Seminar: Unpaired Image-to-image translation (I2I) - CUT++

Image translation between domains while keeping the contents of the source has numerous benefits. Besides being a prominent task within itself (e.g., for artistic purposes), it has many additional use cases in different domains as well as other tasks (e.g., data augmentation and domain transfer). In this presentation, I go over a few image-to-image translation methods following up on ‘Contrastive learning for unpaired image-to-image translation’ (CUT), which propose a non-cyclic means to map from source to target domain images, thus allowing for a more efficient training setup.