公开(公告)号：US20230334321A1

公开(公告)日：2023-10-19

申请号：US17976655

申请日：2022-10-28

Applicant: ELECTRONICS AND TELECOMMUNICATIONS RESEARCH INSTITUTE

Inventor： Su Woong LEE ,  Jong-Gook KO ,  Wonyoung YOO ,  Seungjae LEE ,  Yongsik LEE ,  Juwon LEE ,  Da-Un JUNG

IPC: G06N3/08

CPC classification number: G06N3/082 ,  G06N3/084

Abstract: Disclosed are a deep neural network lightweight device based on batch normalization, and a method thereof. The deep neural network lightweight device based on batch normalization includes a memory that stores at least one data and at least one processor that executes a network lightweight module. When executing the network lightweight module, the processor performs learning on an input neural network based on sparsity regularization to adaptively determine at least one parameter of the sparsity regularization, performs pruning on the learning result, and performs fine tuning on the pruning result.

公开(公告)号：US20240177473A1

公开(公告)日：2024-05-30

申请号：US18496063

申请日：2023-10-27

Applicant: ELECTRONICS AND TELECOMMUNICATIONS RESEARCH INSTITUTE

Inventor： Jaehwan KIM ,  Jung Jae YU ,  Wonyoung YOO ,  Juwon LEE

IPC: G06V10/82 ,  G06T7/73 ,  G06V10/77

CPC classification number: G06V10/82 ,  G06T7/75 ,  G06V10/7715 ,  G06T2207/30196

Abstract: A neural network device for learning dependency of feature data includes: a memory in which at least one program is stored; and a processor that performs a calculation by executing the at least one program, in which the processor is configured to acquire graph information including a data node for a human body; extract feature data corresponding to a plurality of joints constituting the human body from the graph information; acquire a self-attention output corresponding to the feature data based on a self-attention mechanism; and generate result data for a motion of the human body based on the self-attention output, and the self-attention output includes position information acquired based on positional encoding of the feature data and structural information acquired based on geodesic encoding of the feature data.

公开(公告)号：US20230237629A1

公开(公告)日：2023-07-27

申请号：US18077434

申请日：2022-12-08

Applicant: ELECTRONICS AND TELECOMMUNICATIONS RESEARCH INSTITUTE

Inventor： Juwon LEE ,  Jung Jae YU ,  Wonyoung YOO

IPC: G06T5/50 ,  G06V40/16 ,  G06T7/11 ,  G06T3/00

CPC classification number: G06T5/50 ,  G06V40/171 ,  G06T7/11 ,  G06T3/0068 ,  G06T3/0093 ,  G06T2207/30201 ,  G06T2207/20221

Abstract: A method and apparatus for generating a face-harmonized image are disclosed. The method of generating a face-harmonized image includes receiving an input image, extracting facial landmarks from a target image and the input image, generating a face-removed image of the target image based on a facial mask region, extracting a user face image from the input image, transforming the user face image to correspond to the facial mask region, generating a face-blended image by blending the transformed user face image with the target image, extracting a feature map of the face-blended image, generating a combined feature map based on the feature map of the face-blended image and a feature map of the target image, generating a face harmonization result image based on the combined feature map, and providing the generated face harmonization result image.

公开(公告)号：US20220222525A1

公开(公告)日：2022-07-14

申请号：US17554870

申请日：2021-12-17

Applicant: ELECTRONICS AND TELECOMMUNICATIONS RESEARCH INSTITUTE

Inventor： Su Woong LEE ,  Seungjae LEE ,  Jong-Gook KO ,  Wonyoung YOO ,  Jung Jae YU ,  Keun Dong LEE ,  Yongsik LEE ,  Da-Un JUNG

IPC: G06N3/08 ,  G06K9/62

Abstract: Provided are a method and system for training a dynamic deep neural network. The method for training a dynamic deep neural network includes receiving an output of a last layer of the deep neural network and outputting a first loss, receiving an output of a routing module according to an input class of the deep neural network and outputting a second loss, calculating a third loss based on the first loss and the second loss, and updating a weight of the deep neural network by using the third loss.

公开(公告)号：US20210350241A1

公开(公告)日：2021-11-11

申请号：US17242604

申请日：2021-04-28

Applicant: ELECTRONICS AND TELECOMMUNICATIONS RESEARCH INSTITUTE

Inventor： Seungjae LEE ,  Jong-Gook KO ,  Keun Dong LEE ,  Su Woong LEE ,  Yongsik LEE ,  Da-Un JUNG ,  Wonyoung YOO

IPC: G06N3/08 ,  G06N3/04

Abstract: An apparatus and method for searching a neural network architecture may be disclosed. The apparatus may include an architecture searcher and an architecture evaluator. The architecture searcher may search for a topology between nodes included in a basic cell of a network, search for an operation to be applied between the nodes after searching for the topology, and determine the basic cell. The architecture evaluator may evaluate performance of the determined basic cell.