Method Details


Details for method 'MaskRCNN_BOSH'

 

Method overview

name MaskRCNN_BOSH
challenge pixel-level semantic labeling
details MaskRCNN segmentation baseline for Bosh autodrive challenge , using Matterport's implementation of Mask RCNN https://github.com/matterport/Mask_RCNN 55k iterations, default parameters (backbone :resenet 101) each pixel is assigend with label-id based on its highest class score
publication Jin shengtao, Yi zhihao, Liu wei [Our team name is firefly]
project page / code
used Cityscapes data fine annotations
used external data coco
runtime n/a
subsampling no
submission date June, 2018
previous submissions

 

Average results

Metric Value
IoU Classes 41.5943
iIoU Classes 19.2986
IoU Categories 61.2908
iIoU Categories 31.2781

 

Class results

Class IoU iIoU
road 84.0654 -
sidewalk 20.6033 -
building 68.753 -
wall 15.8263 -
fence 12.5805 -
pole 11.1466 -
traffic light 33.7419 -
traffic sign 40.7674 -
vegetation 65.3318 -
terrain 7.45913 -
sky 50.8492 -
person 60.2306 24.1512
rider 38.5047 16.6114
car 81.7904 41.0207
truck 32.8656 11.2989
bus 44.5179 17.4347
train 58.5633 20.3863
motorcycle 28.8212 9.36296
bicycle 33.8738 14.1224

 

Category results

Category IoU iIoU
flat 90.4843 -
nature 61.949 -
object 22.2331 -
sky 50.8492 -
construction 68.2131 -
human 58.2264 24.1195
vehicle 77.0807 38.4368

 

Links

Download results as .csv file

Benchmark page