Sim2Real路侧LiDAR检测的几何精修与分布对齐
原标题：Geometric Refinement and Distributional Alignment for Sim2Real Roadside LiDAR Detection
AI 摘要
哥伦比亚大学Aman Upganlawar参加ECCV 2026 UrbanTwin Sim2Real LiDAR挑战赛LUMPI赛道，仅用合成数据训练PointPillars检测器并在真实路侧LiDAR上评测，最终综合得分0.4793（检测mAP 0.1333、真实感0.9126）。其贡献不在新架构，而在于诊断出小目标失败本质是定位问题，并据此做无参数几何精修（行人AP相对提升86%），同时修复后处理中破坏有效预测的规则（Van类相对提升335%）。作者还提出将平均AP按类分解、使单次提交可承载多个受控实验的方法论，并在提交前将最终分数预测到小数点后四位。
正文节选
Geometric Refinement and Distributional Alignment for Sim2Real Roadside LiDAR Detection UCF UrbanTwin Sim2Real LiDAR Challenge, LUMPI Track, DriveX Workshop, ECCV 2026 Author: Aman Upganlawar, Columbia University Team: AMAN_UPG (single participant) Contact: aman.upg27024@gmail.com Affiliation: Columbia University Track election: Publication track (we do not opt into the non-publication track) Final combined score: 0.4793 Abstract We describe our approach to the LUMPI track of the UrbanTwin Sim2R