ETRI-Knowledge Sharing Plaform

ENGLISH

성과물

논문 검색
구분 SCI
연도 ~ 키워드

상세정보

학술대회 An Intensive Study of Backbone and Architectures with Test Image Augmentation and Box Refinement for Object Detection and Segmentation
Cited 1 time in scopus Download 0 time Share share facebook twitter linkedin kakaostory
저자
이근동, 고종국, 유원영
발행일
201910
출처
International Conference on Information and Communication Technology Convergence (ICTC) 2019, pp.673-677
DOI
https://dx.doi.org/10.1109/ICTC46691.2019.8939591
협약과제
19HS2300, 객체추출 및 실-가상 정합 지원 모바일 AR 기술 개발, 고종국
초록
In this paper, we present an intensive study of various backbones and architectures with test augmentation and box refinement for object detection and segmentation. Recently, several models discovered by Neural Architecture Search achieve the state-of-the-art in ImageNet Classification. However, their robustness on detection task is not yet verified. While various architectures have been proposed to improve the accuracy, they are rarely evaluated with test-time image augmentation and box refinement which can further boost their gains. In this work, we thoroughly evaluate them on challenging COCO dataset to identify the robustness of backbones on detection task, the effect of various architectures and the state-of-the-art accuracy on object detection and segmentation.
키워드
box refinement, deep learning, object detection, object segmentation, test augmentation
KSP 제안 키워드
Detection task, Image Augmentation, Object detection, Object segmentation, deep learning(DL), state-of-The-Art