基于单目摄像机的三维目标检测器的区域感知知识提取

IF 4.2 3区计算机科学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

ICT Express Pub Date : 2025-08-01 DOI:10.1016/j.icte.2025.04.012

Se-Gwon Cheon, Hyuk-Jin Shin, Seung-Hwan Bae

{"title":"基于单目摄像机的三维目标检测器的区域感知知识提取","authors":"Se-Gwon Cheon, Hyuk-Jin Shin, Seung-Hwan Bae","doi":"10.1016/j.icte.2025.04.012","DOIUrl":null,"url":null,"abstract":"<div><div>Recent knowledge distillation (KD) for 3D object detection often involves costly LiDAR or multi-camera data. We focus on monocular camera-based 3D detectors, where missing 3D cues cause large feature gaps. To address this, we propose region-aware KD, aligning object features by matching their scales and pyramid levels. We introduce a probabilistic distribution to weigh region importance. Applied to MonoRCNN++ and MonoDETR on the KITTI and Waymo dataset, our approach achieves reduced complexity and strong performance with a lightweight backbone. Compared to recent KD methods, ours excels in both effectiveness and efficiency.</div></div>","PeriodicalId":48526,"journal":{"name":"ICT Express","volume":"11 4","pages":"Pages 696-702"},"PeriodicalIF":4.2000,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Region-aware knowledge distillation between monocular camera-based 3D object detectors\",\"authors\":\"Se-Gwon Cheon, Hyuk-Jin Shin, Seung-Hwan Bae\",\"doi\":\"10.1016/j.icte.2025.04.012\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>Recent knowledge distillation (KD) for 3D object detection often involves costly LiDAR or multi-camera data. We focus on monocular camera-based 3D detectors, where missing 3D cues cause large feature gaps. To address this, we propose region-aware KD, aligning object features by matching their scales and pyramid levels. We introduce a probabilistic distribution to weigh region importance. Applied to MonoRCNN++ and MonoDETR on the KITTI and Waymo dataset, our approach achieves reduced complexity and strong performance with a lightweight backbone. Compared to recent KD methods, ours excels in both effectiveness and efficiency.</div></div>\",\"PeriodicalId\":48526,\"journal\":{\"name\":\"ICT Express\",\"volume\":\"11 4\",\"pages\":\"Pages 696-702\"},\"PeriodicalIF\":4.2000,\"publicationDate\":\"2025-08-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"ICT Express\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S2405959525000578\",\"RegionNum\":3,\"RegionCategory\":\"计算机科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"COMPUTER SCIENCE, INFORMATION SYSTEMS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"ICT Express","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2405959525000578","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}

引用次数: 0

摘要

最近用于三维目标检测的知识蒸馏（KD）通常涉及昂贵的激光雷达或多相机数据。我们专注于基于单目相机的3D探测器，其中缺少3D线索会导致大的特征空白。为了解决这个问题，我们提出了区域感知KD，通过匹配目标的尺度和金字塔水平来对齐目标特征。我们引入一个概率分布来衡量区域的重要性。应用于KITTI和Waymo数据集上的monorcn++和MonoDETR，我们的方法通过轻量级主干实现了降低复杂性和强大的性能。与最近的KD方法相比，我们的方法在有效性和效率方面都很出色。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

查看原文本刊更多论文

Region-aware knowledge distillation between monocular camera-based 3D object detectors

Recent knowledge distillation (KD) for 3D object detection often involves costly LiDAR or multi-camera data. We focus on monocular camera-based 3D detectors, where missing 3D cues cause large feature gaps. To address this, we propose region-aware KD, aligning object features by matching their scales and pyramid levels. We introduce a probabilistic distribution to weigh region importance. Applied to MonoRCNN++ and MonoDETR on the KITTI and Waymo dataset, our approach achieves reduced complexity and strong performance with a lightweight backbone. Compared to recent KD methods, ours excels in both effectiveness and efficiency.

求助全文

通过发布文献求助，成功后即可免费获取论文全文。去求助

来源期刊

ICT Express Multiple-

CiteScore

10.20

自引率

1.90%

发文量

167

审稿时长

35 weeks

期刊介绍： The ICT Express journal published by the Korean Institute of Communications and Information Sciences (KICS) is an international, peer-reviewed research publication covering all aspects of information and communication technology. The journal aims to publish research that helps advance the theoretical and practical understanding of ICT convergence, platform technologies, communication networks, and device technologies. The technology advancement in information and communication technology (ICT) sector enables portable devices to be always connected while supporting high data rate, resulting in the recent popularity of smartphones that have a considerable impact in economic and social development.