Method Details


Details for method 'SAIT'

 

Method overview

name SAIT
challenge pixel-level semantic labeling
details Given Training Data Only No Post-Processing (like CRF) This method was previously listed as "GSTE".
publication Anonymous
project page / code
used Cityscapes data fine annotations, coarse annotations
used external data ImageNet
runtime 1 s
GPU TITAN X
subsampling no
submission date August, 2016
previous submissions 1, 2, 3, 4, 5, 6

 

Average results

Metric Value
IoU Classes 75.2127
iIoU Classes 50.2307
IoU Categories 88.8176
iIoU Categories 73.6448

 

Class results

Class IoU iIoU
road 98.4241 -
sidewalk 85.061 -
building 91.9958 -
wall 47.0655 -
fence 56.0583 -
pole 59.4381 -
traffic light 67.5766 -
traffic sign 73.5371 -
vegetation 92.8186 -
terrain 71.1442 -
sky 94.7887 -
person 82.5529 60.508
rider 65.499 39.8038
car 94.9804 87.7365
truck 65.8836 34.0043
bus 79.824 43.5017
train 71.595 45.6395
motorcycle 59.4944 35.4441
bicycle 71.3034 55.2075

 

Category results

Category IoU iIoU
flat 98.5414 -
nature 92.5141 -
object 66.4768 -
sky 94.7887 -
construction 92.4111 -
human 82.6909 61.5433
vehicle 94.3 85.7463

 

Links

Download results as .csv file

Benchmark page