Two-dimensional subspace alignment for convolutional activations adaptation
作者：
Highlights：
• Two-dimensional subspace alignment (2DSA) is proposed for domain adaptation.
• The classification performance has low correlation to domain discrepancy measure.
• Local within- and between-class divergences are introduced to compare domains.
• A novel domain adaptation application in agriculture is illustrated.
• A MTFS3-DA dataset with 10 domains is developed for cross-field evaluation.
摘要
•Two-dimensional subspace alignment (2DSA) is proposed for domain adaptation.•The classification performance has low correlation to domain discrepancy measure.•Local within- and between-class divergences are introduced to compare domains.•A novel domain adaptation application in agriculture is illustrated.•A MTFS3-DA dataset with 10 domains is developed for cross-field evaluation.
论文关键词：Visual domain adaptation,Subspace alignment,Convolutional activations,Two-dimensional PCA,Domain divergence measure
论文评审过程：Received 13 July 2016, Revised 22 May 2017, Accepted 7 June 2017, Available online 13 June 2017, Version of Record 21 June 2017.
论文官网地址：https://doi.org/10.1016/j.patcog.2017.06.010
