Laser & Optoelectronics Progress, Volume. 59, Issue 16, 1610004(2022)
Human Instance Segmentation Based on Two-Stream Convolutional Neural Network
Figures & Tables(7)
Table 1. Results of ablation study on COCOPersons validation set
View tableTable 1. Results of ablation study on COCOPersons validation set
Training method AP APM APL One Stream 0.342 0.333 0.385 Two Streams +Concatenate 0.588 0.542 0.690 +Max Pooling 0.591 0.545 0.689 +Avg Pooling 0.584 0.542 0.684 +SENet［ 23 ］0.593 0.545 0.693 +CA［ 29 ］0.592 0.548 0.690 +FFB（Ours） 0.595 0.548 0.697
Table 2. Segmentation results of different algorithms on OCHuman dataset
View tableTable 2. Segmentation results of different algorithms on OCHuman dataset
Method Dataset Backbone AP APH Mask R-CNN［ 1 ］OCH val ResNet50-fpn 0.163 0.113 OCH test ResNet50-fpn 0.169 0.128 Pose2Seg（GT Kpt）［ 16 ］OCH val ResNet50-fpn 0.544 0.491 OCH test ResNet50-fpn 0.552 0.495 Pose2Seg OCH val ResNet50-fpn 0.222 0.150 OCH test ResNet50-fpn 0.238 0.175 Ours OCH val ResNet50-fpn 0.573 0.576 OCH test ResNet50-fpn 0.567 0.570
Table 3. Segmentation results of different algorithms on COCOPersons dataset
View tableTable 3. Segmentation results of different algorithms on COCOPersons dataset
Method Backbone AP APM APL Mask R-CNN［ 1 ］ResNet50-fpn 0.532 0.433 0.648 PersonLab［ 18 ］ResNet101 0.476 0.592 PersonLab ResNet101（ms scale） 0.492 0.621 PersonLab ResNet152 0.483 0.595 PersonLab ResNet152（ms scale） 0.497 0.621 Pose2Seg（GT Kpt）［ 16 ］ResNet50-fpn 0.582 0.539 0.679 Pose2Seg ResNet50-fpn 0.555 0.498 0.670 Ours ResNet50-fpn 0.595 0.548 0.697
Tools
Get Citation
Copy Citation Text
Zitong Ma, Guodong Wang. Human Instance Segmentation Based on Two-Stream Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1610004
Paper Information
Category: Image Processing
Received: May. 18, 2021
Accepted: Jun. 27, 2021
Published Online: Jul. 22, 2022
The Author Email: Guodong Wang (doctorwgd@gmail.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20