We show that cross domain image-to-image translation can be subject to bias due to matching the data distribution of the target domain. We specifically show results when using conditional GAN and CycleGAN.
![](https://mlmed.org/w/wp-content/uploads/tumor-translation-error2.png)
We show that cross domain image-to-image translation can be subject to bias due to matching the data distribution of the target domain. We specifically show results when using conditional GAN and CycleGAN.