Method Details


Details for method 'ADSCNet'

 

Method overview

name ADSCNet
challenge pixel-level semantic labeling
details A lightweight and real-time semantic segmentation method for mobile devices.
publication ADSCNet: Asymmetric Depthwise Separable Convolution for Semantic Segmentation in Real-time
project page / code
used Cityscapes data fine annotations
used external data
runtime 0.013 s
GPU
subsampling no
submission date July, 2018
previous submissions

 

Average results

Metric Value
IoU Classes 64.4849
iIoU Classes 36.7691
IoU Categories 84.8648
iIoU Categories 68.7401

 

Class results

Class IoU iIoU
road 97.2996 -
sidewalk 77.9635 -
building 88.5875 -
wall 39.4805 -
fence 40.0884 -
pole 51.3865 -
traffic light 55.0251 -
traffic sign 60.287 -
vegetation 91.1255 -
terrain 66.8951 -
sky 93.4988 -
person 73.6946 53.4794
rider 50.6123 25.2437
car 91.3804 84.4106
truck 44.8555 16.634
bus 51.7347 25.4595
train 48.0084 24.397
motorcycle 43.3706 19.9086
bicycle 59.919 44.6197

 

Category results

Category IoU iIoU
flat 97.9957 -
nature 90.7353 -
object 57.7493 -
sky 93.4988 -
construction 88.8706 -
human 74.8725 55.1413
vehicle 90.3311 82.3389

 

Links

Download results as .csv file

Benchmark page