Transfer method for complete annotation data and electronic device

Abstract

This case discloses a method and electronic device for transferring complete labeled data. The electronic device executes the method for transferring complete labeled data. This method for transferring complete labeled data includes: annotating multiple image data to obtain labeled data, wherein the labeled data includes an attribute feature and a label range; inputting the image data and labeled data into a first deep learning model for training to generate model weight information; storing the model weight information in a specific file format as a model weight file; and transferring the model weight file to an external device for use by the external device.

Description

[Technical Field]

[0001] This case relates to a method for transferring complete annotation data and an electronic device for performing the transfer of complete annotation data. [Previous Technology]

[0002] Before training a deep learning model, it is usually necessary to perform image annotation on the sample data. The image category and location are recorded using bounding boxes or pixel coordinates, and the annotated data is then fed into the model for training. The annotation process is usually performed manually or with the assistance of traditional image preprocessing.

[0003] To transfer the annotation results to other devices, the traditional method is to save the already annotated results (including images and tags) as a specific file format, such as an XML file, and then transfer them to other devices by copying or importing. This type of transfer operation can only transfer the annotated data and does not include transferring the annotated image features. If you want to add other image tags, you need to redo the image annotation process. [Summary of the Invention]

[0004] This application provides a method for transferring complete labeled data, comprising: performing image annotation on multiple image data to obtain labeled data, wherein the labeled data includes an attribute feature and a label range; inputting the image data and labeled data into a first deep learning model for training to generate model weight information; storing the model weight information in a specific file format as a model weight file; and transferring the model weight file from the storage device to an external device for use by the external device.

[0005] This application also provides a method for transferring complete labeled data, comprising: performing image annotation on multiple image data to obtain labeled data, wherein the labeled data includes an attribute feature and a label range; inputting the image data and labeled data into a first deep learning model for training to generate model weight information; storing the model weight information in a specific file format as a model weight file; selecting a set of images to be labeled, and using the model weight file in the first deep learning model to infer the images to be labeled to generate labeled information including an attribute feature and a label range; storing the labeled information in a fixed file format as a labeled file; and transferring the labeled file to an external device for use by the external device.

[0006] This application further provides an electronic device, comprising a storage device and a processing device. The storage device stores a plurality of image data and corresponding annotation data, the annotation data including an attribute feature and a labeling range. The processing device is electrically connected to the storage device and has a built-in first deep learning model. The processing device inputs the image data and annotation data into the first deep learning model for training to generate model weight information, and stores the model weight information in the storage device in a specific file format as a model weight file. The processing device selects to execute a first process or a second process. In the first process, the model weight file is transferred to an external device for use by the external device. In the second process, a set of images to be labeled is selected, and the model weight file is used in the first deep learning model to infer the images to be labeled to generate annotation information including an attribute feature and a labeling range. The annotation information is stored in a fixed file format as a label file, and then the label file is transferred to an external device for use by the external device.

[0007] In summary, this invention proposes a method and electronic device for transferring complete labeled data. This method allows for the transfer of labeled data containing image annotation features to different devices, and can simultaneously transfer features of more than one image category. Furthermore, it can use these labeled features to infer new annotation results from all unlabeled images to be labeled. Therefore, this invention can transfer image annotation features to different devices or storage spaces, and can also obtain annotation results (annotation files) from more images in a semi-automatic manner for direct use in model training.

Implementation Method

[0009] The following provides a detailed description of preferred embodiments; however, these embodiments are merely illustrative and do not limit the scope of protection sought in this application. Furthermore, some elements are omitted from the drawings in the embodiments to clearly show the technical features of this application. The same reference numerals will be used to denote the same or similar elements in all drawings.

[0010] Referring to Figure 1, an electronic device 10 includes a processing device 12, a storage device 14, and a graphics processor 16. The storage device 14 stores a plurality of image data and corresponding annotation data for each image data. This annotation data is in a uniform format and includes an attribute feature and a label range. In this embodiment, the attribute feature is a category feature, which can be used for multiple categories. The processing device 12 is electrically connected to the storage device 14 and has a built-in first deep learning model 121. The processing device 12 reads the image data and annotation data from the storage device 14 and inputs the image data and annotation data into the first deep learning model 121 for training to generate model weight information with a certain accuracy in recognizing the annotation data. The model weight information is stored in the storage device 14 in a specific file format as a model weight file. After obtaining the model weight file, the processing device 12 can choose to execute a first process or a second process. In the first process, the processing device 12 transfers the model weight file to the external device 18 via network transmission or a USB external device for use by the external device 18. In the second process, the processing device 12 selects another set of images to be labeled and inputs them into the first deep learning model 121. The first deep learning model 121 uses the model weight file to infer from the images to generate labeling information containing attribute features and labeling ranges. This labeling information is stored in the storage device 14 in a fixed file format as a labeling file. This labeling file is a transferable file with a unified labeling format. The processing device 12 then transfers the labeling file to the external device 18 via network transmission or a USB external device for use by the external device 18. Additionally, the graphics processor 16 is electrically connected to the processing device 12. When the processing device 12 performs calculations or training, the graphics processor 16 assists the processing device 12 in performing related calculations, thereby accelerating the overall calculation speed.

[0011] In one embodiment, the external device 18 further has a built-in second deep learning model 181. When the processing device 12 selects to transfer the model weight file to the external device 18 in the first process, the external device 18 will select a set of images to be labeled, and use the model weight file to infer the images to be labeled in the external device 18 to generate labeling information containing attribute features and labeling range, and store this labeling information as a labeling file in a fixed file format, so that the second deep learning model 181 in the external device 18 is trained directly using the labeling file.

[0012] In one embodiment, the electronic device 10 is an electronic device such as a personal computer, a laptop computer, or a tablet computer that can independently perform artificial intelligence model training operations, but this invention is not limited thereto. Similarly, the external device 18 is an electronic device such as a personal computer, a laptop computer, or a tablet computer that can independently perform artificial intelligence model training operations, but this invention is also not limited thereto.

[0013] In one embodiment, the processing device 12 is a central processing unit (CPU), or other general-purpose or special-purpose microprocessor, microcontroller, micro control unit (MCU), digital signal processor (DSP), programmable controller, application-specific integrated circuit (ASIC), or other similar elements or combinations of the above elements, and this invention is not limited thereto.

[0014] In one embodiment, the storage device 14 may be any type of fixed or removable random access memory (RAM), read-only memory (ROM), flash memory, hard disk drive (HDD), solid state drive (SSD) or other similar elements or combinations thereof, for storing any models, pictures or data required by the processing device 12, but this application is not limited thereto.

[0015] In one embodiment, the first deep learning model 121 and the second deep learning model 181 are artificial intelligence (AI) models of the same type or with similar functions, so that the annotation files with a unified format generated by the electronic device 10 can be transferred to the external device 18 for use. The first deep learning model 121 and the second deep learning model 181 can be selected from an object detection model, a segmentation model, a classification model, or an anomaly detection model.

[0016] In one embodiment, the model weight file is a file that stores weight parameters learned by the model during training and determines how the model transforms input data into output results. The model weight file is typically stored in binary format and may have different specific file formats depending on the deep learning framework used. Specific file formats used for the model weight file may include h5 format, weights format, ckpt format, or pth format, but this embodiment is not limited to these. Specifically, h5 format is commonly used with Keras and TensorFlow frameworks, weights format is commonly used with Darknet frameworks, ckpt format is commonly used with TensorFlow frameworks, and pth is commonly used with PyTorch frameworks, enabling the model weight file to be used with the aforementioned deep learning frameworks to reconstruct and utilize the trained model.

[0017] In one embodiment, the fixed file format is a JSON file format, so that the annotation file can be stored as a JSON file format, which is a reusable annotation result. In one embodiment, the format of the annotation file includes an image, an annotation category name, and a range to which the annotation category belongs.

[0018] In the electronic device 10, the processing device 12 can also use software to execute a complete annotation data transfer method including the first process. Please refer to Figures 1 and 2 simultaneously. As shown in step S10, the processing device 12 selects multiple image data. As shown in step S12, the image data is annotated manually or automatically to obtain annotation data. As shown in step S14, the processing device 12 inputs the image data and annotation data into the first deep learning model 121 for training to generate model weight information, and stores the model weight information as a model weight file in a specific file format. After obtaining the model weight file, as shown in step S16, it is determined whether the first deep learning model 121 can identify the category features (annotation category) in the model weight file. If yes, the process continues to the next step S18; if no, it means that the category features cannot be identified, and the process returns to step S12 to re-annotate the image. As shown in step S18, the processing device 12 transfers the model weight file to the external device 18 for use. As shown in step S20, the external device 18 will select another set of images to be labeled, use the model weight file to infer the images to be labeled, generate labeling information that includes attribute features and labeling range, and save this labeling information as a labeling file in a fixed file format, so that the external device 18 can directly use this labeling file, or fine-tune or adjust the labeling file through the labeling tool.

[0019] In the electronic device 10, the processing device 12 uses software to execute a complete annotation data transfer method including the second process. Please refer to Figures 1 and 3 simultaneously. As shown in step S30, the processing device 12 selects multiple image data. As shown in step S32, the image data is annotated manually or automatically to obtain annotation data. As shown in step S34, the processing device 12 inputs the image data and annotation data into the first deep learning model 121 for training to generate model weight information, and stores the model weight information as a model weight file in a specific file format. After obtaining the model weight file, as shown in step S36, it is determined whether the first deep learning model 121 can identify the category features (annotation category) in the model weight file. If yes, the process continues to the next step S38; if not, it returns to step S32 to re-annotate the images. As shown in step S38, the processing device 12 selects another set of images to be labeled, and uses the model weight file in the first deep learning model 121 to infer the images to be labeled, so as to generate labeling information that includes attribute features and labeling range, and saves the labeling information as a labeling file in a fixed file format. As shown in step S40, the processing device 12 transfers the labeling file to the external device 18 for use, so that the external device 18 can directly use the labeling file, or fine-tune or adjust the labeling file through the labeling tool.

[0020] In summary, this invention proposes a method and electronic device for transferring complete labeled data. This method allows the transfer of labeled data, including the features of annotated images, to different devices for use. It can simultaneously transfer features of more than one image category and can use these labeled features to infer new labeled results from all unlabeled images to be labeled. Therefore, this invention can transfer image labeled features to different devices or storage spaces, and can also obtain labeled results (label files) on more images in a semi-automatic manner for direct use in model training.

[0021] The above-described embodiments are only for illustrating the technical ideas and features of this case. Their purpose is to enable those skilled in this technology to understand the content of this case and implement it accordingly. They should not be used to limit the scope of the patent in this case. That is, all equivalent changes or modifications made in accordance with the spirit disclosed in this case should still be covered within the scope of the patent application in this case. [Simplified Explanation of the Diagram]

[0008] Figure 1 is a block diagram of an electronic device according to an embodiment of the present invention. Figure 2 is a flowchart illustrating an electronic device according to an embodiment of the present invention performing a method for transferring complete annotation data including a first process. Figure 3 is a flowchart illustrating an electronic device according to another embodiment of the present invention performing a method for transferring complete annotation data including a second process.

Claims

1. A method for transferring complete labeled data, comprising: performing image annotation on a plurality of image data to obtain labeled data, wherein the labeled data includes an attribute feature and a label range; inputting the image data and the labeled data into a first deep learning model for training to generate model weight information; storing the model weight information in a specific file format as a model weight file; transferring the model weight file from a storage device to an external device for use by the external device; selecting a set of images to be labeled, and using the model weight file on the external device to infer the images to be labeled to generate labeled information including the attribute feature and the label range; storing the labeled information in a fixed file format as a labeled file; and providing the labeled file to the external device for use.

2. The method for transferring complete labeled data as described in Request 1, wherein the attribute feature is a categorical feature.

3. The method for transferring complete labeled data as described in claim 1, wherein the external device further has a built-in second deep learning model, which is trained directly using the labeled data.

4. The method for transferring complete annotation data as described in Request 1, wherein the format of the annotation file includes an image, an annotation category name, and a range to which the annotation category belongs.

5. The method for transferring complete labeled data as described in claim 3, wherein the first deep learning model and the second deep learning model are a target detection model, a segmentation model, a classification model, or an anomaly detection model.

6. An electronic device comprising: a storage device storing a plurality of image data and corresponding annotation data, the annotation data including an attribute feature and a labeling range; and a processing device electrically connected to the storage device and having a built-in first deep learning model, the processing device inputting the image data and the annotation data into the first deep learning model for training to generate model weight information, and storing the model weight information in the storage device in a specific file format as a model weight file; the processing device executing a first process or a second process: in the first process, the processing device transferring the model weight file from the storage device to an external device for use by the external device; or in the second process, selecting a set of images to be labeled, using the model weight file in the first deep learning model to infer the images to be labeled to generate annotation information including the attribute feature and the labeling range, storing the annotation information in a fixed file format as a annotation file, and then transferring the annotation file to the external device for use by the external device.

7. The electronic device as claimed in claim 6, wherein the external device further has a built-in second deep learning model, which is trained directly using the annotation file.

8. The electronic device as described in claim 6, wherein the attribute feature is a category feature.

9. The electronic device as described in claim 6, wherein the format of the annotation file includes an image, an annotation category name, and a range to which the annotation category belongs.

10. The electronic device as claimed in claim 7, wherein the first deep learning model and the second deep learning model are a target detection model, a segmentation model, a classification model, or an anomaly detection model.

11. The electronic device as claimed in claim 6 further includes a graphics processor electrically connected to the processing device, the graphics processor assisting the processing device in performing calculations.

Images