CyCADA :Cycle-Consistent Adversarial Domain Adaptation
2017, Dec 13
One Line Summary
- Using the Adversarial learning methods for domain adaptation using cycle reconstruction at the image or feature level.
Motivation
- Few works have explicitly studied visual domain adaptation for the semantic segmentation task.
Detailed Summary
- Network the cyclic loss for the source image reconsturction
- It also has a semantic consistency loss
- GAN is also present to improve the results.
Novelty and Contributions
- Discriminatively-trained Cycle-Consistent Adversarial Domain Adaptation model
- Applied in a variety of visual recognition and prediction settings
Network Details
- The Network has a cyclic reconstruction for the content of the input to be remained which doing the domain adaptation.
Results
Authors
Judy Hoffman, Eric Tzeng, Taesung Park, Jun-Yan Zhu, Phillip Isola, Kate Saenko, Alexei A. Efros, Trevor Darrell