Method

variant-SSD for 3d detection [3DVSSD]


Submitted on 21 Apr. 2018 05:46 by
renling tong (zongmutech)

Running time:0.06 s
Environment:1 core @ 2.5 Ghz (C/C++)

Method Description:
This method proposes a single shot detector,
which helps detect 3-dimension objects and
estimate orientation. The detector was made by
using a variant SSD, which utilized a different
prior box for 3-dimension bounding box
regression. The model shows good time
performance and low computation complexity.
Parameters:
base_lr = 0.001
weight_dacay = 0.0005
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) 84.39 % 73.39 % 65.64 %
Car (Orientation) 77.22 % 64.72 % 57.56 %
Car (3D Detection) 1.38 % 1.14 % 1.27 %
Car (Bird's Eye View) 2.02 % 2.01 % 1.68 %
This table as LaTeX


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



Orientation estimation results.
This figure as: png eps pdf txt gnuplot



3D object detection results.
This figure as: png eps pdf txt gnuplot



Bird's eye view results.
This figure as: png eps pdf txt gnuplot




eXTReMe Tracker