Refine-and-Contrast: Adaptive Instance-Aware BEV Representations for Multi-UAV Collaborative Object Detection

2025-08-18

AI Summary

To address resource constraints, severe occlusion, and limited wide-area coverage in multi-UAV cooperative 3D detection, this paper proposes AdaBEV—a novel framework for efficient and robust bird’s-eye-view (BEV) representation learning. First, a box-guided refinement module adaptively focuses on foreground instance regions. Second, an instance-background contrastive learning mechanism is introduced to enforce discriminative feature separation directly in BEV space. Third, lightweight BEV optimization—integrated with 2D supervision and spatial subdivision—is employed to generate instance-aware BEV representations from low-resolution inputs. By departing from conventional uniform-grid BEV modeling, AdaBEV significantly enhances occlusion robustness and large-scale scene perception. On the Air-Co-Pred benchmark, AdaBEV achieves state-of-the-art accuracy with substantially lower computational overhead, approaching the performance upper bound of high-resolution methods.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Machine Learning: Multi-instance/Multi-view Learning
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Multi-UAV collaborative 3D detection enables accurate and robust perception by fusing multi-view observations from aerial platforms, offering significant advantages in coverage and occlusion handling, while posing new challenges for computation on resource-constrained UAV platforms. In this paper, we present AdaBEV, a novel framework that learns adaptive instance-aware BEV representations through a refine-and-contrast paradigm. Unlike existing methods that treat all BEV grids equally, AdaBEV introduces a Box-Guided Refinement Module (BG-RM) and an Instance-Background Contrastive Learning (IBCL) to enhance semantic awareness and feature discriminability. BG-RM refines only BEV grids associated with foreground instances using 2D supervision and spatial subdivision, while IBCL promotes stronger separation between foreground and background features via contrastive learning in BEV space. Extensive experiments on the Air-Co-Pred dataset demonstrate that AdaBEV achieves superior accuracy-computation trade-offs across model scales, outperforming other state-of-the-art methods at low resolutions and approaching upper bound performance while maintaining low-resolution BEV inputs and negligible overhead.

Problem

• Enhances multi-UAV collaborative 3D detection accuracy
• Reduces computation for resource-constrained UAV platforms
• Improves foreground-background feature separation in BEV space

Innovation

• Adaptive instance-aware BEV representations
• Box-Guided Refinement Module (BG-RM)
• Instance-Background Contrastive Learning (IBCL)