Method Details

Details for method 'RelationNet_Coarse'


Method overview

name RelationNet_Coarse
challenge pixel-level semantic labeling
details Semantic image segmentation, which assigns labels in pixel level, plays a central role in image understanding. Recent approaches have attempted to harness the capabilities of deep learning. However, one central problem of these methods is that deep convolution neural network gives little consideration to the correlation among pixels. To handle this issue, in this paper, we propose a novel deep neural network named RelationNet, which utilizes CNN and RNN to aggregate context information. Besides, a spatial correlation loss is applied to supervise RelationNet to align features of spatial pixels belonging to same category. Importantly, since it is expensive to obtain pixel-wise annotations, we exploit a new training method for combining the coarsely and finely labeled data. Separate experiments show the detailed improvements of each proposal. Experimental results demonstrate the effectiveness of our proposed method to the problem of semantic image segmentation.
publication RelationNet: Learning Deep-Aligned Representation for Semantic Image Segmentation
Yueqing Zhuang
project page / code
used Cityscapes data fine annotations, coarse annotations
used external data ImageNet
runtime n/a
subsampling no
submission date December, 2017
previous submissions


Average results

Metric Value
IoU Classes 82.3827
iIoU Classes 61.9213
IoU Categories 91.7642
iIoU Categories 81.3671


Class results

Class IoU iIoU
road 98.8209 -
sidewalk 87.8531 -
building 94.0488 -
wall 67.6939 -
fence 64.3664 -
pole 70.2303 -
traffic light 77.0751 -
traffic sign 81.1116 -
vegetation 93.933 -
terrain 73.502 -
sky 95.8116 -
person 87.8062 71.9151
rider 73.3936 55.8321
car 96.4348 91.1962
truck 75.3153 44.1013
bus 89.4443 58.632
train 88.1458 55.5492
motorcycle 72.034 52.6158
bicycle 78.2512 65.5286


Category results

Category IoU iIoU
flat 98.7849 -
nature 93.637 -
object 76.1192 -
sky 95.8116 -
construction 94.1368 -
human 87.9354 73.3237
vehicle 95.9245 89.4105



Download results as .csv file

Benchmark page