Method Details


Details for method 'DSNet'

 

Method overview

name DSNet
challenge pixel-level semantic labeling
details DSNet for Real-Time Driving Scene Semantic Segmentation
publication DSNet for Real-Time Driving Scene Semantic Segmentation
Wenfu Wang
https://arxiv.org/abs/1812.07049
project page / code https://github.com/s7ev3n/DSNet
used Cityscapes data fine annotations, coarse annotations
used external data
runtime 0.027 s
subsampling 2
submission date November, 2018
previous submissions

 

Average results

Metric Value
IoU Classes 69.3494
iIoU Classes 42.8309
IoU Categories 86.0237
iIoU Categories 70.7173

 

Class results

Class IoU iIoU
road 97.1499 -
sidewalk 79.7279 -
building 89.4398 -
wall 37.8159 -
fence 50.4046 -
pole 56.7067 -
traffic light 63.0525 -
traffic sign 68.5348 -
vegetation 91.0261 -
terrain 67.872 -
sky 93.4553 -
person 75.66 57.1239
rider 61.4416 35.0234
car 91.9143 85.3724
truck 50.4037 21.9457
bus 66.7712 37.1416
train 56.768 28.4335
motorcycle 54.0617 30.0738
bicycle 65.4328 47.5332

 

Category results

Category IoU iIoU
flat 97.6695 -
nature 90.5302 -
object 63.1422 -
sky 93.4553 -
construction 89.8891 -
human 76.1455 58.0285
vehicle 91.3344 83.4061

 

Links

Download results as .csv file

Benchmark page