Method Details


Details for method 'GUNet'

 

Method overview

name GUNet
challenge pixel-level semantic labeling
details Guided Upsampling Network for Real-Time Semantic Segmentation
publication Guided Upsampling Network for Real-Time Semantic Segmentation
Davide Mazzini
arxiv
https://arxiv.org/abs/1807.07466
project page / code
used Cityscapes data fine annotations
used external data ImageNet
runtime 0.03 s
Titan Xp
subsampling 2
submission date April, 2018
previous submissions

 

Average results

Metric Value
IoU Classes 70.3817
iIoU Classes 40.8457
IoU Categories 86.8188
iIoU Categories 69.107

 

Class results

Class IoU iIoU
road 98.1598 -
sidewalk 82.6616 -
building 90.5716 -
wall 47.3305 -
fence 45.3676 -
pole 51.9338 -
traffic light 59.0903 -
traffic sign 66.59 -
vegetation 91.7473 -
terrain 68.5239 -
sky 94.8033 -
person 79.0083 53.6719
rider 59.5358 29.9088
car 94.1197 85.5772
truck 60.3168 21.272
bus 71.4334 33.7668
train 54.0003 30.4476
motorcycle 54.8609 25.0472
bicycle 67.197 47.074

 

Category results

Category IoU iIoU
flat 98.3561 -
nature 91.4186 -
object 59.2429 -
sky 94.8033 -
construction 90.8206 -
human 79.6571 55.236
vehicle 93.4326 82.9781

 

Links

Download results as .csv file

Benchmark page