Method Details

Details for method 'Spatial Sampling Net'


Method overview

name Spatial Sampling Net
challenge instance-level semantic labeling
details We propose a network architecture to perform efficient scene understanding. This work presents three main novelties: the first is an Improved Guided Upsampling Module that can replace in toto the decoder part in common semantic segmentation networks. Our second contribution is the introduction of a new module based on spatial sampling to perform Instance Segmentation. It provides a very fast instance segmentation, needing only thresholding as post-processing step at inference time. Finally, we propose a novel efficient network design that includes the new modules and we test it against different datasets for outdoor scene understanding.
publication Spatial Sampling Network for Fast Scene Understanding
Davide Mazzini, Raimondo Schettini
CVPR 2019 Workshop on Autonomous Driving
project page / code
used Cityscapes data fine annotations
used external data ImageNet
runtime 0.00884 s
Titan Xp
subsampling 2
submission date March, 2019
previous submissions


Average results

Metric Value
AP 9.1828
AP50% 16.8466
AP100m 16.4305
AP50m 21.3972


Class results

Class AP AP50% AP100m AP50m
person 8.79089 22.0124 18.5868 19.8783
rider 3.19721 11.6965 5.83988 6.21282
car 24.0375 35.4222 41.001 49.3948
truck 9.95311 13.3939 18.1166 25.2318
bus 13.1632 19.2854 24.436 36.8475
train 8.48218 16.1681 14.3033 23.36
motorcycle 4.35087 11.3931 6.52919 7.50976
bicycle 1.48746 5.40145 2.63146 2.74299



Download results as .csv file

Benchmark page