Method Details

Details for method 'TKCN'


Method overview

name TKCN
challenge pixel-level semantic labeling
details Most existing semantic segmentation methods employ atrous convolution to enlarge the receptive field of filters, but neglect important local contextual information. To tackle this issue, we firstly propose a novel Kronecker convolution which adopts Kronecker product to expand its kernel for taking into account the feature vectors neglected by atrous convolutions. Therefore, it can capture local contextual information and enlarge the field of view of filters simultaneously without introducing extra parameters. Secondly, we propose Tree-structured Feature Aggregation (TFA) module which follows a recursive rule to expand and forms a hierarchical structure. Thus, it can naturally learn representations of multi-scale objects and encode hierarchical contextual information in complex scenes. Finally, we design Tree-structured Kronecker Convolutional Networks (TKCN) that employs Kronecker convolution and TFA module. Extensive experiments on three datasets, PASCAL VOC 2012, PASCAL-Context and Cityscapes, verify the e Submission was previously published with name "TKCNNet".
publication Tree-structured Kronecker Convolutional Networks for Semantic Segmentation
Tianyi Wu, Sheng Tang, Rui Zhang Linghui Li, Yongdong Zhang
project page / code
used Cityscapes data fine annotations
used external data ImageNet
runtime 1 s
subsampling no
submission date April, 2018
previous submissions 1


Average results

Metric Value
IoU Classes 78.8518
iIoU Classes 56.8644
IoU Categories 90.6926
iIoU Categories 77.4714


Class results

Class IoU iIoU
road 98.4268 -
sidewalk 85.3838 -
building 92.9535 -
wall 53.4949 -
fence 62.2114 -
pole 66.4596 -
traffic light 74.9408 -
traffic sign 79.0891 -
vegetation 93.4821 -
terrain 72.5224 -
sky 94.7897 -
person 86.2491 66.6433
rider 69.8628 48.4754
car 95.9119 89.1979
truck 67.8552 40.214
bus 83.753 51.9918
train 76.9482 52.7598
motorcycle 67.4482 43.3514
bicycle 76.4019 62.2819


Category results

Category IoU iIoU
flat 98.6634 -
nature 93.2069 -
object 72.8903 -
sky 94.7897 -
construction 93.4489 -
human 86.5793 67.6839
vehicle 95.2696 87.2589



Download results as .csv file

Benchmark page