Method

ResNet18-based Recurrent Rolling Convolution [R-RRC]


Submitted on 4 Nov. 2017 04:38 by
Hyung-Joon Jeon (Sungkyunkwan University)

Running time:0.09 s
Environment:GPU @ 1.0 Ghz (Python + C/C++)

Method Description:
This is the Recurrent Rolling Convolution with 18-layer
ResNet as the base net.
Parameters:
base_lr=0.0005,
gamma=0.5,
stepsize=25000,
max_iter:100000
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) 89.99 % 82.94 % 72.21 %
This table as LaTeX


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




eXTReMe Tracker