[go: up one dir, main page]

WO2019112385A1 - Procédé de codage d'informations temporelles de caractéristiques spécifiques à une trame de segment d'image en vue d'une reconnaissance de vidéo - Google Patents

Procédé de codage d'informations temporelles de caractéristiques spécifiques à une trame de segment d'image en vue d'une reconnaissance de vidéo Download PDF

Info

Publication number
WO2019112385A1
WO2019112385A1 PCT/KR2018/015554 KR2018015554W WO2019112385A1 WO 2019112385 A1 WO2019112385 A1 WO 2019112385A1 KR 2018015554 W KR2018015554 W KR 2018015554W WO 2019112385 A1 WO2019112385 A1 WO 2019112385A1
Authority
WO
WIPO (PCT)
Prior art keywords
pooling
gradient
standard deviation
feature point
video
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
Application number
PCT/KR2018/015554
Other languages
English (en)
Korean (ko)
Inventor
김영석
권희승
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
POSTECH Academy Industry Foundation
Original Assignee
POSTECH Academy Industry Foundation
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by POSTECH Academy Industry Foundation filed Critical POSTECH Academy Industry Foundation
Publication of WO2019112385A1 publication Critical patent/WO2019112385A1/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
    • H04N19/136—Incoming video signal characteristics or properties

Definitions

  • the present invention relates to frame-integrated encoding of video, and more particularly to encoding for enhancing motion recognition of a person or object in a video.
  • Video recognition like image recognition, is applicable to a variety of industries.
  • the present invention can be applied not only to video retrieval and video surveillance systems most directly involved but also to fields such as medical imaging, autonomous navigation, human-robot interaction, and intelligent robots.
  • General video recognition technology is a technique for understanding human behavior through video. It recognizes and analyzes human motion in short video segments of 5 to 10 seconds. Due to the fact that video is basically composed of images of 25-30 frames per second, video recognition technology takes many of the methodologies used in image recognition technology. Specifically, when extracting feature points for each frame of an image for video recognition, a feature point extraction method for image recognition is used. Then, by integrating the feature points of each frame of the video using the feature point encoding method, a comprehensive feature point of the entire video is created to complete the video recognition technology.
  • feature point extraction method for video recognition feature point extraction method in image recognition field is mainly used.
  • human-designed feature extraction methods are described as follows.
  • HOG Histogram of Oriented Gradient
  • HoF Histogram of Flow
  • MH Motion Boundary Histogram
  • the method of extracting motion related feature points is mainly used for analyzing motion in an image.
  • the method of learning feature points through the Convolutional Neural Network (CNN) is becoming a popular feature extraction method of video recognition. All frames of the video are extracted, and the RGB image of each frame is used as the network input to learn the feature points of each frame.
  • CNN Convolutional Neural Network
  • optical flow image is produced by predicting the dense optical flow of video, and optical flow image is used as network input.
  • all of the proposed methods are based on extracting the features of each frame of the video, and there is a limitation in that the overall time information of the video can not be considered.
  • the feature point encoding methods associated with the above feature point extraction methods are typically the Bag of Words (BoW) technique and the Imporved Fisher Vector (IFV) technique.
  • the BoW technique is a technique for creating a histogram representing the comprehensive characteristics of video through clustering of extracted feature points.
  • the IFV technique clusters feature points in a more flexible manner than the BoW technique, and is a technique for modeling the comprehensive characteristics of video through the relationship of feature points. Both methods take into account the full frame information of the video but do not model the temporal change information such as frame order.
  • the 3D convolution artificial neural network method is a method of three-dimensionally designing the convolution artificial neural network by recognizing the video as a three-dimensional input by considering the time axis of the video as one dimension.
  • the convergence artificial neural network and the circular neural network are designed to recognize the time axis information by fusing a short-term memory (LSTM) circular neural network at the back of the convolution artificial neural network.
  • LSTM short-term memory
  • the present specification aims to provide a video feature point encoding method for representing video temporal information using pooling equations for video recognition.
  • a video feature point encoding method including: (a) receiving feature points generated for each frame of a video; And (b) encoding the feature points in at least one manner selected from the group consisting of Standard deviation Pooling, Gradient Max Pooling, and Gradient Standard Deviation Pooling; . ≪ / RTI >
  • the standard deviation pooling may encode a standard deviation value of feature point values for each frame on a time axis as feature points.
  • the standard deviation pooling may be encoded using the following equation.
  • the gradient max pooling may calculate the gradient value of the feature point values for each frame and encode the maximum value and the minimum value in the time axis as feature points.
  • the gradient maximum pulling can be encoded using the following equation.
  • the gradient standard deviation pooling may calculate the gradient value of the feature point values for each frame and encode the standard deviation value in the time axis as feature points.
  • the gradient standard deviation pooling can be encoded using the following equation.
  • a video feature point encoding method including: (a) receiving feature points generated for each frame of a video; (b) calculating the feature points using a method selected from the group consisting of Sum Pooling, Max Pooling, Gradient Pooling, Standard Deviation Pooling, Gradient Max Pooling and Gradient Standard Deviation Pooling Gradient Standard Deviation Pooling); And (c) concatenating the feature points calculated in the step (b) and calculating a single vector value.
  • the video feature point encoding method according to the present invention may further include (d) normalizing the vector value encoded in the step (c) by dividing the vector value by an average difference subtraction method and a standard deviation.
  • the video feature point encoding method according to the present invention can be implemented in a computer program written in a computer-readable recording medium so as to perform each step of the video feature point encoding method in a computer.
  • a video feature point encoding apparatus including: a feature point receiver for receiving feature points generated for each frame of a video; And encoding the data generated by the minutia generation unit in at least one manner selected from the group consisting of a standard deviation pooling, a gradient max pooling, and a gradient standard deviation pooling. And a pulling section for performing a pulling operation.
  • a video feature point encoding apparatus including: a feature point receiver for receiving feature points generated for each frame of a video; The minutiae are divided into a sum pooling, a max pooling, a gradient pooling, a standard deviation pooling, a gradient max pooling, and a gradient standard deviation Pooling); And a minutiae point connection unit for connecting each of the minutiae encoded in the pulling unit and calculating a single vector value.
  • the normalization unit may divide the vector value calculated by the minutiae point connecting unit by an average difference method and a standard deviation.
  • a video comprising a specific action is compressed and represented as a new integrated feature vector.
  • the integrated feature vector of the video can then achieve high video recognition performance in combination with a general classifier such as a support vector machine.
  • a new feature point pooling method including statistical information of feature points and information on time flow can be used in addition to the existing feature point pooling method. It compresses various temporal information more than the existing method, and is robust to the encoding of the motion which requires heavy shaking motion and the motion which requires long-term observation.
  • the feature point pooling method is used in connection with the feature point extraction method for each frame of video, and is used in a variety of conventional video recognition methods such as feature point extraction method of conventional image recognition, feature point extraction method of convolutional neural network model It is compatible with feature point extraction method.
  • a complete unscreened learning method that does not require learning is available regardless of the amount of data. Therefore, even if the size of the dataset is small, it can be used effectively.
  • FIG. 1 is a flow chart briefly illustrating a video feature point encoding method according to the present disclosure.
  • FIG. 2 is a flow diagram of a method of performing standard deviation pooling in accordance with one embodiment of the present disclosure.
  • FIG. 3 is a flow diagram of a method of performing gradient max-pooling in accordance with another embodiment of the present disclosure.
  • FIG. 4 is a flow diagram of a method of performing gradient max-pooling in accordance with another embodiment of the present disclosure.
  • FIG. 5 is a schematic view showing a video feature point encoding method according to another embodiment of the present invention.
  • the video feature point encoding method according to the present invention is a feature point pooling method.
  • the feature point pooling scheme is a technique that applies simple pooling equations to the extracted feature points on the time axis. It has been proved in several studies that it can be more effective than the conventional BoW and IFV techniques in expressing time axis information.
  • the video feature point encoding method according to the present invention uses a new pulling scheme that is different from the existing feature point pulling scheme, and further, a method of generating an integrated feature vector of video integrated with the existing feature point pulling scheme.
  • FIG. 1 is a flow chart briefly illustrating a video feature point encoding method according to the present disclosure.
  • the video feature point encoding method according to the present invention may include a feature point generating step (S10) and a step (S20) of pooling the generated feature points.
  • the method of generating feature points of the video (S10) can generate feature points for each frame of the video through the existing feature point extraction method.
  • the creation of the minutiae for each frame is a well known technique widely known to those skilled in the art and is not a technical core in the present specification, and thus a detailed description thereof will be omitted.
  • the feature points can be encoded in at least one manner selected from the group consisting of Standard deviation Pooling, Gradient Max Pooling, and Gradient Standard deviation Pooling. have.
  • FIG. 2 is a flow diagram of a method of performing standard deviation pooling in accordance with one embodiment of the present disclosure.
  • the standard deviation pooling may encode statistical information of the feature point values for each frame on the time axis to generate new feature points, and may encode the standard deviation value of the feature point values for each frame on the time axis into new feature points.
  • the standard deviation pooling may be encoded using Equation 1 below.
  • FIG. 3 is a flow diagram of a method of performing gradient max-pooling in accordance with another embodiment of the present disclosure.
  • the gradient maximum pulling is a method of generating new feature points by extracting a change value according to temporal flow of the feature points. After calculating the gradient values of the minutiae values of the extracted frames, the maximum and minimum values in the time axis can be encoded into new minutiae. By observing the maximum value and the minimum value of the gradient value, it is possible to expect an increase in the recognition rate with respect to the moment at which the motion is changed most or the impact moment in the video.
  • the gradient maximum pulling can be encoded using Equation (2) below.
  • FIG. 4 is a flow diagram of a method of performing gradient max-pooling in accordance with another embodiment of the present disclosure.
  • Gradient standard deviation pulling is also a method of generating new feature points by extracting the change value according to the temporal flow of the feature points. After calculating the gradient values of the extracted feature point values, the standard deviation values on the time axis can be encoded as feature points. By observing the standard deviation of the gradient value, it is possible to recognize the feature points having large variation width in each video and extract common feature points for each operation.
  • the gradient standard deviation pooling may be encoded using Equation 3 below.
  • the standard deviation pooling, the gradient maximum pooling, and the gradient standard deviation pooling may be used in combination of two, or all three, if only one method can be used.
  • feature points representing temporal information of the entire image can be created in association with the feature point extraction method for each existing image frame.
  • FIG. 5 is a schematic view showing a video feature point encoding method according to another embodiment of the present invention.
  • the three existing minutiae point pooling methods used in the present specification are respectively Sum Pooling, Max Pooling, and Gradient Pooling.
  • the integrated pooling is a method of making all the extracted minutiae points on a time axis into a single vector by a time axis.
  • the max pooling method is a method of vectorizing minutiae values of extracted minutiae only on a time axis.
  • Gradient pulling is a method of obtaining gradient values of feature points extracted from a frame and then separating positive and negative values along the time axis into a single vector.
  • Equation 4 expresses the integrated pooling as an equation. Represents the value of the kth feature point among the feature points of the t-th frame and the D dimension among the N frames. Are time series numbers of the start frame and the end frame, respectively. Is the result of integrated pooling. As can be seen from Equation (4), the integrated pulling adds feature values along the time axis.
  • Equation (5) expresses MaxPulling as an equation. Represents the result value of the max pooling. The definitions of the remaining values are the same as in Equation (4). As shown in Equation (5), MaxPulling represents the maximum value of the feature values along the time axis.
  • Equation (6) expresses the gradient pooling as an equation. Are the gradient pulling vector result values of the positive and negative values, respectively. As shown in Equation (6), the calculation of the gradient is calculated on the time axis by the difference between the frame feature point and the previous frame feature point.
  • the present invention can be embodied as a computer program recorded on a computer-readable recording medium so as to perform each step of the video feature point encoding method according to the present invention.
  • the computer program may be stored in a computer readable medium such as C, C ++, JAVA, machine language, or the like that can be read by the processor (CPU) of the computer through the device interface of the computer, in order for the computer to read the program, And may include a code encoded in a computer language.
  • Such code may include a functional code related to a function or the like that defines necessary functions for executing the above methods, and includes a control code related to an execution procedure necessary for the processor of the computer to execute the functions in a predetermined procedure can do. Further, such code may further include memory reference related code as to whether the additional information or media needed to cause the processor of the computer to execute the functions should be referred to at any location (address) of the internal or external memory of the computer have.
  • the code may be communicated to any other computer or server remotely using the communication module of the computer
  • a communication-related code for determining whether to communicate, what information or media should be transmitted or received during communication, and the like.
  • the medium to be stored is not a medium for storing data for a short time such as a register, a cache, a memory, etc., but means a medium that semi-permanently stores data and is capable of being read by a device.
  • examples of the medium to be stored include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, but are not limited thereto.
  • the program may be stored in various recording media on various servers to which the computer can access, or on various recording media on the user's computer.
  • the medium may be distributed to a network-connected computer system so that computer-readable codes may be stored in a distributed manner.
  • the video feature point encoding method according to the present invention can be implemented by a video feature point encoding apparatus including a feature point generating unit and a pulling unit.
  • the feature point generation unit may generate feature points for each frame of the video.
  • the pooling unit may be configured to calculate the data generated in the minutia generation unit by at least one method selected from the group consisting of Standard deviation Pooling, Gradient Max Pooling, and Gradient Standard Deviation Pooling. ≪ / RTI >
  • the pooling unit may divide the minutiae points into at least one of a sum pooling, a max pooling, a gradient pooling, a standard deviation pooling, a gradient max pooling, Max Pooling and Gradient Standard deviation Pooling, respectively.
  • the video feature point encoding apparatus includes a feature point connection unit for connecting each feature point encoded in the pulling unit and calculating a vector value, and a normalization unit for dividing the vector value calculated in the feature point connection unit by an average difference method and a standard deviation And the like.
  • the video feature point encoding apparatus described in connection with the embodiments of the present disclosure may be implemented directly in hardware, in software modules executed by hardware, or in a combination thereof.
  • the software module may be a random access memory (RAM), a read only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, a hard disk, a removable disk, a CD- May reside in any form of computer readable recording medium known in the art to which this disclosure belongs.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)

Abstract

La présente invention concerne un procédé de codage de caractéristiques de vidéo selon lequel, en vue d'une reconnaissance de vidéo, des informations temporelles d'une vidéo sont exprimées à l'aide des formules d'agrégation proposées dans la description. Le procédé de codage de caractéristiques de vidéo selon l'invention peut comprendre les étapes consistant : (a) à recevoir des caractéristiques générées en fonction des trames d'une vidéo; (b) à coder les caractéristiques au moyen d'une agrégation par somme, d'une agrégation par valeur maximale, d'une agrégation par gradient, d'une agrégation par écart-type, d'une agrégation par gradient maximal et d'une agrégation par écart-type de gradient, respectivement; et (c) à calculer un vecteur unique par mise en relation des caractéristiques respectives produites à l'étape (b).
PCT/KR2018/015554 2017-12-04 2018-12-07 Procédé de codage d'informations temporelles de caractéristiques spécifiques à une trame de segment d'image en vue d'une reconnaissance de vidéo Ceased WO2019112385A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR10-2017-0165382 2017-12-04
KR1020170165382A KR101936947B1 (ko) 2017-12-04 2017-12-04 비디오 인식을 위한 영상 세그먼트 프레임별 특징점의 시간 정보 인코딩 방법

Publications (1)

Publication Number Publication Date
WO2019112385A1 true WO2019112385A1 (fr) 2019-06-13

Family

ID=65017310

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2018/015554 Ceased WO2019112385A1 (fr) 2017-12-04 2018-12-07 Procédé de codage d'informations temporelles de caractéristiques spécifiques à une trame de segment d'image en vue d'une reconnaissance de vidéo

Country Status (2)

Country Link
KR (1) KR101936947B1 (fr)
WO (1) WO2019112385A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021177628A1 (fr) * 2020-03-04 2021-09-10 Samsung Electronics Co., Ltd. Procédé et appareil de reconnaissance d'action

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102174658B1 (ko) * 2019-03-27 2020-11-05 연세대학교 산학협력단 비디오 행동 인식 및 행동 구간 탐지 장치 및 방법
KR20210058396A (ko) * 2019-11-14 2021-05-24 삼성전자주식회사 전자 장치 및 그 제어 방법

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20110132770A (ko) * 2010-06-03 2011-12-09 한국과학기술원 디지털 포렌직 기술을 이용한 보안 인쇄물 위조식별 시스템 및 그 방법
KR20120065915A (ko) * 2010-12-13 2012-06-21 한국과학기술원 스테레오스코픽 영상 시스템에서의 시각적 불편감 측정 장치 및 방법, 시각적 불편감 측정 방법을 실행하기 위한 프로그램이 기록된 기록매체
US20130103402A1 (en) * 2011-10-25 2013-04-25 At&T Intellectual Property I, L.P. System and method for combining frame and segment level processing, via temporal pooling, for phonetic classification
US20130156304A1 (en) * 2010-07-01 2013-06-20 Telefonica, S.A. Method for classification of videos

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20110132770A (ko) * 2010-06-03 2011-12-09 한국과학기술원 디지털 포렌직 기술을 이용한 보안 인쇄물 위조식별 시스템 및 그 방법
US20130156304A1 (en) * 2010-07-01 2013-06-20 Telefonica, S.A. Method for classification of videos
KR20120065915A (ko) * 2010-12-13 2012-06-21 한국과학기술원 스테레오스코픽 영상 시스템에서의 시각적 불편감 측정 장치 및 방법, 시각적 불편감 측정 방법을 실행하기 위한 프로그램이 기록된 기록매체
US20130103402A1 (en) * 2011-10-25 2013-04-25 At&T Intellectual Property I, L.P. System and method for combining frame and segment level processing, via temporal pooling, for phonetic classification

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
BOUREAU, Y-LAN ET AL.: "A Theoretical Analysis of Feature Pooling in Visual Recognition", PROC. OF 27TH INT'L CONF. ON MACHINE LEARNING, 21 June 2010 (2010-06-21), pages 111 - 118, XP055286960 *
HUANG, Y. ET AL.: "Feature Coding in Image Classification: A Comprehensive Study", IEEE TRANS., ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, vol. 36, no. 3, March 2014 (2014-03-01), XP011538957 *
PIERGIOVANNI, A. J. ET AL.: "Learning Latent Subevents in Activity Videos Using Temporal Attention Filters", THIRTY-FIRST AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, 4 February 2017 (2017-02-04), Hilton San Francisco Union Square in San Francisco, California, USA, XP055616384, Retrieved from the Internet <URL:https://aaai.org/ocs/index.php/AAAI/AAAI17/paper/viewFile/15005/14307> *
WANG, PENG ET AL.: "Temporal Pyramid Pooling-Based Convolutional Neural Network for Action Recognition", IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, vol. 27, no. 12, 7 June 2016 (2016-06-07), pages 2613 - 2622, XP011674534, Retrieved from the Internet <URL:https://ieeexplore.ieee.org/abstract/document/7484698> *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021177628A1 (fr) * 2020-03-04 2021-09-10 Samsung Electronics Co., Ltd. Procédé et appareil de reconnaissance d'action
US11837025B2 (en) 2020-03-04 2023-12-05 Samsung Electronics Co., Ltd. Method and apparatus for action recognition

Also Published As

Publication number Publication date
KR101936947B1 (ko) 2019-01-09

Similar Documents

Publication Publication Date Title
WO2022059969A1 (fr) Procédé de pré-apprentissage de réseau neuronal profond permettant une classification de données d&#39;électrocardiogramme
WO2018228218A1 (fr) Procédé d&#39;identification, dispositif informatique et support de stockage
EP3568828A1 (fr) Appareil et procédé de traitement d&#39;images utilisant une carte de caractéristiques multicanaux
WO2021232985A1 (fr) Procédé et appareil de reconnaissance faciale, dispositif informatique et support de stockage
WO2017164478A1 (fr) Procédé et appareil de reconnaissance de micro-expressions au moyen d&#39;une analyse d&#39;apprentissage profond d&#39;une dynamique micro-faciale
CN113570689B (zh) 人像卡通化方法、装置、介质和计算设备
WO2019119396A1 (fr) Procédé et dispositif de reconnaissance d&#39;expression faciale
WO2019112385A1 (fr) Procédé de codage d&#39;informations temporelles de caractéristiques spécifiques à une trame de segment d&#39;image en vue d&#39;une reconnaissance de vidéo
CN113313241A (zh) 确定深度学习模型的张量信息的方法和计算装置
WO2021241983A1 (fr) Procédé et appareil d&#39;apprentissage semi-supervisé
CN110472499B (zh) 一种行人重识别的方法及装置
CN113298158A (zh) 数据检测方法、装置、设备及存储介质
CN113705307A (zh) 图像处理方法、装置、设备及存储介质
WO2024034984A1 (fr) Procédé et système de prédiction de l&#39;apparition d&#39;une hypotension intradialytique
CN115328319A (zh) 一种基于轻量型手势识别的智能控制方法及装置
KR20210018586A (ko) 등장인물의 생체특징을 이용하는 비디오 콘텐트 식별 방법 및 장치
CN117718969B (zh) 基于视觉听觉融合的家用机器人控制系统及其方法
CN113780066B (zh) 行人重识别方法、装置、电子设备及可读存储介质
CN119625821A (zh) 行为检测方法、存储介质及计算机程序产品
CN115240647A (zh) 声音事件检测方法、装置、电子设备及存储介质
CN113963237A (zh) 模型训练、戴口罩状态检测方法、电子设备及存储介质
CN112989869A (zh) 人脸质量检测模型的优化方法、装置、设备及存储介质
CN113963426A (zh) 模型训练、戴口罩人脸识别方法、电子设备及存储介质
CN114842261B (zh) 图像处理方法、装置、电子设备及存储介质
WO2023090596A1 (fr) Procédé et système de synthèse de visage

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: 18886953

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18886953

Country of ref document: EP

Kind code of ref document: A1