The document describes a new framework called FUNIT for few-shot unsupervised image-to-image translation. FUNIT aims to map an image from a source class to an analogous target class using only a few examples of the target class provided during testing, without requiring examples of the target class during training. During training, FUNIT learns to translate between any two classes from a set of source classes, and during testing can translate an input to a never-before-seen target class by leveraging only a few examples of that new target class. FUNIT achieves this using an adversarial loss, content reconstruction loss, and feature matching loss.