Method

Improved Tiny Vehicle Detector [ITVD]
https://github.com/liuwei16/ITVD_icme

Submitted on 27 Apr. 2018 16:45 by
Wei Liu (National University of Defense Technology)

Running time:0.3 s
Environment:GPU @ 2.5 Ghz (C/C++)

Method Description:
The proposed method consists of a backward feature
enhancement network (BFEN) and a spatial layout
preserving network (SLPN). BFEN aims at generating
high-quality region proposals for vehicles of
various scales, and SLPN is designed to
progressively integrate ROI features, while
preserving the spatial layouts.
Parameters:
NMS=0.35
Latex Bibtex:
@inproceedings{liu2018learning,
title={Improving Tiny Vehicle Detection in
Complex Scenes},
author={Wei Liu, Shengcai Liao, Weidong Hu,
Xuezhi Liang, Yan Zhang},
booktitle={IEEE International Conference on
Multimedia and Expo (ICME)},
year={2018}
}

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) 95.85 % 91.73 % 79.31 %
This table as LaTeX


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




eXTReMe Tracker