Method Details

Details for method 'DCNAS'


Method overview

name DCNAS
challenge pixel-level semantic labeling
details Neural Architecture Search (NAS) has shown great potentials in automatically designing scalable network architectures for dense image predictions. However, existing NAS algorithms usually compromise on restricted search space and search on proxy task to meet the achievable computational demands. To allow as wide as possible network architectures and avoid the gap between target and proxy dataset, we propose a Densely Connected NAS (DCNAS) framework, which directly searches the optimal network structures for the multi-scale representations of visual information, over a large-scale target dataset. Specifically, by connecting cells with each other using learnable weights, we introduce a densely connected search space to cover an abundance of mainstream network designs. Moreover, by combining both path-level and channel-level sampling strategies, we design a fusion module to reduce the memory consumption of ample search space.
publication DCNAS: Densely Connected Neural Architecture Search for Semantic Image Segmentation
Xiong Zhang, Hongmin Xu, Hong Mo, Jianchao Tan, Cheng Yang, Wenqi Ren
project page / code
used Cityscapes data fine annotations, coarse annotations
used external data
runtime n/a
subsampling no
submission date February, 2020
previous submissions


Average results

Metric Value
IoU Classes 83.6389
iIoU Classes 65.0129
IoU Categories 91.8707
iIoU Categories 82.4741


Class results

Class IoU iIoU
road 98.7856 -
sidewalk 87.9505 -
building 94.1501 -
wall 66.0463 -
fence 66.0985 -
pole 72.2417 -
traffic light 78.7213 -
traffic sign 82.7204 -
vegetation 94.1726 -
terrain 73.9007 -
sky 94.0262 -
person 88.2358 74.6373
rider 75.1175 56.9736
car 96.4795 91.2188
truck 82.6244 46.1493
bus 94.0975 61.5398
train 90.883 65.2694
motorcycle 73.6103 55.6609
bicycle 79.2777 68.6541


Category results

Category IoU iIoU
flat 98.7589 -
nature 93.8133 -
object 77.7283 -
sky 94.0262 -
construction 94.3632 -
human 88.3295 75.4401
vehicle 96.0753 89.5081



Download results as .csv file

Benchmark page