Boost 3D Object Detection via Point Clouds Segmentation and Fused 3D GIoU-L1 Loss | |
Yaran Chen2; Haoran Li2; Ruiyuan Gao1; Dongbin Zhao2 | |
刊名 | IEEE Transactions on Neural Networks and Learning Systems |
2021 | |
期号 | 无页码:无 |
关键词 | 3D object detection, GIoU loss, Segmentation |
英文摘要 | 3D object detection is crucial for many real-world applications, attracting many researchers’ attention. Beyond 2D object detection, 3D object detection usually needs to extract appearance, depth, position and orientation information from the LiDAR and camera sensors. However, due to more degrees- of-freedom and vertices, existing detection methods that directly transform from 2D to 3D still face several challenges, such as exploding increase of anchors’ number and inefficient or hard- to-optimize objective. To this end, we present a fast segmentation method for 3D point clouds to reduce anchors, which can largely decrease computing cost. Moreover, taking advantages of 3D GIoU and L1 losses, we propose a fused loss to facilitate the optimization of 3D object detection. A series of experiments show the proposed method has alleviated the issues above effectively |
内容类型 | 期刊论文 |
源URL | [http://ir.ia.ac.cn/handle/173211/40607] |
专题 | 复杂系统管理与控制国家重点实验室_深度强化学习 |
通讯作者 | Dongbin Zhao |
作者单位 | 1.Chinese University of Hong Kong 2.Institute of Automation, Chinese Academy of Sciences |
推荐引用方式 GB/T 7714 | Yaran Chen,Haoran Li,Ruiyuan Gao,et al. Boost 3D Object Detection via Point Clouds Segmentation and Fused 3D GIoU-L1 Loss[J]. IEEE Transactions on Neural Networks and Learning Systems,2021(无):无. |
APA | Yaran Chen,Haoran Li,Ruiyuan Gao,&Dongbin Zhao.(2021).Boost 3D Object Detection via Point Clouds Segmentation and Fused 3D GIoU-L1 Loss.IEEE Transactions on Neural Networks and Learning Systems(无),无. |
MLA | Yaran Chen,et al."Boost 3D Object Detection via Point Clouds Segmentation and Fused 3D GIoU-L1 Loss".IEEE Transactions on Neural Networks and Learning Systems .无(2021):无. |
个性服务 |
查看访问统计 |
相关权益政策 |
暂无数据 |
收藏/分享 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论