Title:
METHOD AND SYSTEM FOR ADAPTIVE FEATURE FUSION IN CONVOLUTIONAL NEURAL NETWORK
Document Type and Number:
WIPO Patent Application WO/2023/077998
Kind Code:
A1
Abstract:
The present disclosure relates to a method and system for adaptive feature fusion in a convolutional neural network. The method comprises: acquiring a weight coefficient of a feature, which is in each scale, of the current feature fusion layer; activating and normalizing the weight coefficient of the feature, which is in at least one scale, of the current feature fusion; performing weighted fusion on the feature, which is in the at least one scale, of the current feature fusion layer; and splicing results obtained after weighted fusion is performed, so as to obtain a result of adaptive feature fusion, thereby completing adaptive feature fusion in a convolutional neural network.
Inventors:
LUO JING (CN)
LIU YANG (CN)
KONG XIANGBIN (CN)
LI HONGYAN (CN)
SHEN ZHIZHONG (CN)
LI JIE (CN)
WANG XUESONG (CN)
MA LIWEN (CN)
CHEN SHUJUN (CN)
LIU YANG (CN)
KONG XIANGBIN (CN)
LI HONGYAN (CN)
SHEN ZHIZHONG (CN)
LI JIE (CN)
WANG XUESONG (CN)
MA LIWEN (CN)
CHEN SHUJUN (CN)
Application Number:
PCT/CN2022/121730
Publication Date:
May 11, 2023
Filing Date:
September 27, 2022
Export Citation:
Assignee:
CRSC COMMUNICATION & INFORMATION GROUP COMPANY LTD (CN)
International Classes:
G06N3/04; G06V10/774; G06V10/80; G06V10/82
Foreign References:
| CN114092760A | 2022-02-25 | |||
| CN112183295A | 2021-01-05 | |||
| CN111753752A | 2020-10-09 | |||
| CN111797779A | 2020-10-20 | |||
| CN113111975A | 2021-07-13 | |||
| US20190236411A1 | 2019-08-01 |
Attorney, Agent or Firm:
TEE&HOWE INTELLECTUAL PROPERTY ATTORNEYS (CN)
Download PDF: