Method

BirdNet+: End-to-End 3D Object Detection in LiDAR Bird's Eye View [la] [BirdNet+ (legacy)]
https://github.com/AlejandroBarrera/BirdNet2

Submitted on 4 Mar. 2020 19:09 by
Carlos Guindel (Universidad Carlos III de Madrid)

Running time:0.1 s
Environment:Titan Xp (PyTorch)

Method Description:
NOTE: This submission corresponds to our ITSC 2020 paper, where a 35 x 70 m (1400x700) BEV is used. Please refer to the other BirdNet+ entry for results on an optimized 50 x 45 m (1000 x 900) BEV.

We present a fully end-to-end 3D object detection
framework that can infer oriented 3D boxes solely
from BEV images by using a two-stage object
detector and ad-hoc regression branches,
eliminating the need for a post-processing stage.
Parameters:
Final model: ResNet-50-FPN + real dimensions
regression + height regression + hybrid yaw
estimation (12 bins + regression).
Latex Bibtex:
@INPROCEEDINGS{9294293,
author={A. {Barrera} and C. {Guindel} and J. {Beltrán} and F. {García}},
booktitle={2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC)},
title={BirdNet+: End-to-End 3D Object Detection in LiDAR Bird’s Eye View},
year={2020},
pages={2985-2990},
doi={10.1109/ITSC45102.2020.9294293}
}

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) 92.10 % 68.05 % 65.61 %
Car (Orientation) 91.82 % 67.65 % 65.11 %
Car (3D Detection) 70.14 % 51.85 % 50.03 %
Car (Bird's Eye View) 84.80 % 63.33 % 61.23 %
Pedestrian (Detection) 51.38 % 41.97 % 40.15 %
Pedestrian (Orientation) 36.76 % 29.56 % 28.10 %
Pedestrian (3D Detection) 37.99 % 31.46 % 29.46 %
Pedestrian (Bird's Eye View) 45.53 % 38.28 % 35.37 %
Cyclist (Detection) 74.97 % 54.61 % 50.29 %
Cyclist (Orientation) 69.92 % 50.94 % 47.01 %
Cyclist (3D Detection) 67.38 % 47.72 % 42.89 %
Cyclist (Bird's Eye View) 72.45 % 52.15 % 46.57 %
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



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



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