基于上下文信息增强和深度引导的单目3D目标检测

于家艺,吴秦

Monocular 3D object detection based on context information enhancement and depth guidance

Jiayi YU,Qin WU

表 2 不同方法组合下的单目3D目标检测精度

Tab.2 Monocular 3D object detection accuracy of different method combinations %

实验 ECIE $ {L}_{r}^{d} $ $ {L}_{r}^{l} $ ${{\mathrm{AP}}_{3{\mathrm{D}}}}|{R_{40}} $ ${{\mathrm{AP}}_{{\mathrm{BEV}}}}|{R_{40}} $ 简单 中等 困难 简单 中等 困难 1 — — — 25.42 17.09 14.08 33.90 23.30 19.51 2 √ — — 26.73 18.25 15.19 34.20 23.72 20.90 3 — √ — 26.03 17.58 14.56 34.61 24.59 21.05 4 — — √ 26.06 17.84 14.72 33.02 24.17 20.63 5 — √ √ 27.28 18.11 14.93 35.35 24.56 20.97 6 √ √ — 27.11 18.23 15.04 35.43 24.69 21.01 7 √ — √ 27.03 18.25 15.00 35.13 23.96 20.97 8 √ √ √ 27.56 18.32 15.13 35.85 24.82 21.19