Method Details


Details for method 'TASCNet-enhanced'

 

Method overview

name TASCNet-enhanced
challenge panoptic semantic labeling
details We proposed a joint network for panoptic segmentation, which is a variation of our previous work, TASCNet. (https://arxiv.org/pdf/1812.01192.pdf) A shared backbone (ResNeXt-101) pretrained on COCO detection is used.
publication Learning to Fuse Things and Stuff
Jie Li, Allan Raventos, Arjun Bhargava, Takaaki Tagawa, Adrien Gaidon
Arxiv
https://arxiv.org/abs/1812.01192
project page / code
used Cityscapes data fine annotations, video
used external data COCO
runtime n/a
subsampling no
submission date May, 2019
previous submissions

 

Average results

Metric AllThingsStuff
PQ 60.7117 53.4201 66.0146
SQ 81.0202 79.6543 82.0136
RQ 73.807 66.9877 78.7665

 

Class results

Class PQ SQ RQ
road 98.1753 98.3049 99.8682
sidewalk 75.4642 84.1438 89.6848
building 87.8064 90.405 97.1257
wall 33.322 74.1578 44.9339
fence 35.2333 72.9537 48.2955
pole 53.9444 67.9571 79.3801
traffic light 52.5096 73.8356 71.117
traffic sign 66.771 77.6877 85.948
vegetation 90.014 91.3204 98.5695
terrain 43.0675 78.2308 55.0518
sky 89.8528 93.1526 96.4576
person 55.1516 77.3714 71.2817
rider 52.5931 74.1513 70.9268
car 66.8824 84.632 79.0273
truck 49.3571 85.3173 57.8512
bus 57.5856 86.5041 66.5698
train 55.6635 81.176 68.5714
motorcycle 46.9393 75.5429 62.1359
bicycle 43.1882 72.5396 59.5374

 

Links

Download results as .csv file

Benchmark page