Data cleaning device and method

Abstract

The invention provides a data cleaning device and method. The method comprises the steps of: receiving, by a processor, an image through a transceiver, wherein the image comprises a picture; when the processor determines that the continuous value corresponding to the image is greater than the continuous value threshold value, the processor performs global continuity detection on the picture to obtain a gradient distribution value corresponding to the picture; and determining, by the processor, whether to perform data cleaning corresponding to the training set on the picture using the gradient distribution value.

Description

Technical Field

[0001] The present invention relates to a device and method for data cleaning. Background Art

[0002] Currently, in the field of artificial intelligence / deep learning, a large amount of training data (pictures) is required for 3D modeling technology, and human labor is needed to perform filtering / cleaning of the training data. How to perform data cleaning more efficiently is one of the goals that those skilled in the art should strive for. Summary of the Invention

[0003] The present invention provides a device and method for data cleaning, which can perform data cleaning more efficiently.

[0004] The data cleaning device of the present invention includes a transceiver and a processor. The processor is coupled to the transceiver, and the processor receives an image through the transceiver, where the image includes pictures; when the processor determines that the continuous value corresponding to the image is greater than the continuous value threshold, the processor performs a global continuity detection on the pictures to obtain a percentile rank (PR) corresponding to the pictures; the processor uses the percentile rank to determine whether to perform data cleaning on the pictures corresponding to the training set.

[0005] The data cleaning method of the present invention includes the following steps: receiving, by the processor through the transceiver, an image, where the image includes pictures; when the processor determines that the continuous value corresponding to the image is greater than the continuous value threshold, performing, by the processor, a global continuity detection on the pictures to obtain a percentile rank corresponding to the pictures; and determining, by the processor, whether to perform data cleaning on the pictures corresponding to the training set using the percentile rank. Brief Description of the Drawings

[0006] Figure 1 is a schematic diagram of a data cleaning device according to an embodiment of the present invention.

[0007] Figure 2 is a flowchart of a data cleaning method according to an embodiment of the present invention.

[0008] Figure 3 is a schematic diagram of obtaining a continuous value corresponding to an image according to an embodiment of the present invention.

[0009] Figure 4A and Figure 4B is a flowchart of a data cleaning method according to another embodiment of the present invention.

[0010] Description of the Reference Numerals

[0011] 100: Data cleaning device

[0012] 110: Storage device

[0013] 130: Transceiver

[0014] 150: Processor

[0015] S210, S220, S230, S410, S420, S430, S440: Steps Detailed implementation manner

[0016] Figure 1 is a schematic diagram of a data cleaning device 100 according to an embodiment of the present invention. Please refer to Figure 1 Figure 1. The device 100 may include a storage device 110, a transceiver 130, and a processor 150.

[0017] The storage device 110 is, for example, 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 similar components, or a combination of the above components, and is used to store multiple modules or various application programs that can be executed by the processor 150.

[0018] The transceiver 130 transmits and receives signals in a wireless or wired manner.

[0019] The processor 150 is, for example, a central processing unit (CPU), or other programmable general-purpose or special-purpose micro control unit (MCU), microprocessor, digital signal processor (DSP), programmable controller, application specific integrated circuit (ASIC), graphics processing unit (GPU), image signal processor (ISP), image processing unit (IPU), arithmetic logic unit (ALU), complex programmable logic device (CPLD), field programmable gate array (FPGA), or other similar components or a combination of the above components. The processor 150 can be coupled to the storage device 110 and the transceiver 130, and access and execute multiple modules and various application programs stored in the storage device 110.

[0020] Figure 2 is a flowchart of a data cleaning method illustrated according to an embodiment of the present invention, wherein the method can be implemented by Figure 1 the device 100 shown. Please refer to Figure 1 and Figure 2 .

[0021] In step S210, the processor 150 can receive an image through the transceiver 130, where the image includes pictures.

[0022] In step S220, when the processor 150 determines that the continuous value corresponding to the image is greater than the continuous value threshold, the processor 150 can perform a global continuity detection on the picture to obtain a gradient distribution value (PR, PercentileRank) corresponding to the picture. In one embodiment, the continuous value threshold can be 0.75. The following will use Figure 3 to illustrate an implementation example of the "continuous value corresponding to the image" in step S220.

[0023] Figure 3 is a schematic diagram of obtaining the continuous value corresponding to the image illustrated according to an embodiment of the present invention. Please refer to Figure 1 , Figure 2and Figure 3 . In this embodiment, the processor 150 can perform image normalization, image eigenvalue extraction, and non-linear layer feature regression operations on the image to obtain continuous values. As Figure 3 shown, the processor 150 can divide each frame of the image into a uniform grid. Then, for the image in each grid, the processor 150 can use a convolutional neural network to perform feature extraction. Specifically, for the grid of the current frame, the processor 150 can utilize the relative position relationship between the positions of different grids to obtain scene features. On the other hand, for the grids across frames, the processor 150 can obtain the motion trajectory, behavior changes, and dynamic features of the entire image sequence in time. After obtaining the features of each grid, the processor 150 can input these features into a regression model to obtain the continuous value corresponding to the image.

[0024] Please return to Figure 2 . In step S230, the processor 150 can use the gradient distribution value to determine whether to perform data cleaning on the picture corresponding to the training set.

[0025] Figure 4A and Figure 4B are flowcharts of a data cleaning method according to another embodiment of the present invention. Please refer to Figure 1 , Figure 2 , Figure 3 , Figure 4A and Figure 4B . After the processor 150 obtains the continuous value corresponding to the image, the processor 150 can determine whether the continuous value corresponding to the image is greater than the continuous value threshold. As Figure 4A shown, when the processor 150 determines that the continuous value is not greater than the continuous value threshold, the processor 150 does not add the picture to the training set. On the other hand, when the processor 150 determines that the continuous value corresponding to the image is greater than the continuous value threshold, as Figure 4AAs shown in step S220, the processor 150 can perform global continuity detection on the picture to obtain a gradient distribution value corresponding to the picture. In this embodiment, the picture may include pixels. Further, the global continuity detection may include picture scaling, picture grayscale conversion, left - right comparison of pixels, front - back comparison of pictures, and global picture clustering. Specifically, picture scaling may be that the processor 150 removes the details of the picture and retains the structure and brightness information of the picture. Picture grayscale conversion may be that the processor 150 converts the picture into grayscale. The left - right comparison of pixels may be that the processor 150 compares all pixel points left and right. If the left pixel point is greater than the right pixel point, the processor 150 marks it as 1. On the other hand, if the left pixel point is less than or equal to the right pixel point, the processor 150 marks it as 0. Thus, the processor 150 can generate a difference value matrix for this picture. The front - back comparison of pictures may be that the processor 150 uses the Hamming distance to calculate the difference degree between the front and back pictures of the image. If the Hamming distance value between the front and back pictures is smaller, it means the similarity between the front and back pictures is higher. On the other hand, if the Hamming distance value between the front and back pictures is larger, it means the similarity between the front and back pictures is lower. Global picture clustering may be that the processor 150 clusters all pictures according to similarity (i.e., similar pictures are divided into the same group). After performing global picture clustering, the processor 150 can perform gradient sorting on the pictures in each group to obtain a gradient distribution value corresponding to the picture.

[0026] Next, as Figure 4A shown in step S230, when the processor 150 determines that the gradient distribution value is the mode, the processor 150 can add the picture to the training set. On the other hand, when the processor 150 determines that the gradient distribution value is an extreme number, the processor 150 does not add the picture to the training set. In one embodiment, the above - mentioned mode is, for example, that the gradient distribution value is between 40 and 60. On the other hand, the above - mentioned extreme number is, for example, that the gradient distribution value is greater than 90 or less than 10.

[0027] Please continue to refer to Figure 4A and Figure 4B . As Figure 4B shown in step S410, when the processor 150 determines that the gradient distribution value is away from the mean, the processor 150 can perform distortion detection on the picture using the No - Reference Image Quality Evaluator (NIQE) to obtain a distortion value corresponding to the picture. In one embodiment, the above - mentioned value away from the mean is, for example, that the gradient distribution value is greater than 10 and less than 40, or the gradient distribution value is greater than 60 and less than 90. When the processor 150 determines that the distortion value is less than the distortion value threshold, the processor 150 can add the picture to the training set. In one embodiment, the distortion value threshold can be 10. On the other hand, as Figure 4BAs shown in step S420, step S430, and step S440, when the processor 150 determines that the distortion value is not less than the distortion value threshold, the processor 150 may perform a correction operation and a filtering operation on the picture to obtain an integrity value corresponding to the picture.

[0028] In one embodiment, in Figure 4B step S420 of, the processor 150 may use DBGAN (Data Balancing Generative Adversarial Network) to perform a correction operation on the picture. Specifically, the processor 150 may use BGAN (Balancing Generative Adversarial Network) to learn the blurring process of the picture, so as to generate a blurred synthetic picture. Then, the processor 150 may input the blurred synthetic picture into DGBAN to transform the picture from blurred to clear.

[0029] In one embodiment, in Figure 4B step S430 of, the processor 150 may use edge detection technology to perform a filtering operation on the picture. Specifically, the processor 150 may use edge detection technology to determine whether the contour of the picture is complete.

[0030] After performing Figure 4B step S420, step S430, and step S440 of, when the processor 150 determines that the integrity value is greater than the integrity threshold, the processor 150 may add the picture to the training set. On the other hand, when the processor 150 determines that the integrity value is not greater than the integrity threshold, the processor 150 does not add the picture to the training set. In one embodiment, the integrity threshold may be 100.

[0031] In summary, after determining that the continuous value corresponding to the image is greater than the continuous value threshold, the data cleaning device and method of the present invention may use the gradient distribution value corresponding to the picture to determine whether to perform data cleaning corresponding to the training set on the picture. In other words, after the present invention determines not to add a specific picture to the training set, this picture can be automatically deleted (data cleaning). Based on this, it is not necessary for humans to perform filtering / cleaning of training data, thereby enabling more efficient data cleaning.

Claims

1. A device for data cleaning, characterized in that, comprising: a transceiver; and a processor, coupled to the transceiver, wherein the processor receives an image through the transceiver, and the image includes pictures; when the processor determines that a continuous value corresponding to the image is greater than a continuous value threshold, the processor performs a global continuity detection on the picture to obtain a gradient distribution value (PR, Percentile Rank) corresponding to the picture; the processor uses the gradient distribution value to determine whether to perform data cleaning corresponding to a training set on the picture.

2. The device according to claim 1, characterized in that, the continuous value threshold is 0.

75.

3. The device according to claim 1, characterized in that, the processor performs image normalization, image eigenvalue extraction, and non-linear layer feature regression operations on the image to obtain the continuous value.

4. The device according to claim 1, characterized in that, when the processor determines that the continuous value is not greater than the continuous value threshold, the processor does not add the picture to the training set.

5. The device according to claim 1, characterized in that, the picture includes pixels, and the global continuity detection includes picture scaling, picture grayscale conversion, left-right comparison of the pixels, front-back comparison of the picture, and picture global clustering.

6. The device according to claim 1, characterized in that, when the processor determines that the gradient distribution value is the mode, the processor adds the picture to the training set.

7. The device according to claim 1, characterized in that, when the processor determines that the gradient distribution value is an extreme number, the processor does not add the picture to the training set.

8. The device according to claim 1, characterized in that, when the processor determines that the gradient distribution value is a deviation from the mean, the processor performs a distortion detection on the picture using a no-reference image quality assessment metric (NIQE) to obtain a distortion value corresponding to the picture; when the processor determines that the distortion value is less than a distortion value threshold, the processor adds the picture to the training set.

9. The device according to claim 8, characterized in that, the distortion value threshold is 10.

10. The device according to claim 8, characterized in that, when the processor determines that the distortion value is not less than the distortion value threshold, the processor performs a correction operation and a filtering operation on the picture to obtain an integrity value corresponding to the picture; when the processor determines that the integrity value is greater than an integrity threshold, the processor adds the picture to the training set; when the processor determines that the integrity value is not greater than the integrity threshold, the processor does not add the picture to the training set.

11. The device according to claim 10, characterized in that, the integrity threshold is 100.

12. The device according to claim 10, characterized in that, The processor uses DBGAN (Data Balancing Generative Adversarial Network) to perform the correction operation on the picture.

13. The apparatus according to claim 10, wherein, the processor uses edge detection technology to perform the filtering operation on the picture.

14. A method for data cleaning, applicable to an apparatus including a transceiver and a processor, wherein, the method includes the following steps: receiving, by the processor through the transceiver, an image, where the image includes a picture; when the processor determines that a continuous value corresponding to the image is greater than a continuous value threshold, performing, by the processor, global continuity detection on the picture to obtain a gradient distribution value corresponding to the picture; and determining, by the processor using the gradient distribution value, whether to perform data cleaning on the picture corresponding to a training set.

15. The method according to claim 14, wherein, the method further includes: performing, by the processor, image normalization, image eigenvalue extraction, and non-linear layer feature regression operation on the image to obtain the continuous value.

16. The method according to claim 14, wherein, the picture includes pixels, and the global continuity detection includes picture scaling, picture grayscale conversion, left-right comparison of the pixels, front-back comparison of the picture, and picture global clustering.

17. The method according to claim 14, wherein, the method further includes: when the processor determines that the gradient distribution value is off-mean, performing, by the processor, distortion detection on the picture using a no-reference picture quality assessment metric to obtain a distortion value corresponding to the picture; and when the processor determines that the distortion value is less than a distortion value threshold, adding, by the processor, the picture to the training set.

18. The method according to claim 17, wherein, the method further includes: when the processor determines that the distortion value is not less than the distortion value threshold, performing, by the processor, a correction operation and a filtering operation on the picture to obtain an integrity value corresponding to the picture; when the processor determines that the integrity value is greater than an integrity threshold, adding, by the processor, the picture to the training set; when the processor determines that the integrity value is not greater than the integrity threshold, not adding, by the processor, the picture to the training set.

19. The method according to claim 18, wherein, when the processor determines that the distortion value is not less than the distortion value threshold, the step of performing, by the processor, the correction operation and the filtering operation on the picture to obtain the integrity value corresponding to the picture includes: using, by the processor, DBGAN (Data Balancing Generative Adversarial Network) to perform the correction operation on the picture.

20. The method according to claim 18, wherein, When the processor determines that the distortion value is not less than the distortion value threshold, the steps for the processor to perform the correction operation and the filtering operation on the picture to obtain the integrity value corresponding to the picture include: The processor performs the filtering operation on the picture by using edge detection technology.

Images