The Pix2pix and CycleGAN losses have vastly improved the qualitative and quantitative visual quality of results in image-to-image translation tasks. We extend this framework by exploring approximately invertible architectures which are well suited to these losses. These architectures are approximately invertible by design and thus partially satisfy cycle-consistency before training even begins. Furthermore, since invertible architectures have constant memory complexity in depth, these models can be built arbitrarily deep. We are able to demonstrate superior quantitative output on the Cityscapes and Maps datasets.
Additionally, we show that the model allows us to perform several memory-intensive medical imaging tasks, including a super-resolution problem on 3D MRI brain volumes. We also demonstrate that our model can perform a 3D domain-adaptation and 3D super-resolution task on chest CT volumes. By doing this, we provide a proof-of-principle for using reversible networks to create a model capable of pre-processing 3D CT scans to high resolution with a standardized appearance.