DRAFE: Domain-Robust Asymmetric Fusion of Heterogeneous Detection Transformers for Cross-City Fine-Grained Traffic Object Detection

2026-08-17

AI Summary

This study addresses the challenge of balancing domain generalization and accuracy in fine-grained traffic object detection across cross-city scenarios by proposing the DRAFE framework. The method employs a heterogeneous asymmetric fusion mechanism integrating LW-DETR and RF-DETR, combined with two-stage training to construct high-quality corpora. During inference, complementary hypotheses are effectively recovered and transfer robustness is enhanced through anchor-conditioned matching, reliability-weighted fusion, and protocol-aware confidence recalibration. Validated on Track 6 of the AI City Challenge 2026, DRAFE achieved a mAP of 0.4022, ranking sixth and outperforming the baseline ensemble by 0.0553 mAP, thereby demonstrating its effectiveness in real-world cross-domain traffic detection tasks.

Abstract

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.

Problem

• Fine-Grained Traffic Object Detection
• Cross-City Domain Generalization
• Intelligent Transportation Systems

Innovation

• Asymmetric Fusion
• Domain-Robust
• Heterogeneous DETR
• Cross-City Generalization
• Fine-Grained Detection