Method Details

Details for method 'Fast-SCNN'


Method overview

name Fast-SCNN
challenge pixel-level semantic labeling
details The encoder-decoder framework is state-of-the-art for offline semantic image segmentation. Since the rise in autonomous systems, real-time computation is increasingly desirable. In this paper, we introduce fast segmentation convolutional neural network (Fast-SCNN), an above real-time semantic segmentation model on high resolution image data (1024x2048px) suited to efficient computation on embedded devices with low memory. Building on existing two-branch methods for fast segmentation, we introduce our `learning to downsample' module which computes low-level features for multiple resolution branches simultaneously. Our network combines spatial detail at high resolution with deep features extracted at lower resolution, yielding an accuracy of 68.0% mean intersection over union at 123.5 frames per second on Cityscapes. We also show that large scale pre-training is unnecessary. We thoroughly validate our metric in experiments with ImageNet pre-training and the coarse labeled data of Cityscapes. Finally, we show even faster computation with competitive results on subsampled inputs, without any network modifications.
publication Fast-SCNN: Fast Semantic Segmentation Network
Rudra PK Poudel, Stephan Liwicki, Roberto Cipolla
project page / code
used Cityscapes data fine annotations, coarse annotations
used external data
runtime 0.0081 s
Nvidia Titan Xp (Pascal)
subsampling no
submission date November, 2018
previous submissions


Average results

Metric Value
IoU Classes 68.0156
iIoU Classes 37.9056
IoU Categories 84.7353
iIoU Categories 63.4613


Class results

Class IoU iIoU
road 97.9461 -
sidewalk 81.5856 -
building 89.6942 -
wall 46.3715 -
fence 48.6384 -
pole 48.3227 -
traffic light 53.0512 -
traffic sign 60.5445 -
vegetation 90.7149 -
terrain 67.175 -
sky 94.3191 -
person 73.9914 45.1417
rider 54.5843 25.5122
car 92.9909 83.255
truck 57.4255 20.045
bus 65.468 32.0083
train 58.217 34.2284
motorcycle 50.0359 20.7084
bicycle 61.2194 42.3458


Category results

Category IoU iIoU
flat 98.1956 -
nature 90.3145 -
object 55.0954 -
sky 94.3191 -
construction 89.6207 -
human 74.1565 46.1312
vehicle 91.4451 80.7914



Download results as .csv file

Benchmark page