发表机构
Iowa State University; University of Arkansas; Parsons Corporation; Jacobs; University of South Florida; Neel-Schaffer, Inc.（爱荷华州立大学; 阿肯色大学; 帕森斯公司; 雅各布斯公司; 南佛罗里达大学; 尼尔-沙弗公司）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对跨城市细粒度交通目标检测中细粒度识别与域泛化难兼顾的问题，提出 DRAFE 模型，通过两阶段训练及多模块融合策略，在 AI City Challenge 2026 Track 6 任务中取得 0.4022 mAP 排名第六的成绩，性能优于初步集成模型。
AI 中文摘要
基于深度学习的目标检测器是智能交通系统的基础，可实现交通监控、车辆分析和基础设施管理。然而，同时实现细粒度车辆识别和鲁棒的跨城市域泛化仍具挑战性。本文提出 Domain-Robust Asymmetric Fusion Ensemble（DRAFE），将独立训练的 LW-DETR 和 RF-DETR 检测器结合用于跨城市细粒度交通目标检测。DRAFE 采用两阶段训练策略：首先利用伪标签扩展和人在回路的标注优化，在多样化公共交通数据集上预训练互补检测器，生成包含 6049 张图像和 203619 个标注的精选语料库；随后在 Project Hafnia Track 6 数据集上进行符合挑战赛要求的微调。推理阶段，DRAFE 采用锚条件类一致匹配、可靠性加权坐标融合、感知一致性的置信度重新校准及互补假设恢复。在 AI City Challenge 2026 Track 6 任务中，DRAFE 达到 0.4022 的 mAP，在 25 支参赛队伍中排名第六，且在相同基准条件下，比初步集成模型的 mAP 提升了 0.0553。
英文摘要
Deep learning-based object detectors are fundamental to intelligent transportation systems, enabling traffic monitoring, vehicle analytics, and infrastructure management. However, achieving both fine-grained vehicle recognition and robust cross-city domain generalization remains challenging. We present the Domain-Robust Asymmetric Fusion Ensemble (DRAFE), which combines independently trained LW-DETR and RF-DETR detectors for cross-city fine-grained traffic object detection. DRAFE employs a two-stage training strategy that first pretrains complementary detectors on diverse public traffic datasets using pseudo-label expansion and human-in-the-loop annotation refinement, producing a curated corpus of 6,049 images and 203,619 annotations, before challenge-compliant fine-tuning on the Project Hafnia Track 6 dataset. At inference, DRAFE applies anchor-conditioned class-consistent matching, reliability-weighted coordinate fusion, agreement-aware confidence recalibration, and complementary hypothesis recovery. On AI City Challenge 2026 Track 6, DRAFE achieves 0.4022 mAP, ranks sixth among 25 participating teams, and improves by 0.0553 mAP over a preliminary ensemble evaluated under identical benchmark conditions.
Comments17 pages, 2 figures, 6 tables. Code available at: https://github.com/dyagbobli/VisionOps-Trainer
