PointLAM: Local Attentive Mamba for Efficient Point-based 3D Object Detection

Xuanming Shang; Weijia Zhang; Chao Ma

cs.CVSep 18, 2026

PointLAM: Local Attentive Mamba for Efficient Point-based 3D Object Detection

Authors: Xuanming Shang, Weijia Zhang, Chao Ma

Organizations: MoE Key Lab of Artificial Intelligence, Institute of AI, Shanghai Jiao Tong University, Shanghai, China

Abstract

3D object detection from LiDAR point clouds faces a fundamental dilemma: voxel-based methods achieve efficiency at the cost of geometric quantization, while point-based methods preserve fidelity but suffer from prohibitive computational bottlenecks. Specifically, point-based architectures are crippled by slow downsampling strategies (e.g., FPS) and expensive dynamic neighbor queries (e.g., k-NN) coupled with costly continuous interactions. To tackle these systemic inefficiencies, we propose PointLAM, a highly efficient and powerful point-based architecture driven by two synergistic innovations. First, to resolve the downsampling bottleneck, we develop the Laplacian Point Sampler (LPS). LPS employs an implicit discrete Laplacian high-pass filter and Doubly Sorted Sampling to achieve fast, structure-aware foreground preservation. Second, to overcome local modeling latency, we design the Local Hadamard Aggregator (LHA). LHA decouples spatial indexing from feature representation using transient grids, and replaces complex continuous interactions with a Hadamard Gating mechanism for topology-aware, attentive modulation. By coupling this local gating with Bi-Directional Mamba (BDM) layers for global sequence modeling, we formulate the Local Attentive Mamba (LAM) block. Powered by this architecture, PointLAM achieves competitive performance on nuScenes and Waymo for point-based detectors. It rivals highly optimized voxel competitors while requiring a fraction of the computational footprint, demonstrating marked superiority in detecting small instances and handling extreme sparsity. Project page: https://pointlam.github.io/.

Figures & tables

Figure 1: Comparison of 3D point cloud processing paradigms. Left: Point-based methods [ 27 , 28 , 33 , 45 ] rely on expensive neighborhood queries, while voxel-based methods [ 21 , 47 , 48 , 39 ] sacrifice fine geometric details for efficiency. Right: PointLAM offers a favorable performance-efficiency trade-off on nuScenes. It is faster and lighter (marker size indicates parameter count) than several strong voxel-based baselines (circles), positioning it as a competitive point-based model (red star).

Figure 2: The overall architecture of PointLAM. Top: The pipeline starts with the Laplacian Point Sampler (LPS) for fast, structure-aware downsampling, followed by a 3D backbone of $N$ blocks. Each block cascades X-axis or Y-axis Bi-Directional Mamba (BDM) layers for global context, paired with a Local Hadamard Aggregator (LHA) module for efficient local topology modeling, followed by BEV projection and detection heads. Bottom-Left: LPS utilizes a Deviation Network (DevNet) to effectively implement an implicit discrete Laplacian high-pass filter, combined with Doubly Sorted Sampling (DSS) to strictly preserve informative foreground structures. Bottom-Center: LHA captures local geometry by spatially indexing via a transient voxel grid, and employs a Hadamard Gating mechanism for topology-aware feature modulation. Bottom-Right: BDM establishes global context by sequentially processing points sorted along the X and Y axes via Mamba blocks.

Method	Representation	NDS	mAP	Car	Truck	Bus	Trailer	C.V.	Ped.	M.C.	Bike	T.C.	Barrier
Results on the validation set
PillarNeXt [ 17 ]	Pillar	68.4	62.2	85.0	57.4	67.6	35.6	20.6	86.8	68.6	53.1	77.3	69.7
DSVT [ 39 ]	Pillar	71.1	66.4	87.4	62.6	75.9	42.1	25.3	88.2	74.8	58.7	77.8	70.9
CenterPoint [ 46 ]	Voxel	66.5	59.2	84.9	57.4	70.7	38.1	16.9	85.1	59.0	42.0	69.8	68.3
TransFusion-L [ 1 ]		70.1	65.5	86.9	60.8	73.1	43.4	25.2	87.5	72.9	57.3	77.2	70.3
VoxelNeXt [ 4 ]		66.7	60.5	83.9	55.5	70.5	38.1	21.1	84.6	62.8	50.0	69.4	69.4

Table 1: Performance comparison on the nuScenes dataset without using any test-time augmentation and model ensemble strategies. ‘C.V.’, ‘Ped.’, ‘M.C.’, and ‘T.C.’ denote construction vehicle, pedestrian, motorcycle, and traffic cone, respectively.

		ALL (3D mAPH)		Vehicle (AP/APH)		Pedestrian (AP/APH)		Cyclist (AP/APH)
Method	Representation	L1	L2	L1	L2	L1	L2	L1	L2
Results on the validation set
PV-RCNN [ 31 ]	Point-Voxel	69.6	63.3	77.5 / 76.9	69.0 / 68.4	75.0 / 65.7	66.0 / 57.6	67.8 / 66.4	65.4 / 64.0
PV-RCNN++ [ 32 ]	Point-Voxel	75.2	68.6	79.1 / 78.6	70.3 / 69.9	80.6 / 74.6	71.9 / 66.3	73.5 / 72.4	70.7 / 69.6
PointPillar [ 15 ]	Pillar	63.3	57.5	71.6 / 71.0	63.1 / 62.5	70.6 / 56.7	62.9 / 50.2	64.4 / 62.3	61.9 / 59.9
PillarNet [ 30 ]		74.6	68.4	79.1 / 78.6	70.9 / 70.5	80.6 / 74.0	72.3 / 66.2	72.3 / 71.2	69.7 / 68.7

Table 2: Performance comparison on the Waymo Open Dataset (single-frame setting) without using any test-time augmentation and model ensemble strategies. Symbol ‘-’ denotes that the result is not available.

Method	Venue	Backbone	Representation	#Parameters (M)	FLOPs (G)	Latency (ms)	NDS	L2 mAPH
PVRCNN [ 31 ]	CVPR’20	spCNN	Point-Voxel	8.5	47.4	281.2	-	63.3
CenterPoint [ 46 ]	CVPR’21		Voxel	2.7	41.8	60.1	66.5	67.6
VoxelRCNN [ 5 ]	AAAI’21		Voxel	11.7	23.3	94.5	-	66.2
VoxelNeXt [ 4 ]	CVPR’23		Voxel	15.5	97.5	192.9	66.7	70.1
PVRCNN++ [ 32 ]	IJCV’23		Point-Voxel	9.5	55.4	97.3	-	68.6
HEDNet [ 47 ]	NeurIPS’23		Voxel	4.6	106.2	96.7	71.4	73.4

Table 3: Comparison of parameter count, computation cost, and inference latency of different 3D detection backbones.

Table 6

Size ( $m^{3}$ )	Ours		LION [ 21 ]		Ours-KNN
Size ( $m^{3}$ )	L1 mAPH	L2 mAPH	L1 mAPH	L2 mAPH	L1 mAPH	L2 mAPH
1.0	3.75	3.28	3.38	2.93	2.51	2.17
1.5	20.70	18.82	20.26	18.37	19.09	17.27
2.0	29.75	27.27	29.36	26.85	27.77	25.34
2.5	35.40	32.69	34.95	32.23	33.76	31.08
3.0	47.12	43.97	46.12	42.97	43.72	40.68

Table 6: Robustness comparison across different object sizes.

Figure 3: Visualization of LPS sampling and LHA feature modulation. Raw input points are in light gray. Points retained by LPS are colored based on their LHA activation magnitude, highlighting high-frequency structural boundaries.

Table 9

Table 10

Table 11

3D Object Detection

3D object detection identifies objects and estimates their locations, sizes, and orientations from images, point clouds, radar, or fused sensors.

See all 180 papers

Aug 4, 2026

Lightweight 3D Object Detection via Mamba-Based Knowledge Distillation

3D object detection using light detection and ranging (LiDAR) sensors requires a balance between accuracy and computational efficiency for onboard perception in autonomous driving and robotic navigation. Many existing LiDAR-based detection methods employ complex architectures to extract features, integrating large amounts of contextual information to enhance accuracy. This often results in significant computational costs, leading to suboptimal performance on resource-constrained embedded devices. In this study, we propose a knowledge distillation framework that transfers object-level voxel representations from a strong teacher model to lightweight student models through selective voxel-space feature alignment. Taking advantage of the linear-time sequence model with selective state spaces (Mamba), we design a multi-branch Mamba teacher backbone and a box-aware feature transfer mechanism that aligns spatially corresponding voxel features between teacher and student networks through a Mamba-based projection module. Experimental results on both a public dataset and real-world data show that our approach significantly reduces computational load while maintaining competitive accuracy compared with state-of-the-art methods.

Quoc Cuong Ninh, Huy Xuan Pham, Anh Tung Nguyen +1

Viettel AI, Viettel Group, Vietnam · Artificial Intelligence in Robotics Laboratory (AiR Lab), Department of Electrical and Computer Engineering, Aarhus University, 8000 Aarhus C, Denmark · VIST, Viettel Group, Vietnam

Lidar-Based 3D Object Detection 3D Object Detection Knowledge Distillation

Ask PDF

Apr 3, 2026

RayMamba: Ray-Aligned Serialization for Long-Range 3D Object Detection

Long-range 3D object detection remains challenging because LiDAR observations become highly sparse and fragmented in the far field, making reliable context modeling difficult for existing detectors. To address this issue, recent state space model (SSM)-based methods have improved long-range modeling efficiency. However, their effectiveness is still limited by generic serialization strategies that fail to preserve meaningful contextual neighborhoods in sparse scenes. To address this issue, we propose RayMamba, a geometry-aware plug-and-play enhancement for voxel-based 3D detectors. RayMamba organizes sparse voxels into sector-wise ordered sequences through a ray-aligned serialization strategy, which preserves directional continuity and occlusion-related context for subsequent Mamba-based modeling. It is compatible with both LiDAR-only and multimodal detectors, while introducing only modest overhead. Extensive experiments on nuScenes and Argoverse 2 demonstrate consistent improvements across strong baselines. In particular, RayMamba achieves up to 2.49 mAP and 1.59 NDS gain in the challenging 40--50 m range on nuScenes, and further improves VoxelNeXt on Argoverse 2 from 30.3 to 31.2 mAP.

Cheng Lu, Mingqian Ji, Shanshan Zhang +2

School of Computer Science and Technology, Nanjing University of Science and Technology, Nanjing 210094, China.

3D Object Detection Lidar-Based 3D Object Detection

Ask PDF

May 15, 2026

3DTMDet: A Dual-Path Synergy Network of Transformer and SSM for 3D Object Detection in Point Clouds

A fundamental challenge in point cloud object detection lies in the conflict between the extreme sparsity of distant points and the need for remote context understanding. The existing methods typically use 1D serialization to expand the receptive field, which inevitably discards already scarce local geometric details and reduces detection of distant and small objects. To address this issue, we propose 3DTMDet, a novel detection network that synergistically combines state space models (Mamba) with Transformers. The core idea is to utilize SSM's linear complexity and advantages in long sequence modeling to effectively capture global interactions between sparse and distant points, while using Transformer modules with local attention to encode fine-grained geometric structures in local point sets, preserving accurate shape information. We propose the 3D Hybrid Mamba Transformer (3DHMT) block, which uses an SSM-Attention-SSM pipeline to balance global context understanding and local detail preservation, effectively alleviating the tension between receptive field enlargement and geometric preservation in remote detection. In addition, we introduced a voxel generation block inspired by LiDAR physics, which diffuses features along the sensor observation direction to reconstruct the complete object structure of occlusion and distant areas. Extensive experiments conducted on the KITTI and ONCE datasets have shown that 3DTMDet outperforms state-of-the-art detectors. The code is available at https://github.com/QiuBingwen/3DTMDet.

Bingwen Qiu, Yuan Liu, Junqi Bai +5

aSchool of Electronic and Optical Engineering, Nanjing University of Science and Technology, Nanjing, 210094, Jiangsu, China · bThe 28th Research Institute of China Electronics Technology Group Corporation, Nanjing, 210007, Jiangsu, China · cCollege of Astronautics, Nanjing University of Aeronautics and Astronautics, Nanjing, 210016, Jiangsu, China +2

3D ViTs 3D Object Detection

Ask PDF

Abstract

3D Object Detection

Explore similar work

Lightweight 3D Object Detection via Mamba-Based Knowledge Distillation

RayMamba: Ray-Aligned Serialization for Long-Range 3D Object Detection

3DTMDet: A Dual-Path Synergy Network of Transformer and SSM for 3D Object Detection in Point Clouds