Method and system for augmenting object-centric data for explainable model learning
The Hann window function-based augmentation method enhances object-centered learning by focusing on central areas, improving model performance and enabling explainable technology to concentrate on objects, addressing the lack of explanatory elements in existing deep learning models.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- KOREA ELECTRONICS TECH INST
- Filing Date
- 2024-11-20
- Publication Date
- 2026-05-07
AI Technical Summary
Existing deep learning models for object detection and classification lack explanatory elements, leading to degraded performance and a failure to focus on objects during data augmentation, especially when explainable technology is applied.
An augmentation method and system that applies the Hann window function to enhance the central area of object regions in image data, excluding the outer edges, and sets weights for scaling, generating augmented image data for object-centered learning.
Improves model performance and enables explainable technology to focus on objects, enhancing the model's ability to provide explanations for its judgments.
Smart Images
Figure KR2024018337_07052026_PF_FP_ABST
Abstract
Description
Object-oriented explainable model training data augmentation method and system
[0001] The present invention relates to a method and system for augmenting training data, and more specifically, to a method and system for augmenting training data used in object-centered explainable model learning.
[0002] Various deep learning models, such as image-based object detection or classification models, lack explanatory elements regarding the basis for their judgments. Consequently, although much research has recently been conducted on visual intelligence explanation techniques, there is a problem where the performance of explainable techniques is degraded by background and other factors.
[0003] Accordingly, while various data augmentation methods exist for object detection or classification models, these methods are not related to explainable technology, which may lead to a problem where the performance of the learning model is not improved and explainable technology fails to focus on objects.
[0004]
[0005] The present invention has been devised to solve the above-mentioned problems, and the objective of the present invention is to provide an augmentation method and system capable of augmenting training data of a visual intelligence model to which explainable technology is applied.
[0006] A method for augmenting object-centered explainable model learning data according to an embodiment of the present invention for achieving the above objective comprises: a system acquiring image data for object detection; and a system applying the acquired image data to an object detection model to form an object region (I ROI A step of extracting ); and a system, wherein the object area (I ROI By applying the Hann window function to the augmented image data (I H Includes the step of generating ).
[0007] And the Hann window function can be configured so that the outer area formed in a curved shape along the border is excluded, and only the central area is reinforced.
[0008] Additionally, the Hann window function can be configured such that the outer area formed in a curved shape along the border is excluded and only the central area is reinforced, so that the applied image has a value within a preset error range of 1 or 1 in the central area and a value within a preset error range of 0 or 0 in the outer area formed in a curved shape along the border.
[0009] And the Hann window function (h(n)) can be calculated through the following Equation 1.
[0010] (Formula 1)
[0011] In addition, the object-oriented explainable model learning data augmentation method according to the present embodiment comprises a system, wherein the object domain (I ROI It may further include a step of setting weights for scaling the center area of the Hann window function before applying the Hann window function to ).
[0012] And the step of setting weight(s) for scaling the center area can be set by referring to Equation 2 below when the default Hann window function is h.
[0013] (Equation 2)
[0014] In addition, the weight(s) for scaling the central region allows for relatively smoother scaling as the value becomes relatively smaller.
[0015] And augmented image data (I H ) can be used as training data or input data for a synthetic face recognition model with explainable technology applied.
[0016] In addition, the object-oriented explainable model learning data augmentation method according to the present embodiment comprises a system, augmented image data (I H It may further include a step of generating a training dataset based on ), wherein the training dataset is, the same object region (I ROI Multiple augmented image data (I) generated by applying different weight(s) to ) H ) may be included.
[0017] Meanwhile, according to another embodiment of the present invention, an object-centered explainable model learning data augmentation system comprises: a communication unit for acquiring image data for object detection; and applying the acquired image data to an object detection model to an object region (I ROI Extract ) and object area (I ROI By applying the Hann window function to the augmented image data (I H Includes a processor that generates ).
[0018] In addition, according to another embodiment of the present invention, an object-centered explainable model learning data augmentation method comprises a system applying image data containing an object to an object detection model to an object region (I ROI A step of extracting ); a step in which the system sets weights for scaling the central region in a Hann window function; and a step in which the system, the object region (I ROI By applying a weighted Hann window function to the augmented image data (I H Includes the step of generating ).
[0019] And according to another embodiment of the present invention, an object-centered explainable model learning data augmentation system applies image data containing an object to an object detection model to an object region (I ROIAn object region extraction module that extracts ); and sets weights for scaling the center region in the Hann window function, and the object region (I ROI By applying a weighted Hann window function to the augmented image data (I H Includes a data augmentation module that generates ).
[0020] As described above, according to the embodiments of the present invention, by providing augmented training data to a visual intelligence model to which explainable technology is applied, the visual intelligence model to which explainable technology is applied is induced to learn in an object-centered manner, thereby contributing to the improvement of model performance and enabling explainable technology to focus more on objects.
[0021] FIG. 1 is a drawing provided for the configuration description of an object-centered explainable model learning data augmentation system according to an embodiment of the present invention,
[0022] FIG. 2 is a drawing provided for a more detailed configuration description of the processor illustrated in FIG. 1.
[0023] FIG. 3 is a drawing provided to explain the process of generating augmented image data through an object-centered explainable model learning data augmentation system according to an embodiment of the present invention.
[0024] FIG. 4 is a drawing provided for explaining a Hann window function used in an object-centered explainable model learning data augmentation system according to an embodiment of the present invention.
[0025] FIG. 5 is a flowchart provided for explaining an object-centered explainable model learning data augmentation method according to an embodiment of the present invention, and
[0026] FIG. 6 is a diagram provided for the description of a visual intelligence model to which an explainable technology is applied, in which augmented image data generated through an object-centered explainable model learning data augmentation method according to an embodiment of the present invention is utilized as learning data.
[0027] The present invention will be described in more detail below with reference to the drawings. To clearly explain the invention, parts unrelated to the description have been omitted from the drawings, and in the drawings, the width, length, thickness, etc., of the components may be exaggerated for convenience.
[0028] FIG. 1 is a diagram provided to describe the configuration of an object-centered explainable model learning data augmentation system according to one embodiment of the present invention.
[0029] The object-oriented explainable model learning data augmentation system according to the present embodiment (hereinafter collectively referred to as the 'system') is provided to augment learning data of a visual intelligence model to which explainable technology is applied.
[0030] For example, to enable object-centered augmentation of training data, the system acquires image data for object detection or classification, applies the image data to an object detection model to extract object regions, and generates augmented image data for the object regions by applying a window that excludes the outer edges of the extracted object regions and enhances only the central part of the object regions. At this time, the generated augmented image data can be utilized as training data for a visual intelligence model to which explainable technology is applied.
[0031] To this end, the system may include a communication unit (100), a processor (200), and a storage unit (300).
[0032] The communication unit (100) is equipped with a communication module connected to a network, and can acquire image data for object detection.
[0033] The storage unit (300) is provided to store programs and data necessary for the operation of the processor (200).
[0034] A processor (200) is provided to handle all matters for augmenting training data.
[0035] Specifically, the processor (200) applies image data to an object detection model to create an object region (I ROI Extract ) and set weights for scaling the center area in the Hann window function, so that the object area (I ROI A weighted Hann window function is applied to ), and through this, augmented image data (I H Can generate ).
[0036] And the processor (200) has augmented image data (I H A training dataset can be generated based on ) and used as training data for a visual intelligence model with explainable technology applied.
[0037] Here, the training dataset is the same object region (I ROI Multiple augmented image data (I) generated by applying different weight(s) to ) H ) may be included.
[0038] FIG. 2 is a drawing provided for a more detailed configuration description of the processor illustrated in FIG. 1, and FIG. 3 is a drawing provided for a description of the process of generating augmented image data through an object-oriented explainable model learning data augmentation system according to an embodiment of the present invention.
[0039] Referring to FIG. 2, the processor (200) applies image data containing an object to an object detection model to an object region (I ROIA weight for scaling the center region is set in an object region extraction module (210) that extracts ) and a Hann window function, and the object region (I ROI By applying a weighted Hann window function to the augmented image data (I H It may include a data augmentation module (220) that generates ).
[0040] The Hann window function can be configured so that the outer area formed in a curved shape along the border, as shown in FIG. 4, is excluded, and only the central area is reinforced.
[0041] Specifically, the Hann window function can be configured such that the outer area formed in a curved shape along the border is excluded and only the central area is reinforced, so that the applied image has a value within a preset error range of 1 or 1 in the central area and a value within a preset error range of 0 or 0 in the outer area formed in a curved shape along the border.
[0042] At this time, the Hann window function (h(n)) can be calculated through the following Equation 1.
[0043] (Formula 1)
[0044] In Formula 1, the cos function is set as a periodic function that has a value close to 1 at the center and a value of 0 in the form of a smooth curve near the outer edge. Also, in Formula 1, 0.5 is a weight value that can adjust the overall height of the window.
[0045] And the Hann window function can set weights by referring to Equation 2 below so that the Hann window can sufficiently contain objects.
[0046] (Equation 2)
[0047] That is, when the basic Hann window function is h, the Hann window can sufficiently contain objects by setting weights by referring to Formula 2 below.
[0048] Here, the weight(s) for scaling the central region allows the scale to be adjusted relatively smoothly as the value becomes smaller.
[0049] In summary, the Hann window function (h(n)) can scale how wide the center is by adding a squared term s to the basic Hann window function (h), and the smaller the value of s, the smoother the scale changes.
[0050] The data augmentation module (220) is the same object area (I ROI Multiple augmented image data (I) to which different weight(s) values are applied (Hann windows of different scales are applied) H You can generate ) and create a training dataset based on it.
[0051] FIG. 5 is a flowchart provided for the description of an object-centered explainable model learning data augmentation method according to an embodiment of the present invention, and FIG. 6 is a diagram provided for the description of a visual intelligence model to which an explainable technology is applied, in which augmented image data generated through an object-centered explainable model learning data augmentation method according to an embodiment of the present invention is utilized as learning data.
[0052] The object-oriented explainable model learning data augmentation method according to the present embodiment can be executed by the system described above with reference to FIGS. 1 to 4.
[0053] Referring to FIG. 5, the system acquires image data for object detection or classification, etc. (S510) to perform object-centered training data augmentation, and applies the image data to an object detection model to obtain an object region (I ROI ) is extracted (S520), weights for scaling the center region are set in the Hann window function (S530), and the extracted object region (I ROI By applying a weighted Hann window function to the augmented image data (I H ) can be generated (S540).
[0054] Subsequently, the system uses augmented image data (I H A training dataset can be generated based on ) and provided to a visual intelligence model with explainable technology applied.
[0055] A visual intelligence model with explainable technology applied may be a synthetic face discrimination model with explainable technology applied, as exemplified in Fig. 6.
[0056] A synthetic face recognition model with explainable technology applied uses augmented image data (I H By utilizing ) as training data or input data, a synthetic face discrimination model with explainable technology can be enabled to distinguish synthetic faces from an object-centered perspective.
[0057] That is, a synthetic face recognition model with explainable technology applied uses augmented image data (I H By utilizing ) as training data or input data, it is possible to determine whether an object in the image data is a real face or a fake synthetic face, and provide the basis (explanation) for the determination along with the determination result.
[0058] So far, preferred embodiments of an object-oriented explainable model learning data augmentation method and system have been described in detail.
[0059] Existing image-based object detection or classification methods may face problems where they fail to improve the performance of training models or allow explainable technologies to focus on objects, as they lack explanatory elements regarding the basis for their judgments or are unrelated to explainable technologies during data augmentation.
[0060] On the other hand, according to an embodiment of the present invention, augmented learning data is provided to a visual intelligence model to which explainable technology is applied, thereby inducing the visual intelligence model to learn in an object-centered manner.
[0061] Through this, model performance improvement and explainable technology can contribute to enabling visual intelligence models to focus on objects during data training using explainable technology, compared to existing data augmentation methods.
[0062] Meanwhile, it goes without saying that the technical concept of the present invention may also be applied to a computer-readable recording medium containing a computer program that enables the device and method according to the present embodiment to perform their functions. Furthermore, the technical concept according to various embodiments of the present invention may be implemented in the form of computer-readable code recorded on a computer-readable recording medium. A computer-readable recording medium may be any data storage device that can be read by a computer and store data. For example, a computer-readable recording medium may be a ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical disk, hard disk drive, etc. Additionally, computer-readable code or a program stored on a computer-readable recording medium may be transmitted through a network connected between computers.
[0063] Furthermore, although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above. Various modifications are possible by those skilled in the art without departing from the essence of the invention as claimed in the claims, and such modifications should not be understood individually from the technical spirit or perspective of the present invention.
Claims
1. A step in which the system acquires image data for object detection; The system applies acquired image data to an object detection model to define object regions (I ROI Step of extracting ); and The system, object area (I ROI By applying the Hann window function to the augmented image data (I H An object-oriented explainable model learning data augmentation method comprising the step of generating ).
2. In Claim 1, Hann window functions are, An object-centered explainable model learning data augmentation method characterized by excluding the outer region formed in a curved shape along the border and setting only the central region to be reinforced.
3. In Claim 2, Hann window functions are, An object-centered explainable model learning data augmentation method characterized by excluding an outer region formed in a curved shape along the border and reinforcing only the central region, wherein the applied image is set to have a value within a preset error range of 1 or 1 in the central region and a value within a preset error range of 0 or 0 in the outer region formed in a curved shape along the border.
4. In Claim 3, The Hann window function (h(n)) is, An object-oriented explainable model learning data augmentation method characterized by being calculated through the following Formula 1. (Formula 1) 5. In Claim 1, The system, object area (I ROI An object-centered explainable model learning data augmentation method characterized by further including the step of setting weights for scaling the central region of the Hann window function before applying the Hann window function to ).
6. In Claim 5, The step of setting weight(s) for scaling the central region is, An object-oriented explainable model training data augmentation method characterized by setting weights by referring to Formula 2 below when the basic Hann window function is h. (Equation 2) 7. In Claim 6, The weight(s) for scaling the central region are, An object-centered explainable model learning data augmentation method characterized by the fact that the scale is adjusted relatively smoothly as the value becomes relatively smaller.
8. In Claim 1, Augmented image data (I H )Is, An object-centered explainable model training data augmentation method characterized by being usable as training data or input data for a synthetic face recognition model to which explainable technology is applied.
9. In Claim 1, The system, augmented image data (I H It further includes the step of generating a training dataset based on ), The training dataset is, Same object area (I ROI Multiple augmented image data (I) generated by applying different weight(s) to ) H An object-oriented explainable model learning data augmentation method characterized by including ).
10. A communication unit for acquiring image data for object detection; and By applying the acquired image data to an object detection model, the object region (I ROI Extract ) and object area (I ROI By applying the Hann window function to the augmented image data (I H An object-oriented explainable model learning data augmentation system comprising a processor that generates ).
11. The system applies image data containing an object to an object detection model to define the object region (I ROI Step of extracting ); The system sets weights for scaling the central region in the Hann window function; and The system, object area (I ROI By applying a weighted Hann window function to the augmented image data (I H An object-oriented explainable model learning data augmentation method comprising the step of generating ).
12. Applying image data containing objects to an object detection model to determine the object region (I ROI Object region extraction module that extracts ); and Set weights for scaling the center area in the Hann window function, and the object area (I ROI By applying a weighted Hann window function to the augmented image data (I H An object-oriented explainable model learning data augmentation system comprising a data augmentation module that generates ).