Method Details


Details for method 'Panoptic-DeepLab w/ SWideRNet [Mapillary Vistas]'

 

Method overview

name Panoptic-DeepLab w/ SWideRNet [Mapillary Vistas]
challenge panoptic semantic labeling
details We revisit the architecture design of Wide Residual Networks. We design a baseline model by incorporating the simple and effective Squeeze-and-Excitation and Switchable Atrous Convolution to the Wide-ResNets. Its network capacity is further scaled up or down by adjusting the width (i.e., channel size) and depth (i.e., number of layers), resulting in a family of SWideRNets (short for Scaling Wide Residual Networks). We demonstrate that such a simple scaling scheme, coupled with grid search, identifies several SWideRNets that significantly advance state-of-the-art performance on panoptic segmentation datasets in both the fast model regime and strong model regime.
publication Scaling Wide Residual Networks for Panoptic Segmentation
Liang-Chieh Chen, Huiyu Wang, Siyuan Qiao
https://arxiv.org/abs/2011.11675
project page / code
used Cityscapes data fine annotations
used external data ImageNet, Mapillary Vistas Research Edition
runtime n/a
subsampling no
submission date November, 2020
previous submissions

 

Average results

Metric AllThingsStuff
PQ 67.8211 60.9113 72.8463
SQ 83.7937 81.7757 85.2613
RQ 80.214 74.4178 84.4295

 

Class results

Class PQ SQ RQ
road 98.6622 98.7272 99.9341
sidewalk 79.9432 86.6463 92.2638
building 90.2671 92.2119 97.8909
wall 46.4352 78.5224 59.1362
fence 48.7728 77.7505 62.7299
pole 69.2895 74.8507 92.5703
traffic light 60.6986 80.3888 75.5062
traffic sign 76.0908 83.2338 91.4182
vegetation 91.7482 92.5039 99.1831
terrain 47.6531 79.7153 59.7791
sky 91.7492 93.3241 98.3125
person 59.7505 78.3998 76.2125
rider 58.2818 75.5672 77.1257
car 71.9981 85.5011 84.2072
truck 59.4246 88.3023 67.2968
bus 69.0872 89.345 77.3263
train 62.8695 85.685 73.3728
motorcycle 55.1678 77.4937 71.19
bicycle 50.7112 73.911 68.6111

 

Links

Download results as .csv file

Benchmark page