Method Details


Details for method 'Foveal Vision for Instance Segmentation of Road Images'

 

Method overview

name Foveal Vision for Instance Segmentation of Road Images
challenge instance-level semantic labeling
details Directly based on 'Pixel-level Encoding for Instance Segmentation'. Adds an improved angular distance measure and a foveal concept to better address small objects at the vanishing point of the road.
publication Foveal Vision for Instance Segmentation of Road Images
Benedikt Ortelt, Christian Herrmann, Dieter Willersinn, J├╝rgen Beyerer
VISAPP 2018
project page / code
used Cityscapes data fine annotations, stereo
used external data ImageNet
runtime n/a
subsampling no
submission date September, 2017
previous submissions

 

Average results

Metric Value
AP 12.5006
AP50% 25.2121
AP100m 20.361
AP50m 22.1157

 

Class results

Class AP AP50% AP100m AP50m
person 13.3986 31.4868 24.512 24.72
rider 11.4223 29.6768 19.6153 20.1893
car 24.4889 39.9936 39.2834 42.4629
truck 9.35526 15.9759 14.5412 17.151
bus 14.497 23.7958 24.1934 27.6372
train 12.1842 21.6991 18.4973 21.8025
motorcycle 7.97874 19.2051 11.0959 11.6755
bicycle 6.67976 19.8637 11.1491 11.2875

 

Links

Download results as .csv file

Benchmark page