KR20230040541A - Out-of-distribution object detection method and system - Google Patents

Abstract

The present invention relates to a method for detecting an OOD object. The method for detecting the OOD object comprises: a step of learning a teacher machine learning model to perform a classification for each image included in a learning image set using the learning image set; a step of learning the teacher machine learning model to perform an OOD object detection for a different image when an image other than the image included in the learning image set is inputted during learning of the teacher machine learning model; and a step of learning a student machine learning model to perform classification and OOD object detection for the image based on the teacher machine learning model learned using knowledge distillation. Therefore, the present invention is capable of effectively improving a detection performance.

Description

OOD object detection method and system {OUT-OF-DISTRIBUTION OBJECT DETECTION METHOD AND SYSTEM}

The present invention relates to a method and system for detecting an OOD object, and more particularly, to a method and system for detecting an OOD object using a plurality of machine learning models.

Out-of-distribution (OOD) object detection may be to distinguish whether a given input belongs to any one of the trained classes. For example, when a deep learning model or the like is over confident in its own output, an OOD object that does not belong to any of the trained classes may be incorrectly classified as being of a specific class. That is, it is important to improve OOD object detection performance to improve the performance of deep learning models.

The present invention provides an OOD object detection method, a computer program stored in a computer readable medium, and an apparatus (system) for solving the above problems.

The present invention can be implemented in a variety of ways, including a method, an apparatus (system), a computer program stored in a computer readable storage medium or a computer readable storage medium in which a computer program is stored.

According to an embodiment of the present invention, an OOD object detection method performed by at least one processor learns a teacher machine learning model to perform classification for each image included in a training image set using a training image set. Step of training the teacher machine learning model to perform OOD object detection for other images when an image other than the image included in the training image set is input during training of the teacher machine learning model and learning using knowledge distillation Based on the trained teacher machine learning model, training a student machine learning model to perform image classification and OOD object detection.

According to an embodiment of the present invention, when an arbitrary image is input, the step of performing OOD detection using the trained student machine learning model is further included.

에 의해 산출된 값이 사전 결정된 임계값 이하인 경우, 임의의 이미지를 OOD 객체로 탐지하는 단계를 포함한다. 여기서,

는 소프트맥스 레이어의 출력값을 나타내고, x는 임의의 이미지를 나타내고, y는 임의의 이미지와 연관된 레이블을 나타내고, k는 임의의 이미지와 연관된 클래스를 나타내고,

은 클래스 n에 대한 학생 기계학습 모델의 특징 출력 레이어의 출력값을 나타낸다.According to an embodiment of the present invention, the step of performing OOD detection using the learned student machine learning model,

and detecting an arbitrary image as an OOD object when the value calculated by is less than or equal to a predetermined threshold. here,

denotes the output value of the softmax layer, x denotes an arbitrary image, y denotes a label associated with an arbitrary image, k denotes a class associated with an arbitrary image,

represents the output value of the feature output layer of the student machine learning model for class n.

According to an embodiment of the present invention, the step of learning the student machine learning model may include using a knowledge distillation loss function and a cross entropy loss function to change the student machine learning model to be similar to the teacher machine learning model. It includes the step of learning.

According to an embodiment of the present invention, the step of training the student machine learning model includes extracting a first feature vector value of a feature output layer of the teacher machine learning model, and a second feature of the feature output layer of the student machine learning model. A step of extracting a vector value and providing a second feature vector value to the projection layer to match the size of the first feature vector value with the size of the second feature vector value.

According to one embodiment of the present invention, the projection layer includes a fully connected layer.

A computer program stored in a computer readable recording medium is provided to execute the method according to an embodiment of the present invention on a computer.

In various embodiments of the present invention, the computing device performs OOD detection using the student machine learning model with improved accuracy without slowing down even when a separate additional parameter is not used to improve the OOD detection performance of the student machine learning model. can do.

In various embodiments of the present invention, unlike methods such as ODIN that require a data set including an unlearned object for tuning of existing hyperparameters, the computing device can perform training without an additional training data set for training a student machine learning model. Detection performance for an image including an OOD object can be effectively improved.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned are clear to those skilled in the art (referred to as "ordinary technicians") from the description of the claims. will be understandable.

BRIEF DESCRIPTION OF THE DRAWINGS Embodiments of the present invention will be described with reference to the accompanying drawings described below, wherein like reference numbers indicate like elements, but are not limited thereto.
1 is a diagram illustrating an example in which a computing device receives an image and generates an OOD detection result according to an embodiment of the present invention.
2 is a functional block diagram showing the internal configuration of a computing device according to an embodiment of the present invention.
3 is a diagram illustrating an example of learning a student machine learning model according to an embodiment of the present invention.
4 is a diagram showing an example of a machine learning model according to an embodiment of the present invention.
5 is a flowchart illustrating an example of an OOD object detection method according to an embodiment of the present invention.
6 is a flowchart illustrating an example of a method for performing knowledge distillation according to an embodiment of the present invention.
7 is a block diagram showing an internal configuration of a computing device according to an embodiment of the present invention.

Hereinafter, specific details for the implementation of the present invention will be described in detail with reference to the accompanying drawings. However, in the following description, if there is a risk of unnecessarily obscuring the gist of the present invention, detailed descriptions of well-known functions or configurations will be omitted.

In the accompanying drawings, identical or corresponding elements are given the same reference numerals. In addition, in the description of the following embodiments, overlapping descriptions of the same or corresponding components may be omitted. However, omission of a description of a component does not intend that such a component is not included in an embodiment.

Advantages and features of the disclosed embodiments, and methods of achieving them, will become apparent with reference to the following embodiments in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and can be implemented in various different forms, only these embodiments make the present invention complete and the scope of the invention to those skilled in the art. It is provided only for complete information.

Terms used in this specification will be briefly described, and the disclosed embodiments will be described in detail. The terms used in this specification have been selected from general terms that are currently widely used as much as possible while considering the functions in the present invention, but these may vary depending on the intention or precedent of a person skilled in the related field, the emergence of new technologies, and the like. In addition, in a specific case, there is also a term arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the description of the invention. Therefore, the term used in the present invention should be defined based on the meaning of the term and the overall content of the present invention, not simply the name of the term.

Expressions in the singular number in this specification include plural expressions unless the context clearly dictates that they are singular. Also, plural expressions include singular expressions unless the context clearly specifies that they are plural. When it is said that a certain part includes a certain component in the entire specification, this means that it may further include other components without excluding other components unless otherwise stated.

In the present invention, the terms "comprise", "comprising" and the like may indicate that features, steps, operations, elements and/or components are present, but may be used when such terms include one or more other functions, It is not excluded that steps, actions, elements, components, and/or combinations thereof may be added.

In the present invention, when a specific element is referred to as being “coupled”, “combined”, “connected”, or “reactive” to any other element, the specific element is directly bonded to, combined with, and/or other elements. or may be linked or reacted, but is not limited thereto. For example, one or more intermediate components may exist between certain components and other components. Also, in the present invention, “and/or” may include each of one or more items listed or a combination of at least a part of one or more items.

In the present invention, terms such as "first" and "second" are used to distinguish a specific component from other components, and the aforementioned components are not limited by these terms. For example, the “first” element may have the same or similar shape as the “second” element.

In the present invention, 'knowledge distillation' may refer to a technique of improving the performance of a small model by transferring the learned knowledge of a large model to a small model. For example, knowledge distillation may be performed using a loss function or the like.

In the present invention, 'loss' and/or 'loss function' may refer to a scale, function, etc. for measuring an error of an object in a machine learning model or the like. A machine learning model or the like may be trained to reduce the error produced by the loss function. For example, the loss function may include a knowledge distillation loss function, a cross entropy loss function, and the like.

1 is a diagram illustrating an example of generating an OOD detection result 120 by receiving an image 110 by a computing device 100 according to an embodiment of the present invention. According to one embodiment, the computing device 100 is a device configured to classify the image 110 into a specific category, a specific class, or perform out-of-distribution (OOD) detection, and one It can be associated with the above machine learning models. Here, the OOD object is an unlearned object that has not been learned in a specific machine learning model, such as semantic shifted OOD (S-OOD), non-semantic shifted OOD (NS-OOD), It may include semantic shift and non-semantic shift OOD (SNS-OOD: semantic shifted + non-semantic shifted OOD). For example, if a trained machine learning model overconfident its own output, it may be difficult to effectively perform OOD detection on unlearned objects.

According to an embodiment, the computing device 100 may improve OOD detection performance by using a plurality of machine learning models. That is, the computing device 100 can greatly improve the OOD object detection accuracy of the student machine learning model by performing knowledge distillation using the learning result of the high-performance teacher machine learning model. Here, knowledge distillation may refer to a technique of improving the performance of a small model by transferring the learned knowledge of a large model to a small model.

According to an embodiment, the computing device 100 may train a teacher machine learning model to perform classification on each image included in the training image set using the training image set. In addition, the computing device 100 may train the teacher machine learning model to perform OOD object detection for another image when an image other than an image included in the training image set is input during learning of the teacher machine learning model. Here, the teacher machine learning model may be a discriminative model used to discriminate data, but is not limited thereto. For example, the computing device 100 uses a training image set including object A (eg, dog) and a training image set including object B (eg, cat) when a specific image is input, the corresponding image A teacher machine learning model that distinguishes whether an object included in is an A object or a B object can be trained. In this case, if an image that does not contain both object A and object B is input to the trained teacher machine learning model, the teacher machine learning model may determine the image and/or an object included in the image as an OOD object. .

After training the teacher machine learning model, the computing device 100 learns a student machine learning model to perform image classification and OOD object detection based on the trained teacher machine learning model using knowledge distillation. can make it Here, the student machine learning model is a small model having fewer parameters than the teacher machine learning model, and may be, for example, a model that receives a low-resolution image and classifies an object. Also, the student machine learning model may be a discrimination model used to discriminate data, but is not limited thereto. That is, the computing device 100 may train the student machine learning model to mimic the output of the teacher machine learning model through knowledge distillation.

Then, when an arbitrary image 110 is input, the computing device 100 may perform OOD detection using the learned student machine learning model and output an OOD detection result 120 . With this configuration, the computing device 100 detects OOD using the student machine learning model with improved accuracy without slowing down even when a separate additional parameter is not used to improve the OOD detection performance of the student machine learning model. can be performed.

2 is a functional block diagram showing the internal configuration of the computing device 100 according to an embodiment of the present invention. As shown, the computing device 100 may include a machine learning model learning unit 210, a knowledge distillation performing unit 220, an OOD detection unit 230, and the like, but is not limited thereto. The computing device 100 communicates with external devices, systems, etc., and may exchange data and/or information necessary for OOD detection and the like.

As described above, the machine learning model learning unit 210 may train a teacher machine learning model to classify each image included in the training image set using the training image set. In addition, the machine learning model learning unit 210 trains the teacher machine learning model to perform OOD object detection for other images when an image other than the image included in the training image set is input during learning of the teacher machine learning model. can For example, the machine learning model learning unit 210 may train a teacher machine learning model to classify K classes. In this case, when a given image is provided, the trained machine learning model may determine whether the corresponding image belongs to any one of K classes or not to all classes included in K classes.

According to an embodiment, the machine learning model learning unit 210 may train a teacher machine learning model to perform OOD object detection using maximum softmax probability (MSP). Here, MSP may be an OOD detection method that calculates the probability that a given image belongs to each class and determines the image as an unlearned object when the largest value of the calculated probability is less than a predetermined value. Additionally or alternatively, the machine learning model learning unit 210 may train a teacher machine learning model to perform OOD object detection using an out-of-distribution detector for neural networks (ODIN). For example, ODIN smooths the output distribution representing the probability that a given image belongs to each class, and if the largest value among the smoothed values is less than a predetermined value, OOD determines the image as an unlearned object. It can be a detection method. Additionally or alternatively, the machine learning model learning unit 210 may train a teacher machine learning model to perform OOD object detection using a Mahalanobis distance based OOD detector (MAHA). Here, the MAHA pre-calculates the feature distribution of the model for each learned class, and when an image is input, if the mahalanobis distance to the feature distribution is greater than a predetermined value, the corresponding image is automatically displayed. It may be an OOD detection method that determines a learning object.

The knowledge distillation performer 220 may train a student machine learning model to perform image classification and OOD object detection based on the teacher machine learning model learned using knowledge distillation. For example, the knowledge distillation unit 220 extracts a first feature vector value of a penultimate layer (eg, a hint layer) of a teacher machine learning model and features of a student machine learning model. After extracting the second feature vector value of the output layer (eg, a guided layer), knowledge distillation may be performed so that the second feature vector value mimics the first feature vector value.

According to an embodiment, the size of the first feature vector value and the second feature vector value may be different. In this case, the knowledge distillation unit 220 provides the second feature vector value to the projection layer, matches the size of the first feature vector value with the size of the second feature vector value, and matches the size of the second feature vector value. Knowledge distillation can be performed so that the values mimic the first feature vector values. Here, the projection layer may include a fully connected layer.

According to an embodiment, the knowledge distillation unit 220 may perform knowledge distillation using two loss functions. For example, the knowledge distillation unit 220 may include a knowledge distillation loss function (KD loss) for changing a student machine learning model to be similar to a teacher machine learning model and a cross entropy loss function (CE) as a common loss. loss: cross entropy loss function) to train a student machine learning model. Using these two loss functions, the final loss can be calculated as in Equation 1 below.

은 최종 손실이며,

는 지식 증류 손실이고,

는 크로스 엔트로피 손실일 수 있다. 또한,

및

는 각각 지식 증류 손실 함수와 크로스 엔트로피 손실 함수의 가중치일 수 있다. 또한, 지식 증류 손실과 크로스 엔트로피 손실은 다음의 수학식 2 및 3과 같이 산출될 수 있다.here,

is the final loss,

is the knowledge distillation loss,

may be the cross entropy loss. also,

and

may be weights of the knowledge distillation loss function and the cross entropy loss function, respectively. In addition, knowledge distillation loss and cross entropy loss can be calculated as in Equations 2 and 3 below.

는 교사 기계학습 모델의 특징 출력 레이어의 평균 풀링 출력(average pooled output)과 학생 기계학습 모델의 특징 출력 레이어의 투사된 평균 풀링 출력(projected average pooled output) 사이의 L2 거리(L2 distance)를 나타낼 수 있다.here,

can represent the L2 distance between the average pooled output of the feature output layer of the teacher machine learning model and the projected average pooled output of the feature output layer of the student machine learning model. there is.

는 정답(ground truth) t와

의 확률 분포 사이의 크로스 엔트로피 손실일 수 있다.here,

is the ground truth t and

It may be a cross entropy loss between probability distributions of .

The OOD detection unit 230 may perform OOD detection using the student machine learning model learned through knowledge distillation. For example, the OOD detection unit 230 and/or the student machine learning model may perform OOD detection by Equation 4 below.

는 소프트맥스 레이어(softmax layer)의 출력값을 나타내고, x는 임의의 이미지를 나타내고, y는 임의의 이미지와 연관된 레이블(label)을 나타내고, k는 임의의 이미지와 연관된 클래스(class)를 나타내고,

은 클래스 n에 대한 학생 기계학습 모델의 특징 출력 레이어(예: 마지막에서 두번째 레이어)의 출력값을 나타낼 수 있다.That is, the OOD detection unit 230 may detect an input image as an OOD object when the value calculated by Equation 4 is less than or equal to a predetermined threshold value. here,

Represents the output value of the softmax layer, x represents an arbitrary image, y represents a label associated with an arbitrary image, k represents a class associated with an arbitrary image,

may represent the output value of the feature output layer (eg, the second to last layer) of the student machine learning model for class n.

In FIG. 2 , each functional configuration included in the computing device 100 has been separately described, but this is only to aid understanding of the present invention, and one computing device may perform two or more functions. With this configuration, the computing device 100 provides additional learning data for training the student machine learning model, unlike conventional methods such as ODIN that require a data set including unlearned objects for hyperparameter tuning. Even without a set, detection performance for images containing OOD objects can be effectively improved.

3 is a diagram showing an example of learning a student machine learning model 340 according to an embodiment of the present invention. According to an embodiment, the teacher machine learning model 320 and the student machine learning model 340 may include a plurality of convolution blocks. For example, the teacher machine learning model 320 and the student machine learning model 340 may receive an image set 310 and be trained to classify each image included in the image set 310 into a class. there is. In the illustrated example, the teacher machine learning model 320 and the student machine learning model 340 are shown to be trained based on the same image set 310, but are not limited thereto, and may be trained using different image sets. there is.

According to an embodiment, an output value (eg, a feature vector value) of the student machine learning model 340 may be trained to be similar to an output value of the teacher machine learning model 320 . In this case, the output value of the feature output layer (eg, the second to last layer) of the student machine learning model 340 and the output value of the feature output layer of the teacher machine learning model 320 may be used. For example, the output value of the feature output layer of the student machine learning model 340 may be provided to the projection layer 350 and changed to have the same size as the output value of the feature output layer of the teacher machine learning model 320 . Then, the output of the student machine learning model 340 can be trained by the KD loss function 330 to be similar to the output of the teacher machine learning model 320. In this way, when knowledge distillation is performed by the KD loss function 330, the performance of the cross entropy loss function can be improved by widening the angle margin.

According to an embodiment, an output value of the student machine learning model 340 may be provided as a cross entropy loss function 380 through a classification layer 360. In this case, the output value of the student machine learning model 340 may be learned by the cross entropy loss function 380 so that a label 370 corresponding to each image included in the image set 310 is determined.

4 is a diagram showing an example of a machine learning model 400 according to an embodiment of the present invention. As noted above, the machine learning model 400 may include a teacher machine learning model and/or a student machine learning model. For example, the machine learning model 400 may be a discriminative model used to discriminate data, but is not limited thereto.

In the illustrated example, the first image 410 may be an image associated with object A (eg, a dog) included in the training image set, and the second image 420 may be associated with object C not included in the training image set. It can be an image. According to an embodiment, when the first image 410 is input, the machine learning model 400 may output a first classification result 412 for classifying the first image 410 as object A. Additionally, when the second image 420 is input, the machine learning model 400 may output a second classification result 422 classifying the second image 420 as an unlearned object (OOD object).

According to an embodiment, when each of the images 410 and 420 is input, the machine learning model 400 may calculate a probability value corresponding to each object from which the images 410 and 420 are learned. In this case, the machine learning model 400 may classify the corresponding image into a class having the highest calculated value. Additionally, the machine learning model 400 may detect an arbitrary image as an OOD object when the highest value among the calculated values is less than or equal to a predetermined threshold value.

5 is a flowchart illustrating an example of an OOD object detection method 500 according to an embodiment of the present invention. The OOD object detection method 500 may be performed by a processor (eg, at least one processor of a computing device). As shown, the OOD object detection method 500 may be initiated by a processor training a teacher machine learning model to perform classification on each image included in the training image set using the training image set (S510). .

The processor may train the teacher machine learning model to perform OOD object detection for another image when an image other than the image included in the training image set is input during training of the teacher machine learning model (S520). In addition, the processor may train the student machine learning model to perform image classification and OOD object detection based on the teacher machine learning model learned using knowledge distillation (S530). In this case, the processor may train the student machine learning model using a knowledge distillation loss function and a cross entropy loss function for changing the student machine learning model to be similar to the teacher machine learning model. Then, when an arbitrary image is input, the processor may perform OOD detection using the trained student machine learning model.

6 is a flow diagram illustrating an example of a method 600 for performing knowledge distillation according to one embodiment of the present invention. Method 600 for performing knowledge distillation may be performed by a processor (eg, at least one processor of a computing device). As shown, the knowledge distillation method 600 may be initiated by a processor extracting a first feature vector value of a feature output layer of a teacher machine learning model (S610).

The processor may extract the second feature vector value of the feature output layer of the student machine learning model (S620). In addition, the processor may provide the second feature vector value to the projection layer to match the size of the first feature vector value with the size of the second feature vector value. Here, the projection layer may include a fully connected layer. Then, the processor can train the student machine learning model using the knowledge distillation loss function and the cross entropy loss function.

7 is a block diagram showing an internal configuration of a computing device 700 according to an embodiment of the present invention. The computing device 700 may include a memory 710 , a processor 720 , a communication module 730 and an input/output interface 740 . As shown in FIG. 7 , the computing device 700 may be configured to communicate information and/or data over a network using a communication module 730 .

Memory 710 may include any non-transitory computer readable storage medium. According to one embodiment, the memory 710 is a non-perishable mass storage device (permanent mass storage device) such as random access memory (RAM), read only memory (ROM), disk drive, solid state drive (SSD), flash memory, and the like. mass storage device). As another example, a non-perishable mass storage device such as a ROM, SSD, flash memory, or disk drive may be included in the computing device 700 as a separate permanent storage device separate from memory. Also, an operating system and at least one program code may be stored in the memory 710 .

These software components may be loaded from a computer-readable recording medium separate from the memory 710 . A recording medium readable by such a separate computer may include a recording medium directly connectable to the computing device 700, for example, a floppy drive, a disk, a tape, a DVD/CD-ROM drive, a memory card, and the like. It may include a computer-readable recording medium. As another example, software components may be loaded into the memory 710 through the communication module 730 rather than a computer-readable recording medium. For example, at least one program may be loaded into the memory 710 based on a computer program installed by files provided by developers or a file distribution system that distributes application installation files through the communication module 730. can

The processor 720 may be configured to process commands of a computer program by performing basic arithmetic, logic, and input/output operations. Commands may be provided to a user terminal (not shown) or other external system by the memory 710 or the communication module 730 .

The communication module 730 may provide a configuration or function for a user terminal (not shown) and the computing device 700 to communicate with each other through a network, and the computing device 700 may provide an external system (for example, a separate cloud system). etc.) may provide a configuration or function to communicate with. For example, control signals, commands, data, etc. provided under the control of the processor 720 of the computing device 700 are transmitted through the communication module 730 and the network to the user terminal and/or to the user terminal through the communication module of the external system. and/or transmitted to an external system.

Also, the input/output interface 740 of the computing device 700 may be connected to the computing device 700 or may be a means for interface with a device (not shown) for input or output that may be included in the computing device 700. . In FIG. 7 , the input/output interface 740 is illustrated as an element separately configured from the processor 720 , but is not limited thereto, and the input/output interface 740 may be included in the processor 720 . Computing device 700 may include many more components than those of FIG. 7 . However, there is no need to clearly show most of the prior art components.

The processor 720 of the computing device 700 may be configured to manage, process, and/or store information and/or data received from a plurality of user terminals and/or a plurality of external systems.

The above-described methods and/or various embodiments may be realized with digital electronic circuits, computer hardware, firmware, software, and/or combinations thereof. Various embodiments of the present invention may be performed by a data processing device, eg, one or more programmable processors and/or one or more computing devices, or as a computer readable recording medium and/or a computer program stored on a computer readable recording medium. can be implemented The above-described computer programs may be written in any form of programming language, including compiled or interpreted languages, and may be distributed in any form, such as a stand-alone program, module, or subroutine. A computer program may be distributed over one computing device, multiple computing devices connected through the same network, and/or distributed over multiple computing devices connected through multiple different networks.

The methods and/or various embodiments described above may be performed by one or more processors configured to execute one or more computer programs that process, store, and/or manage any function, function, or the like, by operating on input data or generating output data. can be performed by For example, the method and/or various embodiments of the present invention may be performed by a special purpose logic circuit such as a Field Programmable Gate Array (FPGA) or an Application Specific Integrated Circuit (ASIC), and the method and/or various embodiments of the present invention may be performed. Apparatus and/or systems for performing the embodiments may be implemented as special purpose logic circuits such as FPGAs or ASICs.

The one or more processors executing the computer program may include a general purpose or special purpose microprocessor and/or one or more processors of any kind of digital computing device. The processor may receive instructions and/or data from each of the read-only memory and the random access memory, or receive instructions and/or data from the read-only memory and the random access memory. In the present invention, components of a computing device performing methods and/or embodiments may include one or more processors for executing instructions, and one or more memory devices for storing instructions and/or data.

According to one embodiment, a computing device may exchange data with one or more mass storage devices for storing data. For example, a computing device may receive/receive data from and transfer data to a magnetic or optical disc. A computer-readable storage medium suitable for storing instructions and/or data associated with a computer program includes semiconductor memory devices such as Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable PROM (EEPROM), and flash memory devices. Any type of non-volatile memory may be included, but is not limited thereto. For example, computer readable storage media may include magnetic disks such as internal hard disks or removable disks, magneto-optical disks, CD-ROM and DVD-ROM disks.

To provide interaction with a user, a computing device includes a display device (eg, a cathode ray tube (CRT), a liquid crystal display (LCD), etc.) It may include a pointing device (eg, a keyboard, mouse, trackball, etc.) capable of providing input and/or commands to, but is not limited thereto. That is, the computing device may further include any other type of device for providing interaction with a user. For example, a computing device may provide any form of sensory feedback to a user for interaction with the user, including visual feedback, auditory feedback, and/or tactile feedback. In this regard, the user may provide input to the computing device through various gestures such as visual, voice, and motion.

In the present invention, various embodiments may be implemented in a computing system including a back-end component (eg, a data server), a middleware component (eg, an application server), and/or a front-end component. In this case, the components may be interconnected by any form or medium of digital data communication, such as a communication network. For example, the communication network may include a local area network (LAN), a wide area network (WAN), and the like.

A computing device based on the example embodiments described herein may be implemented using hardware and/or software configured to interact with a user, including a user device, user interface (UI) device, user terminal, or client device. can For example, the computing device may include a portable computing device such as a laptop computer. Additionally or alternatively, the computing device may include personal digital assistants (PDAs), tablet PCs, game consoles, wearable devices, internet of things (IoT) devices, virtual reality (VR) devices, AR (augmented reality) device, etc. may be included, but is not limited thereto. A computing device may further include other types of devices configured to interact with a user. Further, the computing device may include a portable communication device (eg, a mobile phone, smart phone, wireless cellular phone, etc.) suitable for wireless communication over a network, such as a mobile communication network. A computing device communicates wirelessly with a network server using wireless communication technologies and/or protocols such as radio frequency (RF), microwave frequency (MWF) and/or infrared ray frequency (IRF). It can be configured to communicate with.

The various embodiments herein, including specific structural and functional details, are exemplary. Accordingly, embodiments of the present invention are not limited to those described above and may be implemented in various other forms. In addition, terms used in the present invention are for describing some embodiments and are not construed as limiting the embodiments. For example, the singular and the above may be construed to include the plural as well, unless the context clearly dictates otherwise.

In the present invention, unless defined otherwise, all terms used in this specification, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which such concept belongs. . In addition, terms commonly used, such as terms defined in a dictionary, should be interpreted as having a meaning consistent with the meaning in the context of the related art.

Although the present invention has been described in relation to some embodiments in this specification, various modifications and changes can be made without departing from the scope of the present invention that can be understood by those skilled in the art. Moreover, such modifications and variations are intended to fall within the scope of the claims appended hereto.

100: computing device
110: image
120: OOD detection result

Claims (7)

As an out of distribution (OOD) object detection method performed by at least one processor,
training a teacher machine learning model to perform classification for each image included in the training image set using a training image set;
training the teacher machine learning model to perform OOD object detection for the other image when an image other than the image included in the training image set is input during learning of the teacher machine learning model; and
training a student machine learning model to perform image classification and OOD object detection based on the trained teacher machine learning model using knowledge distillation;
Including, OOD object detection method.
According to claim 1,
If an arbitrary image is input, performing OOD detection using the learned student machine learning model;
Further comprising, OOD object detection method.
According to claim 2,
The step of performing OOD detection using the learned student machine learning model,

When the value calculated by is less than a predetermined threshold value, detecting the arbitrary image as an OOD object,
here,

represents the output value of the softmax layer, x represents the arbitrary image, y represents a label associated with the arbitrary image, and k represents a class associated with the arbitrary image indicate,

Represents an output value of a penultimate layer of the student machine learning model for class n, OOD object detection method.
According to claim 1,
The step of learning the student machine learning model,
Training the student machine learning model using a knowledge distillation loss function and a cross entropy loss function for changing the student machine learning model to be similar to the teacher machine learning model. ;
Including, OOD object detection method.
According to claim 1,
The step of learning the student machine learning model,
extracting a first feature vector value of a penultimate layer of the teacher machine learning model;
extracting a second feature vector value of a feature output layer of the student machine learning model; and
providing the second feature vector value to a projection layer to match the size of the first feature vector value with the size of the second feature vector value;
Including, OOD object detection method.
According to claim 5,
The projection layer includes a fully connected layer, OOD object detection method.
A computer program stored in a computer-readable recording medium to execute the method according to any one of claims 1 to 6 on a computer.

Images