CyCADA: Cycle-Consistent Adversarial Domain Adaptation

Judy Hoffman, Eric Tzeng, Taesung Park, Jun-Yan Zhu, Phillip Isola, Kate Saenko, Alexei Efros, Trevor Darrell
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:1989-1998, 2018.

Abstract

Domain adaptation is critical for success in new, unseen environments. Adversarial adaptation models have shown tremendous progress towards adapting to new environments by focusing either on discovering domain invariant representations or by mapping between unpaired image domains. While feature space methods are difficult to interpret and sometimes fail to capture pixel-level and low-level domain shifts, image space methods sometimes fail to incorporate high level semantic knowledge relevant for the end task. We propose a model which adapts between domains using both generative image space alignment and latent representation space alignment. Our approach, Cycle-Consistent Adversarial Domain Adaptation (CyCADA), guides transfer between domains according to a specific discriminatively trained task and avoids divergence by enforcing consistency of the relevant semantics before and after adaptation. We evaluate our method on a variety of visual recognition and prediction settings, including digit classification and semantic segmentation of road scenes, advancing state-of-the-art performance for unsupervised adaptation from synthetic to real world driving domains.

Cite this Paper

BibTeX
@InProceedings{pmlr-v80-hoffman18a, title = {{C}y{CADA}: Cycle-Consistent Adversarial Domain Adaptation}, author = {Hoffman, Judy and Tzeng, Eric and Park, Taesung and Zhu, Jun-Yan and Isola, Phillip and Saenko, Kate and Efros, Alexei and Darrell, Trevor}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {1989--1998}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/hoffman18a/hoffman18a.pdf}, url = {https://proceedings.mlr.press/v80/hoffman18a.html}, abstract = {Domain adaptation is critical for success in new, unseen environments. Adversarial adaptation models have shown tremendous progress towards adapting to new environments by focusing either on discovering domain invariant representations or by mapping between unpaired image domains. While feature space methods are difficult to interpret and sometimes fail to capture pixel-level and low-level domain shifts, image space methods sometimes fail to incorporate high level semantic knowledge relevant for the end task. We propose a model which adapts between domains using both generative image space alignment and latent representation space alignment. Our approach, Cycle-Consistent Adversarial Domain Adaptation (CyCADA), guides transfer between domains according to a specific discriminatively trained task and avoids divergence by enforcing consistency of the relevant semantics before and after adaptation. We evaluate our method on a variety of visual recognition and prediction settings, including digit classification and semantic segmentation of road scenes, advancing state-of-the-art performance for unsupervised adaptation from synthetic to real world driving domains.} }
Endnote
%0 Conference Paper %T CyCADA: Cycle-Consistent Adversarial Domain Adaptation %A Judy Hoffman %A Eric Tzeng %A Taesung Park %A Jun-Yan Zhu %A Phillip Isola %A Kate Saenko %A Alexei Efros %A Trevor Darrell %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-hoffman18a %I PMLR %P 1989--1998 %U https://proceedings.mlr.press/v80/hoffman18a.html %V 80 %X Domain adaptation is critical for success in new, unseen environments. Adversarial adaptation models have shown tremendous progress towards adapting to new environments by focusing either on discovering domain invariant representations or by mapping between unpaired image domains. While feature space methods are difficult to interpret and sometimes fail to capture pixel-level and low-level domain shifts, image space methods sometimes fail to incorporate high level semantic knowledge relevant for the end task. We propose a model which adapts between domains using both generative image space alignment and latent representation space alignment. Our approach, Cycle-Consistent Adversarial Domain Adaptation (CyCADA), guides transfer between domains according to a specific discriminatively trained task and avoids divergence by enforcing consistency of the relevant semantics before and after adaptation. We evaluate our method on a variety of visual recognition and prediction settings, including digit classification and semantic segmentation of road scenes, advancing state-of-the-art performance for unsupervised adaptation from synthetic to real world driving domains.
APA
Hoffman, J., Tzeng, E., Park, T., Zhu, J., Isola, P., Saenko, K., Efros, A. & Darrell, T.. (2018). CyCADA: Cycle-Consistent Adversarial Domain Adaptation. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:1989-1998 Available from https://proceedings.mlr.press/v80/hoffman18a.html.

Related Material