A data production method, device, equipment, storage medium and product

Abstract

The application discloses a data production method, device, equipment, storage medium and product, and the method comprises the following steps: acquiring source image data and initial labeling information of the source image data; processing and relabeling the source image data to obtain updated labeling information; fusing the initial labeling information and the updated labeling information to obtain fused labeling information; inputting the fused labeling information into a target data production model to obtain a data production result for the source image data. According to the data production method, the source image data is processed and relabeled, then the updated labeling information obtained after relabeling is fused with the initial labeling information of the source image data, and the fused labeling information is taken as the input of the target data production model, so that the quality of data production can be improved.

Description

Technical Field

[0001] This invention relates to the field of data production technology, and in particular to a data production method, apparatus, equipment, storage medium, and product. Background Technology

[0002] Data production refers to the process of generating more relevant data in batches based on acquired raw data. In related technologies, raw data is generally obtained through open-source datasets, internet data, and collected data. However, data obtained through these methods is generally of poor quality and cannot meet the training requirements of data production models, resulting in the inability to achieve high-quality data production even after the model's capabilities are improved. Summary of the Invention

[0003] This invention provides a data production method, apparatus, equipment, storage medium, and product to achieve high-quality data production.

[0004] According to one aspect of the present invention, a data production method is provided, comprising:

[0005] Obtain the source image data and its initial annotation information;

[0006] The source image data is processed and re-annotated to obtain updated annotation information;

[0007] The initial annotation information and the updated annotation information are fused to obtain fused annotation information. The fused annotation information is then input into the target data production model to obtain the data production result for the source image data.

[0008] Furthermore, the source image data is processed and re-annotated to obtain updated annotation information, including:

[0009] The source image data is filtered and classified to obtain the first target image;

[0010] The first target image is fully annotated to obtain full annotation information;

[0011] The first target image is annotated with specific regions to obtain specific region annotation information, and the full image annotation information and the specific region annotation information are used as the updated annotation information.

[0012] Further, the source image data is filtered and classified to obtain the first target image, including:

[0013] Extract the data from the source image data that conforms to the set filtering rules to obtain the filtered result data;

[0014] The filtered result data is categorized using a target classification model, and the categorized filtered result data is used as the first target image.

[0015] Further, specific region annotation is performed on the first target image to obtain specific region annotation information, including:

[0016] Extract the target visual information from the first target image, and filter the first target image according to the target visual information to obtain the second target image;

[0017] Extract the salient region of the target from the second target image, and generate annotation information for the salient region of the target as the specific region annotation information.

[0018] Further, target visual information is extracted from the first target image, and the first target image is filtered based on the target visual information to obtain a second target image, including:

[0019] For each data point in the first target image, extract at least two dimensions of target visual information;

[0020] Cross-validation and filtering are performed on the target visual information corresponding to at least two dimensions for each data in the first target image, and the first target image that matches the target visual information in each dimension is retained. The retained first target image is used as the second target image.

[0021] Furthermore, the initial annotation information and the updated annotation information are fused to obtain fused annotation information, including:

[0022] Obtain the target fusion model and model hints;

[0023] The model prompt information, the initial annotation information, the updated annotation information, and the source image data are input into the target fusion model to obtain the fusion annotation information; wherein, the model prompt information is used to instruct the target fusion model to generate the fusion annotation information based on the initial annotation information, the updated annotation information, and the source image data.

[0024] According to another aspect of the present invention, a data production apparatus is provided, comprising:

[0025] The source image data and initial annotation information acquisition module is used to acquire source image data and the initial annotation information of the source image data;

[0026] The updated annotation information determination module is used to process and re-annotate the source image data to obtain updated annotation information;

[0027] The data production module is used to fuse the initial annotation information with the updated annotation information to obtain fused annotation information, and input the fused annotation information into the target data production model to obtain the data production result for the source image data.

[0028] Optionally, the update annotation information determination module is also used for:

[0029] The source image data is filtered and classified to obtain the first target image;

[0030] The first target image is fully annotated to obtain full annotation information;

[0031] The first target image is annotated with specific regions to obtain specific region annotation information, and the full image annotation information and the specific region annotation information are used as the updated annotation information.

[0032] Optionally, the update annotation information determination module is also used for:

[0033] Extract the data from the source image data that conforms to the set filtering rules to obtain the filtered result data;

[0034] The filtered result data is categorized using a target classification model, and the categorized filtered result data is used as the first target image.

[0035] Optionally, the update annotation information determination module is also used for:

[0036] Extract the target visual information from the first target image, and filter the first target image according to the target visual information to obtain the second target image;

[0037] Extract the salient region of the target from the second target image, and generate annotation information for the salient region of the target as the specific region annotation information.

[0038] Optionally, the update annotation information determination module is also used for:

[0039] For each data point in the first target image, extract at least two dimensions of target visual information;

[0040] Cross-validation and filtering are performed on the target visual information corresponding to at least two dimensions for each data in the first target image, and the first target image that matches the target visual information in each dimension is retained. The retained first target image is used as the second target image.

[0041] Optionally, the data production module is also used for:

[0042] Obtain the target fusion model and model hints;

[0043] The model prompt information, the initial annotation information, the updated annotation information, and the source image data are input into the target fusion model to obtain the fusion annotation information; wherein, the model prompt information is used to instruct the target fusion model to generate the fusion annotation information based on the initial annotation information, the updated annotation information, and the source image data.

[0044] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0045] At least one processor; and

[0046] A memory communicatively connected to the at least one processor; wherein,

[0047] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the data production method according to any embodiment of the present invention.

[0048] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the data production method according to any embodiment of the present invention.

[0049] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program / instructions, which, when executed by a processor, implement the steps of the data production method described in any embodiment of the present invention.

[0050] The data production method disclosed in this invention first obtains source image data and its initial annotation information; then, it processes and re-annotates the source image data to obtain updated annotation information; finally, it fuses the initial annotation information with the updated annotation information to obtain fused annotation information, which is then input into a target data production model to obtain the data production result for the source image data. This data production method improves the quality of source image data through processing and re-annotating it, and then fuses the updated annotation information with the initial annotation information, using the fused annotation information as input to the target data production model, ultimately improving the quality of data production.

[0051] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of a data production method provided according to Embodiment 1 of the present invention;

[0054] Figure 2 This is a flowchart of a data production method provided according to Embodiment 2 of the present invention;

[0055] Figure 3 This is a schematic diagram of the structure of a data production device according to Embodiment 3 of the present invention;

[0056] Figure 4 This is a schematic diagram of the structure of the electronic device that implements Embodiment 4 of the present invention. Detailed Implementation

[0057] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0058] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0059] Example 1

[0060] Figure 1This is a flowchart of a data production method provided in Embodiment 1 of the present invention. This embodiment is applicable to the situation of batch data production based on raw data. The method can be executed by a data production device, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes:

[0061] S110. Obtain the source image data and the initial annotation information of the source image data.

[0062] The source image data consists of raw data obtained through open-source datasets, internet data, and collected data. The initial annotation information consists of descriptions, annotations, and labels included in the source image data.

[0063] In this embodiment, source image data can be obtained by accessing open-source datasets or internet data, or by data collection and other methods. Based on the specific content of the source image data, if there are descriptions, annotations, labels, or other data in the source image data, the descriptions, annotations, labels, or other data are extracted from the source image data as the initial annotation information of the source image data.

[0064] S120. Process and re-annotate the source image data to obtain updated annotation information.

[0065] Among them, the updated annotation information is the new annotation information obtained after re-annotating the source image data.

[0066] In this embodiment, since the source image data may contain noise, the initial annotation information may be inaccurate. Therefore, it is necessary to process and re-annotate the source image data in order to obtain more accurate updated annotation information.

[0067] Optionally, the methods for processing the source image data may include, but are not limited to, image cleaning and filtering. For example, based on set filtering rules, data that does not conform to the rules in the source image data can be removed, while data that conforms to the rules can be retained. Preferably, the set filtering rules may include, but are not limited to: the length and width of the source image data must be greater than 256 pixels; the text length of the initial annotation information in the source image data should be no less than 2; the source image data should not contain watermarks; and the source image data should not contain unsafe images involving violence or politics. In addition, a set algorithm can be used to remove duplicate data from the source image data.

[0068] Furthermore, after processing the source image data, new annotation information can be generated through a re-annotation operation, i.e., updating the annotation information. The re-annotation operation can be to annotate the entire source image data, treating each image as a whole; alternatively, it can extract specific regions from each image and annotate salient detection boxes within those regions. These salient detection boxes can be detection boxes corresponding to a specific object or feature in the source image data. For example, if the specific object in the source image data is a dog, then the salient detection boxes could be detection boxes specifically for that dog.

[0069] S130. The initial annotation information and the updated annotation information are fused to obtain fused annotation information. The fused annotation information is then input into the target data production model to obtain the data production result for the source image data.

[0070] Here, the fused annotation information is the annotation information obtained by fusing the initial annotation information and the updated annotation information of the source image data. The target data production model is a model that produces data, whose input information is image annotation information and whose output information is the data production result. For example, the target data production model can be a large language model such as GPT-4 (fourth-generation generative pre-trained Transformer model), or a multimodal model based on the Transformer architecture.

[0071] In this embodiment, after generating updated annotation information, it can be fused with the initial annotation information of the source image data to generate higher-quality fused annotation information. The fused annotation information can be used as input to the target data production model, thereby obtaining the data production result using the target data production model.

[0072] Optionally, the fusion of initial and updated annotation information can be achieved using a specific information fusion model. The initial and updated annotation information can be input into the model, and prompt words can be set for the model to output high-quality fused annotation information.

[0073] The data production method disclosed in this invention first acquires source image data and its initial annotation information; then, it processes and re-annotates the source image data to obtain updated annotation information; finally, it fuses the initial annotation information and the updated annotation information to obtain fused annotation information, which is then input into a target data production model to obtain the data production result for the source image data. This data production method, by processing and re-annotating the source image data, and then fusing the updated annotation information with the initial annotation information of the source image data, uses the fused annotation information as input to the target data production model. This provides the target data production model with richer and more accurate information, thereby making the data production result from the target data production model more comprehensive and accurate, ultimately improving the quality of data production.

[0074] Example 2

[0075] Figure 2 This is a flowchart of a data production method provided in Embodiment 2 of the present invention. This embodiment is a refinement of the above embodiment, such as... Figure 2 As shown, the method includes:

[0076] S210. Obtain the source image data and the initial annotation information of the source image data.

[0077] The source image data consists of raw data obtained through open-source datasets, internet data, and collected data. The initial annotation information consists of descriptions, annotations, and labels included in the source image data.

[0078] In this embodiment, source image data and initial annotation information of source image data can be obtained by accessing open-source datasets or Internet data, or by data collection and other methods.

[0079] S220. Filter and classify the source image data to obtain the first target image.

[0080] The first target image is the result of filtering and classifying the source image data.

[0081] Optionally, the method for filtering and classifying the source image data to obtain the first target image can be: extracting data from the source image data that conforms to the set filtering rules to obtain the filtered result data; classifying the filtered result data into categories using a target classification model, and using the classified filtered result data as the first target image.

[0082] Specifically, setting filtering rules can include:

[0083] (1) Pixel filtering: Develop image pixel filtering rules according to requirements, such as both length and width need to be greater than 256 pixels, aspect ratio between 1 / 4 and 4, etc.

[0084] Furthermore, since the source image data comes from a wide variety of sources, it may contain many duplicate or highly similar images. Duplicate images can be removed by setting an algorithm. Preferably, the phash algorithm, or perceptual hash algorithm, can be used. This is an algorithm for generating image fingerprints and is widely used in image similarity detection.

[0085] (2) Text filtering: Develop text filtering rules according to requirements, such as the initial annotation text length should not be less than 2, etc.

[0086] Furthermore, since the initial annotation text of the source image data comes from a wide range of sources, the initial annotation text of the source image data can be extracted, and then duplicate text can be removed using a predefined algorithm. Preferably, the simhash algorithm, also known as the Similar Hash Algorithm, can be used. This algorithm can quickly estimate the similarity between two strings or documents.

[0087] (3) Model Filtering: One or more trained models can be used to detect source image data, obtain corresponding evaluation results, and filter the source image data based on the evaluation results. These models for detecting source image data may include, but are not limited to: watermark detection models, which can detect watermarks in images and return watermark probabilities; aesthetic scoring models, which can score the aesthetics of images and return aesthetic score values; Not Safe For Work (NSFW) detection, which can detect whether images contain content that is not suitable for display in the workplace or public environment, such as pornography, violence, or other content that violates community guidelines, and return NSFW probabilities; and safety detection models, which can detect whether images contain unsafe information such as violence or politics, and return safety probabilities.

[0088] Furthermore, after filtering the source image data to obtain the filtered result data, the filtered result data can be further categorized to obtain the first target image. Preferably, a target classification model can be used to classify all filtered result data into several major categories, such as animals, plants, food, vehicles, daily necessities, commodities, buildings, landscapes, animation, and people. By classifying the filtered result data into scenes, on the one hand, the distribution of the filtered result data can be intuitively seen (e.g., a higher proportion of animal data, a lower proportion of animation data, etc.), which facilitates adjusting the proportion of different data during training to improve model performance; on the other hand, relevant data can be quickly organized according to business needs (e.g., animal recognition data can be quickly extracted if needed).

[0089] In this embodiment, by filtering and classifying the source image data, noise in the source image data can be removed, preparing it for subsequent processing, thereby making the final data production result more accurate and effective.

[0090] S230. Perform full icon annotation on the first target image to obtain full icon annotation information.

[0091] Among them, full-image annotation refers to annotating each image in the source image data as a whole, and full-image annotation information is the annotation information of the source image data obtained through full-image annotation.

[0092] In this embodiment, after obtaining the first target image, each image in the source image data can be labeled as a whole to obtain full image annotation information.

[0093] Preferably, when performing full-image annotation, a multimodal large model can be used to perform the annotation task. A multimodal large model refers to a machine learning model capable of understanding and processing various types of data, including text, images, audio, and video. This type of model can integrate information from multiple modalities to perform more complex and intelligent tasks, such as visual question answering, image-text generation, speech recognition and synthesis, and video understanding and generation.

[0094] S240. Perform specific region annotation on the first target image to obtain specific region annotation information, and use the full image annotation information and specific region annotation information as updated annotation information.

[0095] Among them, specific region annotation is to extract specific regions from each image in the source image data, and to annotate the extracted specific regions. The specific region annotation information is the annotation information of the source image data obtained through specific region annotation.

[0096] In this embodiment, in addition to fully annotating the first target image, specific regions in the image can be extracted and annotated. Then, the full annotation information and the specific region annotation information are used as updated annotation information.

[0097] Optionally, the method for annotating specific regions of the first target image to obtain specific region annotation information can be: extracting target visual information from the first target image, filtering the first target image based on the target visual information to obtain a second target image; extracting the target salient region from the second target image, and generating annotation information for the target salient region as specific region annotation information.

[0098] Specifically, after filtering and classifying the source image data to obtain the first target image, its target visual information can be extracted, and then further filtered based on this visual information to obtain the second target image. For the second target image, salient regions can be extracted. These salient regions can be specific areas determined from the second target image through defined rules or detection algorithms. Based on these salient regions, they can be labeled to obtain specific region annotation information. Preferably, when labeling specific regions, a multimodal large model can be used to perform the annotation task.

[0099] Optionally, when annotating a specific region of the first target image to obtain specific region annotation information, the source image data can be filtered and classified first to obtain the first target image, and then the first target image can be annotated with a specific region to obtain specific region annotation information; alternatively, data that conforms to the set filtering rules in the source image data can be extracted first to obtain the filtered result data, and then the filtered result data can be classified by the target classification model. The filtered result data after classification is used as the first target image, and then the first target image can be annotated with a specific region to obtain specific region annotation information.

[0100] Furthermore, the method for extracting target visual information from the first target image and filtering the first target image based on the target visual information to obtain the second target image can be as follows: for each data in the first target image, extract target visual information in at least two dimensions; perform cross-validation and filtering on the target visual information in at least two dimensions corresponding to each data in the first target image, and retain the first target image whose target visual information in each dimension matches as the second target image.

[0101] Specifically, for each data point in the first target image, multiple visual expert models can be used to extract target visual information from multiple dimensions. Preferably, the extracted target visual information may include, but is not limited to, fine-grained labels, image classification labels, image detection results, image segmentation results, and Optical Character Recognition (OCR) information. Among these, fine-grained labels refer to labels that provide detailed annotations for different subclasses within the same broad category in image classification tasks; image classification labels are tags or keywords used to describe and classify image content; image detection results refer to the identification and location of target objects of interest in an image or video; image segmentation results are multiple meaningful regions or objects obtained by dividing an image; and OCR information is editable and searchable electronic text converted from text information in an image.

[0102] Optionally, to obtain the second target image, the source image data can be filtered and classified first to obtain the first target image. Then, for each data in the first target image, at least two dimensions of target visual information are extracted. The target visual information corresponding to each data in the first target image is then cross-validated and filtered. The first target image that matches the target visual information in each dimension is retained, and the retained first target image is used as the second target image.

[0103] Optionally, to obtain the second target image, data that conforms to the set filtering rules can be extracted from the source image data to obtain the filtered result data. Then, the filtered result data is classified by a target classification model. The filtered result data after classification is used as the first target image. Then, for each data in the first target image, at least two dimensions of target visual information are extracted. Then, cross-validation and filtering are performed on the target visual information corresponding to at least two dimensions of each data in the first target image. The first target image that matches the target visual information of each dimension is retained, and the retained first target image is used as the second target image.

[0104] Furthermore, after obtaining target visual information in multiple dimensions for each data point in the first target image, cross-validation can be performed using at least two dimensions of target visual information to remove data that cannot be matched, thereby further improving data quality. For example, for each image in the first target image, if the image detection result and image segmentation result for the same image deviate significantly, it indicates a problem with the detection or segmentation result and can be filtered out; or, if, compared with the image classification label and the image detection result, the image detection result corresponding to a certain label does not match the label information, it indicates that there is illusory information in that label and can be filtered out.

[0105] In this embodiment, by annotating the first target image with the second specific region, more useful information beyond the initial annotation information can be extracted from the source image data. This information can be used as updated annotation information to generate fused annotation information, which can provide more information for the target data production model, thus making the final data production result richer.

[0106] S250. Obtain the target fusion model and model prompt information. Input the model prompt information, initial annotation information, updated annotation information and source image data into the target fusion model to obtain fusion annotation information.

[0107] Among them, the model prompt information is used to instruct the target fusion model to generate fusion annotation information based on the initial annotation information, the updated annotation information and the source image data.

[0108] In this embodiment, the target fusion model is the model that generates fusion annotation information, and the model prompt information is instruction information for the target fusion model. The target fusion model can complete the task of generating fusion annotation information according to the instructions of the model prompt information. For example, the model prompt information can be:

[0109] "--You are a professional multimodal data expert, responsible for generating text descriptions for images, which need to be described in as much detail as possible;

[0110] In addition to the image, we will provide some external information to help you understand the image, including the original image description (initial annotation information) and updated annotation information (including object detection results, OCR results, attribute labels, etc.).

[0111] The original image description may contain a wealth of world knowledge related to the image, which needs to be reflected in the final image description, but it may also contain information unrelated to the image, which needs to be identified.

[0112] The object detection results include the object category and the corresponding detection bounding box information;

[0113] OCR results include the text content that appears on the image;

[0114] Attribute labels contain some category labels for the image.

[0115] --Please use this external information in your image description. Additionally, there may be errors in the external information, including missing or incorrect details. You can ignore these errors. Please note that external information such as bounding boxes are for reference only, and certain details such as bounding boxes should not appear in the final title as they are not common information in titles. If you do not use external information, please do not state why you did not use it.

[0116] --The specific external information is as follows:

[0117] "

[0118] Furthermore, the external information input to the target fusion model includes initial annotation information and updated annotation information, for example, it could be:

[0119] ##Original image description: {original caption}

[0120] ##Object detection results:

[0121] {OBJECT A}:[x1,y1,x2,y2]

[0122] {OBJECT B}:[x1,y1,x2,y2]

[0123] ##OCR:

[0124] {SENTENCE A}

[0125] {SENTENCE B}

[0126] ## Attribute tags: {tag1, tag2, ...}

[0127] "

[0128] Wherein, {original caption} represents the original image description, i.e., the initial annotation information; {OBJECT A}:[x1,y1,x2,y2] and {OBJECT B}:[x1,y1,x2,y2] represent the object detection results, where OBJECT A and OBJECT B are the specific objects detected, and x1,y1,x2,y2 are the specific features contained in the objects; SENTENCE A and SENTENCE B represent the OCR results; tag1, tag2,… represent attribute labels; the object detection results, OCR results, and attribute labels are the updated annotation information.

[0129] Under the guidance of the model prompts, the target fusion model can generate fused annotation information based on the input initial annotation information, updated annotation information and source image data.

[0130] Optionally, to obtain the fused annotation information, the source image data and its initial annotation information can be acquired first. Then, the source image data can be processed and re-annotated to obtain updated annotation information. Finally, the target fusion model and model hint information can be acquired. The model hint information, initial annotation information, updated annotation information, and source image data are then input into the target fusion model to obtain the fused annotation information. The model hint information is used to instruct the target fusion model to generate fused annotation information based on the initial annotation information, updated annotation information, and source image data.

[0131] Optionally, to obtain the fused annotation information, the source image data and its initial annotation information can be acquired first. Then, the source image data is filtered and classified to obtain the first target image. Next, the first target image is fully annotated to obtain full-image annotation information. Then, specific regions of the first target image are annotated to obtain specific-region annotation information. The full-image annotation information and the specific-region annotation information are then used as updated annotation information. Finally, the target fusion model and model hint information are acquired. The model hint information, initial annotation information, updated annotation information, and source image data are input into the target fusion model to obtain the fused annotation information. The model hint information is used to instruct the target fusion model to generate the fused annotation information based on the initial annotation information, updated annotation information, and source image data.

[0132] Optionally, to obtain the fused annotation information, the source image data and its initial annotation information can be acquired first. Then, data conforming to the set filtering rules in the source image data can be extracted to obtain the filtered result data. The filtered result data can then be categorized using a target classification model. The categorized filtered result data can be used as the first target image. The first target image can then be fully annotated to obtain full annotation information. Specific regions of the first target image can then be annotated to obtain specific region annotation information. The full annotation information and specific region annotation information can then be used as updated annotation information. Finally, the target fusion model and model hint information can be obtained. The model hint information, initial annotation information, updated annotation information, and source image data can be input into the target fusion model to obtain the fused annotation information. The model hint information is used to instruct the target fusion model to generate fused annotation information based on the initial annotation information, updated annotation information, and source image data.

[0133] Optionally, to obtain the fused annotation information, the source image data and its initial annotation information can be acquired first. Then, the source image data is filtered and classified to obtain a first target image. The first target image is then fully annotated to obtain full annotation information. Next, target visual information is extracted from the first target image. Based on this target visual information, the first target image is filtered to obtain a second target image. The salient regions of the target are then extracted from the second target image, and annotation information for these salient regions is generated as specific region annotation information. The full annotation information and specific region annotation information are then used as updated annotation information. Finally, the target fusion model and model hint information are acquired. The model hint information, initial annotation information, updated annotation information, and source image data are input into the target fusion model to obtain the fused annotation information. The model hint information is used to instruct the target fusion model to generate fused annotation information based on the initial annotation information, updated annotation information, and source image data.

[0134] Optionally, to obtain the fused annotation information, the source image data and its initial annotation information can be acquired first. Then, the source image data is filtered and classified to obtain a first target image. The first target image is then fully annotated to obtain full annotation information. For each data point in the first target image, at least two dimensions of target visual information are extracted. The target visual information corresponding to each data point in the first target image is then cross-validated and filtered, retaining the first target image whose target visual information matches across all dimensions. This retained first target image is used as the second target image. The salient regions of the target are then extracted from the second target image, and annotation information for these salient regions is generated as specific region annotation information. The full annotation information and specific region annotation information are then used as updated annotation information. Finally, the target fusion model and model hint information are obtained. The model hint information, initial annotation information, updated annotation information, and source image data are input into the target fusion model to obtain the fused annotation information. The model hint information is used to instruct the target fusion model to generate fused annotation information based on the initial annotation information, updated annotation information, and source image data.

[0135] In this embodiment, by using a target fusion model to fuse the initial annotation information, the updated annotation information, and the source image data, the resulting fused annotation information can include not only the initial annotation information from the source image data but also the newly generated updated annotation information. This makes the fused annotation information targeted while containing more information, providing more information for the target data production model, thereby making the final data production result richer in content.

[0136] S260. Input the fused annotation information into the target data production model to obtain the data production result for the source image data.

[0137] In this embodiment, after obtaining the fused annotation information, it can be input into the target data production model, thereby using the target data production model to obtain the data production result.

[0138] Furthermore, in order to control production costs and improve data production efficiency, a small batch of data can be produced first using a data production model, and then a larger batch of data can be produced using a lower-cost model.

[0139] For example, consider an image containing people, background, and text. The initial annotation for this source image data is "Welcome everyone to the McNamara Alumni Center's 'Game Day Party' carnival." The data production result obtained through the data production method provided in this embodiment is: "The image is a promotional image for the McNamara Alumni Center's 'Game Day Party.' The top of the image displays a stylized illustration of the McNamara Alumni Center, adorned with musical notes and the bold text 'Game Day Party.' Below the image are two people, both wearing maroon sweatshirts with 'MINNESOTA' written in yellow. One of them is holding a beer bottle. The background is a clear blue sky and a group of people, suggesting an outdoor event. The text at the bottom reads 'Open 2 hours prior to all home football games' and prompts 'learn more.'" The data production result obtained through the data production method disclosed in this invention contains much more information.

[0140] The data production method disclosed in this invention can improve the quality of source image data by processing and re-annotating the source image data. Then, the updated annotation information obtained after re-annotation is fused with the initial annotation information of the source image data, and the fused annotation information is used as the input of the target data production model, which can ultimately improve the quality of data production.

[0141] Example 3

[0142] Figure 3 This is a schematic diagram of the structure of a data production device provided in Embodiment 3 of the present invention, as shown below. Figure 3 As shown, the device includes: a source image data and initial annotation information acquisition module 310, an updated annotation information determination module 320, and a data production module 330.

[0143] The source image data and initial annotation information acquisition module 310 is used to acquire source image data and initial annotation information of source image data.

[0144] The updated annotation information determination module 320 is used to process and re-annotate the source image data to obtain updated annotation information.

[0145] The data production module 330 is used to fuse the initial annotation information with the updated annotation information to obtain fused annotation information. The fused annotation information is then input into the target data production model to obtain the data production result for the source image data.

[0146] Optionally, the update annotation information determination module 320 is also used for:

[0147] The source image data is filtered and classified to obtain the first target image; the first target image is fully annotated to obtain full annotation information; specific regions of the first target image are annotated to obtain specific region annotation information, and the full annotation information and specific region annotation information are used as the updated annotation information.

[0148] Optionally, the update annotation information determination module 320 is also used for:

[0149] Extract data from the source image data that conforms to the set filtering rules to obtain the filtered result data; classify the filtered result data into categories using a target classification model, and use the classified filtered result data as the first target image.

[0150] Optionally, the update annotation information determination module 320 is also used for:

[0151] Extract the target visual information from the first target image, filter the first target image based on the target visual information to obtain the second target image; extract the target salient region from the second target image, and generate annotation information for the target salient region as specific region annotation information.

[0152] Optionally, the update annotation information determination module 320 is also used for:

[0153] For each data point in the first target image, extract at least two dimensions of target visual information; perform cross-validation and filtering on the target visual information corresponding to at least two dimensions for each data point in the first target image, and retain the first target image that matches the target visual information in each dimension as the second target image.

[0154] Optionally, the data production module 330 is also used for:

[0155] Obtain the target fusion model and model prompt information; input the model prompt information, initial annotation information, updated annotation information and source image data into the target fusion model to obtain fusion annotation information; wherein, the model prompt information is used to instruct the target fusion model to generate fusion annotation information based on the initial annotation information, updated annotation information and source image data.

[0156] The data production apparatus provided in the embodiments of the present invention can execute the data production method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the method.

[0157] Example 4

[0158] Figure 4 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0159] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0160] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0161] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as data production methods.

[0162] In some embodiments, the data production method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or mounted on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the data production described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the data production method by any other suitable means (e.g., by means of firmware).

[0163] Various embodiments of the systems and techniques described above 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), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0164] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0165] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0166] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, 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 sound input, voice input, or tactile input).

[0167] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0168] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

Claims

1. A data production method, characterized in that, include: Obtain the source image data and its initial annotation information; The source image data is processed and re-annotated to obtain updated annotation information; The initial annotation information and the updated annotation information are fused to obtain fused annotation information. The fused annotation information is then input into the target data production model to obtain the data production result for the source image data.

2. The method according to claim 1, characterized in that, The source image data is processed and re-annotated to obtain updated annotation information, including: The source image data is filtered and classified to obtain the first target image; The first target image is fully annotated to obtain full annotation information; The first target image is annotated with specific regions to obtain specific region annotation information, and the full image annotation information and the specific region annotation information are used as the updated annotation information.

3. The method according to claim 2, characterized in that, The source image data is filtered and classified to obtain the first target image, including: Extract the data from the source image data that conforms to the set filtering rules to obtain the filtered result data; The filtered result data is categorized using a target classification model, and the categorized filtered result data is used as the first target image.

4. The method according to claim 2 or 3, characterized in that, The first target image is annotated with specific regions to obtain specific region annotation information, including: Extract the target visual information from the first target image, and filter the first target image according to the target visual information to obtain the second target image; Extract the salient region of the target from the second target image, and generate annotation information for the salient region of the target as the specific region annotation information.

5. The method according to any one of claims 2-4, characterized in that, The step of extracting target visual information from the first target image and filtering the first target image based on the target visual information to obtain the second target image includes: For each data point in the first target image, extract at least two dimensions of target visual information; Cross-validation and filtering are performed on the target visual information corresponding to at least two dimensions for each data in the first target image, and the first target image that matches the target visual information in each dimension is retained. The retained first target image is used as the second target image.

6. The method according to any one of claims 1-5, characterized in that, The initial annotation information and the updated annotation information are merged to obtain merged annotation information, including: Obtain the target fusion model and model hints; The model prompt information, the initial annotation information, the updated annotation information, and the source image data are input into the target fusion model to obtain the fusion annotation information; wherein, the model prompt information is used to instruct the target fusion model to generate the fusion annotation information based on the initial annotation information, the updated annotation information, and the source image data.

7. A data production apparatus, characterized in that, include: The source image data and initial annotation information acquisition module is used to acquire source image data and the initial annotation information of the source image data; The updated annotation information determination module is used to process and re-annotate the source image data to obtain updated annotation information; The data production module is used to fuse the initial annotation information with the updated annotation information to obtain fused annotation information, and input the fused annotation information into the target data production model to obtain the data production result for the source image data.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor to enable the at least one processor to perform the data production method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that are used to cause a processor to execute the data production method according to any one of claims 1-6.

10. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the data production method as described in any one of claims 1-6.

Images