Method Details

Details for method 'DeepLabv2-CRF'


Method overview

name DeepLabv2-CRF
challenge pixel-level semantic labeling
details DeepLabv2-CRF is based on three main methods. First, we employ convolution with upsampled filters, or ‘atrous convolution’, as a powerful tool to repurpose ResNet-101 (trained on image classification task) in dense prediction tasks. Atrous convolution allows us to explicitly control the resolution at which feature responses are computed within DCNNs. It also allows us to effectively enlarge the field of view of filters to incorporate larger context without increasing the number of parameters or the amount of computation. Second, we propose atrous spatial pyramid pooling (ASPP) to robustly segment objects at multiple scales. ASPP probes an incoming convolutional feature layer with filters at multiple sampling rates and effective fields-of-views, thus capturing objects as well as image context at multiple scales. Third, we improve the localization of object boundaries by combining methods from DCNNs and fully connected Conditional Random Fields (CRFs). The model is only trained on train set.
publication DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs
Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, Alan L. Yuille
arXiv preprint
project page / code
used Cityscapes data fine annotations
used external data ImageNet
runtime n/a
subsampling no
submission date May, 2016
previous submissions


Average results

Metric Value
IoU Classes 70.3816
iIoU Classes 42.5971
IoU Categories 86.4017
iIoU Categories 67.696


Class results

Class IoU iIoU
road 97.8649 -
sidewalk 81.3219 -
building 90.35 -
wall 48.7736 -
fence 47.3634 -
pole 49.5789 -
traffic light 57.8685 -
traffic sign 67.2847 -
vegetation 91.8508 -
terrain 69.4396 -
sky 94.192 -
person 79.8312 51.4918
rider 59.8495 31.2131
car 93.7134 85.3921
truck 56.5019 26.5142
bus 67.4976 37.8275
train 57.4574 34.4958
motorcycle 57.6633 27.3617
bicycle 68.8479 46.4802


Category results

Category IoU iIoU
flat 98.2689 -
nature 91.4789 -
object 57.2914 -
sky 94.192 -
construction 90.7687 -
human 80.2267 52.5053
vehicle 92.5854 82.8867



Download results as .csv file

Benchmark page