CoLC: Communication-Efficient Collaborative Perception with LiDAR Completion

Yushan Han; Hui Zhang; Qiming Xia; Yi Jin; Yidong Li

cs.CVFeb 28, 2026

CoLC: Communication-Efficient Collaborative Perception with LiDAR Completion

Authors: Yushan Han, Hui Zhang, Qiming Xia, Yi Jin, Yidong Li

Organizations: Key Laboratory of Big Data & Artificial Intelligence in Transportation, Ministry of Education · School of Computer Science and Technology, Beijing Jiaotong University · Xiamen University

Abstract

Collaborative perception empowers autonomous agents to share complementary information and overcome perception limitations. While early fusion offers more perceptual complementarity and is inherently robust to model heterogeneity, its high communication cost has limited its practical deployment, prompting most existing works to favor intermediate or late fusion. To address this, we propose a communication-efficient early Collaborative perception framework that incorporates LiDAR Completion to restore scene completeness under sparse transmission, dubbed as CoLC. Specifically, the CoLC integrates three complementary designs. First, each neighbor agent applies Foreground-Aware Point Sampling (FAPS) to selectively transmit informative points that retain essential structural and contextual cues under bandwidth constraints. The ego agent then employs Completion-Enhanced Early Fusion (CEEF) to reconstruct dense pillars from the received sparse inputs and adaptively fuse them with its own observations, thereby restoring spatial completeness. Finally, the Dense-Guided Dual Alignment (DGDA) strategy enforces semantic and geometric consistency between the enhanced and dense pillars during training, ensuring consistent and robust feature learning. Experiments on both simulated and real-world datasets demonstrate that CoLC achieves superior perception-communication trade-offs and remains robust under heterogeneous model settings. The code is available at https://github.com/CatOneTwo/CoLC.

Figures & tables

Figure 1: Comparative performance of early fusion when transmitting foreground (FG) points, foreground with surrounding (FG+SUR) points, background (BG) points, or all points. Transmitting FG or FG+SUR points leads to significant degradation due to missing contextual cues, demonstrating the necessity of incorporating both FG and BG points in early fusion.

Figure 2: The overall architecture of CoLC. (a) Each neighbor agent applies Foreground-Aware Point Sampling (FAPS) to select and transmit informative points to the ego agent for efficient communication. (b) The ego agent performs Completion-Enhanced Early Fusion (CEEF), where the received sparse points are pillarized $\Phi_{\text{p}}(\cdot)$ , completed into dense pillars, and adaptively fused with the initial sparse fusion to enhance structural detail. (c) During training, Dense-Guided Dual Alignment (DGDA) enforces semantic distribution and geometric direction consistency, further improving the alignment of the fused representation.

Figure 4: Adaptive complementary fusion.

Methods	V2XSim	OPV2V	V2XSet	DAIR-V2X
No Fusion	74.73 $/\penalty$ 73.72 $/\penalty$ 61.65	77.65 $/\penalty$ 74.42 $/\penalty$ 54.52	78.41 $/\penalty$ 74.18 $/\penalty$ 57.43	67.15 $/\penalty$ 64.32 $/\penalty$ 53.27
Late Fusion	89.23 $/\penalty$ 86.87 $/\penalty$ 78.78	94.22 $/\penalty$ 92.51 $/\penalty$ 77.82	91.81 $/\penalty$ 89.05 $/\penalty$ 75.72	74.12 $/\penalty$ 62.05 $/\penalty$ 42.82
V2VNet [ 29 ]	88.16 $/\penalty$ 87.11 $/\penalty$ 69.73	97.09 $/\penalty$ 95.89 $/\penalty$ 83.56	92.70 $/\penalty$ 90.59 $/\penalty$ 74.52	78.15 $/\penalty$ 70.24 $/\penalty$ 43.71
DiscoNet [ 14 ]	80.73 $/\penalty$ 77.46 $/\penalty$ 64.03	95.45 $/\penalty$ 93.45 $/\penalty$ 82.95	89.63 $/\penalty$ 86.28 $/\penalty$ 71.03	80.32 $/\penalty$ 73.21 $/\penalty$ 57.38
AttFusion [ 38 ]	77.08 $/\penalty$ 74.51 $/\penalty$ 64.86	94.93 $/\penalty$ 93.13 $/\penalty$ 80.66	90.07 $/\penalty$ 87.32 $/\penalty$ 72.26	79.69 $/\penalty$ 73.46 $/\penalty$ 58.82
V2X-ViT [ 37 ]	90.42 $/\penalty$ 87.77 $/\penalty$ 70.86	96.35 $/\penalty$ 95.01 $/\penalty$ 85.84	93.93 $/\penalty$ 90.28 $/\penalty$ 74.75	78.34 $/\penalty$ 70.62 $/\penalty$ 54.02

Table 1: Performance comparison ([email protected]

/\penalty

0.5

/\penalty

0.7) on collaborative 3D object detection methods.

Panel of Figure 5 — Figure 5: Evaluation on V2XSim. In (b)-(d), CoLC transmits $100\%$ of the LiDAR data, while CoLC* transmits only $20\%$ of foreground points and $50\%$ of background points. In (e)-(h), when perform FAPS, $R^{fg}$ is fixed at 0.2, and $R^{bg}$ is varied from 0.5 to 0.01.

Panel of Figure 6 — Figure 6: Detection results of CoLC under different point sampling ratios. In (b)-(f), the foreground sampling ratio $R^{\text{fg}}$ is fixed at 0.2. Black and red boxes represent ground truth and model predictions, respectively.

$K$	$D_{c}$	$D$	$L$	IoU $\uparrow$	MSE $\downarrow$	MB $\downarrow$	[email protected]/0.7 $\uparrow$
128	128	128	3	0.473	0.060	39.08	88.14 / 75.93
128	128	128	6	0.585	0.052	44.07	89.02 / 78.26
256	256	128	6	0.590	0.052	46.53	88.93 / 78.11
512	256	128	6	0.592	0.046	49.28	88.91 / 78.37
512	256	256	6	0.617	0.047	79.17	89.01 / 78.59
512	512	256	6	0.626	0.043	85.93	88.89 / 79.28

Table 2: Ablation study on the codebook size (

K

), codebook dimension (

D_{c}

), sparse encoder dimension (

D

) and depth (

L

Method	IoU $\uparrow$	MSE $\downarrow$	[email protected]/0.7 $\uparrow$
MAE-based	0.633	0.057	88.17 / 77.55
VQ-based	0.626	0.043	88.89 / 79.28

Table 3: Comparison of VQ-based and MAE-based completion.

$\mathcal{L}_{\text{sda}}$	$\mathcal{L}_{\text{gda}}$	IoU $\uparrow$	MSE $\downarrow$	[email protected]/0.7 $\uparrow$
		0.765	0.415	88.48 / 77.03
✓		0.831	0.140	88.75 / 77.88
	✓	0.836	0.129	88.16 / 76.82
✓	✓	0.834	0.118	88.89 / 79.28

Table 4: Ablation on

\mathcal{L}_{\text{sda}}

and

\mathcal{L}_{\text{gda}}

, where IoU and MSE quantify the similarity between

\hat{\mathcal{P}}_{i}^{\text{de}}

and

\mathcal{P}_{i}^{\text{de}}

Explore similar work

May 26, 2026

Adaptation-Free Heterogeneous Collaborative Perception with Unseen Agent Configurations

Collaborative perception improves 3D object detection by enabling agents to share complementary observations, but most existing methods assume fixed or known collaborator encoder configurations, limiting deployment in practice. In this work, we consider an open-world setting in which auxiliary agents with unseen configurations may appear after deployment, such as different LiDAR beam counts or encoder architectures. To address this challenge, we propose ALF, a collaborative perception framework that enables zero-adaptation collaboration with unseen agent configurations by lifting lightweight box-level messages into ego-compatible auxiliary features. ALF converts auxiliary box-level messages into pseudo-BEV maps and synthesizes ego-compatible latent features by combining object-centric cues with scene context from the ego feature. On V2X-Real, under a zero-shot evaluation across 64 case studies, ALF outperforms the strongest prior baseline by 35.91% in relative [email protected] while requiring only 120 bytes per agent per frame (approximately 9.6 Kbps bandwidth at 10 Hz).

Hyunchul Bae, Heejin Ahn

School of Electrical Engineering Korea Advanced Institute of Science and Technology (KAIST) Daejeon, Korea 34141

Zero-Shot Learning Communication-Efficient Collaborative Perception 3D Object Detection V2X Communication Cooperative Perception Heterogeneous Collaborative Perception

Ask PDF

Sep 1, 2026

CERF: Communication-Efficient and Retraining-Free Collaborative Perception

Collaborative perception shares information among multiple agents to obtain a comprehensive scene representation, enhancing the perceptual capability of individual agents. However, most existing methods rely on transmitting and fusing dense feature maps for collaboration, which incurs inevitable communication overhead and heterogeneity challenges, limiting their practicality for real-world deployment. To address these challenges, we propose CERF, a novel Communication-Efficient and Retraining-Free framework for open heterogeneous collaborative perception. In CERF, we introduce a new virtual modality (termed Poture), which is generated from the perception outputs of other agents, to augment the extracted Bird's Eye View (BEV) features of the ego agent. To mitigate transmission delays, we employ a Kalman-filter based tracker and a motion forecasting model to derive the current predictions from historical perception results. Extensive experiments demonstrate that CERF achieves performance comparable to mainstream intermediate-collaboration methods while reducing communication overhead by 95% across various downstream tasks. Furthermore, CERF enables seamless integration of unknown heterogeneous agents into the existing collaborative framework without additional retraining costs. Code is available at https://github.com/uestchjw/CERF.

Jiuwu Hao, Ziyi Ni, Liguo Sun +5

School of Artificial Intelligence, University of Chinese Academy of Sciences · Institute of Automation, Chinese Academy of Sciences

Communication-Efficient Collaborative Perception Autonomous Driving Perception Cooperative Perception Heterogeneous Collaborative Perception

Ask PDF

Jun 22, 2026

UECP: Uncertainty-Enhanced Collaborative Perception

Collaborative perception serves as a pivotal solution to enhance the perception capability of individual agents in autonomous driving, where a core challenge lies in seeking reliable evidence to quantify and weight the contribution of each participating agent. Existing methods typically rely on a confidence map, which is co-trained with the detection head, but it is inherently correlated with the detection results and thus fails to provide unbiased physical evidence. Furthermore, how to deeply integrate evidence into the cooperative fusion process remains an open question. To address these issues, this paper first proposes an uncertainty map, a physically grounded and unambiguous metric for evaluating perception quality. This map is directly supervised by real-time sensor signals, i.e., LiDAR point density, ensuring decoupling from detection noise and thereby providing physical scenario-aware evidence for weighting agent contribution. Based on this map, we develop the Uncertainty-Enhanced Collaborative Perception (UECP) framework, centered on the Uncertainty-Aware Pyramid Fusion (UAPF) module. UAPF uses a coarse-to-fine strategy, with two key components: Uncertainty-Weighted Downsampling (UWD) for high-fidelity feature preservation, and Uncertainty-Guided Residual Fusion (UGRF) to reinforce ego features, suppressing noise and ensuring robust fusion. Extensive experiments on real-world datasets show UECP outperforms state-of-the-art methods in effectiveness and robustness by embedding the uncertainty map into fusion. Code will be publicly available.

Kang Yang, Tianci Bu, Peng Wang +3

Renmin University of China, Beijing, China · College of Systems Engineering, National University of Defense Technology, Changsha, China · Intelligent Shipping Center, China Waterborne Transport Research Institute, Beijing, China

Uncertainty Quantification Autonomous Driving Perception Multi-Scale Feature Fusion Cooperative Perception

Ask PDF