We propose a novel method, CAPE, for 3D object detection by using local camera-view coordinates, not global. This extends to temporal modeling, boosting detection, and leads to SOTA LiDAR-free performance. #CVPR2023 Paper: https://
arxiv.org/abs/2303.10209 Code: https://
github.com/PaddlePaddle/P
addle3D
…

→ View original post on X — @baiduresearch