Multi-Scale Context Aggregation by Dilated Convolutions

23 Nov 2015Fisher YuVladlen Koltun

State-of-the-art models for semantic segmentation are based on adaptations of convolutional networks that had originally been designed for image classification. However, dense prediction and image classification are structurally different... (read more)

PDF Abstract

Evaluation results from the paper


Task Dataset Model Metric name Metric value Global rank Compare
Semantic Segmentation ADE20K DilatedNet Validation mIoU 32.31 # 5
Semantic Segmentation CamVid Dilated Convolutions Mean IoU 65.3% # 4
Real-Time Semantic Segmentation CamVid Dilation10 mIoU 65.3% # 4
Real-Time Semantic Segmentation CamVid Dilation10 Time (ms) 227 # 5
Real-Time Semantic Segmentation CamVid Dilation10 Frame (fps) 4.4 # 5
Semantic Segmentation Cityscapes Dilation10 Mean IoU 67.1% # 13
Real-Time Semantic Segmentation Cityscapes Dilation10 mIoU 67.1% # 6
Real-Time Semantic Segmentation Cityscapes Dilation10 Time (ms) 4000 # 8
Real-Time Semantic Segmentation Cityscapes Dilation10 Frame (fps) 0.25 # 10
Semantic Segmentation PASCAL VOC 2012 Dilated Convolutions Mean IoU 67.6% # 17