Method

Instance Clustering on Fused Feature Maps [FCN-Depth]
None

Submitted on 12 Jun. 2017 06:39 by
Yecheng Lyu (Worcester Polytechnic Institute)

Running time:1 s
Environment:GPU @ 1.5 Ghz (Matlab + C/C++)

Method Description:
The proposed detector system contains following
parts: a Convolutional Neural Network generates
class map from camera image, a projector maps
point clouds to dense depth map, a object
instance cluster creates bounding box
candidates from fused feature maps, a scorer
evaluates the confidence of every candidates,
and a non-maximum suppression removes all
nearby candidates with lower confidence.
Parameters:
NMS is set to 0.7
Latex Bibtex:
None

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) 50.55 % 25.66 % 24.95 %
This table as LaTeX


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




eXTReMe Tracker