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 3 高效的上下文信息增强模块中不同卷积核大小的单目3D目标检测精度

Tab.3 Monocular 3D object detection accuracy of different kernel sizes in efficient contextual information enhancement module %

7 11 21 ${{\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 √ — — 24.54 17.12 14.14 32.65 22.95 20.04 — √ — 26.02 17.35 14.38 32.63 22.69 19.99 — — √ 25.18 17.36 14.35 33.29 23.20 20.39 √ √ — 25.88 17.64 14.52 33.58 23.26 20.42 √ — √ 26.60 17.73 14.68 32.67 22.80 19.37 — √ √ 26.62 17.88 14.72 34.38 23.58 20.67 √ √ √ 26.73 18.25 15.19 34.20 23.72 20.90