Method Details


Details for method 'pruned & dilated inception-resnet-v2 (PD-IR2)'

 

Method overview

name pruned & dilated inception-resnet-v2 (PD-IR2)
challenge pixel-level semantic labeling
details
publication Anonymous
project page / code https://github.com/MalteOeljeklaus/incept_resnet2_segmentation
used Cityscapes data fine annotations, coarse annotations
used external data ImageNet
runtime 0.69 s
Nvidia 12GB GTX Titan X (Maxwell)
subsampling no
submission date May, 2017
previous submissions

 

Average results

Metric Value
IoU Classes 67.2944
iIoU Classes 42.0699
IoU Categories 86.5336
iIoU Categories 68.2818

 

Class results

Class IoU iIoU
road 97.8825 -
sidewalk 81.8671 -
building 90.183 -
wall 39.5394 -
fence 47.7424 -
pole 54.7841 -
traffic light 58.1162 -
traffic sign 69.8859 -
vegetation 91.2957 -
terrain 70.3779 -
sky 94.3589 -
person 76.9729 53.0649
rider 51.8995 35.2961
car 92.8552 82.8592
truck 40.6886 20.8436
bus 54.2683 36.4657
train 55.2191 32.1092
motorcycle 45.5229 28.5122
bicycle 65.1336 47.4078

 

Category results

Category IoU iIoU
flat 98.2599 -
nature 91.0179 -
object 61.9435 -
sky 94.3589 -
construction 90.1621 -
human 78.403 55.3341
vehicle 91.5902 81.2295

 

Links

Download results as .csv file

Benchmark page