WO2018174623A1 - 가상 3차원 심층 신경망을 이용하는 영상 분석 장치 및 방법 - Google Patents
가상 3차원 심층 신경망을 이용하는 영상 분석 장치 및 방법 Download PDFInfo
- Publication number
- WO2018174623A1 WO2018174623A1 PCT/KR2018/003404 KR2018003404W WO2018174623A1 WO 2018174623 A1 WO2018174623 A1 WO 2018174623A1 KR 2018003404 W KR2018003404 W KR 2018003404W WO 2018174623 A1 WO2018174623 A1 WO 2018174623A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- dimensional
- image
- neural network
- data
- virtual
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/64—Three-dimensional [3D] objects
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—Three-dimensional [3D] image rendering
- G06T15/10—Geometric effects
- G06T15/20—Perspective computation
- G06T15/205—Image-based rendering
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three-dimensional [3D] modelling for computer graphics
- G06T17/30—Polynomial surface description
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/06—Topological mapping of higher dimensional structures onto lower dimensional surfaces
- G06T3/067—Reshaping or unfolding three-dimensional [3D] tree structures onto two-dimensional [2D] planes
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/60—Rotation of whole images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Definitions
- An embodiment of the present invention relates to an image analysis technique using image reconstruction, and more particularly, to an image analysis apparatus and method using a virtual three-dimensional deep neural network.
- ANNs Artificial neural networks
- an artificial neural network is composed of an input layer, a hidden layer, and an output layer.
- Each layer is composed of neurons, and each layer of neurons is connected to the output of neurons of the previous layer.
- Each output of the neurons in the previous layer and the corresponding product of the corresponding link weights, plus a bias, are typically added to a non-linear activation function. And pass the output to the next level of neurons.
- the convolutional neural network has gained much attention in the field of image recognition, overwhelming the performance of existing machine learning techniques.
- the structure of a convolutional neural network is almost the same as that of a general artificial neural network, and additional components include a convolutional layer and a pooling layer.
- the structure of a general convolutional neural network consists of alternating convolutional and consolidation layers, followed by two or three fully-connected layers, followed by an output layer. Neurons of the convolutional layer are local connectivity only to a small area of the previous layer, unlike artificial neural networks that are fully-connected with all neurons of the previous layer.
- neurons belonging to a slice such as a feature map
- This operation is a composite product, and the set of weights applied is called a filter or kernel.
- the multiplicative neural network can effectively extract features from the image and reduce the number of parameters to prevent overfitting and to improve generalization performance.
- the integration layer is located between the convolutional layers and serves to reduce the spatial size of the feature map. This process also serves to prevent overfitting by reducing the number of parameters.
- the most common form is the max-pooling method, in which 2x2 filters are applied at intervals of two. This process reduces the size of the feature map by half for the width and height directions.
- An object of the present invention for solving the problems of the prior art as described above, reconstructs the two-dimensional image to three-dimensional data in three-dimensional space and rotates the reconstructed three-dimensional data to generate another three-dimensional data, the plurality of generated
- the present invention provides an image analysis apparatus and method for easily analyzing three-dimensional image data in a deep neural network by applying a two-dimensional composite product neural network to each of three-dimensional data.
- an image analyzing apparatus using a virtual three-dimensional deep neural network including: an image obtaining unit stacking a plurality of two-dimensional image data in a preset order; A three-dimensional image generation unit generating a plurality of three-dimensional data based on a plurality of pieces of information of different forms of the plurality of two-dimensional image data of the stacked form from the image acquisition unit; And a deep learning algorithm analyzer configured to apply a 2D convolutional neural network to the plurality of 3D data from the 3D image generator and to combine the results of applying the 2D convolutional neural network to the plurality of 3D data.
- the three-dimensional image generator a zero-mean or unit-variance calculation for each of the plurality of two-dimensional image data before generating the plurality of three-dimensional data Can be performed.
- the plurality of pieces of information of different types may include recognizing a pattern corresponding to a change in movement or shape according to time or position of the stacked 2D image data.
- the deep learning algorithm analysis unit outputs the results of applying the two-dimensional convolutional neural network to the plurality of three-dimensional data convolutional layer, fully-connected layer, output layer ( output layer and decision level fusion, which averages the final results.
- an image analysis method using a virtual three-dimensional deep neural network comprising: stacking a plurality of two-dimensional image data in a predetermined order; Generating, by the 3D image generating unit, a plurality of 3D data based on a plurality of pieces of information of different forms of the plurality of 2D image data in a stacked form; And in the deep learning algorithm analysis unit, applying a 2D convolutional neural network to each of the plurality of 3D data and combining the results of applying the 2D convolutional neural network to the plurality of 3D data.
- the generating may include performing a zero-mean or unit-variance operation on each of the plurality of two-dimensional image data before generating the plurality of three-dimensional data. Can be done.
- the merging may include applying a result of applying the two-dimensional convolutional neural network to the plurality of three-dimensional data to a convolutional layer, a fully-connected layer, and an output layer. output layer and decision level fusion, which averages the final results.
- an image analyzing apparatus using a virtual three-dimensional deep neural network an image acquisition unit for stacking two-dimensional images in the order of shooting position or time;
- the first 3D image data is generated from the 2D images received from the image acquisition unit, and the axis representing the photographing position or time in the first 3D image data corresponds to any one of the other two axes.
- a three-dimensional image generator which generates rotated second three-dimensional image data;
- a deep learning algorithm analysis unit configured to apply a 2D composite product neural network to each of the plurality of 3D data received from the 3D image generator and to combine the application results of the 3D data.
- the 3D image generating unit generates additional 3D data based on differences between frames of the 2D images or other 2D images obtained by rotating the 2D images obtained through optical flow. can do.
- An image analysis method using a virtual three-dimensional deep neural network for solving the technical problem, the step of stacking the two-dimensional images in the image pickup position or time sequence;
- the 3D image generating unit generates first 3D image data from the 2D images received from the image obtaining unit, and an axis indicating the shooting position or time in the first 3D image data is one of the remaining two axes.
- Generating second 3D image data rotated to match any one;
- the generating may include generating additional three-dimensional data based on differences between frames of the two-dimensional images or other two-dimensional images obtained by rotating the two-dimensional images obtained through optical flow. Can be.
- the 3D data can be more efficiently learned and image analyzed using the 2D convolutional neural network having fewer parameters than the conventional 3D convolutional neural network method.
- the present invention since the number of parameters is very large, it takes up a lot of memory, takes a long time to learn, and solves the problem of the 3D convolutional neural network model, which requires a long computation time when using the trained model. It is possible to provide a new image analysis model that can perform efficient learning and image analysis on the data.
- FIG. 1 is a block diagram of an image analyzing apparatus using a virtual three-dimensional deep neural network according to an embodiment of the present invention.
- FIG. 2 is a schematic diagram illustrating an operating principle of the image analyzing apparatus of FIG. 1.
- FIG. 3 is an exemplary diagram for explaining a conventional operating principle of a 2D convolutional neural network that may be employed in the image analyzing apparatus of FIG. 1.
- FIG. 4 is an exemplary view for explaining a principle of operation of the three-dimensional composite product neural network according to a comparative example.
- FIG. 5 is a flowchart of an image analysis method using a virtual three-dimensional deep neural network according to another embodiment of the present invention.
- FIG. 6 is a block diagram of an image analyzing apparatus using a virtual three-dimensional deep neural network according to another embodiment of the present invention.
- FIG. 1 is a block diagram of an image analysis apparatus using a virtual three-dimensional deep neural network according to an embodiment of the present invention.
- the image analyzing apparatus 100 includes an image acquirer 110, a 3D image generator 120, and a deep learning algorithm analyzer 130.
- the image acquisition unit 110 prepares 2D images stacked in order according to a photographing angle or time of the 2D images.
- the image acquisition unit 110 may be connected to a camera, a controller, a communication unit, and the like.
- the 3D image generator 120 generates a plurality of 3D data from the 2D images received from the image acquirer 110.
- the 3D image generating unit 120 stacks 2D images and converts the 2D images into first 3D data, and converts the converted first 3D data at an arbitrary angle in a 3D space.
- a plurality of three-dimensional data may be reconstructed such that any one of the three axes (x, y, z) in the three-dimensional space rotates to the position of the other axis to generate the second three-dimensional data.
- a plurality of two-dimensional image data stacked along a time axis according to a predetermined criterion can obtain a plurality of three-dimensional data having different forms in accordance with a change in relative time or position with respect to the plurality of two-dimensional image data.
- the 3D data may include 3D image data.
- two-dimensional image data including information about a moving video such as a cell image may have a shape in which a morphology may change and a location may change in each of the two-dimensional image data.
- the image recognition apparatus extracts differences in which the outline is changed or the position is slightly changed from the 2D image data according to the position or time, and converts the 2D image data into 3D data.
- a pattern corresponding to a change in motion or a change in shape may be recognized based on the extracted information.
- the image recognition device may perform pattern recognition by using volumetric or the like.
- the stacked two-dimensional image data has a three-dimensional data form, where the stacked two-dimensional image data are viewed from above and from the side.
- the difference can be recognized as the main difference of morphology, and when viewing 2D image data from the side, the difference depends on the temporal difference It can be recognized as a change in its position.
- a plurality of data that is, a plurality of virtual three-dimensional data recognized as different forms of the stacked two-dimensional image data are obtained and used.
- the deep learning algorithm analysis unit 130 applies a 2D convolutional neural network (CNN) to each of the plurality of reconstructed 3D data and analyzes the 3D image by combining the application results of each 3D data. do.
- CNN 2D convolutional neural network
- the main technical feature of the present embodiment is to perform a three-dimensional image analysis by learning different types of information on the two-dimensional image data accumulated after the two-dimensional image data is accumulated by the two-dimensional composite product neural network.
- components 110 to 130 may be configured in hardware, but are not limited thereto.
- Components of the image analyzing apparatus 100 are stored in a storage device such as a memory in the form of a software module, and a processor connected to the storage device executes the software module to effectively learn 3D image data based on a virtual 3D deep neural network. And analysis.
- FIG. 2 is a schematic diagram illustrating an operating principle of the image analyzing apparatus of FIG. 1.
- the image acquisition unit may accumulate two-dimensional images received or acquired from the outside or stored in an external or internal memory of the image analysis apparatus according to a photographing position or a photographing time.
- the 3D image generating unit generates a plurality of 3D data using the 2D images received from the image obtaining unit.
- the 3D data may include 3D image data.
- the 3D image generating unit may generate 3D image data by stacking a plurality of 2D images in a photographing position or a time sequence.
- the 3D image generator may generate additional 3D image data by rotating the generated 3D image data at a predetermined angle.
- a plurality of three-dimensional image data can be generated by the following process. That is, if two axes of the 2D image are called x and y and the axis indicating the shooting position or time of the 2D images is z, the 3D data Dxyz (first 3D data) formed by stacking 2D images in the z-axis order ) And 3D data Dyzx and 3D data Dzxy created by rotating Dxyz in two different axial directions, respectively.
- the 3D image generating unit may generate and use another 3D data in addition to the plurality of 3D data described above. That is, the 3D image generating unit may generate a plurality of 3D images by applying the above-described method to other images obtained through a pre-prepared calculation from the original 2D images. For example, a plurality of three-dimensional data may be generated through the above-described method after normalization to have zero-mean and unit-variance for each two-dimensional image.
- the 3D image generator may generate additional 3D images by rotating the images and the images obtained through calculation of a difference between frames or an optical flow.
- the deep learning algorithm analyzer may generate a plurality of two-dimensional data sets by dividing and projecting the plurality of three-dimensional data received from the three-dimensional image generator at arbitrary intervals as needed. A plurality of two-dimensional data sets may be included in the three-dimensional data.
- the deep learning algorithm analyzer may apply a 2D composite product neural network to each of the plurality of 3D data received from the 3D image generator, and obtain the image analysis result by combining them.
- each 2D convolutional neural network is combined into a convolutional layer or a fully-connected layer or an output layer or an average of the final results. There may be decision level fusion.
- FIG. 3 is an exemplary view for explaining a principle of operation of a two-dimensional composite product neural network that may be employed in the image analyzing apparatus of FIG. 1.
- 4 is an exemplary view for explaining a principle of operation of the three-dimensional composite product neural network according to a comparative example.
- Equation 1 The composite product calculation structure of the two-dimensional composite product neural network may be expressed by Equation 1 below.
- Equation 1 Denotes the value of the position (x, y) in the j th feature map of the i th layer
- m represents the index of the feature map of the (i-1) th layer
- the two-dimensional convolutional neural network described above shows excellent performance in image recognition. However, since the convolution performed only calculates two-dimensional spatial features, when only two-dimensional convolutional neural networks are used, information about depth or time in a three-dimensional image including several two-dimensional images is collected. Can't learn
- the general three-dimensional convolutional neural network learns three-dimensional filters to analyze three-dimensional images. It takes a long time (see Equation 2).
- a composite product calculation structure using a 2D composite product neural network and a composite product calculation structure using a 3D composite product neural network are combined and used in a new manner.
- the composite product calculation structure of the three-dimensional composite product neural network coupled to the composite product calculation structure of the two-dimensional composite product neural network may be illustrated as shown in FIG. 4, and may be expressed as in Equation 2 below.
- Equation 2 Denotes the value of position (x, y, z) in the j th feature map of the i th layer
- m represents the index of the feature map of the (i-1) th layer
- , Denotes the size of the kernel in the vertical, horizontal, and depth (or time) directions, respectively.
- the conventional technique using only the above-described three-dimensional convolutional neural network model occupies a lot of memory because it has a large number of parameters, and it takes a long time to train, and furthermore, a calculation time is long when using the trained model. Therefore, in the present exemplary embodiment, the 2D convolutional neural network having fewer parameters than the 3D convolutional neural network can be used to efficiently learn and analyze the 3D image data.
- the deep learning algorithm analyzer includes applying a 2D composite product neural network to each of a plurality of 2D data sets (a plurality of 3D data) received from the 3D image generator, and combining the application results.
- Image analysis results by 'virtual 3D deep neural network' can be derived.
- FIG. 5 is a flowchart of an image analysis method using a virtual three-dimensional deep neural network according to another embodiment of the present invention.
- a step of stacking two-dimensional images of a specific group by a photographing position or time is performed by an image acquisition unit in an image analysis apparatus (S51).
- Generating a three-dimensional image (first three-dimensional data) by using the two-dimensional images generating second three-dimensional data obtained by rotating the first three-dimensional data (S52), and a plurality of three-dimensional images.
- the image analysis method using the virtual three-dimensional deep neural network uses the two-dimensional convolutional neural network with fewer parameters than the conventional three-dimensional convolutional neural network method to learn three-dimensional data more efficiently and to analyze the image. Applicable This method may be called a method by a virtual 3D deep neural network.
- FIG. 6 is a block diagram of an image analyzing apparatus using a virtual three-dimensional deep neural network according to another embodiment of the present invention.
- the image analyzing apparatus 100 may include a communication unit 160, a controller 170, and a memory 180.
- the image analyzing apparatus 100 may include a controller or a computing device.
- the image analyzing apparatus 100 may be connected to an input / output device 190 for processing data or a signal according to an input from a user, an administrator, a control terminal, or the like and outputting the result.
- the image analysis apparatus 100 may be connected to a database system 200 having a database.
- the database may include at least one of identification information, access information, and authentication information of a device that provides an image to be analyzed.
- the input / output device 190 and the database system 200 are illustrated in a form not included in the image analysis device 100 in the present embodiment, the present invention is not limited to such a configuration, and the input / output device 190 may be implemented according to an implementation. And it may be implemented to further include at least one or more of the database system 200.
- the communication unit 160 connects the image analyzing apparatus 100 to a communication network.
- the communication unit 160 may receive information or a signal related to an image or an image analysis from a user terminal, a server, an administrator terminal, and the like, which are accessed through a network.
- the communicator 160 may include one or more wired and / or wireless communication subsystems that support one or more communication protocols.
- Wired communication subsystems include public switched telephone networks (PSTN), Asymmetric Digital Subscriber Line (ADSL) or Very High-data Rate Digital Subscriber Line (VDSL) networks, subsystems for PSTN Emulation Service (PES), and Internet Protocol (IP).
- PSTN public switched telephone networks
- ADSL Asymmetric Digital Subscriber Line
- VDSL Very High-data Rate Digital Subscriber Line
- PES PSTN Emulation Service
- IP Internet Protocol
- Multimedia subsystem IMS and the like.
- the wireless communication subsystem may include a radio frequency (RF) receiver, an RF transmitter, an RF transceiver, an optical (eg, infrared) receiver, an optical transmitter, an optical transceiver, or a combination thereof.
- RF radio frequency
- a wireless network basically refers to Wi-Fi, but is not limited thereto.
- the communication unit 160 may use various wireless networks, for example, Global System for Mobile Communication (GSM), Enhanced Data GSM Environment (EDGE), Code Division Multiple Access (CDMA), and W-Code Division Multiple (W-CDMA). It may be implemented to support at least one selected from Access, Long Term Evolution (LTE), LET-A (LET-Advanced), Orthogonal Frequency Division Multiple Access (OFDMA), WiMax, Wireless Fidelity (Wi-Fi), Bluetooth, and the like. Can be.
- GSM Global System for Mobile Communication
- EDGE Enhanced Data GSM Environment
- CDMA Code Division Multiple Access
- W-CDMA Wideband Code Division Multiple Access
- W-CDMA Wideband Code Division Multiple Access
- LTE Long Term Evolution
- LET-A LET-Advanced
- OFDMA Orthogonal Frequency Division Multiple Access
- WiMax Wireless Fidelity
- Wi-Fi Wireless Fidelity
- the controller 170 may implement an image analysis method by performing a software module or program stored in the internal memory or the memory 180.
- the controller 170 may be referred to as a processor, for example, and may perform a series of procedures illustrated in FIG. 5.
- the controller 170 may be implemented as a processor or a microprocessor including at least one central processing unit (CPU) or a core.
- the central processing unit or core is a register that stores the instructions to be processed, an arithmetic logical unit (ALU) that is responsible for comparison, determination, and operation, and the CPU internally to interpret and execute the instructions. It may be provided with a control unit (control unit) for controlling, and an internal bus connecting them.
- the CPU or core may be implemented as a system on chip (SOC) in which a micro control unit (MCU) and a peripheral device (an integrated circuit for an external expansion device) are arranged together, but is not limited thereto.
- SOC system on chip
- the controller 170 may include one or more data processors, an image processor, or a codec, but is not limited thereto.
- the controller 170 may include a peripheral device interface and a memory interface.
- the peripheral interface may connect an input / output system such as the controller 170 and the input / output device 190 or another peripheral device, and the memory interface may connect the controller 170 and the memory 180.
- the memory 180 may store a software module for analyzing an image using a virtual 3D deep neural network.
- the software module may include first to third modules that perform steps S51 to S53 of FIG. 5, respectively.
- the above-described memory 180 is a non-volatile random access memory (NVRAM), semiconductor memory, such as dynamic random access memory (DRAM), which is a typical volatile memory, hard disk drive (HDD), optical storage It may be implemented as a device, flash memory, or the like.
- NVRAM non-volatile random access memory
- DRAM dynamic random access memory
- HDD hard disk drive
- the memory 180 may store an operating system, a program, a command set, etc. in addition to software modules for analyzing an image using a virtual 3D deep neural network.
- the image analysis method according to the present embodiment may be implemented in the form of program instructions that can be executed by various computer means may be recorded on a computer readable medium.
- Computer-readable media may include, alone or in combination with the program instructions, data files, data structures, and the like.
- the program instructions recorded on the computer readable medium may be those specially designed and constructed for the present invention, or may be known and available to those skilled in computer software.
- Examples of computer readable media include hardware devices that are specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like.
- Examples of program instructions include machine language code, such as produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like.
- the hardware device described above may be configured to operate with at least one software module to perform the operations of the present invention, and vice versa.
- the present embodiment provides a method for constructing a deep neural network structure for analyzing 3D image data.
- the virtual three-dimensional deep neural network structure according to the present embodiment may be utilized for analyzing three-dimensional image data such as diagnosing a disease in an input medical image, finding a location of a lesion, or recognizing human behavior in a video. .
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Mathematical Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Algebra (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Complex Calculations (AREA)
Abstract
Description
Claims (7)
- 복수의 2차원 영상 데이터를 미리 설정된 순서대로 쌓는 영상 획득부;상기 영상 획득부로부터의 쌓은 형태의 복수의 2차원 영상 데이터에 대한 서로 다른 형태들의 복수의 정보에 기초하여 복수의 3차원 데이터를 생성하는 3차원 영상 생성부; 및상기 3차원 영상 생성부로부터의 복수의 3차원 데이터 각각에 대해 2차원 합성곱 신경망을 적용하고 상기 복수의 3차원 데이터에 대한 2차원 합성곱 신경망의 적용 결과들을 합치는 딥러닝 알고리즘 분석부를 포함하는,가상 3차원 심층 신경망을 이용하는 영상 분석 장치.
- 청구항 1에 있어서,상기 3차원 영상 생성부는, 상기 복수의 3차원 데이터를 생성하기 전에 상기 복수의 2차원 영상 데이터 각각에 대해 제로-평균(zero-mean) 또는 단위-변화(unit-variance) 연산을 수행하는, 가상 3차원 심층 신경망을 이용하는 영상 분석 장치.
- 청구항 1에 있어서,상기 서로 다른 형태들의 복수의 정보는 상기 쌓은 2차원 영상 데이터의 시간 또는 위치에 따른 움직임이나 모양의 변화에 대응하는 패턴을 인식한 것을 포함하는, 가상 3차원 심층 신경망을 이용하는 영상 분석 장치.
- 청구항 1에 있어서,상기 딥러닝 알고리즘 분석부는 상기 복수의 3차원 데이터에 대한 상기 2차원 합성곱 신경망의 적용 결과들을 합성곱 레이어(convolutional layer), 전결합 레이어(fully-connected layer), 출력 레이어(output layer) 및 최종 결과들의 평균을 내는 결정 수준 융합(decision level fusion) 중 어느 하나에서 합치는, 가상 3차원 심층 신경망을 이용하는 영상 분석 장치.
- 영상 획득부에서, 복수의 2차원 영상 데이터를 미리 설정된 순서대로 쌓는 단계;3차원 영상 생성부에서, 쌓은 형태의 상기 복수의 2차원 영상 데이터에 대한 서로 다른 형태들의 복수의 정보에 기초하여 복수의 3차원 데이터를 생성하는 단계; 및딥러닝 알고리즘 분석부에서, 상기 복수의 3차원 데이터 각각에 대해 2차원 합성곱 신경망을 적용하고 상기 복수의 3차원 데이터에 대한 2차원 합성곱 신경망의 적용 결과들을 합치는 단계;를 포함하는 가상 3차원 심층 신경망을 이용하는 영상 분석 방법.
- 청구항 5에 있어서,상기 생성하는 단계는, 상기 복수의 3차원 데이터를 생성하기 전에 상기 복수의 2차원 영상 데이터 각각에 대해 제로-평균(zero-mean) 또는 단위-변화(unit-variance) 연산을 수행하는, 가상 3차원 심층 신경망을 이용하는 영상 분석 방법.
- 청구항 5에 있어서,상기 합치는 단계는, 상기 복수의 3차원 데이터에 대한 상기 2차원 합성곱 신경망의 적용 결과들을 합성곱 레이어(convolutional layer), 전결합 레이어(fully-connected layer), 출력 레이어(output layer) 및 최종 결과들의 평균을 내는 결정 수준 융합(decision level fusion) 중 어느 하나에서 합치는, 가상 3차원 심층 신경망을 이용하는 영상 분석 방법.
Priority Applications (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201880027104.8A CN110574077B (zh) | 2017-03-24 | 2018-03-23 | 利用虚拟三维深层神经网络的影像分析装置及方法 |
| JP2019552542A JP6979664B2 (ja) | 2017-03-24 | 2018-03-23 | 仮想3次元深層ニューラルネットワークを利用する画像解析装置及び方法 |
| US16/496,960 US10970520B1 (en) | 2017-03-24 | 2018-03-23 | Apparatus and method for image analysis using virtual three-dimensional deep neural network |
| EP18771852.3A EP3605472A4 (en) | 2017-03-24 | 2018-03-23 | IMAGE ANALYSIS APPARATUS AND METHOD USING A VIRTUAL THREE-DIMENSIONAL DEEP NEURONAL NETWORK |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| KR10-2017-0037958 | 2017-03-24 | ||
| KR20170037958 | 2017-03-24 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2018174623A1 true WO2018174623A1 (ko) | 2018-09-27 |
Family
ID=63585872
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/KR2018/003404 Ceased WO2018174623A1 (ko) | 2017-03-24 | 2018-03-23 | 가상 3차원 심층 신경망을 이용하는 영상 분석 장치 및 방법 |
Country Status (6)
| Country | Link |
|---|---|
| US (1) | US10970520B1 (ko) |
| EP (1) | EP3605472A4 (ko) |
| JP (1) | JP6979664B2 (ko) |
| KR (1) | KR102061408B1 (ko) |
| CN (1) | CN110574077B (ko) |
| WO (1) | WO2018174623A1 (ko) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11276249B2 (en) | 2020-05-14 | 2022-03-15 | International Business Machines Corporation | Method and system for video action classification by mixing 2D and 3D features |
Families Citing this family (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11190944B2 (en) | 2017-05-05 | 2021-11-30 | Ball Aerospace & Technologies Corp. | Spectral sensing and allocation using deep machine learning |
| US11386900B2 (en) * | 2018-05-18 | 2022-07-12 | Deepmind Technologies Limited | Visual speech recognition by phoneme prediction |
| KR102107182B1 (ko) * | 2018-10-23 | 2020-05-06 | 전남대학교 산학협력단 | 손 제스처 인식 시스템 및 방법 |
| KR102250163B1 (ko) * | 2018-11-23 | 2021-05-10 | 네이버웹툰 유한회사 | 딥러닝 기술을 이용하여 비디오 영상을 3d 비디오 영상으로 변환하는 방법 및 장치 |
| RU2703327C1 (ru) * | 2018-12-10 | 2019-10-16 | Самсунг Электроникс Ко., Лтд. | Способ обработки двухмерного изображения и реализующее его вычислительное устройство пользователя |
| US11893681B2 (en) | 2018-12-10 | 2024-02-06 | Samsung Electronics Co., Ltd. | Method for processing two-dimensional image and device for executing method |
| KR102263005B1 (ko) * | 2019-01-15 | 2021-06-08 | 포항공과대학교 산학협력단 | 동적으로 3d cnn을 이용하는 고속 영상 인식 방법 및 장치 |
| KR102263017B1 (ko) * | 2019-01-15 | 2021-06-08 | 포항공과대학교 산학협력단 | 3d cnn을 이용한 고속 영상 인식 방법 및 장치 |
| US11851217B1 (en) * | 2019-01-23 | 2023-12-26 | Ball Aerospace & Technologies Corp. | Star tracker using vector-based deep learning for enhanced performance |
| US11412124B1 (en) | 2019-03-01 | 2022-08-09 | Ball Aerospace & Technologies Corp. | Microsequencer for reconfigurable focal plane control |
| CN111988666B (zh) * | 2019-05-23 | 2022-04-26 | 阿里巴巴集团控股有限公司 | 视频检测、3d卷积与映射方法、设备及存储介质 |
| US11488024B1 (en) | 2019-05-29 | 2022-11-01 | Ball Aerospace & Technologies Corp. | Methods and systems for implementing deep reinforcement module networks for autonomous systems control |
| US11303348B1 (en) | 2019-05-29 | 2022-04-12 | Ball Aerospace & Technologies Corp. | Systems and methods for enhancing communication network performance using vector based deep learning |
| KR102081854B1 (ko) * | 2019-08-01 | 2020-02-26 | 전자부품연구원 | 3d edm을 이용한 수어 또는 제스처 인식 방법 및 장치 |
| US11828598B1 (en) | 2019-08-28 | 2023-11-28 | Ball Aerospace & Technologies Corp. | Systems and methods for the efficient detection and tracking of objects from a moving platform |
| KR102219364B1 (ko) * | 2019-09-19 | 2021-02-25 | 주식회사 싸인텔레콤 | 영상감지장치 기반의 버스정류장 조명제어 시스템 |
| TWI730452B (zh) * | 2019-10-16 | 2021-06-11 | 逢甲大學 | 立體類神經網路系統 |
| KR102166835B1 (ko) | 2019-10-28 | 2020-10-16 | 주식회사 루닛 | 신경망 학습 방법 및 그 장치 |
| KR102581941B1 (ko) * | 2019-12-27 | 2023-09-22 | 권세기 | 딥러닝을 이용한 반려견의 입마개 착용 여부 감시 시스템 및 방법 |
| US11830227B2 (en) | 2020-05-12 | 2023-11-28 | Lunit Inc. | Learning apparatus and learning method for three-dimensional image |
| CN111612689B (zh) * | 2020-05-28 | 2024-04-05 | 上海联影医疗科技股份有限公司 | 医学图像处理方法、装置、计算机设备和可读存储介质 |
| US20210398338A1 (en) * | 2020-06-22 | 2021-12-23 | Nvidia Corporation | Image generation using one or more neural networks |
| KR102453834B1 (ko) * | 2020-07-15 | 2022-10-11 | 한국로봇융합연구원 | 다수의 열화상 및 영상 카메라의 출력 정보를 심층신경망 모델의 입력데이터로 구조화하기 위한 방법 |
| CN111985618B (zh) * | 2020-08-14 | 2024-03-05 | 杭州海康威视数字技术股份有限公司 | 3d卷积神经网络在神经网络处理器上的处理方法和装置 |
| KR102505994B1 (ko) * | 2020-09-28 | 2023-03-07 | (주)제이엘케이 | 경량 3차원 볼륨 데이터 획득 시스템 및 방법 |
| KR102575224B1 (ko) * | 2021-01-07 | 2023-09-08 | 충북대학교 산학협력단 | 가변 합성곱 신경망을 이용한 객체 검출 시스템 및 그 방법 |
| US20220391752A1 (en) * | 2021-06-08 | 2022-12-08 | X Development Llc | Generating labeled synthetic images to train machine learning models |
| KR20230027948A (ko) | 2021-08-20 | 2023-02-28 | 선문대학교 산학협력단 | 이미지 초고해상도를 위한 컨볼루션 신경망 시스템 |
| KR102769440B1 (ko) | 2021-08-20 | 2025-02-17 | 선문대학교 산학협력단 | 초고해상도 컨볼루션 신경망 시스템의 연산량 감소 장치 및 방법 |
| JP7745813B1 (ja) * | 2024-04-30 | 2025-09-29 | 三菱電機株式会社 | 物体認識装置、物体認識方法、および、物体認識システム |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000105838A (ja) * | 1998-09-29 | 2000-04-11 | Toshiba Corp | 画像表示方法及び画像処理装置 |
| KR20160061856A (ko) * | 2014-11-24 | 2016-06-01 | 삼성전자주식회사 | 객체 인식 방법 및 장치, 및 인식기 학습 방법 및 장치 |
| KR20160101973A (ko) * | 2013-12-19 | 2016-08-26 | 아비질론 포트리스 코퍼레이션 | 비제약형 매체에 있어서 얼굴을 식별하는 시스템 및 방법 |
| KR20160122452A (ko) | 2015-04-14 | 2016-10-24 | (주)한국플랫폼서비스기술 | 비주얼 콘텐츠기반 영상 인식을 위한 딥러닝 프레임워크 및 영상 인식 방법 |
Family Cites Families (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000293704A (ja) | 1999-04-02 | 2000-10-20 | Kazuo Yamada | 3次元データ表示装置及び3次元データ表示方法並びに情報記録媒体 |
| JP2010051691A (ja) | 2008-08-29 | 2010-03-11 | Shimadzu Corp | 超音波診断装置 |
| JP2010072910A (ja) * | 2008-09-18 | 2010-04-02 | Nippon Telegr & Teleph Corp <Ntt> | 顔の3次元モデル生成装置、顔の3次元モデル生成方法および顔の3次元モデル生成プログラム |
| US8866845B2 (en) * | 2010-03-10 | 2014-10-21 | Empire Technology Development Llc | Robust object recognition by dynamic modeling in augmented reality |
| US11282287B2 (en) * | 2012-02-24 | 2022-03-22 | Matterport, Inc. | Employing three-dimensional (3D) data predicted from two-dimensional (2D) images using neural networks for 3D modeling applications and other applications |
| US10095917B2 (en) * | 2013-11-04 | 2018-10-09 | Facebook, Inc. | Systems and methods for facial representation |
| US9898804B2 (en) * | 2014-07-16 | 2018-02-20 | Samsung Electronics Co., Ltd. | Display driver apparatus and method of driving display |
| CN110110843B (zh) * | 2014-08-29 | 2020-09-25 | 谷歌有限责任公司 | 用于处理图像的方法和系统 |
| US9928410B2 (en) | 2014-11-24 | 2018-03-27 | Samsung Electronics Co., Ltd. | Method and apparatus for recognizing object, and method and apparatus for training recognizer |
| KR102276339B1 (ko) | 2014-12-09 | 2021-07-12 | 삼성전자주식회사 | Cnn의 근사화를 위한 학습 장치 및 방법 |
| GB2543893A (en) | 2015-08-14 | 2017-05-03 | Metail Ltd | Methods of generating personalized 3D head models or 3D body models |
| US11425866B2 (en) * | 2015-11-03 | 2022-08-30 | Keith Charles Burden | Automated pruning or harvesting system for complex morphology foliage |
| CN105787439B (zh) * | 2016-02-04 | 2019-04-05 | 广州新节奏智能科技股份有限公司 | 一种基于卷积神经网络的深度图像人体关节定位方法 |
| US9836820B2 (en) * | 2016-03-03 | 2017-12-05 | Mitsubishi Electric Research Laboratories, Inc. | Image upsampling using global and local constraints |
| US11055063B2 (en) * | 2016-05-02 | 2021-07-06 | Marvell Asia Pte, Ltd. | Systems and methods for deep learning processor |
| CN106407903A (zh) * | 2016-08-31 | 2017-02-15 | 四川瞳知科技有限公司 | 基于多尺度卷积神经网络的实时人体异常行为识别方法 |
| US10282918B2 (en) * | 2016-09-20 | 2019-05-07 | Siemens Healthcare Gmbh | Two-dimensional cinematic medical imaging in color based on deep learning |
| US10460511B2 (en) * | 2016-09-23 | 2019-10-29 | Blue Vision Labs UK Limited | Method and system for creating a virtual 3D model |
| HK1224513A2 (zh) * | 2016-10-14 | 2017-08-18 | 智能3D有限公司 | 通过机器学习技术改进2d至3d的自动转换质量的方法 |
| JP2018067154A (ja) * | 2016-10-19 | 2018-04-26 | ソニーセミコンダクタソリューションズ株式会社 | 演算処理回路および認識システム |
| US10176551B2 (en) * | 2017-04-27 | 2019-01-08 | Apple Inc. | Configurable convolution engine for interleaved channel data |
| WO2018227105A1 (en) * | 2017-06-08 | 2018-12-13 | The United States Of America, As Represented By The Secretary, Department Of Health And Human Services | Progressive and multi-path holistically nested networks for segmentation |
| CN107730503B (zh) * | 2017-09-12 | 2020-05-26 | 北京航空航天大学 | 三维特征嵌入的图像对象部件级语义分割方法与装置 |
| EP3462373A1 (en) * | 2017-10-02 | 2019-04-03 | Promaton Holding B.V. | Automated classification and taxonomy of 3d teeth data using deep learning methods |
| EP3474192A1 (en) * | 2017-10-19 | 2019-04-24 | Koninklijke Philips N.V. | Classifying data |
| US10762637B2 (en) * | 2017-10-27 | 2020-09-01 | Siemens Healthcare Gmbh | Vascular segmentation using fully convolutional and recurrent neural networks |
| US11636668B2 (en) * | 2017-11-10 | 2023-04-25 | Nvidia Corp. | Bilateral convolution layer network for processing point clouds |
| US10824862B2 (en) * | 2017-11-14 | 2020-11-03 | Nuro, Inc. | Three-dimensional object detection for autonomous robotic systems using image proposals |
| US10552664B2 (en) * | 2017-11-24 | 2020-02-04 | International Business Machines Corporation | Image feature classification and localization using discriminative representations for robotic surgical control |
| US11132797B2 (en) * | 2017-12-28 | 2021-09-28 | Topcon Corporation | Automatically identifying regions of interest of an object from horizontal images using a machine learning guided imaging system |
| CN108198145B (zh) * | 2017-12-29 | 2020-08-28 | 百度在线网络技术(北京)有限公司 | 用于点云数据修复的方法和装置 |
| KR102106694B1 (ko) * | 2018-05-17 | 2020-05-04 | 한국과학기술원 | 뉴럴 네트워크를 이용하는 영상 처리 장치 및 상기 장치가 수행하는 방법 |
| US11234666B2 (en) * | 2018-05-31 | 2022-02-01 | Canon Medical Systems Corporation | Apparatus and method for medical image reconstruction using deep learning to improve image quality in position emission tomography (PET) |
| CN110163048B (zh) * | 2018-07-10 | 2023-06-02 | 腾讯科技(深圳)有限公司 | 手部关键点的识别模型训练方法、识别方法及设备 |
| US10297070B1 (en) * | 2018-10-16 | 2019-05-21 | Inception Institute of Artificial Intelligence, Ltd | 3D scene synthesis techniques using neural network architectures |
| US11436743B2 (en) * | 2019-07-06 | 2022-09-06 | Toyota Research Institute, Inc. | Systems and methods for semi-supervised depth estimation according to an arbitrary camera |
-
2018
- 2018-03-22 KR KR1020180033533A patent/KR102061408B1/ko active Active
- 2018-03-23 US US16/496,960 patent/US10970520B1/en active Active
- 2018-03-23 JP JP2019552542A patent/JP6979664B2/ja active Active
- 2018-03-23 EP EP18771852.3A patent/EP3605472A4/en not_active Withdrawn
- 2018-03-23 WO PCT/KR2018/003404 patent/WO2018174623A1/ko not_active Ceased
- 2018-03-23 CN CN201880027104.8A patent/CN110574077B/zh active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000105838A (ja) * | 1998-09-29 | 2000-04-11 | Toshiba Corp | 画像表示方法及び画像処理装置 |
| KR20160101973A (ko) * | 2013-12-19 | 2016-08-26 | 아비질론 포트리스 코퍼레이션 | 비제약형 매체에 있어서 얼굴을 식별하는 시스템 및 방법 |
| KR20160061856A (ko) * | 2014-11-24 | 2016-06-01 | 삼성전자주식회사 | 객체 인식 방법 및 장치, 및 인식기 학습 방법 및 장치 |
| KR20160122452A (ko) | 2015-04-14 | 2016-10-24 | (주)한국플랫폼서비스기술 | 비주얼 콘텐츠기반 영상 인식을 위한 딥러닝 프레임워크 및 영상 인식 방법 |
Non-Patent Citations (3)
| Title |
|---|
| LEE, BEOM-JIN ET AL.: "RGB-D-T Face Recognition Using Convolutional-recursive Deep Learning", 2014 KOREAN INSTITUTE OF INFORMATION SCIENTISTS AND ENGINEERS THE 41 ST ANNUAL MEETING AND WINTER CONFERENCE, December 2014 (2014-12-01), pages 616 - 618, XP009516851 * |
| See also references of EP3605472A4 |
| TTH, BLINT PI ET AL.: "Deep Learning and SVM Classification for Plant Recognition in Content-Based Large Scale Image Retrieval", WORKING NOTES OF CLEF 2016-CONFERENCE AND LABS OF THE EVALUATION FORUM, 5 September 2016 (2016-09-05), pages 569 - 578, XP055559756 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11276249B2 (en) | 2020-05-14 | 2022-03-15 | International Business Machines Corporation | Method and system for video action classification by mixing 2D and 3D features |
Also Published As
| Publication number | Publication date |
|---|---|
| KR20180108501A (ko) | 2018-10-04 |
| CN110574077B (zh) | 2023-08-01 |
| EP3605472A4 (en) | 2020-12-23 |
| US20210103716A1 (en) | 2021-04-08 |
| JP2020513124A (ja) | 2020-04-30 |
| JP6979664B2 (ja) | 2021-12-15 |
| US10970520B1 (en) | 2021-04-06 |
| KR102061408B1 (ko) | 2019-12-31 |
| CN110574077A (zh) | 2019-12-13 |
| EP3605472A1 (en) | 2020-02-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2018174623A1 (ko) | 가상 3차원 심층 신경망을 이용하는 영상 분석 장치 및 방법 | |
| CN112236779B (zh) | 基于卷积神经网络的图像处理方法和图像处理装置 | |
| CN111291809B (zh) | 一种处理装置、方法及存储介质 | |
| EP3779800B1 (en) | Image recognition method, apparatus, electronic device and storage medium | |
| CN109543549B (zh) | 用于多人姿态估计的图像数据处理方法及装置、移动端设备、服务器 | |
| US11151361B2 (en) | Dynamic emotion recognition in unconstrained scenarios | |
| WO2020071701A1 (ko) | 딥러닝 네트워크 모델에 의한 실시간 객체 검출 방법 및 장치 | |
| WO2021103731A1 (zh) | 一种语义分割方法、模型训练方法及装置 | |
| WO2018217019A1 (ko) | 신경망 학습 기반의 변종 악성 코드를 탐지하기 위한 장치, 이를 위한 방법 및 이 방법을 수행하기 위한 프로그램이 기록된 컴퓨터 판독 가능한 기록매체 | |
| EP3500979A1 (en) | Computer device for training a deep neural network | |
| CN110163111A (zh) | 基于人脸识别的叫号方法、装置、电子设备及存储介质 | |
| WO2022165675A1 (zh) | 一种手势识别方法、装置、终端设备及可读存储介质 | |
| CN114663593A (zh) | 三维人体姿态估计方法、装置、设备及存储介质 | |
| JP7104546B2 (ja) | 情報処理装置、情報処理方法 | |
| CN114463840B (zh) | 基于骨架的移位图卷积网络人体行为识别方法 | |
| WO2019132588A1 (ko) | 영상의 특징 및 맥락에 기초한 영상 분석 장치 및 방법 | |
| WO2021049784A2 (ko) | Gan을 이용한 의료 영상의 광강도 분포 일반화 기법 | |
| CN120708092A (zh) | 一种轻量级遥感目标检测方法及系统 | |
| CN116888605B (zh) | 神经网络模型的运算方法、训练方法及装置 | |
| WO2025161249A1 (zh) | 一种人脸图像转换方法、装置及终端设备 | |
| CN115169537B (zh) | 一种神经网络的模型固件生成方法和图像识别方法 | |
| CN110889438A (zh) | 一种图像处理方法、装置、电子设备及存储介质 | |
| CN114373205B (zh) | 一种基于卷积宽度网络的人脸检测和识别方法 | |
| CN110021036A (zh) | 红外目标检测方法、装置、计算机设备和存储介质 | |
| WO2023096133A1 (ko) | 경량화된 자세 추정 모델 제공 장치 및 방법 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 18771852 Country of ref document: EP Kind code of ref document: A1 |
|
| ENP | Entry into the national phase |
Ref document number: 2019552542 Country of ref document: JP Kind code of ref document: A |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2018771852 Country of ref document: EP |
|
| ENP | Entry into the national phase |
Ref document number: 2018771852 Country of ref document: EP Effective date: 20191024 |

