Skip to content

Latest commit

 

History

History
59 lines (42 loc) · 11.1 KB

README.md

File metadata and controls

59 lines (42 loc) · 11.1 KB
Raw

EncNet

Context Encoding for Semantic Segmentation

Introduction

Official Repo

Code Snippet

Abstract

Recent work has made significant progress in improving spatial resolution for pixelwise labeling with Fully Convolutional Network (FCN) framework by employing Dilated/Atrous convolution, utilizing multi-scale features and refining boundaries. In this paper, we explore the impact of global contextual information in semantic segmentation by introducing the Context Encoding Module, which captures the semantic context of scenes and selectively highlights class-dependent featuremaps. The proposed Context Encoding Module significantly improves semantic segmentation results with only marginal extra computation cost over FCN. Our approach has achieved new state-of-the-art results 51.7% mIoU on PASCAL-Context, 85.9% mIoU on PASCAL VOC 2012. Our single model achieves a final score of 0.5567 on ADE20K test set, which surpass the winning entry of COCO-Place Challenge in 2017. In addition, we also explore how the Context Encoding Module can improve the feature representation of relatively shallow networks for the image classification on CIFAR-10 dataset. Our 14 layer network has achieved an error rate of 3.45%, which is comparable with state-of-the-art approaches with over 10 times more layers. The source code for the complete system are publicly available.

Results and models

Cityscapes

Method Backbone Crop Size Lr schd Mem (GB) Inf time (fps) Device mIoU mIoU(ms+flip) config download
EncNet R-50-D8 512x1024 40000 8.6 4.58 V100 75.67 77.08 config model | log
EncNet R-101-D8 512x1024 40000 12.1 2.66 V100 75.81 77.21 config model | log
EncNet R-50-D8 769x769 40000 9.8 1.82 V100 76.24 77.85 config model | log
EncNet R-101-D8 769x769 40000 13.7 1.26 V100 74.25 76.25 config model | log
EncNet R-50-D8 512x1024 80000 - - V100 77.94 79.13 config model | log
EncNet R-101-D8 512x1024 80000 - - V100 78.55 79.47 config model | log
EncNet R-50-D8 769x769 80000 - - V100 77.44 78.72 config model | log
EncNet R-101-D8 769x769 80000 - - V100 76.10 76.97 config model | log

ADE20K

Method Backbone Crop Size Lr schd Mem (GB) Inf time (fps) Device mIoU mIoU(ms+flip) config download
EncNet R-50-D8 512x512 80000 10.1 22.81 V100 39.53 41.17 config model | log
EncNet R-101-D8 512x512 80000 13.6 14.87 V100 42.11 43.61 config model | log
EncNet R-50-D8 512x512 160000 - - V100 40.10 41.71 config model | log
EncNet R-101-D8 512x512 160000 - - V100 42.61 44.01 config model | log

Citation

@InProceedings{Zhang_2018_CVPR,
author = {Zhang, Hang and Dana, Kristin and Shi, Jianping and Zhang, Zhongyue and Wang, Xiaogang and Tyagi, Ambrish and Agrawal, Amit},
title = {Context Encoding for Semantic Segmentation},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2018}
}