Method Details


Details for method 'EffPS_b1bs4_RVC'

 

Method overview

name EffPS_b1bs4_RVC
challenge panoptic semantic labeling
details EfficientPS with EfficientNet-b1 backbone. Trained with a batch size of 4.
publication EfficientPS: Efficient Panoptic Segmentation
Rohit Mohan, Abhinav Valada
https://arxiv.org/abs/2004.02307
project page / code http://panoptic.cs.uni-freiburg.de/
used Cityscapes data fine annotations
used external data ImageNet, RVC
runtime n/a
subsampling no
submission date August, 2020
previous submissions

 

Average results

Metric AllThingsStuff
PQ 48.0239 40.8388 53.2494
SQ 78.32 78.7236 78.0265
RQ 59.1518 51.8745 64.4444

 

Class results

Class PQ SQ RQ
road 97.1688 97.3936 99.7692
sidewalk 64.5762 80.056 80.6638
building 84.0502 87.2672 96.3137
wall 22.0883 72.9639 30.273
fence 20.5339 66.8619 30.7109
pole 28.5536 61.9783 46.0703
traffic light 20.4107 63.3331 32.2275
traffic sign 42.7707 71.4515 59.8598
vegetation 86.8316 89.4053 97.1213
terrain 30.8747 74.5404 41.4201
sky 87.8851 93.0404 94.4591
person 47.939 78.2932 61.2302
rider 45.4879 71.554 63.5714
car 61.993 84.421 73.4332
truck 42.6916 85.7714 49.7738
bus 48.5959 86.1241 56.4254
train 8.77989 80.775 10.8696
motorcycle 36.7274 72.4081 50.7227
bicycle 34.4954 70.4418 48.97

 

Links

Download results as .csv file

Benchmark page