Method Details

Details for method 'DCNAS+ASPP [Mapillary Vistas]'


Method overview

name DCNAS+ASPP [Mapillary Vistas]
challenge pixel-level semantic labeling
details Existing NAS algorithms usually compromise on restricted search space or search on proxy task to meet the achievable computational demands. To allow as wide as possible network architectures and avoid the gap between realistic and proxy setting, we propose a novel 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 without proxy. 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 and mixture layer to reduce the memory consumption of ample search space, hence favor the proxyless searching. Compared with contemporary works, experiments reveal that the proxyless searching scheme is capable of bridge the gap between searching and training environments.
publication Anonymous
project page / code
used Cityscapes data fine annotations, coarse annotations
used external data mapillary vistas
runtime n/a
subsampling no
submission date November, 2020
previous submissions


Average results

Metric Value
IoU Classes 85.2893
iIoU Classes 69.9665
IoU Categories 93.0985
iIoU Categories 85.3329


Class results

Class IoU iIoU
road 98.9842 -
sidewalk 89.3682 -
building 94.8792 -
wall 71.7304 -
fence 69.0601 -
pole 75.5973 -
traffic light 81.9585 -
traffic sign 84.9182 -
vegetation 94.4953 -
terrain 75.2817 -
sky 96.3114 -
person 89.8726 78.2871
rider 79.0985 62.765
car 96.9086 92.6239
truck 81.6465 58.0365
bus 95.2768 71.539
train 86.9529 62.7616
motorcycle 77.1169 61.9473
bicycle 81.0402 71.7716


Category results

Category IoU iIoU
flat 98.9226 -
nature 94.3205 -
object 80.6266 -
sky 96.3114 -
construction 94.9196 -
human 90.0045 79.2064
vehicle 96.584 91.4593



Download results as .csv file

Benchmark page