Method Details


Details for method 'PBRNet'

 

Method overview

name PBRNet
challenge pixel-level semantic labeling
details modified MobileNetV2 backbone + Prediction and Boundary attention-based Refinement Module (PBRM)
publication Anonymous
project page / code
used Cityscapes data fine annotations
used external data ImageNet
runtime 0.0107 s
subsampling no
submission date May, 2020
previous submissions

 

Average results

Metric Value
IoU Classes 72.3846
iIoU Classes 51.9104
IoU Categories 88.6518
iIoU Categories 77.3402

 

Class results

Class IoU iIoU
road 98.0149 -
sidewalk 81.8296 -
building 91.3396 -
wall 46.2638 -
fence 51.548 -
pole 58.5298 -
traffic light 69.0943 -
traffic sign 74.0982 -
vegetation 92.9202 -
terrain 70.9592 -
sky 94.7021 -
person 82.6371 65.9348
rider 64.7344 46.0485
car 94.5257 89.4437
truck 57.7377 33.8801
bus 62.5232 45.1849
train 48.9849 30.8701
motorcycle 61.9379 44.5851
bicycle 72.9265 59.3365

 

Category results

Category IoU iIoU
flat 98.503 -
nature 92.5597 -
object 66.2229 -
sky 94.7021 -
construction 91.9319 -
human 83.221 67.0381
vehicle 93.4223 87.6424

 

Links

Download results as .csv file

Benchmark page