Method

Pruned ResNet18-based Recurrent Rolling Convolution [Pruned ResNet-RRC]


Submitted on 10 Jan. 2018 03:07 by
Hyung-Joon Jeon (Sungkyunkwan University)

Running time:0.058 s
Environment:GPU @ 1.0 Ghz (Python + C/C++)

Method Description:
This is the fast version of Recurrent Rolling
Convolution, with 18-layer ResNet as the base net.
The weights of the original ResNet-RRC have been
pruned.
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) 89.23 % 80.80 % 71.47 %
This table as LaTeX


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




eXTReMe Tracker