hustvl / PolarDETR

PolarDETR

Polar Parametrization for Vision-based Surround-View 3D Detection

by
Shaoyu Chen, Xinggang Wang^†, Tianheng Cheng, Qian Zhang, Chang Huang, Wenyu Liu.
(†: corresponding author)

[Paper]

Overview

We reformulate position parametrization, velocity decomposition, perception range, label assignment and loss function in polar coordinate system. Polar parametrization establishes explicit associations between image patterns and prediction targets, exploiting the view symmetry of surround-view cameras.

Framework

PolarDETR ranks 1st on the nuScenes benchmark at the submission time. Following are snapshots of nuScenes benchmark on Mar. 6th, 2022.

Citation

@article{PolarDETR,
  author    = {Chen, Shaoyu and and Wang, Xinggang and Cheng, Tianheng and Zhang, Qian and Huang, Chang and Liu, Wenyu},
  title     = {Polar Parametrization for Vision-based Surround-View 3D Detection},
  journal   = {arXiv:2206.10965},
  year      = {2022},
}