Abstract:The semantic segmentation of high-resolution (HR) remote sensing images (RSIs) has been extensively researched in various applications. However, segmentation networks are prone to significant performance degradation on unlabeled data due to domain shift, such as data distribution shifts arising from distinct geographic locations. To address this issue, we propose a multilevel-guided curriculum domain adaptation (MuGCDA) approach for joint samplewise, categorywise, and pixelwise tasks, which facilitates the final fine-grained segmentation task by guiding the target domain to acquire samplewise and categorywise domain-robust properties. Concretely, at the sample level, we formulate a sample spatial relationship consistency guidance (SSCG) loss that guides the target domain to acquire similar sample spatial relationship properties to the source domain. At the category level, we propose a category layout structure consistency guidance (CLCG) module that guides the target domain to acquire consistent layout properties. At the pixel level, we design an adaptive hierarchical pseudolabel weight setting (AHPWS) method with a self-training (ST) paradigm to reduce the effect of label noise while improving the quality of the generated pseudolabels. Furthermore, to improve the stability of the training process, we use a momentum network (MN) as the teacher network to obtain the property knowledge and pseudolabels, and then guide the whole domain transfer process of the segmentation network, which acts as the student network. Extensive comparison and ablation experiments are conducted in several cross-space and cross-spectral scenes, and the results show that our method achieves significant performance improvements in cross-domain scenes for HR RSIs.

PDA: Progressive Domain Adaptation for Semantic Segmentation

ADeLA: Automatic Dense Labeling with Attention for Viewpoint Shift in Semantic Segmentation

Domain Adaptation for Remote Sensing Image Semantic Segmentation: An Integrated Approach of Contrastive Learning and Adversarial Learning

Semi-Supervised Domain Adaptation via Adaptive and Progressive Feature Alignment

MultiDAN: Unsupervised, Multistage, Multisource and Multitarget Domain Adaptation for Semantic Segmentation of Remote Sensing Images

Partial Domain Adaptation on Semantic Segmentation

Progressive Domain Adaptation for Object Detection

Pixel-Level Domain Adaptation: A New Perspective for Enhancing Weakly Supervised Semantic Segmentation

Affinity Space Adaptation for Semantic Segmentation Across Domains

Optimal transport-based domain adaptation for semantic segmentation of remote sensing images

SSF-DAN: Separated Semantic Feature Based Domain Adaptation Network for Semantic Segmentation

IterDANet: Iterative Intra-Domain Adaptation for Semantic Segmentation of Remote Sensing Images

A Fine-Grained Unsupervised Domain Adaptation Framework for Semantic Segmentation of Remote Sensing Images

MADAN: Multi-source Adversarial Domain Aggregation Network for Domain Adaptation

Domain Adaptive and Generalizable Network Architectures and Training Strategies for Semantic Image Segmentation

Semi-supervised Domain Adaptation for Semantic Segmentation

DiGA: Distil to Generalize and then Adapt for Domain Adaptive Semantic Segmentation

A Multilevel-Guided Curriculum Domain Adaptation Approach to Semantic Segmentation for High-Resolution Remote Sensing Images.

EDAPS: Enhanced Domain-Adaptive Panoptic Segmentation

Domain-Agnostic Prior for Transfer Semantic Segmentation.

Multi-source Domain Adaptation for Semantic Segmentation