Method

faster RCNN with ohem [FRO]


Submitted on 3 Jun. 2017 06:40 by
guo zhixian (Nanchang Hangkong University)

Running time:0.19 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
faster RCNN with ohem
Parameters:
We used 10 scales and 7 aspect ratios in the RPN
and ohem with fast RCNN
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) 70.96 % 53.78 % 46.00 %
Pedestrian (Detection) 57.56 % 45.43 % 40.50 %
Cyclist (Detection) 59.96 % 42.98 % 38.97 %
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