KR20230029470A - Method and device for performing a task with a deep learning model for an abnormal behavior detection - Google Patents

Abstract

According to an embodiment of the present disclosure for realizing the above object, disclosed is a method for performing a task through a deep learning model implemented by a computing device including at least one processor. The method includes the steps of: determining a use stage of the deep learning model; and processing input data through the deep learning model according to the determined use stage. The deep learning model may include a plurality of convolutional layers, and the deep learning model may adjust a usage ratio of parameters included in each of the plurality of convolutional layers according to the use stage.

Description

Method and apparatus for performing a task using a deep learning model for detecting abnormal behavior in images

The present invention relates to machine learning for image analysis, and more particularly, to a method and apparatus for performing a task through a deep learning model capable of variably processing input data according to a plurality of use stages.

When a person directly checks whether abnormal behavior (assault, fall, etc.) occurs while watching CCTV images, it takes a lot of time. Accordingly, many machine learning studies are being conducted to develop algorithms that can automatically detect abnormal behaviors, and currently, most of the abnormal behavior detection studies are being conducted using convolutional networks that exhibit superior performance. In addition, studies on abnormal behavior detection using a lightweight network that can be used on mobile devices that can utilize real-time applications but have relatively small computational capacity are also being conducted.

In general, a lightweight network has a fast detection speed, but has a disadvantage in that it has relatively low performance compared to a large network. In order to solve the trade-off between inference speed and network performance, a method of including several convolutional structures in one network has been used in previous studies. For example, there has been a study that allows users to select and use a network with good performance and a network with a fast inference speed according to the desired situation. However, these existing studies do not suggest network design and learning methods for 4-dimensional image data. In particular, since 3D convolution used to process images requires much more operations than 2D convolution used to process images, 4D data cannot be processed in image processing. A lightweight network is very important.

A lightweight network has a fast inference speed but a problem with relatively low performance. There is a need for a network model to mitigate this performance degradation problem.

An object of the present disclosure is to provide a method and apparatus for performing a task through a deep learning model capable of variably processing input data according to a plurality of use steps.

According to an embodiment of the present disclosure for realizing the above object, a method for performing a task through a deep learning model implemented by a computing device including at least one processor is disclosed. The method includes determining a step of using a deep learning model; processing input data through the deep learning model according to the determined use stage; Including, the deep learning model includes a plurality of convolutional layers, the deep learning model can adjust the use ratio of parameters included in each of the plurality of convolutional layers according to the use stage.

According to an alternative embodiment, determining the use phase of the deep learning model may include: measuring GPU memory of the computing device to determine device memory usage; determining a ratio of the device memory usage and base memory usage; and determining a use step of the deep learning model based on the determined ratio.

According to an alternative embodiment, the base memory usage may be the GPU memory usage used when running the base network.

According to an alternative embodiment, the using step includes a first using step and a second using step, and the number of parameters used in the plurality of convolutional layers according to the second using step depends on the first using step. It may be more than the number of parameters used in the plurality of convolution layers.

According to an alternative embodiment, parameters used in the plurality of convolution layers according to the second use step may include parameters used in the plurality of convolution layers according to the first use step.

According to an alternative embodiment, the using step further includes a third using step, and the number of parameters used in the plurality of convolutional layers according to the third using step depends on the plurality of convolution layers according to the second using step. It can be more than the number of parameters used in the solution layer.

According to an alternative embodiment, parameters used in the plurality of convolution layers according to the third use step may include parameters used in the plurality of convolution layers according to the second use step.

According to an alternative embodiment, the number of parameters used in at least some of the plurality of convolutional layers according to the second use step is the number of parameters used in at least some of the plurality of convolutional layers according to the first use step. It may be n _- ² times of .

According to an alternative embodiment, the task may be to detect an abnormal behavior of an object included in an image that is the input data.

According to another embodiment of the present disclosure for realizing the above object, a computing device that performs a task through a deep learning model is disclosed. The computing device includes a processor including at least one core; a memory containing program codes executable by the processor; And a network unit for receiving image input data; includes,

The processor: determines a use step of a deep learning model, and processes input data through the deep learning model according to the determined use step, the deep learning model includes a plurality of convolution layers, and the deep learning model includes a plurality of convolutional layers. The model may adjust the use ratio of the parameters included in each of the plurality of convolution layers according to the use stage.

A computer program stored in a computer readable storage medium is disclosed according to another embodiment of the present disclosure for realizing the above object. When the computer program is executed on one or more processors, the following operations of performing a task through a deep learning model are performed, and the operations include: determining a use stage of the deep learning model; and processing input data through the deep learning model according to the determined use stage. Including, the deep learning model includes a plurality of convolutional layers, the deep learning model can adjust the use ratio of parameters included in each of the plurality of convolutional layers according to the use stage.

The present disclosure may provide a method and apparatus for performing a task through a deep learning model capable of variably processing input data according to a plurality of use steps.

1 is a block diagram of a computing device for performing a task through a deep learning model according to an embodiment of the present disclosure.
2 is a flowchart of a method of performing a task through a deep learning model according to some embodiments of the present disclosure.
3 is a diagram for explaining a deep learning model for performing a task according to some embodiments of the present disclosure.
4 is a diagram for explaining parameters included in a plurality of convolution layers according to some embodiments of the present disclosure.
5 is another diagram for describing parameters included in a plurality of convolution layers according to some embodiments of the present disclosure.
6 is a simplified and general schematic diagram of an exemplary computing environment in which embodiments of the present disclosure may be implemented.

Various embodiments are now described with reference to the drawings. In this specification, various descriptions are presented to provide an understanding of the present disclosure. However, it is apparent that these embodiments may be practiced without these specific details.

The terms “component,” “module,” “system,” and the like, as used herein, refer to a computer-related entity, hardware, firmware, software, a combination of software and hardware, or an execution of software. For example, a component may be, but is not limited to, a procedure, processor, object, thread of execution, program, and/or computer running on a processor. For example, both an application running on a computing device and a computing device may be components. One or more components may reside within a processor and/or thread of execution. A component can be localized within a single computer. A component may be distributed between two or more computers. Also, these components can execute from various computer readable media having various data structures stored thereon. Components may be connected, for example, via signals with one or more packets of data (e.g., data and/or signals from one component interacting with another component in a local system, distributed system) to other systems and over a network such as the Internet. data being transmitted) may communicate via local and/or remote processes.

In addition, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or”. That is, unless otherwise specified or clear from the context, “X employs A or B” is intended to mean one of the natural inclusive substitutions. That is, X uses A; X uses B; Or, if X uses both A and B, "X uses either A or B" may apply to either of these cases. Also, the term "and/or" as used herein should be understood to refer to and include all possible combinations of one or more of the listed related items.

Also, the terms "comprises" and/or "comprising" should be understood to mean that the features and/or components are present. However, it should be understood that the terms "comprises" and/or "comprising" do not exclude the presence or addition of one or more other features, elements, and/or groups thereof. Also, unless otherwise specified or where the context clearly indicates that a singular form is indicated, the singular in this specification and claims should generally be construed to mean "one or more".

In addition, the term “at least one of A or B” should be interpreted as meaning “when only A is included”, “when only B is included”, and “when A and B are combined”.

Those skilled in the art will further understand that the various illustrative logical blocks, components, modules, circuits, means, logics, and algorithm steps described in connection with the embodiments disclosed herein may be implemented using electronic hardware, computer software, or combinations of both. It should be recognized that it can be implemented as To clearly illustrate the interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented in hardware or as software depends on the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application. However, such implementation decisions should not be interpreted as causing a departure from the scope of this disclosure.

The description of the presented embodiments is provided to enable any person skilled in the art to use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art of this disclosure. The general principles defined herein may be applied to other embodiments without departing from the scope of this disclosure. Thus, the present invention is not limited to the embodiments presented herein. The present invention is to be accorded the widest scope consistent with the principles and novel features set forth herein.

In the present disclosure, network functions, artificial neural networks, and neural networks may be used interchangeably.

On the other hand, the term "image" or "image data" used throughout the description and claims of the present invention refers to multidimensional data composed of discrete image elements (e.g., pixels in a two-dimensional image), in other words. , is a term that refers to a visible object (e.g., displayed on a video screen) or a digital representation of that object.

Throughout this specification, deep learning = model, network model, computational model, neural network, network function, and neural network may be used interchangeably. A neural network may consist of a set of interconnected computational units, which may generally be referred to as nodes. These nodes may also be referred to as neurons. A neural network includes one or more nodes. Nodes (or neurons) constituting neural networks may be interconnected by one or more links. A node may be composed of a combination of channels, filters, and nonlinear functions.

A set of parameters (or weights) included in the convolution operation may exist as many as the number of times the dimension of the input data multiplied by the number of input channels and the number of output channels. In this case, the entire convolution operation may be interpreted as the number of convolution operations of one input channel to one output channel multiplied by the number of channels and the number of output channels. A convolution operation for one input channel and one output channel may be defined as unit convolution. Unit convolution may refer to a minimum set of operations among all convolution operations. A unit convolution may have at least one value obtained/changed through a convolution operation. Here, a set of at least one value processed in unit convolution may be referred to as a parameter of a convolution layer. The number of weights included in unit convolutions derived from the same convolution may all be the same. Here, unit convolution can be used in the same sense as node.

In a neural network, one or more nodes connected through a link may form a relative relationship of an input node and an output node. The concept of an input node and an output node is relative, and any node in an output node relationship with one node may have an input node relationship with another node, and vice versa. As described above, an input node to output node relationship may be created around a link. More than one output node can be connected to one input node through a link, and vice versa.

In a relationship between an input node and an output node connected through one link, the value of data of the output node may be determined based on data input to the input node. Here, a link interconnecting an input node and an output node may have a weight. The weight may be variable, and may be changed by a user or an algorithm in order to perform a function desired by the neural network. For example, when one or more input nodes are interconnected by respective links to one output node, the output node is set to a link corresponding to values input to input nodes connected to the output node and respective input nodes. An output node value may be determined based on the weight.

As described above, in the neural network, one or more nodes are interconnected through one or more links to form an input node and output node relationship in the neural network. Characteristics of the neural network may be determined according to the number of nodes and links in the neural network, an association between the nodes and links, and a weight value assigned to each link. For example, when there are two neural networks having the same number of nodes and links and different weight values of the links, the two neural networks may be recognized as different from each other.

A neural network may be composed of a set of one or more nodes. A subset of nodes constituting a neural network may constitute a layer. Some of the nodes constituting the neural network may form one layer based on distances from the first input node. For example, a set of nodes having a distance of n from the first input node may constitute n layers. The distance from the first input node may be defined by the minimum number of links that must be passed through to reach the corresponding node from the first input node. However, the definition of such a layer is arbitrary for explanation, and the order of a layer in a neural network may be defined in a method different from the above. For example, a layer of nodes may be defined by a distance from a final output node.

An initial input node may refer to one or more nodes to which data is directly input without going through a link in relation to other nodes among nodes in the neural network. Alternatively, in a relationship between nodes based on a link in a neural network, it may mean nodes that do not have other input nodes connected by a link. Similarly, the final output node may refer to one or more nodes that do not have an output node in relation to other nodes among nodes in the neural network. Also, the hidden node may refer to nodes constituting the neural network other than the first input node and the last output node.

In the neural network according to an embodiment of the present disclosure, the number of nodes in the input layer may be the same as the number of nodes in the output layer, and the number of nodes decreases and then increases again as the number of nodes progresses from the input layer to the hidden layer. can In addition, the neural network according to another embodiment of the present disclosure may be a neural network in which the number of nodes of the input layer may be less than the number of nodes of the output layer and the number of nodes decreases as the number of nodes increases from the input layer to the hidden layer. there is. In addition, the neural network according to another embodiment of the present disclosure is a neural network in which the number of nodes in the input layer may be greater than the number of nodes in the output layer, and the number of nodes increases as the number of nodes increases from the input layer to the hidden layer. can A neural network according to another embodiment of the present disclosure may be a neural network in the form of a combination of the aforementioned neural networks.

A deep neural network (DNN) may refer to a neural network including a plurality of hidden layers in addition to an input layer and an output layer. Deep neural networks can reveal latent structures in data. In other words, it can identify the latent structure of a photo, text, video, sound, or music (e.g., what objects are in the photo, what the content and emotion of the text are, what the content and emotion of the audio are, etc.). . Deep neural networks include convolutional neural networks (CNNs), recurrent neural networks (RNNs), auto encoders, generative adversarial networks (GANs), and restricted boltzmann machines (RBMs). machine), a deep belief network (DBN), a Q network, a U network, a Siamese network, a Generative Adversarial Network (GAN), and the like. The description of the deep neural network described above is only an example, and the present disclosure is not limited thereto.

The neural network may be trained using at least one of supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning. Learning of the neural network may be a process of applying knowledge for the neural network to perform a specific operation to the neural network.

A neural network can be trained in a way that minimizes output errors. In the learning of the neural network, the learning data is repeatedly input into the neural network, the output of the neural network for the training data and the error of the target are calculated, and the error of the neural network is transferred from the output layer of the neural network to the input layer in the direction of reducing the error. It is a process of updating the weight of each node of the neural network by backpropagating in the same direction. In the case of teacher learning, the learning data in which the correct answer is labeled is used for each learning data (ie, the labeled learning data), and in the case of comparative teacher learning, the correct answer may not be labeled in each learning data. That is, for example, learning data in the case of teacher learning regarding data classification may be data in which each learning data is labeled with a category. Labeled training data is input to a neural network, and an error may be calculated by comparing an output (category) of the neural network and a label of the training data. As another example, in the case of comparative history learning for data classification, an error may be calculated by comparing input learning data with a neural network output. The calculated error is back-propagated in a reverse direction (ie, from the output layer to the input layer) in the neural network, and the connection weight of each node of each layer of the neural network may be updated according to the back-propagation. The amount of change in the connection weight of each updated node may be determined according to a learning rate. The neural network's computation of input data and backpropagation of errors can constitute a learning cycle (epoch). The learning rate may be applied differently according to the number of iterations of the learning cycle of the neural network. For example, a high learning rate may be used in the early stage of neural network training to increase efficiency by allowing the neural network to quickly obtain a certain level of performance, and a low learning rate may be used in the late stage to increase accuracy.

In neural network learning, generally, training data can be a subset of real data (ie, data to be processed using the trained neural network), and therefore, errors for training data are reduced, but errors for real data are reduced. There may be incremental learning cycles. Overfitting is a phenomenon in which errors for actual data increase due to excessive learning on training data. For example, a phenomenon in which a neural network that has learned a cat by showing a yellow cat does not recognize that it is a cat when it sees a cat other than yellow may be a type of overfitting. Overfitting can act as a cause of increasing the error of machine learning algorithms. Various optimization methods can be used to prevent such overfitting. To prevent overfitting, methods such as increasing the training data, regularization, inactivating some nodes in the network during learning, and using a batch normalization layer should be applied. can

1 is a block diagram of a computing device for performing a task through a deep learning model according to an embodiment of the present disclosure.

The configuration of the computing device 100 shown in FIG. 1 is only a simplified example. In one embodiment of the present disclosure, the computing device 100 may include other components for performing a computing environment of the computing device 100, and only some of the disclosed components may constitute the computing device 100.

The computing device 100 may include a processor 110 , a memory 130 , and a network unit 150 .

The processor 110 may include one or more cores, and includes a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), and a tensor processing unit (TPU) of a computing device. unit), data analysis, and processors for deep learning. The processor 110 may read a computer program stored in the memory 130 and process data for machine learning according to an embodiment of the present disclosure. According to an embodiment of the present disclosure, the processor 110 may perform an operation for learning a neural network. The processor 110 is used for neural network learning, such as processing input data for learning in deep learning (DL), extracting features from input data, calculating errors, and updating neural network weights using backpropagation. calculations can be performed. At least one of the CPU, GPGPU, and TPU of the processor 110 may process learning of the network function. For example, the CPU and GPGPU can process learning of network functions and specific tasks using network functions. In addition, in one embodiment of the present disclosure, learning of a network function and a specific task using a network function may be processed by using processors of a plurality of computing devices together. In addition, a computer program executed in a computing device according to an embodiment of the present disclosure may be a CPU, GPGPU or TPU executable program.

According to an embodiment of the present disclosure, the processor 110 may perform a predetermined task based on an image including at least one object of interest using a pretrained deep learning model. Here, the task may be understood as a task of predicting or determining a class related to a state, attribute, or characteristic of a target of interest. For example, the processor 110 may detect an abnormal behavior of a target of interest included in an image by using a pretrained deep learning model.

The deep learning model used by the processor 110 extracts a feature vector based on image input data including at least one object of interest, and estimates a probability value corresponding to a predetermined task based on the derived feature vector. models can be included. In this case, the deep learning model may include a convolution neural network optimized for extracting a feature vector from an image. Here, the optimized convolutional neural network may be a 3D convolutional neural network for “horizontal values,” “vertical values,” and “RGAB” channel “values” of the video (or “image). The deep learning model of the present disclosure may adjust the number of parameters used in the convolutional neural network according to one of a plurality of use steps. In addition, the deep learning model may include a fully-connected neural network that is connected to the convolutional neural network to generate an output value suitable for the task. For example, a deep learning model is a convolutional neural network that extracts a feature vector based on an image including at least one object of interest and a complete probability value for deviant behavior based on a feature vector derived from the convolutional neural network. -Can include a connecting neural network. In addition, “deep learning” models can “process” “values” output from each of “plural” convolutional “layers” through “non-linear” functions (for example, “RACE” functions). In addition to the description above, various types of neural networks suitable for classification tasks in the computer vision field may be applied to the neural network included in the deep learning model of the present disclosure.

According to an embodiment of the present disclosure, the memory 130 may store any type of information generated or determined by the processor 110 and any type of information received by the network unit 150 .

According to an embodiment of the present disclosure, the memory 130 is a flash memory type, a hard disk type, a multimedia card micro type, or a card type memory (eg, SD or XD memory, etc.), RAM (Random Access Memory, RAM), SRAM (Static Random Access Memory), ROM (Read-Only Memory, ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory) -Only Memory), a magnetic memory, a magnetic disk, and an optical disk may include at least one type of storage medium. The computing device 100 may operate in relation to a web storage that performs a storage function of the memory 130 on the Internet. The above description of the memory is only an example, and the present disclosure is not limited thereto.

The network unit 150 according to an embodiment of the present disclosure may use any type of known wired or wireless communication system.

The network unit 150 may receive an image expressing a target of interest from an external system. For example, the network unit 150 may receive an image of a CCTV installed at the entrance of a building from a CCTV image transmission system. The CCTV image may be data for learning or inference of a neural network model to be learned as a 2D feature or a 3D feature. Here, the CCTV image may be an image of several people passing through the building entrance. The image expressing the object of interest is not limited to the above example, and may include all images expressing the object of interest acquired through various photographing devices.

In addition, the network unit 150 may transmit and receive information processed by the processor 110, a user interface, and the like through communication with other terminals. For example, the network unit 150 may provide a user interface generated by the processor 100 to a client (eg, a user terminal). In addition, the network unit 150 may receive an external input of a user authorized as a client and transmit it to the processor 110 . In this case, the processor 100 may process operations such as outputting, correcting, changing, adding, and the like of information provided through the user interface based on the user's external input received from the network unit 150 .

Meanwhile, the computing device 100 according to an embodiment of the present disclosure may include a server as a computing system that transmits and receives information through communication with a client. In this case, the client may be any type of terminal capable of accessing the server. For example, the computing device 100 serving as a server may receive an image expressing a target of interest from an image capturing terminal, detect a specific action of the target of interest, and provide a user interface including the detected result to the user terminal. . At this time, the user terminal may output a user interface received from the computing device 100 as a server, and receive or process information through interaction with the user.

In an additional embodiment, the computing device 100 may include any type of terminal that receives data resources generated by any server and performs additional information processing.

2 is a flowchart of a method of performing a task through a deep learning model according to some embodiments of the present disclosure. 3 is a diagram for explaining a deep learning model for performing a task according to some embodiments of the present disclosure. 4 is a diagram for explaining parameters included in a plurality of convolution layers according to some embodiments of the present disclosure. 5 is another diagram for describing parameters included in a plurality of convolution layers according to some embodiments of the present disclosure.

The deep learning model according to the present disclosure may perform a task by processing an image including a target of interest. For example, a task performed by the deep learning model according to the present disclosure may be detecting an abnormal behavior of an object included in an image that is input data. A deep learning model according to the present disclosure may include a 3D convolution layer for processing 4D data of an image. Since a 3D convolution layer for processing 4D data processes much more data than a 2D convolution layer for processing images, it is important to reduce the weight of networks to be used on various devices. A lightweight network has a trade-off of high inference speed but relatively low performance. The deep learning model according to the present disclosure can make an optimal choice in the trade-off between performance and inference speed by implementing a multi-level lightweight network in one network.

According to some embodiments of the present disclosure, a method of performing a task through a deep learning model may include determining a use step of the deep learning model (s110).

As described above, the deep learning model of the present disclosure may process input data for an image, which is 4-dimensional data, in order to perform a task. For example, input data for an image may be 4-dimensional data including vertical, horizontal, RGB channel values, and frame number values for single image data. In addition, the deep learning model can process multiple images in parallel. In this case, the input data may be 5-dimensional data including vertical, horizontal, RGB channel values, frame information values, and image information values of a plurality of image data.

The deep learning model according to the present disclosure may process input data according to a suitable use stage among a plurality of use stages according to the performance of a driven computing device. The use step may be determined according to the performance of the computing device on which the deep learning model is driven. In general, since a component of a computing device used to drive a deep learning model is a GPU, the method of the present disclosure may use a method of measuring the GPU memory of the computing device to determine the use level. Several embodiments of using GPU memory to determine usage levels are described below.

According to some embodiments of the present disclosure, determining the use of the deep learning model includes determining device memory usage by measuring GPU memory of a computing device, determining a ratio between device memory usage and base memory usage, and determining the driving performance based on the determined ratio.

Specifically, the device memory usage may include information about the memory of the computing device. For example, device memory usage may be determined by measuring GPU memory of a computing device. Since the driving of the deep learning model is performed using the GPU, the driving performance can be quickly determined using the GPU memory of the computing device. As another example, device memory usage may be determined by measuring GPU memory usage by running an underlying network on a computing device. In this case, the overall performance of the computing device, such as the CPU and the input/output device, can be evaluated, so that the device memory usage can be more accurately determined. Here, the base network may include a 3D convolution of a predetermined configuration as a task model for measuring GPU memory. The underlying network may be used to determine the driving performance of the deep learning model of the computing device by comparing the amount of GPU memory used in each device by running on various devices.

In order for the underlying network to be suitable for determining the usage phase, the usage phase of the deep learning model needs to be designed to easily distinguish the performance of the device according to the driving in the underlying network. For example, in order to design a deep learning model with multiple use steps, a constant of proportionality between the network model and the memory used by the GPU, which can vary depending on the batch size of the deep learning network, the driving framework, and the type of parameters, can be calculated. there is. In addition, the amount of GPU memory used by running the base network under given conditions can be measured. The GPU memory usage measured in this way may be referred to as a base memory. In other words, the base memory usage may be the GPU memory usage used when driving the base network. In addition, it is possible to investigate the GPU memory of various devices that can drive deep learning models. The GPU memory measured in this way may be device memory usage. In addition, the number of channels of the plurality of convolution layers included in the network suitable for each network may be designed according to 'the number of channels of the convolution layer included in the base network x GPU memory/base memory of each device'. Since the number of parameters may be determined according to the number of channels of the convolution layer, the number of suitable use stages and the corresponding number of parameters may be derived by arranging the device memory usage of various devices. The device memory usage for the GPU memory of the deep learning model can be effectively used to determine the use level of the deep learning model on the computing device.

A usage level of the deep learning model may be determined by determining a ratio of device memory usage and base memory usage. Since the base memory usage, which is the GPU memory usage of the base network, provides the same standard for multiple devices, the ratio of the device memory usage to the base memory usage can be used to determine the driving performance of the deep learning model of the computing device. For example, the higher the 'device memory usage/based memory', the higher the driving performance of the computing device. In this case, the using step may be determined to use a larger number of parameters. Also, the smaller the 'device memory usage/based memory', the lower the driving performance of the computing device. In this case, the using step may be determined to use a smaller number of parameters.

Specifically, for example, the use step of the deep learning model may include a first use step, a second use step, and a third use step. As the number of parameters used increases from the first use step to the third use step, the number of parameters used may increase (eg, n for the first use step, 2n for the second use step, and 3n for the third use step). parameters). In this case, when 'device memory usage/based memory' is less than 1, the use step of the deep learning model may be determined as the first use step. In addition, when 'device memory usage/based memory' is greater than 1 and less than 2, the use step of the deep learning model may be determined as the second use step. Also, when 'device memory usage/based memory' is greater than 2, the use step of the deep learning model may be determined as the third use step. However, it is not limited thereto, and the use stage may be determined in various ways.

According to some embodiments of the present disclosure, a method of performing a task through a deep learning model may include processing input data through a deep learning model according to a determined use stage (s120). Here, the deep learning model includes a plurality of convolutional layers, and the deep learning model may adjust a usage ratio of parameters included in each of the plurality of convolutional layers according to a use stage.

As described above, since the use stage of the deep learning model is determined according to the driving performance of the computing device on which the deep learning model is driven, the deep learning model may exhibit optimal performance by processing input data according to the use stage. The detailed structure of the deep learning model of the present disclosure for performing inference with various performance according to various usage stages is described below.

According to some embodiments of the present disclosure, the using step includes a first using step and a second using step, and the number of parameters used in the plurality of convolution layers according to the second using step is plural according to the first using step. It can be more than the number of parameters used in the convolutional layer.

Referring to FIG. 4 , a convolution structure for input data is shown. Here, the 3D convolution 300 is shown as including 27 unit convolutions. Since the input data 11 is image data, the input data may have three input channels according to RGB channels. The exemplary next layer 230 may have nine channels as shown. The number of unit convolutions each having a parameter may be determined by multiplying the number of channels of the previous layer and the number of channels of the next layer. As described above, a set of at least one value processed in unit convolution may be referred to as a parameter of a convolution layer. Therefore, the number of unit convolutions may be equal to the number of parameters. Accordingly, in this embodiment, the number of parameters of the 3D convolution 300 may be 27 in total.

In addition, convolution may be performed using 27 convolution filters while moving in the vertical, horizontal, and frame directions of the layer. In this case, the number of parameters 310 used in the 3D convolution 300 according to the first use step may be nine. Also, the number of parameters 320 used in the 3D convolution 300 according to the second use step may be 18. Also, according to the third use step, the number of parameters in the 3D convolution 300 may be 27 (330). Accordingly, the number of parameters used may increase as the use level increases.

Referring to FIG. 5, a convolution structure in the second layer or higher is shown. Here, the three-dimensional convolution 400 is shown as including 81 unit convolutions. An exemplary previous layer 240 may have 9 channels as shown. Also, the exemplary next layer 250 may have nine channels as shown. As described above, the number of unit convolutions each having a parameter may be determined by multiplying the number of channels of the previous layer and the number of channels of the next layer. Also, as described above, a set of at least one value processed in unit convolution may be referred to as a parameter of a convolution layer. Therefore, the number of unit convolutions may be equal to the number of parameters. Accordingly, the number of parameters of the 3D convolution 400 may be 81 in total.

Similarly, convolution may be performed using 81 or less convolution filters while moving in the vertical, horizontal, and frame directions of the layer. In this case, the number of parameters 410 used in the 3D convolution 400 according to the first use step may be nine. Also, the number of parameters 320 used in the 3D convolution 300 according to the second use step may be 36. Also, according to the third use step, the number of parameters in the 3D convolution 300 may be 81 (330). Accordingly, the number of parameters used may increase as the use level increases.

Therefore, in the deep learning model, the number of parameters used in the plurality of convolution layers according to the second use step is greater than the number of parameters used in the plurality of convolution layers according to the first use step, and the number of parameters used in the plurality of convolution layers according to the third use step The number of parameters used in the convolution layer may be greater than the number of parameters used in the plurality of convolution layers according to the second use step. Therefore, the deep learning model can perform inference with various amounts of computation depending on the stage of use.

According to some embodiments of the present disclosure, parameters used in the plurality of convolution layers according to the second use step may include parameters used in the plurality of convolution layers according to the first use step. Also, parameters used in the plurality of convolution layers according to the third use step may include parameters used in the plurality of convolution layers according to the second use step.

The deep learning model of the present disclosure may have overlapping used parts as the use step increases. For example, referring to FIG. 4 , the parameter 320 used according to the second use step may include the parameter 310 used according to the first use step. Therefore, the parameters used in the first use step and the second use step may have an overlapping part with the parameter 310 used according to the first use step. In addition, the parameter 330 used according to the third use step may include the parameter 320 used according to the second use step.

Also, referring to FIG. 5 , the parameter 420 used according to the second use step may include the parameter 410 used according to the first use step. Accordingly, the parameters used in the first use step and the second use step may have an overlapping part with the parameter 410 used according to the first use step. In addition, the parameter 430 used according to the third use step may include the parameter 420 used according to the second use step. Therefore, the parameters used in the first use step, the second use step, and the third use step may have the parameter 410 used according to the first use step as an overlapping part. This overlapping part makes it possible to increase the learning efficiency of the deep learning model even when the stages of use gradually increase, and it can operate to maintain high accuracy of inference at low stages.

According to some embodiments of the present disclosure, the number of parameters used in at least some of the plurality of convolution layers according to the second use step is the number of parameters used in at least some of the plurality of convolution layers according to the first use step. It may be n _- ² times of . Here, at least some of the plurality of convolution layers may be layers other than the convolution layers connected to the input data. As shown in FIG. 5 , the number of parameters 410 used in the 3D convolution 400 according to the first use step may be nine. According to the second use step, the number of parameters 420 used in the 3D convolution 400 may be 36. Therefore, the number of parameters 420 used in the 3D convolution 400 according to the second use step can be 2 ² times the number 410 used in the 3D convolution 400 according to the first use step. there is. Also, the number of parameters 430 used in the 3D convolution 400 according to the third use step may be 91. Therefore, the number of parameters 430 used in the 3D convolution 400 according to the third use step may be 3 ² times the number 410 used in the 3D convolution 400 according to the first use step. there is. The method in which the number of parameters increases by an integer squared can facilitate the method of extending the network. For example, the network can be easily expanded by increasing the number of channels in the layer by an integer multiple.

The deep learning model, in which the design of the deep learning model is completed according to the above-described structure, may be learned by, for example, gradient descent. A deep learning model according to the present disclosure can be trained by defining a loss function to perform gradient descent. For example, a loss function for detecting anomalous behavior may be defined as a cross-entropy between a probability value predicted by a network and a label indicating whether or not there is an anomalous behavior in an actual corresponding video. The deep learning model can infer predictions for tasks at each step, and a loss function for each step can be defined. Therefore, the overall loss function can be defined as the sum of the loss functions for each step. In this case, values of the convolution filter may be learned by performing back-propagation learning in a direction of decreasing the value of the entire loss function through gradient descent.

Referring to FIG. 3, an overview of the operation of the entire deep learning model is shown. Image input data including continuous images of the deep learning model may be processed by a plurality of convolution layers 200 . In this case, values output from each convolution layer may be processed as a nonlinear function. Here, the ratio of parameters used in each convolution layer may be adjusted according to each use stage, as shown in FIG. 3 . As shown, the parameters used in the first use step may be smaller than the parameters used in the second use step, and the parameters used in the second use step may be smaller than the parameters used in the third use step. In addition, parameters used in the first use step may be overlapping parts used in all use steps. Output values of the plurality of convolution layers output according to one use step among the plurality of use steps may be processed by a fully-connected neural network. The output value 20 output from the fully-connected neural network may be a two-dimensional vector representing a probability value for a task. The output value output according to the first use stage may be the result 21 of the first stage, and the output value output according to the second use stage may be the result 22 of the second stage. The first step result 21 may be derived at a higher speed than the second step result 22 . However, the first stage result 21 may have lower accuracy than the second stage result 22 . An output value output according to the third use step may be the result 23 of the third step. The third step result 23 may be derived at a slower rate than the first step result 22 and the third step result 22 . However, the third step result 23 may have higher accuracy than the first step result 21 and the second step result 22 . Therefore, the deep learning model according to the present disclosure can perform fast inference or achieve high performance as needed by processing input data according to various stages of use.

The deep learning model according to the present disclosure can be used not only on IoT and mobile devices with small computational capacity, but also on high-performance desktops and servers. In addition, the deep learning model can solve the problem of finding abnormal behavior in already observed images, which requires a large amount of computation, and the problem of finding abnormal behavior in real-time CCTV surveillance, which requires fast inference. Therefore, the deep learning model of the present disclosure can effectively solve the trade-off problem between performance and inference speed.

6 is a simplified and general schematic diagram of an exemplary computing environment in which embodiments of the present disclosure may be implemented.

Although the present disclosure has been described above as being generally embodied by a computing device, those skilled in the art will understand that the present disclosure may be combined with computer-executable instructions and/or other program modules that may be executed on one or more computers and/or may be implemented in hardware and software. It will be appreciated that it can be implemented as a combination.

Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In addition, those skilled in the art will understand that the methods of the present disclosure can be applied to single-processor or multiprocessor computer systems, minicomputers, mainframe computers as well as personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, and the like ( It will be appreciated that each of these may be implemented with other computer system configurations, including those that may be operative in connection with one or more associated devices.

The described embodiments of the present disclosure may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

Computers typically include a variety of computer readable media. Computer readable media can be any medium that can be accessed by a computer, including volatile and nonvolatile media, transitory and non-transitory media, removable and non-transitory media. Includes removable media. By way of example, and not limitation, computer readable media may include computer readable storage media and computer readable transmission media. Computer readable storage media are volatile and nonvolatile media, transitory and non-transitory, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. includes media Computer readable storage media may include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital video disk (DVD) or other optical disk storage device, magnetic cassette, magnetic tape, magnetic disk storage device or other magnetic storage device. device, or any other medium that can be accessed by a computer and used to store desired information.

A computer readable transmission medium typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism. Including all information delivery media. The term modulated data signal means a signal that has one or more of its characteristics set or changed so as to encode information within the signal. By way of example, and not limitation, computer readable transmission media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, and other wireless media. Combinations of any of the above are also intended to be included within the scope of computer readable transmission media.

An exemplary environment 1100 implementing various aspects of the present disclosure is shown including a computer 1102, which includes a processing unit 1104, a system memory 1106, and a system bus 1108. do. System bus 1108 couples system components, including but not limited to system memory 1106 , to processing unit 1104 . Processing unit 1104 may be any of a variety of commercially available processors. Dual processor and other multiprocessor architectures may also be used as the processing unit 1104.

System bus 1108 may be any of several types of bus structures that may additionally be interconnected to a memory bus, a peripheral bus, and a local bus using any of a variety of commercial bus architectures. System memory 1106 includes read only memory (ROM) 1110 and random access memory (RAM) 1112 . A basic input/output system (BIOS) is stored in non-volatile memory 1110, such as ROM, EPROM, or EEPROM, and is a basic set of information that helps transfer information between components within computer 1102, such as during startup. contains routines. RAM 1112 may also include high-speed RAM, such as static RAM, for caching data.

The computer 1102 may also include an internal hard disk drive (HDD) 1114 (eg, EIDE, SATA) - the internal hard disk drive 1114 may also be configured for external use within a suitable chassis (not shown). Yes—a magnetic floppy disk drive (FDD) 1116 (e.g., for reading from or writing to a removable diskette 1118), and an optical disk drive 1120 (e.g., a CD-ROM) for reading disc 1122 or reading from or writing to other high capacity optical media such as DVDs). The hard disk drive 1114, magnetic disk drive 1116, and optical disk drive 1120 are connected to the system bus 1108 by a hard disk drive interface 1124, magnetic disk drive interface 1126, and optical drive interface 1128, respectively. ) can be connected to The interface 1124 for external drive implementation includes at least one or both of USB (Universal Serial Bus) and IEEE 1394 interface technologies.

These drives and their associated computer readable media provide non-volatile storage of data, data structures, computer executable instructions, and the like. In the case of computer 1102, drives and media correspond to storing any data in a suitable digital format. Although the description of computer readable media above refers to HDDs, removable magnetic disks, and removable optical media such as CDs or DVDs, those skilled in the art can use zip drives, magnetic cassettes, flash memory cards, cartridges, etc. It will be appreciated that other tangible media readable by the computer, such as the like, may also be used in the exemplary operating environment and any such media may include computer executable instructions for performing the methods of the present disclosure.

A number of program modules may be stored on the drive and RAM 1112, including an operating system 1130, one or more application programs 1132, other program modules 1134, and program data 1136. All or portions of the operating system, applications, modules and/or data may also be cached in RAM 1112. It will be appreciated that the present disclosure may be implemented in a variety of commercially available operating systems or combinations of operating systems.

A user may enter commands and information into the computer 1102 through one or more wired/wireless input devices, such as a keyboard 1138 and a pointing device such as a mouse 1140. Other input devices (not shown) may include a microphone, IR remote control, joystick, game pad, stylus pen, touch screen, and the like. Although these and other input devices are often connected to the processing unit 1104 through an input device interface 1142 that is connected to the system bus 1108, a parallel port, IEEE 1394 serial port, game port, USB port, IR interface, may be connected by other interfaces such as the like.

A monitor 1144 or other type of display device is also connected to the system bus 1108 through an interface such as a video adapter 1146. In addition to the monitor 1144, computers typically include other peripheral output devices (not shown) such as speakers, printers, and the like.

Computer 1102 may operate in a networked environment using logical connections to one or more remote computers, such as remote computer(s) 1148 via wired and/or wireless communications. Remote computer(s) 1148 may be a workstation, computing device computer, router, personal computer, handheld computer, microprocessor-based entertainment device, peer device, or other common network node, and generally includes It includes many or all of the components described for, but for simplicity, only memory storage device 1150 is shown. The logical connections shown include wired/wireless connections to a local area network (LAN) 1152 and/or a larger network, such as a wide area network (WAN) 1154 . Such LAN and WAN networking environments are common in offices and corporations and facilitate enterprise-wide computer networks, such as intranets, all of which can be connected to worldwide computer networks, such as the Internet.

When used in a LAN networking environment, computer 1102 connects to local network 1152 through wired and/or wireless communication network interfaces or adapters 1156. Adapter 1156 may facilitate wired or wireless communications to LAN 1152, which also includes a wireless access point installed therein to communicate with wireless adapter 1156. When used in a WAN networking environment, computer 1102 may include a modem 1158, be connected to a communicating computing device on WAN 1154, or establish communications over WAN 1154, such as over the Internet. have other means. A modem 1158, which may be internal or external and a wired or wireless device, is connected to the system bus 1108 through a serial port interface 1142. In a networked environment, program modules described for computer 1102 or portions thereof may be stored on remote memory/storage device 1150 . It will be appreciated that the network connections shown are exemplary and other means of establishing a communication link between computers may be used.

Computer 1102 is any wireless device or entity that is deployed and operating in wireless communication, eg, printers, scanners, desktop and/or portable computers, portable data assistants (PDAs), communication satellites, wireless detectable tags associated with It operates to communicate with arbitrary equipment or places and telephones. This includes at least Wi-Fi and Bluetooth wireless technologies. Thus, the communication may be a predefined structure as in conventional networks or simply an ad hoc communication between at least two devices.

Wi-Fi (Wireless Fidelity) makes it possible to connect to the Internet without wires. Wi-Fi is a wireless technology, such as a cell phone, that allows such devices, eg, computers, to transmit and receive data both indoors and outdoors, i.e. anywhere within coverage of a base station. Wi-Fi networks use a radio technology called IEEE 802.11 (a, b, g, etc.) to provide secure, reliable, and high-speed wireless connections. Wi-Fi can be used to connect computers to each other, to the Internet, and to wired networks (using IEEE 802.3 or Ethernet). Wi-Fi networks can operate in the unlicensed 2.4 and 5 GHz radio bands, for example, at 11 Mbps (802.11a) or 54 Mbps (802.11b) data rates, or in products that include both bands (dual band) .

Those skilled in the art will understand that information and signals may be represented using any of a variety of different technologies and techniques. For example, data, instructions, instructions, information, signals, bits, symbols and chips that may be referenced in the above description are voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields s or particles, or any combination thereof.

Those skilled in the art will understand that the various illustrative logical blocks, modules, processors, means, circuits, and algorithm steps described in connection with the embodiments disclosed herein are electronic hardware, (for convenience) , may be implemented by various forms of program or design code (referred to herein as software) or a combination of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the particular application and the design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

Various embodiments presented herein may be implemented as a method, apparatus, or article of manufacture using standard programming and/or engineering techniques. The term article of manufacture includes a computer program, carrier, or media accessible from any computer-readable storage device. For example, computer-readable storage media include magnetic storage devices (eg, hard disks, floppy disks, magnetic strips, etc.), optical disks (eg, CDs, DVDs, etc.), smart cards, and flash memory devices (eg, EEPROM, cards, sticks, key drives, etc.), but are not limited thereto. Additionally, various storage media presented herein include one or more devices and/or other machine-readable media for storing information.

It is to be understood that the specific order or hierarchy of steps in the processes presented is an example of example approaches. Based upon design priorities, it is to be understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of this disclosure. The accompanying method claims present elements of the various steps in a sample order, but are not meant to be limited to the specific order or hierarchy presented.

The description of the presented embodiments is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art of this disclosure, and the general principles defined herein may be applied to other embodiments without departing from the scope of this disclosure. Thus, the present disclosure is not to be limited to the embodiments presented herein, but is to be interpreted in the widest scope consistent with the principles and novel features presented herein.

Claims (11)

A method of performing a task through a deep learning model implemented by a computing device including at least one processor,
determining the stage of use of the deep learning model; and
processing input data through the deep learning model according to the determined use stage;
including,
The deep learning model includes a plurality of convolutional layers, and the deep learning model adjusts a usage ratio of parameters included in each of the plurality of convolutional layers according to a use step.
method.
According to claim 1,
The steps to determine the stage of use of a deep learning model are:
determining device memory usage by measuring GPU memory of the computing device;
determining a ratio of the device memory usage and base memory usage; and
determining a use step of the deep learning model based on the determined ratio;
including,
method.
According to claim 2,
The base memory usage is the GPU memory usage used when driving the base network,
method.
According to claim 1,
The use step includes a first use step and a second use step, and the number of parameters used in the plurality of convolution layers according to the second use step is determined in the plurality of convolution layers according to the first use step. More than the number of parameters used,
method.
According to claim 4,
The parameters used in the plurality of convolution layers according to the second use step include parameters used in the plurality of convolution layers according to the first use step,
method.
According to claim 5,
The using step further includes a third using step,
The number of parameters used in the plurality of convolution layers according to the third use step is greater than the number of parameters used in the plurality of convolution layers according to the second use step.
method.
According to claim 6,
The parameters used in the plurality of convolution layers according to the third use step include parameters used in the plurality of convolution layers according to the second use step,
method.
According to claim 4,
The number of parameters used in at least some of the plurality of convolution layers according to the second use step is n _- ² times the number of parameters used in at least some of the plurality of convolution layers according to the first use step;
method.
According to claim 1,
The task is to detect abnormal behavior of an object included in an image that is the input data.
method.
As a computing device that performs a task through a deep learning model,
a processor comprising at least one core;
a memory containing program codes executable by the processor; and
a network unit for receiving video input data;
including,
The processor:
determine the stage of use of the deep learning model; and
Processing input data through the deep learning model according to the determined use stage;
The deep learning model includes a plurality of convolutional layers, and the deep learning model adjusts a usage ratio of parameters included in each of the plurality of convolutional layers according to a use step.
Device.
A computer program stored on a computer-readable storage medium, which, when executed on one or more processors, causes the following operations to perform a task through a deep learning model, the operations comprising:
determining the use stage of the deep learning model; and
processing input data through the deep learning model according to the determined use stage;
including,
The deep learning model includes a plurality of convolutional layers, and the deep learning model adjusts a usage ratio of parameters included in each of the plurality of convolutional layers according to a use step.
A computer program stored on a computer-readable storage medium.

Images