ETRI-Knowledge Sharing Plaform

ENGLISH

성과물

논문 검색
구분 SCI
연도 ~ 키워드

상세정보

학술대회 GSR-MAR: Global Super-Resolution for Person Multi-Attribute Recognition
Cited 0 time in scopus Download 2 time Share share facebook twitter linkedin kakaostory
저자
시아다리, 한미경, 윤현진
발행일
201910
출처
International Conference on Computer Vision Workshops (ICCVW) 2019, pp.1098-1103
DOI
https://dx.doi.org/10.1109/ICCVW.2019.00140
협약과제
19MH1500, 5G 기반의 스마트시티 서비스 개발 및 실증, 한미경
초록
Person attribute recognition aims to predict attribute labels based on person's appearance usually captured from surveillance cameras. It is a challenging problem in computer vision due to poor imaging quality with complex background clutter and unconstrained viewing conditions from various angles and distances between person and surveillance cameras. In this paper, we address such a problem using an end-to-end network called Global Super-Resolution for Multi Attribute Recognition (GSR-MAR). GSR-MAR integrates a conversion process of low-resolution input images into high-resolution images and predicts person attributes from input images. Before performing the classification process, GSR-MAR not only converts low-resolution images to high-resolution images to recover details of image textures but also captures larger context information by using large separable convolutional layers. The experiment results on two popular benchmark datasets demonstrate the performance improvement and effectiveness of our GSR-MAR model over competing baselines.
키워드
Attribute, Global, Person, Recognition, Super resolution
KSP 제안 키워드
Attribute recognition, Background clutter, Benchmark datasets, Classification process, Complex background, Computer Vision(CV), Context Information, End to End(E2E), Experiment results, Low-resolution images, Super resolution