DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs

2 Jun 2016Liang-Chieh ChenGeorge PapandreouIasonas KokkinosKevin MurphyAlan L. Yuille

In this work we address the task of semantic image segmentation with Deep Learning and make three main contributions that are experimentally shown to have substantial practical merit. First, we highlight convolution with upsampled filters, or 'atrous convolution', as a powerful tool in dense prediction tasks... (read more)

PDF Abstract

Evaluation results from the paper

Task Dataset Model Metric name Metric value Global rank Compare
Semantic Segmentation Cityscapes DeepLab-CRF (ResNet-101) Mean IoU (class) 70.4% # 18
Semantic Segmentation PASCAL Context DeepLabV2 mIoU 45.7 # 8
Semantic Segmentation PASCAL VOC 2012 DeepLab-CRF (ResNet-101) Mean IoU 79.7% # 11