文献类型：期刊文献

英文题名：CLEAN: Category Knowledge-Driven Compression Framework for Efficient 3D Object Detection

作者：Zhang, Haonan[1];Liu, Longjun[2,3];Hui, Fei[1];Zhang, Bo[4];Zhang, Hengmin[5];Zha, Zhiyuan[6]

机构：[1]Changan Univ, Sch Elect & Control Engn, Xian 710018, Peoples R China;[2]Xi An Jiao Tong Univ, Natl Key Lab Human Machine Hybrid Augmented Intell, Xian 710049, Peoples R China;[3]Xi An Jiao Tong Univ, Inst Artificial Intelligence & Robot, Xian 710049, Peoples R China;[4]Sichuan Univ, Sch Mech Engn, Chengdu 610065, Peoples R China;[5]East China Univ Sci & Technol, Sch Informat Sci & Engn, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[6]Jilin Univ, Coll Commun Engn, Changchun 130012, Peoples R China

年份：2025

卷号：47

期号：10

起止页码：8740

外文期刊名：IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE

收录：;EI(收录号：20252718725154);WOS:【SCI-EXPANDED(收录号:WOS:001571487700037)】；

基金：This work was supported in part by the National Natural Science Foundation of China under Grant NSFC 62088102 and Grant NSFC 52172380, in part by the Fundamental Research Funds for the Central Universities (CHD) under Grant 300102325106, in part by the Fundamental Research Funds for the Central Universities (XJTU) under Grant xzy012024066, in part by the National Basic Strengthen Research Program of ReRAM under Grant 2022-00-03, and in part by the Open Project Program of Shanghai Key Laboratory of Data Science.

语种：英文

外文关键词：Detectors; Three-dimensional displays; Point cloud compression; Accuracy; Object detection; Feature extraction; Electronic mail; Training; Image coding; Artificial intelligence; Model compression; knowledge distillation; network pruning; LiDAR-based 3D object detection

摘要：Deep neural networks (DNNs) are potent in LiDAR-based 3D object detection (LiDAR-3DOD), yet their deployment remains daunting due to their cumbersome parameters and computations. Knowledge distillation (KD) is promising for compressing DNNs in LiDAR-3DOD. However, most existing KD methods transfer inadequate knowledge between homogeneous detectors, and do not thoroughly explore optimal student architectures, resulting in insufficient gains for compact student detectors. To this end, we propose a category knowledge-driven compression framework to achieve efficient LiDAR-based 3D detectors. Firstly, we distill knowledge from two-stage teacher detectors to one-stage student detectors, overcoming the limitations of homogeneous pairs. To conduct KD in these heterogeneous pairs, we explore the gap between heterogeneous detectors, and introduce category knowledge-driven KD (CaKD), which includes both student-oriented distillation and two-stage-oriented label assignment distillation. Secondly, to search for the optimal architecture of compact student detectors, we introduce a masked category knowledge-driven structured pruning scheme. This scheme evaluates filter importance by analyzing the changes in category predictions related to foreground regions before and after filter removal, and prunes the less important filters accordingly. Finally, we propose a modified IoU-aware redundancy elimination module to remove redundant false positive samples, thereby further improving the accuracy of detectors. Experiments on various point cloud datasets demonstrate that our method delivers impressive results. For example, on KITTI, several compressed one-stage detectors outperform two-stage detectors in both efficiency and accuracy. Besides, on WOD-mini, our framework reduces the memory footprint of CenterPoint by 5.2x and improves the L2 mAPH by 0.55$\%$%.