Multi-level adversarial network for domain adaptive semantic segmentation

Recent progresses in domain adaptive semantic segmentation demonstrate the effectiveness of adversarial learning (AL) in unsupervised domain adaptation. However, most adversarial learning based methods align source and target distributions at a global image level but neglect the inconsistency around...

Full description

Saved in:
Bibliographic Details
Main Authors: Huang, Jiaxing, Guan, Dayan, Xiao, Aoran, Lu, Shijian
Other Authors: School of Computer Science and Engineering
Format: Article
Language:English
Published: 2022
Subjects:
Online Access:https://hdl.handle.net/10356/157049
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-157049
record_format dspace
spelling sg-ntu-dr.10356-1570492022-05-01T06:16:56Z Multi-level adversarial network for domain adaptive semantic segmentation Huang, Jiaxing Guan, Dayan Xiao, Aoran Lu, Shijian School of Computer Science and Engineering Engineering::Computer science and engineering Unsupervised Domain Adaptation Semantic Segmentation Recent progresses in domain adaptive semantic segmentation demonstrate the effectiveness of adversarial learning (AL) in unsupervised domain adaptation. However, most adversarial learning based methods align source and target distributions at a global image level but neglect the inconsistency around local image regions. This paper presents a novel multi-level adversarial network (MLAN) that aims to address inter-domain inconsistency at both global image level and local region level optimally. MLAN has two novel designs, namely, region-level adversarial learning (RL-AL) and co-regularized adversarial learning (CR-AL). Specifically, RL-AL models prototypical regional context-relations explicitly in the feature space of a labelled source domain and transfers them to an unlabelled target domain via adversarial learning. CR-AL fuses region-level AL and image-level AL optimally via mutual regularization. In addition, we design a multi-level consistency map that can guide domain adaptation in both input space (i.e., image-to-image translation) and output space (i.e., self-training) effectively. Extensive experiments show that MLAN outperforms the state-of-the-art with a large margin consistently across multiple datasets. Submitted/Accepted version 2022-05-01T06:16:55Z 2022-05-01T06:16:55Z 2022 Journal Article Huang, J., Guan, D., Xiao, A. & Lu, S. (2022). Multi-level adversarial network for domain adaptive semantic segmentation. Pattern Recognition, 123, 108384-. https://dx.doi.org/10.1016/j.patcog.2021.108384 0031-3203 https://hdl.handle.net/10356/157049 10.1016/j.patcog.2021.108384 2-s2.0-85118151273 123 108384 en Pattern Recognition © 2021 Elsevier Ltd. All rights reserved. This paper was published in Pattern Recognition and is made available with permission of Elsevier Ltd. application/pdf
institution Nanyang Technological University
building NTU Library
continent Asia
country Singapore
Singapore
content_provider NTU Library
collection DR-NTU
language English
topic Engineering::Computer science and engineering
Unsupervised Domain Adaptation
Semantic Segmentation
spellingShingle Engineering::Computer science and engineering
Unsupervised Domain Adaptation
Semantic Segmentation
Huang, Jiaxing
Guan, Dayan
Xiao, Aoran
Lu, Shijian
Multi-level adversarial network for domain adaptive semantic segmentation
description Recent progresses in domain adaptive semantic segmentation demonstrate the effectiveness of adversarial learning (AL) in unsupervised domain adaptation. However, most adversarial learning based methods align source and target distributions at a global image level but neglect the inconsistency around local image regions. This paper presents a novel multi-level adversarial network (MLAN) that aims to address inter-domain inconsistency at both global image level and local region level optimally. MLAN has two novel designs, namely, region-level adversarial learning (RL-AL) and co-regularized adversarial learning (CR-AL). Specifically, RL-AL models prototypical regional context-relations explicitly in the feature space of a labelled source domain and transfers them to an unlabelled target domain via adversarial learning. CR-AL fuses region-level AL and image-level AL optimally via mutual regularization. In addition, we design a multi-level consistency map that can guide domain adaptation in both input space (i.e., image-to-image translation) and output space (i.e., self-training) effectively. Extensive experiments show that MLAN outperforms the state-of-the-art with a large margin consistently across multiple datasets.
author2 School of Computer Science and Engineering
author_facet School of Computer Science and Engineering
Huang, Jiaxing
Guan, Dayan
Xiao, Aoran
Lu, Shijian
format Article
author Huang, Jiaxing
Guan, Dayan
Xiao, Aoran
Lu, Shijian
author_sort Huang, Jiaxing
title Multi-level adversarial network for domain adaptive semantic segmentation
title_short Multi-level adversarial network for domain adaptive semantic segmentation
title_full Multi-level adversarial network for domain adaptive semantic segmentation
title_fullStr Multi-level adversarial network for domain adaptive semantic segmentation
title_full_unstemmed Multi-level adversarial network for domain adaptive semantic segmentation
title_sort multi-level adversarial network for domain adaptive semantic segmentation
publishDate 2022
url https://hdl.handle.net/10356/157049
_version_ 1734310248366211072