Method for improving the efficiency of deep learning-based feature map compression

The present invention provides a method for improving the efficiency of deep learning-based feature map compression. According to the present invention, an encoder can selectively transmit a feature map, restore the image quality of the transmitted feature map using a deep neural network, and then p...

Full description

Saved in:
Bibliographic Details
Main Authors KWONSEONG, SIM DONG GYU
Format Patent
LanguageEnglish
Korean
Published 01.09.2023
Subjects
CALCULATING
COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
COMPUTING
COUNTING
ELECTRIC COMMUNICATION TECHNIQUE
ELECTRICITY
IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
PHYSICS
PICTORIAL COMMUNICATION, e.g. TELEVISION
Online AccessGet full text

Cover

More Information
Summary:The present invention provides a method for improving the efficiency of deep learning-based feature map compression. According to the present invention, an encoder can selectively transmit a feature map, restore the image quality of the transmitted feature map using a deep neural network, and then predict and generate a feature map that has not been transmitted. Therefore, effective transmission of a feature pyramid can be performed. 본 발명은 딥러닝 기반 피쳐 맵 압축 효율 향상을 위한 방법일 수 있다. 선택적으로 부호화기에서 피쳐 맵을 전송할 수 있고, 심층 신경망 네트워크를 이용하여 전송된 피쳐 맵의 화질을 복원한 후, 전송하지 않은 피쳐 맵을 예측하여 생성할 수 있다.
Bibliography:Application Number: KR20220108356