域适应方法顶刊论文列表 近两年(2020-2021)
作者:joey chang
链接:https://zhuanlan.zhihu.com/p/435500997
来源:知乎
著作权归作者所有。商业转载请联系作者获得授权,非商业转载请注明出处。
本文主要整理域适应(Domain Adaptation)方法近两年在各大顶刊上发表的论文进行整理,主要包含TPAMI、TNNLS以及TIP三大刊物。然后总结相关研究热点和趋势。由于个人精力有限,个别论文可能有疏漏,还请路过的学霸们多多指出~
一、TPAMI
2021年
Zero-shot Deep Domain Adaptation with Common Representation Learning (提出可迁移到任务不相关的目标域)
Learning across Tasks for Zero-Shot Domain Adaptation from a Single Source Domain (Zero-shot DA+域相关)
Source Data-absent Unsupervised Domain Adaptation through Hypothesis Transfer and Labeling Transfer (信息最大化+自监督 实现没有源域数据的情况下实现域特征对齐)
Towards Uncovering the Intrinsic Data Structures for Unsupervised Domain Adaptation using Structurally Regularized Deep Clustering (探索不破坏数据结构的情况实现UDA)
Generalized Domain Conditioned Adaptation Network (域差异大情况下性能下降归因于 domain-specialized low-level特征提取不充分 + 为每个域做实现相应的通道激活)
Category-Level Adversarial Adaptation for Semantic Segmentation using Purified Features (考虑与目标域无关的特征(噪声)对域特征对齐的影响)
Divergence-agnostic Unsupervised Domain Adaptation by Adversarial Attacks (考虑源域和目标域数据均不可用的情况下,通过对抗攻击提高模型泛化能力)
Self-Supervised Learning Across Domains (通过设计自监督任务提升模型泛化能力)
Contradistinguisher: A Vapnik's Imperative to Unsupervised Domain Adaptation (通过学习contrastive features实现DA,避免域对齐)
Triple Adversarial Learning and Multi-view Imaginative Reasoning for Unsupervised Domain Adaptation Person Re-identification (行人重识别中的DA问题+多视图推理+行人间的域差异)
Learning to Adapt Invariance in Memory for Person Re-Identification (主要解决ReID中的目标域中的三种不变形:Examplar-Invariance, Camera-Invariance 以及 Neighborhood-Invariance)
2020年
Maximum Density Divergence for Domain Adaptation (结合adversarial training 和 metric learning减少域差异)
Deep Residual Correction Network for Partial Domain Adaptation (部分域适应+plugged residual block)
Contrastive Adaptation Network for Single- and Multi-Source Domain Adaptation (类内域差异+类间域差异)
Heterogeneous Graph Attention Network for Unsupervised Multiple-Target Domain Adaptation (单源域多目标域问题+通过异构图注意力网络为所有域学习统一的子空间)
MultiDIAL: Domain Alignment Layers for (Multisource) Unsupervised Domain Adaptation (设计的域对齐layers不仅可以对齐域特征,还可学到每层特征的匹配程度)
Domain Stylization: A Fast Covariance Matching Framework Towards Domain Adaptation (模拟数据到真实数据的DA,边缘分布covariance对齐+条件分布covariance对齐)
Unsupervised Domain Adaptation for Depth Prediction from Images (解决深度预测中的DA问题+Stereo Matching)
Map-Guided Curriculum Domain Adaptation and Uncertainty-Aware Evaluation for Semantic Nighttime Image Segmentation (白天到夜间图像分割中的DA问题+课程学习)
二、TNNLS
2021年
Open Set Domain Adaptation: Theoretical Bound and Algorithm (提出open set difference度量目标分类器在未知类别上的risk)
Adversarial Entropy Optimization for Unsupervised Domain Adaptation (最小化最大化熵地对抗学习域不变特征)
Open Set Domain Adaptation With Soft Unknown-Class Rejection (传统OSDA采用阈值区别未知类,本文提出Soft Rejection的方法)
Cross-Domain Graph Convolutions for Adversarial Unsupervised Domain Adaptation (学习域不变特征可导致样本和类别结构发生扭曲,本文结合图的方法对齐样本和类别结构)
Cross-Domain Graph Convolutions for Adversarial Unsupervised Domain Adaptation
Learning Smooth Representation for Unsupervised Domain Adaptation
2020年
Incremental Unsupervised Domain-Adversarial Training of Neural Networks (增量UDA问题+伪标签)
Deep Subdomain Adaptation Network for Image Classification (提出 local maximum mean discrepancy 解决class-level的分布对齐)
三、TIP
2021年
Knowledge Exchange Between Domain-Adversarial and Private Networks Improves Open Set Image Classification (在学习域不变特征的基础上,提出Private Network 区分已知类别和未知类别)
Neutral Cross-Entropy Loss Based Unsupervised Domain Adaptation for Semantic Segmentation (基于熵最小化的UDA方法中存在over-sharpen预测分布的问题,提出neutral cross-entropy loss)
Progressive Modality Cooperation for Multi-Modality Domain Adaptation (多模态DA问题)
2020年
Attention Guided Multiple Source and Target Domain Adaptation (基于注意力的多源域多目标域域适配)
Generating Target Image-Label Pairs for Unsupervised Domain Adaptation (通过生成目标图像标签对实现Class-level分布对齐)
四、文献总结
通过回顾近两年的文献可以看出以下研究热点和趋势:
(1)由单域向多域进行探索(包括源域和目标域)
(2)由标签闭集向标签开集研究
(3)从domain-level的分布差异对齐到sample-level以及class-level的分布差异对齐
(4)考虑负迁移(噪声等因素)以及欠迁移(特征提取不充分)的影响
(5)源域或者目标域数据训练时不可用的场景
- 点赞
- 收藏
- 关注作者
评论(0)