FastFCN: Rethinking Dilated Convolution in the Backbone for Semantic Segmentation

28 Mar 2019Huikai WuJunge ZhangKaiqi HuangKongming LiangYizhou Yu

Modern approaches for semantic segmentation usually employ dilated convolutions in the backbone to extract high-resolution feature maps, which brings heavy computation complexity and memory footprint. To replace the time and memory consuming dilated convolutions, we propose a novel joint upsampling module named Joint Pyramid Upsampling (JPU) by formulating the task of extracting high-resolution feature maps into a joint upsampling problem... (read more)

PDF Abstract

Evaluation results from the paper

Task Dataset Model Metric name Metric value Global rank Compare
Semantic Segmentation ADE20K EncNet + JPU Validation mIoU 44.34 # 3
Semantic Segmentation ADE20K EncNet + JPU Test Score 0.5584 # 1
Semantic Segmentation PASCAL Context Joint Pyramid Upsampling + EncNet mIoU 53.1 # 1