The KITTI Vision Benchmark Suite

Method

Categorical Depth Distribution Network for Monocular 3D Object Detection [CaDDN]
https://github.com/TRAILab/CaDDN

Submitted on 26 Oct. 2020 16:19 by
Cody Reading (University of Toronto)

Running time:		0.63 s
Environment:		GPU @ 2.5 Ghz (Python)

Method Description:

We propose Categorical Depth Distribution
Network (CaDDN), which uses a predicted
categorical depth distribution for each pixel
to project rich contextual feature information
to the appropriate depth interval in 3D space.
We then use the computationally efficient
bird’s-eye-view projection and single-stage
detector to produce the final output bounding
boxes. We design CaDDN as a fully
differentiable end-to-end approach for joint
depth estimation and object detection.

Parameters:

TBD

Latex Bibtex:

@article{CaDDN,
author = {Cody Reading and
Ali Harakeh and
Julia Chae and
Steven L. Waslander},
title = {Categorical Depth Distribution
Network for Monocular 3D Object Detection},
journal = {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)	93.61 %	80.73 %	71.09 %
Car (Orientation)	78.28 %	67.31 %	59.52 %
Car (3D Detection)	19.17 %	13.41 %	11.46 %
Car (Bird's Eye View)	27.94 %	18.91 %	17.19 %
Pedestrian (Detection)	46.35 %	32.42 %	29.98 %
Pedestrian (Orientation)	24.45 %	17.13 %	15.79 %
Pedestrian (3D Detection)	12.87 %	8.14 %	6.76 %
Pedestrian (Bird's Eye View)	14.72 %	9.41 %	8.17 %
Cyclist (Detection)	40.03 %	27.13 %	23.23 %
Cyclist (Orientation)	30.35 %	19.96 %	17.38 %
Cyclist (3D Detection)	7.00 %	3.41 %	3.30 %
Cyclist (Bird's Eye View)	9.67 %	5.38 %	4.75 %