Reversible GANs for Memory-efficient Image-to-Image Translation
This work addresses memory efficiency for researchers and practitioners in computer vision, offering an incremental improvement over existing GAN frameworks.
The paper tackled the problem of high memory usage in image-to-image translation by proposing reversible GAN architectures that are approximately invertible by design, achieving superior quantitative results on Cityscapes and Maps datasets with near constant memory complexity.
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 at near constant memory budget.