Method

CONV-BOX [la] [CONV-BOX]
[Anonymous Submission]

Submitted on 12 Aug. 2018 11:17 by
[Anonymous Submission]

Running time:0.2 s
Environment:Tesla V100

Method Description:
Single-stage detector with lidar-camera fusion.
Predictions of all three classes (car, pedestrian, cyclist)
are produced with a single model
Parameters:
TBD
Latex Bibtex:

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) 90.35 % 89.20 % 87.88 %
Car (3D Detection) 79.98 % 70.47 % 64.49 %
Car (Bird's Eye View) 87.54 % 84.56 % 77.79 %
Pedestrian (Detection) 63.98 % 55.23 % 54.18 %
Pedestrian (3D Detection) 47.74 % 41.01 % 35.98 %
Pedestrian (Bird's Eye View) 52.71 % 45.09 % 43.90 %
Cyclist (Detection) 72.62 % 63.84 % 56.69 %
Cyclist (3D Detection) 68.27 % 54.45 % 52.26 %
Cyclist (Bird's Eye View) 71.60 % 61.84 % 55.03 %
This table as LaTeX


2D object detection 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



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



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