Image-to-Image Translation with Conditional Adversarial Networks
2017, Dec 17
One Line Summary
- Image to Image translation where the loss formulation is difficult
Motivation
- Image to Image translation tasks the loss formulation is not straight forward, but as a community, we no longer hand-engineer our mapping functions, and this work suggests we can achieve reasonable results without hand-engineering our loss functions either.
Detailed Summary
- Conditional adversarial networks are a promising approach for many image to-image translation tasks.
- These networks learn a loss adapted to the task and data at hand, which makes them applicable in a wide variety of settings.
Novelty and Contributions
- Using the deep conditional adversarial network the task of image to image translation.
- No explicit loss formulation.
Network Details
- Uses the encoder decoder architecture, for translating the image from the source domain to the target domain.
- A variant of the original architecture uses the skip connections on the encoder-decoder architecture(U -Net)
Results
Authors
Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, Alexei A. Efros