Method

DeepLineEncoding [DLE]
https://github.com/cnexah/DeepLineEncoding

Submitted on 22 Oct. 2021 23:39 by
Ce Liu (ETH Zurich)

Running time:0.06 s
Environment:NVIDIA Tesla V100

Method Description:
We propose the deep line encoding to make better use of the line information in scenes.
Parameters:
TBD
Latex Bibtex:
@inproceedings{ce21dle,
author = {Ce Liu and Shuhang Gu and Luc Van Gool and Radu Timofte},
booktitle = {Proceedings of the British Machine Vision Conference ({BMVC})},
title = {Deep Line Encoding for Monocular 3D Object Detection and Depth Prediction},
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) 94.66 % 84.45 % 62.10 %
Car (Orientation) 94.06 % 83.19 % 61.13 %
Car (3D Detection) 24.23 % 14.33 % 10.30 %
Car (Bird's Eye View) 31.09 % 19.05 % 14.13 %
This table as LaTeX


2D object detection results.
This figure as: png eps pdf txt gnuplot



Orientation estimation results.
This figure as: png eps pdf txt gnuplot



3D object detection results.
This figure as: png eps pdf txt gnuplot



Bird's eye view results.
This figure as: png eps pdf txt gnuplot




eXTReMe Tracker