Method Details


Details for method 'ECFD'

 

Method overview

name ECFD
challenge pixel-level semantic labeling
details backbone: ConvNext-Large
publication Anonymous
project page / code https://github.com/songyang-xiaobai/ECFD-main
used Cityscapes data fine annotations
used external data ImageNet
runtime n/a
subsampling no
submission date November, 2023
previous submissions

 

Average results

Metric Value
IoU Classes 82.0998
iIoU Classes 63.5979
IoU Categories 92.2222
iIoU Categories 82.1443

 

Class results

Class IoU iIoU
road 98.7203 -
sidewalk 87.6399 -
building 94.115 -
wall 63.6181 -
fence 67.2108 -
pole 72.1356 -
traffic light 78.5576 -
traffic sign 82.4505 -
vegetation 94.1024 -
terrain 73.2192 -
sky 96.0032 -
person 88.7692 73.992
rider 76.3611 55.1865
car 96.313 90.8412
truck 79.2632 48.327
bus 85.0274 61.8646
train 71.2962 55.6435
motorcycle 75.5993 55.9584
bicycle 79.4951 66.9697

 

Category results

Category IoU iIoU
flat 98.7418 -
nature 93.8277 -
object 77.7846 -
sky 96.0032 -
construction 94.4297 -
human 88.8843 74.885
vehicle 95.8837 89.4036

 

Links

Download results as .csv file

Benchmark page