Abstract:Although semantic segmentation models based on deep neural networks (DNNs) have achieved excellent results, generalizing well from one remote sensing dataset (source domain) to another dataset with different acquisition conditions (target domain) remains a major challenge. Many domain adaptation (DA) approaches have been proposed to address this problem. DA aims to help DNNs learn a generalizable representation space in which source and target domains have similar feature distributions, but most of the existing DA approaches have difficulty in aligning the high-dimensional image representations of two domains directly. In this study, we proposed a model integrating contrastive learning and adversarial learning in a unified framework for aligning two domains in both representation space and spatial layout. Specifically, the model consists of a semantic segmentation network for feature extraction and two branches for DA. The first branch is used for adaptation in representation space directly by a proposed pixelwise contrastive loss, while the second branch is used for adaptation in predicted results to help two domains have similar spatial layouts through a novel but simple entropy-based similarity discriminator. Additionally, a training strategy called category similarity matching sampling was proposed to provide source and target image pairs with similar category composition for each training iteration, which can help the two branches work better. Extensive experiments indicated that the two branches can benefit each other to gain a superior performance and DA pretraining by our methods can achieve impressive results with only a small number of target labeled samples.

Drop to Adapt: Learning Discriminative Features for Unsupervised Domain Adaptation

Exploring Dropout Discriminator for Domain Adaptation

Joint Domain Alignment and Discriminative Feature Learning for Unsupervised Deep Domain Adaptation

Curriculum based Dropout Discriminator for Domain Adaptation

Learning Transferable and Discriminative Features for Unsupervised Domain Adaptation

Domain-Invariant Adversarial Learning for Unsupervised Domain Adaption.

Improving Target Discriminability for Unsupervised Domain Adaptation

DAST: Unsupervised Domain Adaptation in Semantic Segmentation Based on Discriminator Attention and Self-Training.

Deep Transfer Network: Unsupervised Domain Adaptation

When Unsupervised Domain Adaptation Meets Tensor Representations.

Finding Robust Transfer Features for Unsupervised Domain Adaptation.

Class Discriminative Adversarial Learning for Unsupervised Domain Adaptation

Discriminative Label Consistent Domain Adaptation

Discriminative Partial Domain Adversarial Network

Classification Constrained Discriminator for Domain Adaptive Semantic Segmentation

Transferable Adversarial Training: A General Approach to Adapting Deep Classifiers

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

CLDA: an Adversarial Unsupervised Domain Adaptation Method with Classifier-Level Adaptation

Domain Adaptation for Structured Output via Discriminative Patch Representations

Unsupervised Transductive Domain Adaptation

Learning Transferable Features with Deep Adaptation Networks