Laser & Optoelectronics Progress, Volume. 56, Issue 24, 241005(2019)
Image Semantic Segmentation Based on Hierarchical Context Information
Figures & Tables(12)
Table 1. Performance of each model on Cityscapes validation set
View tableTable 1. Performance of each model on Cityscapes validation set
Method With/without model M iou /%PAM DCM-4 DCM-3 EM CCM Baseline √ 59.82 Ours-1 √ √ 60.73 Ours-2 √ √ 61.16 Ours-3 √ √ √ 61.78 Ours-4 √ √ √ √ 62.11
Table 2. Performance of each model on Cityscapes test set%
View tableTable 2. Performance of each model on Cityscapes test set%
Method Accuracy M iouRoa Sid Bui Wal Fen Pol TLi TSi Veg Ter Sky Per Rid Car Tru Bus Tra Mot Bic Base- line 96.5 72.4 85.8 38.1 35.7 30.6 41.7 50.6 86.3 58.9 90.4 65.2 45.8 88.0 55.1 64.5 49.8 36.3 51.9 60.17 Ours-2 96.7 73.5 86.1 38.2 36.8 33.3 42.7 51.3 87.1 61.1 91.6 66.1 47.8 90.3 54.8 67.2 50.3 36.7 53.0 61.30 Ours-3 96.7 73.0 86.4 44.2 37.5 30.7 41.5 51.5 87.4 58.4 91.7 66.9 47.3 90.5 56.6 68.0 56.9 40.3 52.4 61.99 Ours-4 96.9 74.5 86.7 44.2 38.2 31.2 42.8 52.1 87.8 59.7 91.7 66.7 48.3 90.4 55.1 65.4 55.8 40.5 53.5 62.19
Table 3. Network performance on Cityscapes dataset
View tableTable 3. Network performance on Cityscapes dataset
Model Backbone M iou /%Val set Test set Baseline* ResNet-50 76.34 - Baseline ResNet-50 75.81 - Ours-4 ResNet-50 77.20 - Deeplab-V2[ 10 ]ResNet-101 - 70.40 Refinenet[ 20 ]ResNet-101 - 73.60 Gcn[ 14 ]ResNet-101 - 76.90 Baseline ResNet-50 - 75.46 Ours-4 ResNet-50 - 76.73 Baseline ResNet-101 - 76.67 Ours-4 ResNet-101 - 78.01
Tools
Get Citation
Copy Citation Text
Shiyi Yue. Image Semantic Segmentation Based on Hierarchical Context Information[J]. Laser & Optoelectronics Progress, 2019, 56(24): 241005
Paper Information
Category: Image Processing
Received: Apr. 25, 2019
Accepted: Jun. 13, 2019
Published Online: Nov. 26, 2019
The Author Email: Yue Shiyi (shiyiyue@tju.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20