Spatial-aware Network using Wasserstein Distance for Unsupervised Domain Adaptation

In a general scenario, the purpose of Unsupervised Domain Adaptation (UDA) is to classify unlabeled target domain data as much as possible, but the source domain data has a large number of labels. To address this situation, this paper introduces the optimal transport theory into the transfer learning, and proposes a deep adaptation network based on the second-order Wasserstein distance, which can measure the discrepancy between the two distributions. In addition, in order to retain the spatial structure information of features, the network is combined with convolutional auto-encoder. Experiments show that our method has achieved good results.