Skip to content
Tensorflow codes for ICML2018, Learning Semantic Representations for Unsupervised Domain Adaptation
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
digit-dataset Update Jul 14, 2018
introduction Add files via upload Jul 14, 2018
office Update Jul 14, 2018 Update Jul 14, 2018

Moving Semantic Transfer Network (ICML2018)

Based on adversarial adaptation, we propose a Pseudo Centroid Alignment Objective to enforce Semantic Transfer. If you are limited to use a relative-small batch size (64 for 31-classification or 100 for 1000-classification), you might be interested in our Moving Centroid Alignment.

Welcome to check our paper.


If you find this useful for your research, we would be appreciated if you cite the following papers:

  title={Learning Semantic Representations for Unsupervised Domain Adaptation},
  author={Xie, Shaoan and Zheng, Zibin and Chen, Liang and Chen, Chuan},
  booktitle={International Conference on Machine Learning},

Tips for your reproduction of DANN and our work MSTN.

My work is based on DANN. During my reimplementation of DANN, I noticed following problems worth attention for reproduing DANN and our work MSTN. Hope these could help you. :)

  • Data Preprocessing
    1. Scale image to 256x256.
    2. When training, source and target images would be "Random Cropping". When testing, target images would be "Center Cropping" (Caffe only uses crop command but the inner implemenattion actually random crop when training while center crop when testing).
  • Implementing Moving Centroid
    1. We can not just update the centroid by "moving_centroid=theta*moving_centroid+(1-theta)*current_centroid". This is just a update_op and need to be executed by specifying. (Look our code for more detail.)
    2. Using tensorboard to check whether centroids have been updated.


    If you have any problem about this library, please create an Issue or send us an Email at:

    Acknowledgement for the code

    For digits dataset, the code is modified from Here. For real-world dataset, the code is based on Here.

    You can’t perform that action at this time.