Method Details


Details for method 'HANet (fine-train only)'

 

Method overview

name HANet (fine-train only)
challenge pixel-level semantic labeling
details We use only fine-training data.
publication TBA
Anonymous
project page / code
used Cityscapes data fine annotations
used external data ImageNet
runtime n/a
subsampling no
submission date November, 2019
previous submissions

 

Average results

Metric Value
IoU Classes 80.8526
iIoU Classes 58.6192
IoU Categories 91.2303
iIoU Categories 79.5198

 

Class results

Class IoU iIoU
road 98.7232 -
sidewalk 87.17 -
building 93.617 -
wall 62.2647 -
fence 62.1925 -
pole 67.5584 -
traffic light 75.1976 -
traffic sign 79.1547 -
vegetation 93.7657 -
terrain 73.0622 -
sky 95.7709 -
person 87.0903 68.7882
rider 71.7266 48.5897
car 96.1732 91.5072
truck 72.6568 39.0423
bus 88.6371 54.1077
train 85.6915 55.8711
motorcycle 69.0237 47.7729
bicycle 76.7227 63.2747

 

Category results

Category IoU iIoU
flat 98.7671 -
nature 93.5506 -
object 73.7665 -
sky 95.7709 -
construction 93.8785 -
human 87.1208 69.5473
vehicle 95.7576 89.4922

 

Links

Download results as .csv file

Benchmark page