Dense Coverage, Sparse Refinement: Byte-Constrained Cooperative Perception

2026-09-24

AI Summary

This study addresses the challenge that dense feature exchange in collaborative perception struggles to satisfy stringent V2X bandwidth constraints. To this end, we propose a "cover-refine" transmission framework. The method first transmits a full bird's-eye-view (BEV) coarse-grained feature layer and subsequently introduces a task-aware utility selector that ranks high-resolution feature units according to their downstream perceptual utility. The remaining bandwidth budget is then precisely allocated to critical patches, enabling deterministic sparse refinement and zero-retraining bandwidth adaptation. Evaluated on the DAIR-V2X dataset, our approach achieves 0.60 AP@0.7 with merely 1.87 KB of communication overhead, significantly outperforming uniform compression baselines. These results demonstrate that the proposed framework effectively optimizes the trade-off between perception accuracy and communication load under strict bandwidth limitations.

Abstract

Collaborative perception improves autonomous perception by sharing intermediate Bird's-Eye-View (BEV) features across connected agents, but dense feature exchange is difficult to deploy under strict Vehicle-to-Everything (V2X) bandwidth limits. Existing efficient methods typically either compress the full feature map uniformly, spending bits on low-value background, or sparsify communication, risking the loss of useful context. We propose a coverage-refinement design for byte-constrained cooperative perception: each agent transmits a highly compressed coarse layer over the full BEV map and allocates the remaining budget to selected high-resolution patches. A Task-Aware Benefit Selector ranks cells by estimated downstream utility, enabling deterministic budgeted refinement and zero-retraining adaptation to changing bandwidth. The receiver reconstructs a dense BEV tensor compatible with standard fusion modules. Experiments on DAIR-V2X and OPV2V show strong accuracy-payload trade-offs at kilobyte-scale budgets. On DAIR-V2X, our method reaches 0.60 AP@0.7 at only 1.87 KB per non-ego agent, compared with 0.52 at 4.61 KB for uniform SimVQ compression. Controlled diagnostics further show that the gain arises from coverage-refinement allocation rather than quantization alone. Code will be published.

Problem

• Cooperative Perception
• Bandwidth Constraint
• Bird's-Eye-View Features
• V2X Communication

Innovation

• Cooperative Perception
• Byte-Constrained Communication
• Bird's-Eye-View (BEV)
• Coverage-Refinement
• Task-Aware Benefit Selector