TY - JOUR
T1 - A Componentwise Approach to Weakly Supervised Semantic Segmentation Using Dual-Feedback Network
AU - Zhang, Zhengqiang
AU - Peng, Qinmu
AU - Fu, Sichao
AU - Wang, Wenjie
AU - Cheung, Yiu-Ming
AU - Zhao, Yue
AU - Yu, Shujian
AU - You, Xinge
N1 - Funding information:
This work was supported in part by the National Natural Science Foundation of China under Grant 61772220 and Grant 62101179, in part by the Key Program for International S&T Cooperation Projects of China under Grant 2016YFE0121200, in part by the Special Projects for Technology Innovation of Hubei Province under Grant 2018ACA135, in part by the Key Science and Technology Innovation Program of Hubei Province under Grant 2017AAA017, in part by the Key Research and Development Program of Hubei Province under Grant 2020BAB027, in part by the Natural Science Foundation of Hubei Province under Grant 2021CFB332, in part by fund from the Science, Technology and Innovation Commission of Shenzhen Municipality under Grant JSGG20180507182030600, and in part by the School Project of Hubei University under Grant 202011903000002.
Publisher copyright:
© 2022 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
PY - 2023/10
Y1 - 2023/10
N2 - Recent weakly supervised semantic segmentation methods generate pseudolabels to recover the lost position information in weak labels for training the segmentation network. Unfortunately, those pseudolabels often contain mislabeled regions and inaccurate boundaries due to the incomplete recovery of position information. It turns out that the result of semantic segmentation becomes determinate to a certain degree. In this article, we decompose the position information into two components: high-level semantic information and low-level physical information, and develop a componentwise approach to recover each component independently. Specifically, we propose a simple yet effective pseudolabels updating mechanism to iteratively correct mislabeled regions inside objects to precisely refine high-level semantic information. To reconstruct low-level physical information, we utilize a customized superpixel-based random walk mechanism to trim the boundaries. Finally, we design a novel network architecture, namely, a dual-feedback network (DFN), to integrate the two mechanisms into a unified model. Experiments on benchmark datasets show that DFN outperforms the existing state-of-the-art methods in terms of intersection-over-union (mIoU).
AB - Recent weakly supervised semantic segmentation methods generate pseudolabels to recover the lost position information in weak labels for training the segmentation network. Unfortunately, those pseudolabels often contain mislabeled regions and inaccurate boundaries due to the incomplete recovery of position information. It turns out that the result of semantic segmentation becomes determinate to a certain degree. In this article, we decompose the position information into two components: high-level semantic information and low-level physical information, and develop a componentwise approach to recover each component independently. Specifically, we propose a simple yet effective pseudolabels updating mechanism to iteratively correct mislabeled regions inside objects to precisely refine high-level semantic information. To reconstruct low-level physical information, we utilize a customized superpixel-based random walk mechanism to trim the boundaries. Finally, we design a novel network architecture, namely, a dual-feedback network (DFN), to integrate the two mechanisms into a unified model. Experiments on benchmark datasets show that DFN outperforms the existing state-of-the-art methods in terms of intersection-over-union (mIoU).
KW - Componentwise approach
KW - dual-feedback network (DFN)
KW - weakly supervised semantic segmentation
KW - deep learning
UR - http://www.scopus.com/inward/record.url?scp=85124186933&partnerID=8YFLogxK
U2 - 10.1109/TNNLS.2022.3144194
DO - 10.1109/TNNLS.2022.3144194
M3 - Journal article
SN - 2162-237X
VL - 34
SP - 7541
EP - 7554
JO - IEEE Transactions on Neural Networks and Learning Systems
JF - IEEE Transactions on Neural Networks and Learning Systems
IS - 10
ER -