task execution result. The task execution result is output via a result output terminal 506, thereby achieving the technical effect of reducing the limitations of the model in processing image data and resolving the technical problem of the model's large limitations in processing image data. The development of various granular visual tasks in the field of computer vision indicates that building an open-world, unified, and interactive general visual model will become an indispensable trend in the journey towards general artificial intelligence. However, due to the time-consuming and labor-intensive labeling process in building visual models, the scope of existing visual understanding tasks remains very limited, far smaller than the vocabulary humans use to describe the real world. This, coupled with the limitations of learning objectives, hinders the scalability of the models, hindering their widespread adoption and preventing them from handling richer and more general semantics. Furthermore, traditional task-specific visual models are trained using data from specific tasks and are unable to develop models for new tasks.Related technologies primarily focus on universality and openness. For example, new object detection frameworks like Pix2Seq and fine art painting software like Painter simultaneously address multiple visual understanding tasks within a unified framework. Alternatively, efforts like open-vocabulary segmentation and open-vocabulary object detection attempt to imbue task-specific models with open-vocabulary capabilities. However, these tools rarely achieve a "universal open vocabulary" within a comprehensive and universal framework. As an alternative, partial unification is achieved within similar tasks. For example, the Language-Image Pretraining (GLIP) model transforms object detection into a visual localization problem, achieving a certain degree of universality in the open vocabulary of detection and localization. While related technologies have, to a certain extent, unified semantic, instance, and panoptic segmentation tasks using open-vocabulary capabilities, these approaches require task-specific data for model training, resulting in significant limitations in the way visual models process image data. To address the above technical issues, this embodiment considers the cognitive differences between visual concepts in datasets, the inherent task domain gaps in multi-task joint training, and the structural differences between different tasks due to varying output granularity. A universal, unified framework for open visual understanding is proposed. This framework integrates the three segmentation tasks of open vocabulary segmentation, localization, and detection, thereby addressing the technical issue of the significant limitations of visual models in processing image data and achieving the technical effect of reducing these limitations. The following further describes the universal, unified framework for open visual understanding proposed in this embodiment. As an optional embodiment, to address the cognitive differences between visual concepts in a dataset, a concept-relationship corpus based on a multimodal dataset is introduced. This concept-relationship corpus unifies the semantic information of data elements and enriches visual concepts, providing important semantic support for multi-task learning, thereby improving the model's generalization and multi-task performance. For example, a concept-relationship corpus containing approximately 18 kilobytes of semantic information and a hierarchical relationship graph of this semantic information can be constructed. This constructed corpus can be used to unify and differentiate open vocabulary learning across multiple tasks. Alternatively, in multi-task learning model training on multiple datasets, data elements (which may be visual perception elements) in different datasets may have significant semantic differences, resulting in inconsistent definitions of data element categories, which in turn affects the effectiveness of model training.To address this issue, this embodiment introduces a hierarchical, multi-granular concept-relation corpus. This concept-relation corpus unifies the type definitions of data elements in different datasets, thereby guiding model training. Optionally, different datasets corresponding to different tasks are obtained, where these datasets may include open vocabularies for segmentation and / or monitoring. Hierarchical relationships between these concepts can be constructed using a knowledge graph, and online discriminative sampling can be performed on the semantic information of the data elements recorded in the different datasets to obtain the semantic information of the data elements recorded in the different datasets. A corpus is constructed using this semantic information and hierarchical relationships. For example, different datasets corresponding to different tasks can be obtained to obtain an original vocabulary containing a large number of data elements: "window, wall, cat, table, sofa, person, potted plant, plant." Online discriminative sampling is performed on the semantic information of the original vocabulary to determine the negative and positive categories within the data elements, thereby obtaining a corpus. Data elements in the negative category can serve as negative samples, and data elements in the positive category can serve as positive samples. Optionally, the relationships between concepts may include hyponymy, partiality, negative relationships, positive relationships, hierarchical relationships, and so on. These are merely examples and do not impose specific limitations on the types and methods of determining relationships between concepts. For example, a multimodal dataset containing multiple data formats, such as images, videos, and text descriptions, may be collected for different tasks. For example, an animal monitoring dataset may include multimodal data such as animal images, names, and descriptions. Semantic information of the data elements and the associations between the semantic information of the data elements recorded in different datasets can be sampled and integrated into a knowledge graph. For example, an animal knowledge graph can be constructed, which includes information such as the names, characteristic descriptions, and genus-species relationships of different animals. By establishing a knowledge graph, the hierarchical relationships between different animals can be clearly displayed, such as the classification of mammals and birds, or the relationship between cats and canines, thereby generating a relationship-concept corpus for the multimodal visual dataset. In this embodiment, a knowledge graph containing rich visual concepts and their hierarchical relationships is constructed. Through a hierarchical multi-granularity concept-relationship corpus, the visual perception elements in multiple datasets are unified to alleviate the problem of annotation differences between different datasets.As an optional embodiment, due to the inherent task domain gaps in visual tasks, existing frameworks often suffer significant performance degradation when training multiple tasks simultaneously. Therefore, to address the inherent task gaps in multi-task joint training, this embodiment introduces the concept of task-conditional joint training. This concept utilizes a concept-relation corpus to develop unified instruction-conditional training. Using a fixed-format task description (i.e., a generation template) as an instruction, the semantic information samples corresponding to the text sample data and the semantic information samples of the associated task are fused to generate task description information samples. The task description information samples are then used to guide the visual model in processing the image sample data, thereby alleviating the task processing gap. The unified instruction condition can be a task description, which can be used to provide the model with specific instructions on how to handle the task. Alternatively, this embodiment proposes strong instruction generation that combines visual-linguistic descriptions. By prepending the task name (i.e., the semantic information samples corresponding to the text sample data) to the description, and incorporating multi-granularity concepts (i.e., semantic information of the associated task), the gap between different visual tasks is effectively bridged. For another example, assume that the generated template is in the form of "[task name], [concept name]." The task name can be used to represent the target task, such as segmentation, localization, or monitoring. The concept name can be used to represent the target of the operation or processing to be performed, such as cars, trees, or animals. Based on the generated template, the semantic information samples corresponding to the text sample data and the semantic information samples of the associated tasks are fused together, combining the concepts and relationship information in the concept-relationship corpus, to generate directive task description information samples. For example, for the segmentation task, the instruction "[segmentation], [car]" can be generated to guide the visual model in performing car segmentation. The generated task description information samples can also be used in model training as instructive information to guide visual model training, helping the model to better understand different tasks and achieve joint training of multiple tasks. In this embodiment, strong directive instructions combining visual and language descriptions are generated. Through these strong directive instructions, the model can be informed of the task for each training session, thereby achieving the purpose of joint training task conditions. By prepending the task name in the description and incorporating the concept of multi-granularity, the gap between different vision tasks is effectively bridged. As an optional embodiment, vision-language alignment and online discriminative sampling are proposed. By combining unified instructions with task query generation, foreground and background objects are decoupled to enhance the performance of multi-granularity outputs, thereby addressing the structural differences between tasks caused by varying output granularity. In this embodiment, online discriminative sampling is proposed when constructing an open-vocabulary vision-language alignment loss.By introducing hierarchically similar and potentially confusing categories from the constructed concept-relationship corpus as negative samples, and / or categories with significant semantic differences as negative samples, training balance and generalization are maintained. The concepts of positive and negative samples encompass various elements in the alignment training process. Optionally, online discriminative sampling introduces a selective negative sample strategy. This dynamically selects semantic information that is similar in concept hierarchy to the positive samples (i.e., the similarity between the hierarchical relationships on the knowledge graph exceeds a first similarity threshold) and / or significantly different in semantics (i.e., the semantic similarity is less than a second similarity threshold) as negative samples in training, thereby improving the model's discriminability and robustness. Optionally, because negative samples are more similar to positive samples and are more likely to confuse the model, introducing negative samples of similar hierarchical level can enhance the model's ability to distinguish between positive and negative samples. For example, when constructing the vision-language alignment loss function, an online discriminative sampling method can be introduced to identify multiple semantic information with hierarchical relationships on the knowledge graph greater than a first similarity threshold, and / or, within the knowledge graph, multiple semantic information with semantic similarity less than a second similarity threshold. These multiple semantic information are used as negative samples, thereby dynamically selecting appropriate negative samples. By utilizing highly discriminative negative samples, training balance can be maintained and the model's generalization ability can be enhanced. As an optional embodiment, a dynamic cross-task concept collaboration method is also proposed, which enables collaborative learning of visual elements and semantics across tasks. This method can dynamically adapt and learn from available visual elements and semantic information during training. In this embodiment, mask embeddings and region embeddings can be extracted for the input segmentation image, localization image, and detection image, and these pseudo-predictions can be incorporated into vision-language alignment training. This cross-task concept collaboration stimulates and enhances overall open vocabulary learning capabilities. The segmentation image, localization image, and detection image can be the same image but assigned to different tasks. Optionally, the cross-task collaborative learning of visual elements and semantics proposed in this embodiment implements a dynamic collaborative learning mechanism. Under this mechanism, a visual model can be trained using target training samples and initial training samples corresponding to associated tasks. This allows the visual model to share and learn visual elements and semantic information across different tasks, enhancing the model's performance in multi-task learning scenarios. The visual elements can refer to images, and the semantic information can be the categories of objects in the images. These categories can be unified into the corpus.For example, FIG6 is a schematic diagram of a general open visual understanding framework according to an embodiment of the present disclosure. As shown in FIG6 , suppose we wish to train a general visual model that can simultaneously perform the three tasks of semantic segmentation, instance segmentation, and panoptic segmentation. Visual concepts from different datasets can be collected to obtain a raw vocabulary 601. Online discriminative sampling can be performed on the raw vocabulary to determine the relationships between the raw vocabulary, forming a concept-relationship corpus 602. A unified instruction condition 603 can be designed for each task, such as "semantic segmentation - person." Concepts in the concept-relationship corpus 602 can be used to enrich the task description and obtain mandatory instructions, such as "semantic segmentation - person - motorcycle." During the training process, images 604 to be semantically segmented, images 605 to be instance segmented, and images 606 to be panoptically segmented can be obtained. Image encoding is performed on images 604, 605, and 606 to obtain image features 610. At the same time, task text 609 is obtained, tokenized, and segmented to accurately obtain text features corresponding to the text data. The text features and image features are concatenated to obtain concatenated features 613. Concatenated features 613 can be input into a decoder in the visual model. The task description information is used to guide the decoder to perform decoding operations corresponding to the target task on the image features, thereby obtaining mask embeddings 611 and region embeddings 612. The decoding operations may include scene decoding 607 and object decoding 608. Scene decoding 607 and object decoding 608 represent the semantic information of the image and the object instances in the image, respectively. Optionally, based on mask embeddings 611 and region embeddings 612, types with similar conceptual hierarchies or significant semantic differences to the positive samples are selected as negative samples. For example, when training "semantic segmentation - person," "semantic segmentation - dog" is selected as a negative sample. Optionally, when constructing the open vocabulary visual-linguistic alignment loss, online discriminative sampling (ODS) is used to introduce hierarchically similar and potentially confusing categories from the constructed concept-relation corpus as positive samples, and / or categories with significant semantic differences as negative samples. As shown in FIG6 , negative samples 614, 615, 616, 617, 618, 619, and 620 are introduced to maintain training balance and generalization. The concepts of positive and negative samples encompass various elements in the alignment training process. It should be noted that the number of positive and negative samples introduced here is for illustrative purposes only and is not a specific limitation.Optionally, as shown in Figure 6 , mask embeddings 611 and region embeddings 612 corresponding to images 604, 605, and 606 are obtained, respectively. These pseudo-predictions are then incorporated into vision-language alignment training to align positive and negative samples. This cross-task conceptual collaboration stimulates and enhances overall open vocabulary learning capabilities. This dynamic cross-task conceptual collaboration approach enables collaborative learning of visual elements and semantics across tasks, leveraging available visual elements and semantic information during training. Optionally, visual elements are extracted for different tasks, such as mask embeddings and region embeddings, and the extracted information can be integrated into vision-language alignment to promote collaborative learning across different tasks. Through the above steps, the model can learn general capabilities across the three tasks of semantic segmentation, instance segmentation, and object detection, while also exhibiting relatively good generalization performance, thereby achieving a truly universal visual model. As shown in Figure 6 , the unified instruction condition 603 is processed based on the concept-relation corpus 602 to obtain a highly directive instruction. By encoding mandatory instructions through a text encoder, segmentation positive features 621, monitoring positive features 622, positioning positive features 623, and discriminant positive features 624 can be determined within the mandatory instructions. During visual model training, text feature samples from the text sample data and image feature samples from the image data samples can be obtained. These text feature samples and image feature samples are concatenated to obtain concatenated feature samples. Simultaneously, features of the semantic information of the data elements are obtained from the corpus. The features of the semantic information of the data elements and the concatenated feature samples can be used to perform comparative learning training on the visual model. In this embodiment, by constructing a hierarchical, multi-granular concept-relation corpus, the semantic differences of visual perception elements across different datasets are effectively addressed, providing a unified, fine-grained, and hierarchical target understanding for multi-task learning. Leveraging this constructed concept-relation corpus, the new concept of task-conditional joint training is introduced, generating highly prescriptive instructions to guide the model to better understand and align the learning process and learning time of different tasks, significantly bridging the gap between tasks. Furthermore, through online discriminative sampling (ODS) and a cross-task concept collaboration strategy, appropriate concepts can be dynamically selected for alignment, thereby introducing more discriminative negative samples during training and enhancing the model's robustness and generalization capabilities. Through the above, this embodiment significantly improves the model's performance in open vocabulary scenarios and enhances its processing capabilities for multi-task visual understanding, thereby achieving the technical effect of reducing the model's limitations in image data processing and resolving the technical issue of the model's significant limitations in image data processing.The method embodiments provided in the above embodiments of the present disclosure can also be executed in a mobile terminal, a computer terminal, or a similar computing device. FIG7 is a hardware block diagram of a computer terminal (or mobile device) according to an image data processing method of an embodiment of the present disclosure. As shown in FIG7 , the computer terminal 70 (or mobile device) may include one or more processors 702 (illustrated as 702a, 702b, ..., 702n in the figure) (the processor 702 may include, but is not limited to, a microprocessor (MCU) or a field programmable gate array (FPGA)), a memory 704 configured to store data, and a transmission device 706 configured to perform communication functions. In addition, the computer terminal 700 may include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS), a network interface, a power supply, and / or a camera. Those skilled in the art will appreciate that the structure shown in FIG7 is merely illustrative and does not limit the structure of the electronic device described above. For example, the computer terminal 70 may include more or fewer components than those shown in FIG7 , or have a configuration different from that shown in FIG7 . The hardware structure block diagram shown in FIG7 can serve not only as an exemplary block diagram of the computer terminal 70 (or mobile device) described above, but also as an exemplary block diagram of the server described above. In an alternative embodiment, FIG2 illustrates, as a block diagram, an embodiment using the computer terminal 70 (or mobile device) shown in FIG7 as a computing node in the computing environment 201 . Memory 704 can be configured to store software programs and components of application software, such as the program instructions / data storage device corresponding to the data processing method in the embodiments of the present disclosure. The processor executes the software programs and components stored in memory 704 to execute various functional applications and data processing, thereby implementing the data processing method described above. Memory 704 can include high-speed random access memory (RAM) and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 704 may further include memories remotely located relative to the processor, and these remote memories may be connected to the computer terminal 70 via a network. Examples of the aforementioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.Transmission device 706 is configured to receive or transmit data via a network. A specific example of such a network may include a wireless network provided by the communications provider of computer terminal 70. In one example, transmission device 706 includes a network interface controller (NIC), which can connect to other network devices via a base station to enable communication with the Internet. In one example, transmission device 706 may be a radio frequency (RF) component configured to communicate with the Internet wirelessly. The display may be, for example, a touch-screen liquid crystal display (LCD), which enables a user to interact with the user interface of computer terminal 70 (or mobile device). It should be noted that all subject information (including but not limited to subject device information, subject personal information, etc.) and data (including but not limited to data configured for analysis, storage, and display, etc.) referred to in this disclosure are authorized by the subject or fully authorized by all parties. The collection, use, and processing of such data must comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation portals are provided for the subject to choose to authorize or deny. It should be noted that, for simplicity of description, the aforementioned method embodiments are described as a series of combined actions. However, those skilled in the art should be aware that the present disclosure is not limited by the order of the actions described, as certain steps may be performed in a different order or simultaneously, according to the present disclosure. Furthermore, those skilled in the art should also be aware that the embodiments described in this specification are preferred embodiments, and the actions and components involved are not necessarily required for the present disclosure. Through the above description of the embodiments, those skilled in the art will clearly understand that the methods according to the aforementioned embodiments can be implemented using software and a necessary general-purpose hardware platform, or alternatively, hardware. Based on this understanding, the technical solution of the present disclosure, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (e.g., ROM / RAM, a magnetic disk, or an optical disk) and includes instructions for enabling a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of the various embodiments of the present disclosure. According to embodiments of the present disclosure, an image data processing device configured to implement the image data processing method shown in FIG. 2 is also provided.FIG8 is a schematic diagram of an image data processing device according to an embodiment of the present disclosure. As shown in FIG8 , the image data processing device 800 may include a first monitoring component 802, a first acquisition component 804, a second acquisition component 806, and a first processing component 808. The first monitoring component 802 is configured to monitor image data to be processed. The first acquisition component 804 is configured to acquire text data corresponding to the image data, wherein the text data is configured to record at least one type of target task. The second acquisition component 806 is configured to acquire task description information for the target task, wherein the task description information is configured to describe semantic information of the target task and semantic information of associated tasks, wherein the semantic information of the target task and the semantic information of the associated tasks are associated with each other. The first processing component 808 is configured to input the task description information and image data into a visual model and use the task description information to guide the visual model to execute the target task on the image data to obtain a task execution result. Here, the first monitoring component 802, first acquisition component 804, second acquisition component 806, and first processing component 808 correspond to steps S202 to S208 in the above embodiment. The four components implement the same examples and application scenarios as the corresponding steps, but are not limited to the contents disclosed in the above embodiment. It should be noted that the above components may be hardware components or software components stored in a memory (e.g., memory 704) and processed by one or more processors (e.g., processors 702a, 702b, ..., 702n). The above components may also be part of an apparatus and run in the computer terminal 70 provided in Example 3. According to an embodiment of the present disclosure, a model generation device configured to implement the model generation method shown in FIG. 3 is also provided. FIG. 9 is a schematic diagram of a model generation device according to an embodiment of the present disclosure. As shown in FIG. 9, the model generation device 900 may include a query component 902, a generation component 904, and a training component 906. The query component 902 is configured to query the corpus for semantic information samples of associated tasks based on semantic information samples and association relationships of the type tasks. The corpus includes at least semantic information samples of different type tasks and association relationships between the semantic information samples of different type tasks. The type tasks are recorded in the text data samples, and there is an association relationship between the semantic information samples of the type tasks and the semantic information samples of the associated tasks.The generation component 904 is configured to combine semantic information samples of the type task with semantic information samples of the associated tasks to form a task description information sample that matches the text data sample. The task description information sample is configured to describe the semantic information sample of the type task and the semantic information sample of the associated tasks. The training component 906 is configured to train a visual model using the image data samples corresponding to the task description information sample and the text data sample. The visual model is configured to perform the type task on the input image data. It should be noted that the query component 902, the generation component 904, and the training component 906 correspond to steps S302 to S306 in the above embodiment. The examples and application scenarios implemented by these three components and the corresponding steps are the same, but are not limited to the contents disclosed in the above embodiment. It should be noted that the above components can be hardware components or software components stored in a memory (e.g., memory 704) and processed by one or more processors (e.g., processors 702a, 702b, ..., 702n). The above components can also be part of an apparatus and run in the computer terminal 70 provided in Example 3. According to an embodiment of the present disclosure, another image data processing device configured to implement the image data processing method shown in FIG. 4 is also provided. FIG. 10 is a schematic diagram of another image data processing device according to an embodiment of the present disclosure. As shown in FIG. 10 , the image data processing device 1000 may include a first display component 1002, a second display component 1004, and a third display component 1006. The first display component 1002 is configured to display image data to be processed on the operation interface in response to an input operation set as an action on the operation interface. The second display component 1004 is configured to display text data corresponding to the image data on the operation interface, wherein the text data is configured to record at least one type of target task. The third display component 1006 is configured to display a task execution result corresponding to the image data on the operation interface in response to a task execution operation set as an action on the operation interface. The task execution result is obtained by guiding a visual model to perform the target task on the image data using task description information of the target task. The task description information is configured to describe semantic information of the target task and semantic information of associated tasks, and there is an association between the semantic information of the target task and the semantic information of the associated tasks. It should be noted that the first display component 1002, the second display component 1004 and the third display component 1006 correspond to steps S402 to S406 in the above embodiment. The three components and the corresponding steps implement the same instances and application scenarios, but are not limited to the contents disclosed in the above embodiment.It should be noted that the above-mentioned components may be hardware components or software components stored in a memory (e.g., memory 704) and processed by one or more processors (e.g., processors 702a, 702b, ..., 702n). The above-mentioned components may also be part of a device and run in the computer terminal 70 provided in Example 3. In the image data processing device of this embodiment, text data corresponding to the image data is obtained, and task description information corresponding to the input image data can be generated. This task description information is used to guide the visual model to better understand the learning process of different tasks, thereby better performing the target task on the image data. This provides a unified, fine-grained, and hierarchical target understanding for multi-task learning, thereby achieving the technical effect of reducing the limitations of the model in processing image data and resolving the technical problem of the large limitations of the model in processing image data. The embodiments of the present disclosure may provide a computer terminal, which may be any computer terminal device in a computer terminal group. Optionally, in this embodiment, the above-mentioned computer terminal may be replaced with a terminal device such as a mobile terminal. Optionally, in this embodiment, the computer terminal may be located in at least one of multiple network devices in a computer network. In this embodiment, the computer terminal may execute program code for the following steps in the image data processing method: monitoring image data to be processed; obtaining text data corresponding to the image data, wherein the text data describes at least one type of target task; obtaining task description information for the target task, wherein the task description information describes semantic information of the target task and semantic information of associated tasks, with the semantic information of the target task and the semantic information of the associated tasks being associated with each other; inputting the task description information and the image data into a visual model, and using the task description information to guide the visual model to execute the target task on the image data, thereby obtaining a task execution result. Optionally, FIG11 is a block diagram of the structure of a computer terminal according to an embodiment of the present disclosure. As shown in FIG11 , the computer terminal A may include one or more (only one is shown) processors 1102, a memory 1104, and a transmission device 1106. The memory may be configured to store software programs and components, such as the program instructions / components corresponding to the image data processing method and apparatus in the embodiments of the present disclosure. The processor executes the software programs and components stored in the memory to perform various functional applications and data processing, thereby implementing the aforementioned image data processing method. The memory may include high-speed random access memory (RAM) and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory.In some examples, the memory may further include memory remotely located relative to the processor, and these remote memories may be connected to the computer terminal A via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. The processor may access information and applications stored in the memory via a transmission device to perform the following steps: monitoring image data to be processed; obtaining text data corresponding to the image data, wherein the text data is used to describe at least one type of target task; obtaining task description information for the target task, wherein the task description information is used to describe semantic information of the target task and semantic information of associated tasks, wherein the semantic information of the target task and the semantic information of the associated tasks are associated with each other; inputting the task description information and the image data into a visual model, and using the task description information to guide the visual model to execute the target task on the image data, thereby obtaining a task execution result. Optionally, the processor may further execute program code for the following steps: identifying at least one type of target task from the text data corresponding to the image data; searching a corpus for semantic information of associated tasks based on the semantic information and association relationships of the target task, wherein the corpus includes at least semantic information of different types of tasks and association relationships between the semantic information of different types of tasks; combining the semantic information of the target task with the semantic information of the associated tasks to form task description information that matches the text data. Optionally, the processor may further execute program code for the following steps: if a task of a type matching the type of the target task is identified in the corpus, using the semantic information of the matched task as the semantic information of the target task; determining an association relationship in the corpus corresponding to the semantic information of the target task; and searching the corpus based on the association relationship based on the semantic information of the target task to obtain semantic information of associated tasks associated with the target task. Optionally, the processor may further execute program code of the following steps: obtaining text features of text data and image features of image data; concatenating the text features and the image features to obtain concatenated features; determining, from features of semantic information of different types of tasks, a target feature that matches the concatenated features; and determining, in a corpus, a type of task corresponding to the target feature as a type of task that matches the type of the target task.Optionally, the processor may further execute program code for the following steps: obtaining text features of the text data and image features of the image data; concatenating the text features and image features to obtain concatenated features; and if the different data sets include the text data set to which the text data belongs, querying the corpus for semantic information of the target task, including: if the features of the semantic information of the data elements include concatenated features, searching the corpus for semantic information of the target task. Optionally, the processor may further execute program code for the following steps: invoking an encoder in the visual model to extract image features from the image data. Optionally, the processor may further execute program code for the following steps: invoking a generation template that matches the visual model to fuse the semantic information of the target task with the semantic information of related tasks to obtain task description information, wherein the generation template is used to indicate the format satisfied by the task description information. Optionally, the processor may further execute program code for the following steps: inputting the task description information and the image features of the image data into a decoder in the visual model; and using the task description information to guide the decoder to perform decoding operations corresponding to the target task on the image features to obtain a task execution result. Optionally, the processor may further execute program code for the following steps: updating model parameters of a visual model using task execution results. The processor may call information and an application stored in a memory via a transmission device to perform the following steps: querying a corpus for semantic information samples of associated tasks based on semantic information samples and association relationships of task types, wherein the corpus includes at least semantic information samples of different task types and association relationships between semantic information samples of different task types, wherein task types are recorded in text data samples, and association relationships exist between semantic information samples of task types and semantic information samples of associated tasks; combining the semantic information samples of task types and semantic information samples of associated tasks to form task description information samples that match the text data samples, wherein the task description information samples are used to describe the semantic information samples of the task types and the semantic information samples of the associated tasks; and training a visual model using image data samples corresponding to the task description information samples and the text data samples, wherein the visual model is used to execute the task types on input image data. Optionally, the processor may further execute program code of the following steps: sampling semantic information of text data samples corresponding to different types of tasks respectively; determining the semantic information of the sampled text data samples as semantic information of the corresponding type of task; and constructing a corpus based on the semantic information of the different types of tasks.Optionally, the processor may further execute program code for the following steps: constructing a knowledge graph from semantic information of different types of tasks and the associations between semantic information of different types of tasks, wherein the corpus includes the knowledge graph. Optionally, the processor may further execute program code for the following steps: determining, in the knowledge graph, multiple pieces of semantic information whose hierarchical relationships on the knowledge graph have similarities greater than a first similarity threshold; and / or determining, in the knowledge graph, multiple pieces of semantic information whose semantic similarity is less than a second similarity threshold; and using the multiple pieces of semantic information as negative samples to perform alignment training on the visual model. Optionally, the processor may further execute program code for the following steps: invoking a generation template that matches the visual model, fusing semantic information samples of the type task and semantic information samples of the associated tasks to obtain task description information samples that match the text data sample, wherein the generation template is used to indicate the format satisfied by the task description information samples. Optionally, the processor may further execute program code for the following steps: determining the task description information samples and the image data samples as target training samples corresponding to the associated task; training a visual model using the target training samples and the initial training samples corresponding to the associated task, wherein the initial training samples include initial task description information samples and initial image data samples, the initial task description information samples being used to describe the semantic information samples of the associated task and the semantic information samples of the type task; the text data samples corresponding to the initial image data samples being used to record the associated task; and the visual model being used to perform the associated task on the input image data. Optionally, the processor may further execute program code for the following steps: obtaining text feature samples of the text data samples and image feature samples of the image data samples; concatenating the text feature samples and the image feature samples to obtain concatenated feature samples; and performing comparative learning training on the visual model using features of semantic information samples of different types of tasks in the corpus and the concatenated feature samples. The processor can call information and applications stored in the memory through a transmission device to perform the following steps: in response to an input operation on an operation interface, display image data to be processed on the operation interface; display text data corresponding to the image data on the operation interface, wherein the text data is used to record at least one type of target task; in response to a task execution operation on the operation interface, display a task execution result corresponding to the image data on the operation interface, wherein the task execution result is obtained by using task description information of the target task to guide a visual model to perform the target task on the image data, and the task description information is used to describe semantic information of the target task and semantic information of associated tasks, and there is an association relationship between the semantic information of the target task and the semantic information of the associated tasks.Using the embodiments of the present disclosure, a hierarchical, multi-granular corpus is constructed. Using this corpus, task description information corresponding to the input image data can be generated to guide the visual model to better understand the learning process of different tasks, thereby better performing tasks on image data. This provides a unified, fine-grained, and hierarchical understanding of the objectives for multi-task learning, thereby achieving the technical effect of reducing the limitations of the model in processing image data and resolving the technical problem of the model's significant limitations in processing image data. Those skilled in the art will appreciate that the structure shown in FIG11 is merely illustrative, and computer terminal A may also be a smartphone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, or a mobile internet device (MID), PAD, or other terminal device. FIG11 does not limit the structure of the aforementioned computer terminal A. For example, computer terminal A may include more or fewer components (such as a network interface, a display device, etc.) than those shown in FIG11 , or may have a configuration different from that shown in FIG11 . Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments can be completed by a program instructing the hardware associated with the terminal device. The program can be stored in a computer-readable storage medium, which may include a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. The embodiments of the present disclosure also provide a computer-readable storage medium. Optionally, in this embodiment, the computer-readable storage medium can be configured to store program code executed by the image data processing method provided in the first embodiment. Optionally, in this embodiment, the computer-readable storage medium can be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.Optionally, in this embodiment, the computer-readable storage medium is configured to store program code configured to perform the following steps: monitoring image data to be processed; obtaining text data corresponding to the image data, wherein the text data is used to record at least one type of target task; obtaining task description information for the target task, wherein the task description information is used to describe semantic information of the target task and semantic information of associated tasks, and there is an association relationship between the semantic information of the target task and the semantic information of the associated tasks; inputting the task description information and the image data into a visual model, and using the task description information to guide the visual model to perform the target task on the image data to obtain a task execution result. Optionally, the computer-readable storage medium may also be configured to perform the following program code: identifying at least one type of target task from the text data corresponding to the image data; querying a corpus for semantic information of associated tasks based on the semantic information and association relationships of the target task, wherein the corpus includes at least semantic information of different types of tasks and association relationships between the semantic information of different types of tasks; and combining the semantic information of the target task with the semantic information of the associated tasks to form task description information that matches the text data. Optionally, the computer-readable storage medium may further include program code for executing the following steps: if a type task matching the target task's type is identified in the corpus, using the semantic information of the matching type task as the semantic information of the target task; determining, in the corpus, an association relationship corresponding to the semantic information of the target task; and, based on the semantic information of the target task, performing a query in the corpus according to the association relationship to obtain semantic information of associated tasks associated with the target task. Optionally, the computer-readable storage medium may further include program code for executing the following steps: obtaining text features of text data and image features of image data; concatenating the text features and image features to obtain concatenated features; determining, from features of semantic information of different type tasks, a target feature that matches the concatenated feature; and determining, in the corpus, the type task corresponding to the target feature as the type task matching the target task's type. Optionally, the computer-readable storage medium may further execute program code for the following steps: obtaining text features of text data and image features of image data; splicing the text features and the image features to obtain splicing features; if the different data sets include a text data set to which the text data belongs, searching the corpus for semantic information of the target task, including: if the features of the semantic information of the data element include the splicing features, searching the corpus for semantic information of the target task.Optionally, the computer-readable storage medium may further execute program code for the following steps: invoking an encoder in the visual model to extract image features from the image data. Optionally, the computer-readable storage medium may further execute program code for the following steps: invoking a generation template that matches the visual model to fuse the semantic information of the target task with the semantic information of associated tasks to obtain task description information, wherein the generation template is used to indicate the format that the task description information satisfies. Optionally, the computer-readable storage medium may further execute program code for the following steps: inputting the task description information and image features of the image data into a decoder in the visual model; using the task description information to guide the decoder to perform decoding operations corresponding to the target task on the image features to obtain task execution results. Optionally, the computer-readable storage medium may further execute program code for the following steps: updating model parameters of the visual model using the task execution results. As an optional example, a computer-readable storage medium is configured to store program code configured to execute the following steps: querying semantic information samples of associated tasks in a corpus based on semantic information samples and association relationships of type tasks, wherein the corpus includes at least semantic information samples of different types of tasks and association relationships between semantic information samples of different types of tasks, type tasks are recorded in text data samples, and there is an association relationship between the semantic information samples of the type tasks and the semantic information samples of the associated tasks; combining the semantic information samples of the type tasks and the semantic information samples of the associated tasks to form a task description information sample that matches the text data sample, wherein the task description information sample is used to describe the semantic information sample of the type task and the semantic information sample of the associated task; and training a visual model using the task description information samples and the image data samples corresponding to the text data samples, wherein the visual model is used to perform the type task on the input image data. Optionally, the computer-readable storage medium may further include program code for executing the following steps: sampling semantic information from text data samples corresponding to different types of tasks; determining the semantic information of the sampled text data samples as semantic information for the corresponding types of tasks; and constructing a corpus based on the semantic information for the different types of tasks. Optionally, the computer-readable storage medium may further include program code for executing the following steps: constructing a knowledge graph from the semantic information for the different types of tasks, as well as the associations between the semantic information for the different types of tasks, wherein the corpus includes the knowledge graph.Optionally, the computer-readable storage medium may further execute program code for the following steps: calling a generation template that matches the visual model, fusing the semantic information samples of the type task and the semantic information samples of the associated task, and obtaining a task description information sample that matches the text data sample. The generation template is used to indicate the format satisfied by the task description information sample. Optionally, the computer-readable storage medium may further execute program code for the following steps: determining the task description information sample and the image data sample as target training samples corresponding to the associated task; and training a visual model using the target training samples and initial training samples corresponding to the associated task. The initial training samples include initial task description information samples and initial image data samples. The initial task description information samples are used to describe the semantic information samples of the associated task and the semantic information samples of the type task. The text data samples corresponding to the initial image data samples are used to record the associated task. The visual model is used to perform the associated task on the input image data. Optionally, the computer-readable storage medium may further execute program code for the following steps: obtaining text feature samples from text data samples and image feature samples from image data samples; concatenating the text feature samples and image feature samples to obtain concatenated feature samples; and performing comparative learning and training on a visual model using features of semantic information samples from different types of tasks in a corpus and the concatenated feature samples. Optionally, the computer-readable storage medium may further execute program code for the following steps: obtaining text feature samples from text sample data and image feature samples from image data samples; concatenating the text feature samples and image feature samples to obtain concatenated feature samples; and performing comparative learning and training on a visual model using features of semantic information of data elements and the concatenated feature samples. As an optional example, a computer-readable storage medium is configured to store program code configured to execute the following steps: in response to an input operation acting on an operation interface, displaying image data to be processed on the operation interface; displaying text data corresponding to the image data on the operation interface, wherein the text data is used to record at least one type of target task; in response to a task execution operation acting on the operation interface, displaying a task execution result corresponding to the image data on the operation interface, wherein the task execution result is obtained by using task description information of the target task to guide a visual model to perform the target task on the image data, and the task description information is used to describe the semantic information of the target task and the semantic information of associated tasks, and there is an association relationship between the semantic information of the target task and the semantic information of the associated tasks.In embodiments of the present disclosure, a hierarchical, multi-granular corpus is constructed. Using this corpus, task description information corresponding to input image data can be generated to guide the visual model to better understand the learning process of different tasks, thereby better performing tasks on image data. This provides a unified, fine-grained, and hierarchical understanding of the objectives for multi-task learning, thereby achieving the technical effect of reducing the limitations of the model in processing image data and resolving the technical problem of the large limitations of the model in processing image data. Embodiments of the present disclosure may provide an electronic device that may include a memory and a processor. Figure 12 is a block diagram of an electronic device for an image data processing method according to an embodiment of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptops, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components, their connections and relationships, and their functions shown herein are provided as examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein. As shown in FIG12 , device 1200 includes a computing component 1201, which can perform various appropriate actions and processes based on computer programs stored in a read-only memory (ROM) 1202 or loaded from a storage component 1208 into a random access memory (RAM) 1203. RAM 1203 may also store various programs and data required for the operation of device 1200. Computing component 1201, ROM 1202, and RAM 1203 are interconnected via a bus 1204. An input / output (I / O) interface 1205 is also connected to bus 1204. Multiple components in device 1200 are connected to I / O interface 1205, including: input component 1206, such as a keyboard and mouse; output component 1204, such as various types of displays and speakers; storage component 1208, such as a magnetic disk and optical disk; and communication component 1209, such as a network card, modem, or wireless communication transceiver. Communication component 1209 allows device 1200 to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunication networks. Computing component 1201 can be any general-purpose or specialized processing component with processing and computing capabilities.Some examples of computing component 1201 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing components that run machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. Computing component 1201 performs the various methods and processes described above, such as the data verification method. For example, in some embodiments, the data verification method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage component 1208. In some embodiments, part or all of the computer program can be loaded and / or installed on device 1200 via ROM 1202 and / or communication component 1209. When the computer program is loaded into RAM 1203 and executed by the computing component 1201, one or more steps of the data verification method described above may be performed. Alternatively, in other embodiments, the computing component 1201 may be configured to perform the data verification method in any other appropriate manner (e.g., via firmware). According to an embodiment of the present disclosure, a method for processing image data is provided. It should be noted that the steps shown in the flowcharts of the accompanying drawings may be executed in a computer system, such as a set of computer-executable instructions. Furthermore, although the flowcharts illustrate a logical sequence, in some cases, the steps shown or described may be executed in a different order than that illustrated or described.Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-a-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device. Program code configured to implement the methods of the present disclosure can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that, when executed by the processor or controller, the program code implements the functions / operations specified in the flowcharts and / or block diagrams. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server. In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by, or in conjunction with, an instruction execution system, device, or apparatus. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing.More specific examples of machine-readable storage media would include electrical connections based on one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. To provide for user interaction, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display, monitor) configured to display information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be configured to provide for user interaction; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input). The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an embodiment of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected via any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet. A computer system may include a client and a server. The client and server are generally remote from each other and typically interact via a communication network. The client-server relationship is established by computer programs running on the respective computers and establishing a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain. Embodiments of the present disclosure also provide a computer program product comprising computer instructions that, when executed by a processor, implement the database product status detection method provided in embodiments of the present disclosure.In this embodiment, the aforementioned computer instructions may be stored in a read-only memory (ROM) or loaded from a storage component into a random access memory (RAM), enabling a processor to execute various appropriate actions and processes in the database product status detection method. In some embodiments, some or all of the aforementioned computer instructions may be loaded and / or installed on an electronic device via a read-only memory and / or a communication component. When the computer instructions are loaded into the random access memory and executed by the computing component, one or more steps in the database product status detection method described above may be performed. It should be noted that the serial numbers of the aforementioned embodiments of the present disclosure are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the aforementioned embodiments of the present disclosure, the descriptions of each embodiment have their respective responsibilities. For portions not detailed in one embodiment, reference should be made to the relevant descriptions of other embodiments. In the several embodiments provided in this disclosure, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of components represents only one logical functional division. In actual implementation, different divisions may be employed. For example, multiple components or components may be combined or integrated into another system, or some features may be omitted or not implemented. Furthermore, the coupling, direct coupling, or communication connection shown or discussed may be through interfaces, or the indirect coupling or communication connection between components or components may be electrical or otherwise. Components described as separate parts may or may not be physically separate, and components shown as components may or may not be physical components, i.e., they may be located in one location or distributed across multiple network components. Some or all of these components may be selected to achieve the objectives of the present embodiments as needed. Furthermore, the functional components in the various embodiments of the present disclosure may be integrated into a single processing component, each component may exist physically separately, or two or more components may be integrated into a single component. These integrated components may be implemented in either hardware or software functional components. If the integrated components are implemented as software functional components and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of the present disclosure, or the portion that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes instructions for causing a computer device (such as a personal computer, server, or network device) to execute all or part of the steps of the various embodiments of the present disclosure.The aforementioned storage media include various media capable of storing program code, such as USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard drives, magnetic disks, or optical disks. The above are merely preferred embodiments of the present disclosure. It should be noted that those skilled in the art may make various improvements and modifications without departing from the principles of the present disclosure, and such improvements and modifications should be considered within the scope of protection of the present disclosure. Industrial Applicability The solution provided by the embodiments of the present disclosure can be applied to the image processing process to monitor image data to be processed; obtain text data corresponding to the image data, wherein the text data is used to record at least one type of target task; obtain task description information of the target task, wherein the task description information is used to describe the semantic information of the target task and the semantic information of associated tasks, and there is an association relationship between the semantic information of the target task and the semantic information of the associated tasks; input the task description information and image data into a visual model, and use the task description information to guide the visual model to perform the target task on the image data to obtain the task execution result, thereby solving the technical problem that the model has large limitations in processing image data.

Claims

34 Claims 1. A method for processing image data, comprising: monitoring image data to be processed; Obtain text data corresponding to the image data, wherein the text data is used to record at least one type of target task; obtain task description information of the target task, wherein the task description information is used to describe semantic information of the target task and semantic information of associated tasks, and there is an association relationship between the semantic information of the target task and the semantic information of the associated tasks; input the task description information and the image data into a visual model, and use the task description information to guide the visual model to execute the target task on the image data to obtain a task execution result.

2. The method according to claim 1, wherein: Obtaining task description information of the target task includes: identifying at least one type of target task from text data corresponding to the image data; querying, in a corpus, semantic information of the associated task based on the semantic information of the target task and the association relationship, wherein the corpus includes at least semantic information of different types of tasks and the association relationship between the semantic information of the different types of tasks; and combining the semantic information of the target task with the semantic information of the associated task to form the task description information that matches the text data.

3. The method according to claim 2, wherein: In a corpus, based on the semantic information of the target task and the association relationship, the semantic information of the associated task is queried, including: if a type task matching the type of the target task is identified in the corpus, the semantic information of the matched type task is used as the semantic information of the target task; in the corpus, the association relationship corresponding to the semantic information of the target task is determined; based on the semantic information of the target task, the semantic information of the associated task associated with the target task is searched in the corpus according to the association relationship to obtain the semantic information of the associated task.

4. The method according to claim 3, wherein: The method further includes: acquiring text features of the text data and image features of the image data; concatenating the text features and the image features to obtain concatenated features; determining a target feature that matches the concatenated features among the features of the semantic information of the different types of tasks; and determining, in the corpus, a type task corresponding to the target feature as a type task that matches the type of the target task.

5. The method according to claim 4, wherein: Acquiring image features of the image data includes: calling an encoder in the visual model to extract the image features from the image data. 35 6. The method according to claim 2, wherein: Combining the semantic information of the target task and the semantic information of the associated tasks to form the task description information that matches the text data includes: calling a generation template that matches the visual model, fusing the semantic information of the target task and the semantic information of the associated tasks to obtain the task description information, wherein the generation template is used to represent the format satisfied by the task description information.

7. The method according to claim 1, wherein: Inputting the task description information and the image data into a visual model, and using the task description information to guide the visual model to perform the target task on the image data to obtain a task execution result, including: inputting the task description information and image features of the image data into a decoder in the visual model; using the task description information to guide the decoder to perform a decoding operation corresponding to the target task on the image features to obtain the task execution result.

8. The method according to any one of claims 1 to 7, wherein: The method further includes: updating model parameters of the visual model using the task execution result.

9. A method for generating a model, comprising: In a corpus, based on semantic information samples and association relationships of type tasks, semantic information samples of associated tasks are queried, wherein the corpus includes at least semantic information samples of different types of tasks and the association relationships between the semantic information samples of the different types of tasks, the type tasks are recorded in text data samples, and the association relationships exist between the semantic information samples of the type tasks and the semantic information samples of the associated tasks; the semantic information samples of the type tasks and the semantic information samples of the associated tasks are combined to form task description information samples that match the text data samples, wherein the task description information samples are used to describe the semantic information samples of the type tasks and the semantic information samples of the associated tasks; a visual model is trained using the task description information samples and image data samples corresponding to the text data samples, wherein the visual model is used to perform the type tasks on input image data.

10. The method according to claim 9, wherein: The method further includes: sampling semantic information of text data samples corresponding to the different types of tasks respectively; determining the semantic information of the sampled text data samples as semantic information of the corresponding types of tasks; and constructing the corpus based on the semantic information of the different types of tasks.

11. The method according to claim 10, wherein: Constructing the corpus based on the semantic information of the different types of tasks includes: constructing the semantic information of the different types of tasks and the association relationship between the semantic information of the different types of tasks into a knowledge graph, wherein the corpus includes the knowledge graph.

12. The method according to claim 11, wherein: The method further comprises: In the knowledge graph, multiple semantic information having similarities between hierarchical relationships on the knowledge graph greater than a first similarity threshold is determined; and / or, in the knowledge graph, multiple semantic information having semantic similarities less than a second similarity threshold is determined; and the multiple semantic information are used as negative samples to perform alignment training on the visual model.

13. The method according to claim 9, wherein: The semantic information sample of the type task and the semantic information sample of the associated task are combined to form a generated task description information sample that matches the text data sample, including: calling a generation template that matches the visual model, fusing the semantic information sample of the type task and the semantic information sample of the associated task to obtain the task description information sample that matches the text data sample, wherein the generation template is used to represent a format satisfied by the task description information sample.

14. The method according to claim 9, wherein: The visual model is trained using the task description information samples and the image data samples, including: determining the task description information samples and the image data samples as target training samples corresponding to the associated task; and training the visual model using the target training samples and initial training samples corresponding to the associated task, wherein the initial training samples include initial task description information samples and initial image data samples, the initial task description information samples are used to describe the semantic information samples of the associated task and the semantic information samples of the type task, the text data samples corresponding to the initial image data samples are used to record the associated task, and the visual model is used to perform the associated task on the input image data.

15. The method according to claim 9, wherein: The method further includes: obtaining text feature samples of the text data sample and image feature samples of the image data sample; splicing the text feature samples and the image feature samples to obtain spliced ​​feature samples; and performing comparative learning training on the visual model using features of semantic information samples of different types of tasks in the corpus and the spliced ​​feature samples.

16. A method for processing image data, comprising: In response to an input operation on the operation interface, displaying image data to be processed on the operation interface; Displaying text data corresponding to the image data on the operation interface, wherein the text data is used to record at least one type of target task; in response to a task execution operation performed on the operation interface, displaying a task execution result corresponding to the image data on the operation interface, wherein the task execution result is obtained by guiding a visual model to execute the target task on the image data using task description information of the target task, wherein the task description information is used to describe semantic information of the target task and semantic information of associated tasks, and the target task There is an association relationship between the semantic information of the service and the semantic information of the associated task.

17. A system for processing image data, comprising: A data input terminal is configured to monitor image data to be processed; A task execution end is configured to obtain text data corresponding to the image data, wherein the text data is used to record at least one type of target task; obtain task description information of the target task, wherein the task description information is used to describe the semantic information of the target task and the semantic information of the associated tasks, and there is an association relationship between the semantic information of the target task and the semantic information of the associated tasks; input the task description information and the image data into a visual model, and use the task description information to guide the visual model to execute the target task on the image data to obtain a task execution result; a result output end is configured to output the task execution result.

18. An electronic device, comprising: memory and processor; The memory is configured to store computer-executable instructions, and the processor is configured to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 16 are implemented.

19. A computer-readable storage medium comprising a stored program, wherein: When the program is running, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 16.

20. A computer program product, comprising computer instructions, wherein when the computer instructions are executed by a processor, the method according to any one of claims 1 to 16 is implemented.

Images