Method Details


Details for method 'tek-Ifly'

 

Method overview

name tek-Ifly
challenge pixel-level semantic labeling
details using a fusion strategy of three single models, the best result of a single model is 80.01%´╝îmulti-scale
publication Iflytek
Iflytek-yin
project page / code
used Cityscapes data fine annotations
used external data ImageNet
runtime n/a
subsampling no
submission date August, 2017
previous submissions

 

Average results

Metric Value
IoU Classes 81.101
iIoU Classes 60.0928
IoU Categories 90.9173
iIoU Categories 79.5542

 

Class results

Class IoU iIoU
road 98.6023 -
sidewalk 86.258 -
building 93.4542 -
wall 61.1689 -
fence 64.1169 -
pole 66.0061 -
traffic light 75.5915 -
traffic sign 79.0729 -
vegetation 93.6575 -
terrain 72.8298 -
sky 95.5689 -
person 86.2588 69.8963
rider 69.919 48.3333
car 96.0435 90.056
truck 76.8006 44.5732
bus 90.7303 55.3552
train 86.7945 57.7367
motorcycle 70.9944 50.2468
bicycle 77.051 64.5448

 

Category results

Category IoU iIoU
flat 98.6927 -
nature 93.3537 -
object 72.8601 -
sky 95.5689 -
construction 93.7422 -
human 86.636 70.7497
vehicle 95.5678 88.3586

 

Links

Download results as .csv file

Benchmark page