语义分割背景下UDA（无监督域自适应）的三个方向

一、基于对抗学习

AdaptSegNet是经典的基于对抗学习的域适应方法。这一类方法训练一个判别器来使得目标域的分布在像素空间（output space)或者特征空间(feature map)上进行对齐（AdaptSegNet证明语义分割任务而言，在像素空间上的对齐优于特征空间），从而使得分割模型的性能从源域泛化到目标域上。除了AdaptSeg之外，这一方向上的经典工作还有：

1、FCNs in the wild: Pixel-level adversarial and constraint-based adaptation（2016 CVPR） 

原文： https://arxiv.org/abs/1612.02649

数据集：Cityscapes、SYNTHIA、GTA5

Hoffman J, Wang D, Yu F, et al. Fcns in the wild: Pixel-level adversarial and constraint-based adaptation[J]. arXiv preprint arXiv:1612.02649, 2016.

2、Learning to adapt structured output space for semantic segmentation（2018 CVPR）（AdaptSegNet）

原文：https://arxiv.org/abs/1802.10349

代码：https://github.com/wasidennis/AdaptSegNet

数据集：Cityscapes、SYNTHIA、GTA5、Synscapes

Tsai Y H, Hung W C, Schulter S, et al. Learning to adapt structured output space for semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 7472-7481.

3、Advent: Adversarial entropy minimization for domain adaptation in semantic segmentation（2019 CVPR）

原文：https://arxiv.org/abs/1811.12833

代码：https://github.com/valeoai/ADVENT

数据集： Cityscapes、SYNTHIA、GTA5

Vu T H, Jain H, Bucher M, et al. Advent: Adversarial entropy minimization for domain adaptation in semantic segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 2517-2526.

4、Unsupervised intra-domain adaptation for semantic segmentation through self-supervision（2020 CVPR）​​​​​​ 

原文：https://arxiv.org/abs/2004.07703

代码：https://github.com/feipan664/IntraDA

数据集： Cityscapes、SYNTHIA、GTA5、Synscapes

Pan F, Shin I, Rameau F, et al. Unsupervised intra-domain adaptation for semantic segmentation through self-supervision[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020: 3764-3773

二、基于风格迁移

1、ColorMap GAN: Unsupervised domain adaptation for semantic segmentation using color mapping generative adversarial networks（2020 TGRS）

原文：https://arxiv.org/abs/1907.12859

Tasar O, Happy S L, Tarabalka Y, et al. ColorMapGAN: Unsupervised domain adaptation for semantic segmentation using color mapping generative adversarial networks[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(10): 7178-7193.

三、自监督学习

1、Dual Path Learning for Domain Adaptation of Semantic Segmentation（2021 ICCV）

原文：​​​​​​https://arxiv.org/abs/2108.06337 

代码：GitHub - royee182/DPL

数据集： Cityscapes、SYNTHIA、GTA5

Cheng Y, Wei F, Bao J, et al. Dual Path Learning for Domain Adaptation of Semantic Segmentation[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021: 9082-9091.