Method

Geometry-aware data augmentation [GA-Aug]


Submitted on 10 Nov. 2020 01:49 by
LIAN QING (HKUST)

Running time:0.04 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
Wait to update
Parameters:
Wait to update
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) 94.55 % 91.46 % 84.85 %
Car (Orientation) 93.61 % 90.06 % 83.39 %
Car (3D Detection) 17.46 % 11.67 % 9.69 %
Car (Bird's Eye View) 24.64 % 16.45 % 14.15 %
Pedestrian (Detection) 78.06 % 63.53 % 57.80 %
Pedestrian (Orientation) 61.17 % 48.40 % 43.69 %
Pedestrian (3D Detection) 8.04 % 4.89 % 4.32 %
Pedestrian (Bird's Eye View) 8.83 % 5.86 % 4.77 %
Cyclist (Detection) 68.81 % 54.46 % 47.71 %
Cyclist (Orientation) 46.84 % 36.22 % 31.76 %
Cyclist (3D Detection) 1.99 % 1.20 % 1.09 %
Cyclist (Bird's Eye View) 2.71 % 1.73 % 1.57 %
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