The KITTI Vision Benchmark Suite

Method

Monocular 3D Object Detection by Reconstruction and Uncertainty Propagation [MonoRUn]
https://github.com/tjiiv-cprg/MonoRUn

Submitted on 15 Nov. 2020 15:42 by
Hansheng Chen (Tongji University)

Running time:		0.07 s
Environment:		GPU @ 2.5 Ghz (Python + C/C++)

Method Description:

We propose a novel monocular 3D detection framework based on dense 2D-3D correspondences，which learns the object 3D geometry in a self-supervised manner and locates the object using PnP algorithm. The framework is fully uncertainty-aware, i.e., it outputs the aleatoric uncertainty of the reconstructed geometry, which is subsequently propagated through the uncertainty-aware PnP algorithm, yielding object pose with covariance.

Parameters:

TBD

Latex Bibtex:

@inproceedings{monorun,
author = {Hansheng Chen and Yuyao Huang and Wei Tian and Zhong Gao and Lu Xiong},
title = {MonoRUn: Monocular 3D Object Detection by Reconstruction and Uncertainty Propagation},
booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year = 2021
}

Detailed Results

Object detection and orientation estimation results. Results for object detection are given in terms of average precision (AP) and results for joint object detection and orientation estimation are provided in terms of average orientation similarity (AOS).

Benchmark	Easy	Moderate	Hard
Car (Detection)	95.48 %	87.91 %	78.10 %
Car (Orientation)	95.44 %	87.64 %	77.75 %
Car (3D Detection)	19.65 %	12.30 %	10.58 %
Car (Bird's Eye View)	27.94 %	17.34 %	15.24 %
Pedestrian (Detection)	73.05 %	56.40 %	51.40 %
Pedestrian (Orientation)	63.28 %	47.82 %	43.23 %
Pedestrian (3D Detection)	10.88 %	6.78 %	5.83 %
Pedestrian (Bird's Eye View)	11.70 %	7.59 %	6.34 %
Cyclist (Detection)	67.47 %	49.13 %	43.41 %
Cyclist (Orientation)	49.04 %	34.36 %	30.22 %
Cyclist (3D Detection)	1.01 %	0.61 %	0.48 %
Cyclist (Bird's Eye View)	1.14 %	0.73 %	0.66 %