Method Details


Details for method 'PSPNet'

 

Method overview

name PSPNet
challenge pixel-level semantic labeling
details Using training set only (2975 images) without adding validation set or coarse set, no post-processsing like CRF.
publication Anonymous
project page / code
used Cityscapes data fine annotations
used external data ImageNet
runtime n/a
subsampling no
submission date August, 2016
previous submissions 1, 2, 3

 

Average results

Metric Value
IoU Classes 75.9299
iIoU Classes 53.4795
IoU Categories 90.0164
iIoU Categories 77.7225

 

Class results

Class IoU iIoU
road 98.5237 -
sidewalk 85.6017 -
building 92.4006 -
wall 43.3227 -
fence 56.4226 -
pole 63.142 -
traffic light 73.0311 -
traffic sign 76.757 -
vegetation 92.9574 -
terrain 72.4608 -
sky 95.0343 -
person 85.6044 66.8283
rider 68.5121 45.8225
car 95.5645 89.6901
truck 58.8694 31.3224
bus 74.6218 49.0834
train 66.0457 38.0967
motorcycle 67.9292 45.7296
bicycle 75.8672 61.263

 

Category results

Category IoU iIoU
flat 98.6628 -
nature 92.6885 -
object 70.1037 -
sky 95.0343 -
construction 92.6468 -
human 86.1412 67.9135
vehicle 94.8375 87.5316

 

Links

Download results as .csv file

Benchmark page