Method

ResNet18-based Recurrent Rolling Convolution with Disparity-Extended Images (Advanced Hardware) [ResNet-RRC w/RGBD]


Submitted on 6 Jul. 2018 06:26 by
Hyung-Joon Jeon (Sungkyunkwan University)

Running time:0.057 s
Environment:GPU @ 1.5 Ghz (Python + C/C++)

Method Description:
This is the result entry for ResNet18-based
Recurrent Rolling Convolution, in which the input
images are extended with disparity channel. The
disparity channel is pre-computed from left-eye and
right-eye image pairs.

Note that HW, OS, and driver specs are same as that
of ResNet-RRC trained in advanced hardware.
Parameters:
Same as ResNet-RRC
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) 89.91 % 81.09 % 71.78 %
This table as LaTeX


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




eXTReMe Tracker