论文阅读笔记：(2021.06 iccv) Geometry-based Distance Decomposition for Monocular 3D Object Detection

好思路：不直接regression深度，而是regression 3D框高度在图片上的投影，再用相似三角形算出深度；

paper:http://openaccess.thecvf.com//content/ICCV2021/papers/Shi_Geometry-Based_Distance_Decomposition_for_Monocular_3D_Object_Detection_ICCV_2021_paper.pdfhttp://openaccess.thecvf.com//content/ICCV2021/papers/Shi_Geometry-Based_Distance_Decomposition_for_Monocular_3D_Object_Detection_ICCV_2021_paper.pdf

code:GitHub - Rock-100/MonoDet: MonoRCNN based Monocular 3D Object Detection for Automonous Driving, ICCV 2021

精度对比：KITTI Cars Moderate Benchmark (Monocular 3D Object Detection) | Papers With CodeThe current state-of-the-art on KITTI Cars Moderate is DD3D. See a full comparison of 20 papers with code.https://paperswithcode.com/sota/monocular-3d-object-detection-on-kitti-cars

实现：