Method Details


Details for method 'SwiftNet RN18 pyr sepBN MVD'

 

Method overview

name SwiftNet RN18 pyr sepBN MVD
challenge pixel-level semantic labeling
details
publication Efficient semantic segmentation with pyramidal fusion
M Oršić, S Šegvić
Pattern Recognition 2020
https://www.sciencedirect.com/science/article/abs/pii/S0031320320304143
project page / code https://github.com/orsic/swiftnet
used Cityscapes data fine annotations
used external data ImageNet, Vistas
runtime 0.029 s
GTX 1080Ti
subsampling no
submission date December, 2019
previous submissions

 

Average results

Metric Value
IoU Classes 76.3926
iIoU Classes 52.9273
IoU Categories 90.2813
iIoU Categories 77.8563

 

Class results

Class IoU iIoU
road 98.5036 -
sidewalk 85.2552 -
building 92.7291 -
wall 53.4133 -
fence 55.2264 -
pole 66.1326 -
traffic light 73.115 -
traffic sign 77.0777 -
vegetation 93.3805 -
terrain 70.4315 -
sky 95.7646 -
person 83.8686 66.2019
rider 65.7271 43.641
car 95.4156 90.0362
truck 64.6411 33.1843
bus 78.6245 48.3216
train 65.9622 42.2764
motorcycle 63.027 41.6389
bicycle 73.164 58.1182

 

Category results

Category IoU iIoU
flat 98.5832 -
nature 93.0049 -
object 72.3022 -
sky 95.7646 -
construction 93.0807 -
human 84.2984 67.6821
vehicle 94.9349 88.0306

 

Links

Download results as .csv file

Benchmark page