Designing of loss function for 3d pedestrian detection using centernet

Kim, C.Y.; Lee, D.H.; Kim, H.J.; Memon, A.A.; Iqbal, E.; Choi, Kwang Nam

doi:10.1145/3442536.3442538

상세 보기

Designing of loss function for 3d pedestrian detection using centernet

Kim, C.Y.;
Lee, D.H.;
Kim, H.J.;
Memon, A.A.;
Iqbal, E.;
... Choi, Kwang Nam

Citations

SCOPUS

0

초록

Pedestrian detection has been a popular research topic in the last decade. In the past, anchor-based networks, for example, 1-stage and 2-stage detector, were famous for pedestrian detection. However, keypoint-based networks among anchor-free networks have been proposed recently and show high performance compared to anchor-based networks. CenterNet is a kind of keypoint-based network used for object detection. We modified the loss Function of CenterNet and proposed a weight function to train an object's height and width for 3D pedestrian detection. The evaluation of 3D pedestrian detection with the modified loss function is performed using the KITTI dataset's monocular images. The proposed loss function improves accuracy in the 3D pedestrian detection network compared to the original loss function. © 2020 ACM.

키워드

3D Object Detection; Deep Learning; Monocular 3D Object Detection; Object Detection; Pedestrian Detection

제목: Designing of loss function for 3d pedestrian detection using centernet

저자: Kim, C.Y.; Lee, D.H.; Kim, H.J.; Memon, A.A.; Iqbal, E.; Choi, Kwang Nam

DOI: 10.1145/3442536.3442538

발행일: 2020-12

유형: Conference Paper

저널명: ACM International Conference Proceeding Series

페이지: 5 ~ 10

언어: ENG

출판사: Association for Computing Machinery

분량: 6 페이지

ISSN: P 0000-0000