Conditional Invertible Neural Networks for Diverse Image-to-Image Translation
This work addresses the need for diverse and stable image-to-image translation methods in computer vision, offering an incremental improvement over existing INN models by integrating preprocessing for better conditioning.
The authors tackled the problem of diverse image-to-image translation for natural images, which is challenging with existing invertible neural networks (INNs), by introducing a conditional invertible neural network (cINN) that combines INNs with a feed-forward network for preprocessing conditioning images, resulting in demonstrated applications like day-to-night translation and image colorization with properties like immunity to mode collapse.
We introduce a new architecture called a conditional invertible neural network (cINN), and use it to address the task of diverse image-to-image translation for natural images. This is not easily possible with existing INN models due to some fundamental limitations. The cINN combines the purely generative INN model with an unconstrained feed-forward network, which efficiently preprocesses the conditioning image into maximally informative features. All parameters of a cINN are jointly optimized with a stable, maximum likelihood-based training procedure. Even though INN-based models have received far less attention in the literature than GANs, they have been shown to have some remarkable properties absent in GANs, e.g. apparent immunity to mode collapse. We find that our cINNs leverage these properties for image-to-image translation, demonstrated on day to night translation and image colorization. Furthermore, we take advantage of our bidirectional cINN architecture to explore and manipulate emergent properties of the latent space, such as changing the image style in an intuitive way.