Method Details


Details for method 'MSNET'

 

Method overview

name MSNET
challenge pixel-level semantic labeling
details previously also listed as "MultiPathJoin" and "MultiPath_Scale".
publication Anonymous
project page / code
used Cityscapes data fine annotations
used external data ImageNet
runtime 0.2 s
gtx1080ti
subsampling no
submission date November, 2017
previous submissions 1, 2, 3

 

Average results

Metric Value
IoU Classes 76.8296
iIoU Classes 57.1467
IoU Categories 90.5758
iIoU Categories 81.6312

 

Class results

Class IoU iIoU
road 98.3243 -
sidewalk 85.0102 -
building 92.5138 -
wall 48.6087 -
fence 56.6653 -
pole 67.5266 -
traffic light 75.5175 -
traffic sign 78.4134 -
vegetation 93.33 -
terrain 72.8307 -
sky 94.8734 -
person 85.8023 74.1783
rider 71.1161 51.4051
car 95.295 89.7395
truck 59.8229 37.6028
bus 73.2612 52.7529
train 65.9408 38.2619
motorcycle 68.4808 46.6128
bicycle 76.4294 66.6206

 

Category results

Category IoU iIoU
flat 98.6062 -
nature 93.0638 -
object 73.4817 -
sky 94.8734 -
construction 93.0216 -
human 86.153 74.9921
vehicle 94.831 88.2702

 

Links

Download results as .csv file

Benchmark page