Method

ResNet18-based Recurrent Rolling Convolution (All objects) [ResNet-RRC]


Submitted on 27 Jul. 2020 09:53 by
Hyung-Joon Jeon (Sungkyunkwan University)

Running time:0.11 s
Environment:GPU @ 1.5 Ghz (Python + C/C++)

Method Description:
This is the ResNet-RRC trained on all objects (car,
pedestrian, cyclist).
Parameters:
Same as ResNet-RRC (Car only), except that
Number_of_Iterations=200000
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) 89.43 % 84.81 % 73.18 %
Pedestrian (Detection) 66.44 % 52.09 % 47.51 %
Cyclist (Detection) 58.72 % 42.88 % 37.74 %
This table as LaTeX


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



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



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




eXTReMe Tracker