Method Details


Details for method 'TuSimple'

 

Method overview

name TuSimple
challenge pixel-level semantic labeling
details
publication Understanding Convolution for Semantic Segmentation
Panqu Wang, Pengfei Chen, Ye Yuan, Ding Liu, Zehua Huang, Xiaodi Hou, Garrison Cottrell
https://arxiv.org/abs/1702.08502
project page / code https://github.com/TuSimple/TuSimple-DUC
used Cityscapes data fine annotations
used external data ImageNet
runtime n/a
subsampling no
submission date November, 2016
previous submissions 1, 2, 3

 

Average results

Metric Value
IoU Classes 77.6005
iIoU Classes 53.5518
IoU Categories 90.1338
iIoU Categories 75.2317

 

Class results

Class IoU iIoU
road 98.4685 -
sidewalk 85.5064 -
building 92.7747 -
wall 58.5983 -
fence 55.4993 -
pole 65.0465 -
traffic light 73.4678 -
traffic sign 77.8542 -
vegetation 93.2555 -
terrain 72.0277 -
sky 95.2064 -
person 84.7767 62.7137
rider 68.4852 43.9108
car 95.3674 88.1879
truck 70.9224 38.5075
bus 78.753 49.7417
train 68.6771 40.274
motorcycle 65.9145 43.6938
bicycle 73.8075 61.3854

 

Category results

Category IoU iIoU
flat 98.5911 -
nature 92.9609 -
object 71.9139 -
sky 95.2064 -
construction 92.9901 -
human 84.7784 63.9606
vehicle 94.4958 86.5028

 

Links

Download results as .csv file

Benchmark page